University of Southern California Dissertations and Theses

Investigating the role of climate in affecting residential electricity consumption through high spatiotemporal resolution observations

(USC Thesis Other)

Investigating the role of climate in affecting residential electricity consumption through high spatiotemporal resolution observations

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Investigating the Role of Climate in Aecting Residential Electricity Consumption through High Spatiotemporal Resolution Observations by Mo Chen A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulllment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (Engineering (Environmental Engineering)) August 2020 Copyright 2020 Mo Chen Acknowledgements The author acknowledges the partial nancial support of the Teh Fu (Dave) Yen Fellowship in Environmental Engineering, the Theodore and Wen-Hui Chen Fellowship from USC, as well as the Teaching Assistantship and Research Assistantship supported by the CEE department and my advisors. I would like to thank my two advisors Dr. Kelly Sanders and Dr. George Ban-Weiss, for your guidance and help and support and inspiration throughout each stage of my PhD study. You have set up a role model for me, not only in academia, but also how to be a good person, mentally and morally. It has been such an honor to work with both of you. I would like to thank all my screening exam, qualifying exam, and defense committee members, Dr. Bistra Dilkina, Dr. Lucio Soibelman, and Dr. John Wilson, for their insightful comments, advice, and guidance throughout my degree. I would like to thank all my friends who have helped me along the way. A special thanks to all my buddies in S3 and Ban-Weiss groups, for your countless support and help with revising the slides and manuscripts, practicing presentations, doing journal clubs, hanging out (academically) at conferences, etc. Without you guys, the journey would not have been so much enjoyable. I would like to thank my parents, who have loved and supported me endlessly since I was born, without asking anything for return. I hope what I have done will make you proud. Finally, I would like to thank my dear wife, Mengyi (Maggie) Yuan. You gave me the courage to start this long march. You gave me the accompany to pass the most dicult time on my way. You gave me the support to nish the journey. A PhD is not something I can accomplish alone. Thank you for never giving up on me, even when I was about to. ii Table of Contents Acknowledgements ii List Of Tables v List Of Figures vi Abstract viii Chapter 1: Introduction 1 1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Present state of knowledge and identication of existing knowledge gaps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 1.3 Denition of research goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 1.4 Structure of document and resulting publications to date . . . . . . . . . . . . . . . . . . . . 9 Chapter 2: The role of household level electricity data in improving estimates of the impacts of climate on building electricity use 11 2.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 2.2 Literature review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 2.2.1 Models and observations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 2.2.2 Electricity data metrics and resolution . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 2.2.3 Temperature data metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 2.2.4 Stationary point temperatures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 2.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 2.3.1 Datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 2.3.2 Statistical models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 2.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 2.4.1 Sensitivity and stationary point temperature distributions . . . . . . . . . . . . . . . . 30 2.4.2 Impact of temperature indicators on computed electricity-temperature sensitivities . . 32 2.4.3 Impact of spatial aggregation on computed electricity-temperature sensitivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 2.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 2.5.1 Advantages of utilizing high spatiotemporal resolution data . . . . . . . . . . . . . . . 36 2.5.2 Roles of temperature indicators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 2.5.3 Comparing computed electricity-temperature sensitivities to previous studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 Chapter 3: A new method utilizing smart meter data for identifying the existence of air conditioning in residential homes 45 3.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 3.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 3.2.1 Datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 iii 3.2.2 Statistical model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 3.3 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 3.3.1 AC penetration rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 3.3.2 Comparison of computed AC penetration rates to other studies . . . . . . . . . . . . . 55 3.3.3 Applicability to other regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 Chapter 4: Utilizing smart-meter data to project impacts of urban warming on residential electricity use for vulnerable populations in Southern California 59 4.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 4.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 4.3 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 4.3.1 Spatial trends in the electricity use response to increases in ambient temperature . . . 66 4.3.2 Impact of auence on the electricity use response to increases in ambient temperature 69 4.3.3 Response of electricity use to increases in ambient temperature in Southern California: combined eects of climate and auence levels . . . . . . . . . . 69 4.3.4 Locating the most temperature-sensitive hotspots and most heat-vulnerable communities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 4.3.5 Identifying communities that may be most vulnerable to ambient temperature increases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 4.3.6 Further considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 4.4 Conclusion and future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 Chapter 5: Conclusion 78 Reference List 81 Appendix A Supplementary material for Chapter 2: the role of household level electricity data in improving estimates of the impacts of climate on building electricity use . . . . . . . . . . . . . . . . . . 91 A.1 Correlation between electricity-temperature sensitivity and stationary point temperature . . . 91 Appendix B Supplementary information for Chapter 3: a new method utilizing smart meter data for identifying the existence of air conditioning in residential homes . . . . . . . . . . . . . . . . . . . . . . . 94 B.1 Method used to calculate statistically representative sample sizes . . . . . . . . . . . . . . . . 94 B.2 Discussion of potential method shortcomings . . . . . . . . . . . . . . . . . . . . . . . . . . . 94 B.3 Additional study statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 Appendix C Supplementary information for Chapter 4: utilizing smart-meter data to project impacts of urban warming on residential electricity use for vulnerable populations in Southern California . . . 99 C.1 Probability density distributions of electricity-temperature sensitivities and stationary point temperatures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 C.2 Single variable and multivariable linear regression . . . . . . . . . . . . . . . . . . . . . . . . . 101 C.3 Discussion about Climate Zone 16 (Big Bear Lake area) . . . . . . . . . . . . . . . . . . . . . 102 C.4 Spatial span and distribution of data included in this study . . . . . . . . . . . . . . . . . . . 103 C.5 Map showing average electricity-temperature sensitivity values per census tract in Southern California (including only households identied as having AC) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104 C.6 Determination of expected prevalence p in sample size calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 C.7 List of identied vulnerable census tracts in Chapter 4 . . . . . . . . . . . . . . . . . . . . . . 106 iv List Of Tables 2.1 Literature review of studies that investigate the in uence of climate on electricity consumption 15 2.2 Electricity-temperature sensitivities and stationary point temperatures computed with various spatial aggregations using electricity consumption data for 1245 California households. . . . . 35 3.1 Comparison of AC penetration rates acquired from this study and publicly available survey data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 B.1 Statistics of annual-averaged electricity consumption for data records in this study by climate zones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 C.1 Linear regression between E-T sensitivity and AC penetration rates versus climatic and socioe- conomic variables at the census track level. Values in columns under \Explanatory variables" refer to regression coecients of explanatory variables. Rows with only one value under \Ex- planatory variables" columns refer to single variable linear regression. Rows with multiple values under \Explanatory variables" columns refer to multivariable linear regressions includ- ing the corresponding variables in the columns with values. Values under column \r 2 " refer to coecients of determination of linear regressions in corresponding rows. . . . . . . . . . . . 101 C.2 Determined sample size n to be statistically representative as a function of parameters . . . . 105 C.3 List of identied vulnerable census tracts in Chapter 4 . . . . . . . . . . . . . . . . . . . . . . 106 v List Of Figures 1.1 Illustration of positive feedback loops in energy-climate nexus . . . . . . . . . . . . . . . . . . 3 2.1 Map showing locations of 1245 residnetial electricity customers and 145 CIMIS weather stations 25 2.2 Example of a home in Clovis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 2.3 Probability density distribution of electricity-temperature sensitivities and stationary point temperatures of 1245 homes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 2.4 Comparison of eects of dierent temperature indicators on a single household and the average of households in the City of San Jose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 2.5 Two case studies illustrating how spatial aggregation choices have an eect on computed electricity-temperature sensitivities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 2.6 Histogram of time of day corresponding to peak energy for each household during summertime 39 3.1 Three example homes in Southern California illustrating the AC identication method . . . . 50 3.2 Choropleth map of AC penetration rates for all census tracts powered by Southern California Edison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 3.3 Probability density distributions of electricity-temperature sensitivities for all 180,476 homes and separately for homes with/without air conditioners . . . . . . . . . . . . . . . . . . . . . 54 4.1 Choropleth map of electricity-temperature sensitivities, stationary point temperature, AC penetration rates, and poverty indices for all census tracts powered by Southern California Edison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 4.2 Box-and-whisker plot of household level electricity-temperature sensitivities for only house- holds identied as having AC, grouped by climate zone . . . . . . . . . . . . . . . . . . . . . . 70 4.3 Bar chart of air conditioner penetration rate versus poverty percentile, grouped by climate zone, in Southern California . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 4.4 Map of (a) historically observed and (b) end-of-century projected number of extreme heat days at the census track level in Southern California, with potentially vulnerable communities highlighted . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 A.1 Correlation between electricity-temperature sensitivity and stationary point temperature across dierent spatial aggregation levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 vi A.2 Electric power throughout 09/10/2015 averaged for 1,245 households and standard deviation 93 B.1 Schema of relationships between energy use and temperature dened in ASHRAE's inverse model toolkit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 B.2 shows the relationships of slope l eftandslope r ightforalldatarecordsanalyzedinChapter3(n = 180; 476) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 B.3 Probability density distributions of stationary point temperatures (SPT ) of all data records analyzed in Chapter 3 (n=180,476) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 C.1 Probability density distributions of electricity-temperature sensitivities and stationary point temperatures of 180,476 homes investigated in Chapter 4 . . . . . . . . . . . . . . . . . . . . 99 C.2 Map showing locations of the 180,476 residential electricity customers and weather stations considered in Chapter 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 C.3 Map showing average electricity-temperature sensitivity values per census tract Southern Cal- ifornia (including only households identied as having AC). . . . . . . . . . . . . . . . . . . . 104 vii Abstract The electricity grid and climate are two systems connected and in uencing each other. A warmer climate will exacerbate the energy demands of global cooling in the future, while meeting such cooling demand by fossil fuels can exacerbate global climate change. These positive feedback loops pose challenges for warming, as well as energy systems. Of all electricity-consuming sectors, the residential sector is of particular interest in this body of work, for its leading contribution to the entire US electricity consumption and larger variations under the in uence of climate. The higher uncertainties in prediction has added an additional layer of diculty to research energy-climate nexus in the residential sector. Although this relationship is understood, prior to this work, methods to quantify the functional relationship between residential electricity consumption and climate variables using highly spatiotemporal resolved data, as well as how the resolution itself impacts such quantication, have not been well studied. This body of work rst quanties the relationship between resi- dential electricity consumption and ambient temperature, determines the most appropriate spatial/temporal resolutions and forms of data for use in quantifying such relationships; then develops method to identify the existence of air conditioning (AC) at household level, so that AC penetration rates (an important driver in energy-climate relationships) can be calculated with high spatial resolution; and lastly explore how climate aects residential electricity consumption in the study area, Southern California, analyzes the spatial and socioeconomic patterns of such climate impacts and then uses acquired insights to identify future energy hotspots and potentially vulnerable communities in the context of more and more frequent extreme heat events. Such knowledge is vital to preparing society to adapt to the impacts of climate change and urban heat island, by informing more targeted energy conservation and other climate/heat mitigation strategies, as well as helping to address energy and environmental justice issues faced by vulnerable groups of people. viii Chapter 1 Introduction 1.1 Motivation The world we live in is warming up. Data from the National Aeronautics and Space Administration (NASA) in 2020 shows that global mean surface temperature has risen more than 1°C compared to pre-industrial levels [55] [96]. At the same time, the world we live in is urbanizing. According to 2018 data from the United Nations, 55% of the total population in the world lives in cities, and this fraction is expected to rise to 68% by mid-century [146]. The percentage of the population living in urban areas in developing regions like Asia and Africa is expected to double from around 30% to 60% by 2050 [146]. Even in regions where urbanization is already high (over 70%), e.g., North America and Europe, the fraction of the population living in cities is expected to exceed 85% on average by mid century [146]. As more population moves into cities, a larger group of inhabitants will be under the in uence of urban micro-climates. In addition to global climate change, the urban heat island (UHI) eect describes a warming trend that aects urban areas. UHI describes the phenomenon where human-made materials in cities, such as buildings and pavements, replace open land and vegetation, and consequantly, absorb heat more quickly during the day and release heat much more slowly after sunset. This eect results in a higher mean temperature in cities than their less dense, rural counterparts. Even worse, research has shown that UHI eect can be exacerbated by global climate change [128]. 1 The combined trends of global climate change, urbanization, and urban heat island eects, coupled with population growth and economic development, has resulted in a large increase global cooling capacity. A report from Organisation for Economic Co-operation and Development (OECD) and International Energy Agency (IEA) illustrated that the world's residential cooling capacity in 2018 was already three times that of 1990, and this trend is expected to accelerate in the coming decades in both developing and developed economics [76]. Based on the trends discussed above, we anticipate two positive feedback loops, as illustrated in Figure 1.1, that will exacerbate the energy demands of global cooling in the future. 1) in the face of a warmer climate, there will be a large expansion of AC unit installations, as well as an increase in the intensity of which they are used, increasing electricity demands. Currently, the majority of electricity, both in the United States (US) and globally, is generated by burning fossil fuels [156] [75], which emits greenhouse gases into the atmosphere. Increased greenhouse gas emissions will exacerbate global climate change, closing the rst positive feed back loop. On the other hand, using more ACs releases heat from indoor environment to the outdoor environment, which becomes a source of anthropogenic heat. Previous research has shown that anthropogenic heat due to AC usage can enhance the urban heat island eect and contribute to higher temperatures in cities [162], which forms another positive feedback loop. These positive feedback loops pose challenges for warming, as well as energy systems. First, increases in residential electricity consumption due to more AC usage can put pressure on current infrastructure (e.g., transmission lines) and peak energy management, especially during summertime [3] [9] [102] [94]. Another challenge from an environmental justice point of view is that not everyone will be impacted by climate change equally. Some populations may be more vulnerable to more and more frequent extreme heat events [134] [130] [120] [89] [99] [104] [61] [69] [62] [84] [103] [70], that lack the nancial resources to run their exisiting air conditioners or simply lack of access to cooling equipment all together. Both reasons can bring potential public health concerns to these vulnerable groups [60]. Of all electricity-consuming sectors, the residential sector is of particular interest in this body of work, for several reasons. First, the residential sector is the largest electricity consumer, accounting for over 38% of total electricity consumption in the US in 2019 [155]. Second, space cooling and heating collectively represent 2 Figure 1.1 Urban warming can result in more electricity consumption for cooling, which can further exacerbate warming, creating positive feedback loops the largest electricity-consuming end uses in US homes, representing more than 20% of total household electricity consumption. Hence, climate plays an important role in determining residential electricity demand. Furthermore, residential homes tend to have larger variations in electricity consumption patterns compared to other sectors, introducing higher uncertainties in prediction [93], possibly due to factors like highly variable building stocks characteristics, appliances usage and selection, occupants behaviors, energy sources and prices, and socioeconomic indicators [11] [135] [39] [14] [82]. These variabilities add an additional layer of diculty to research energy-climate nexus in the residential sector. To tackle these challenges, this dissertation focuses on developing a quantitative understanding of how residential electricity use is aected by climate. Such knowledge is vital to preparing society to adapt to the impacts of climate change and urban heat island , by informing more targeted energy conservation and other climate/heat mitigation strategies, as well as helping to address energy and environmental justice issues faced by vulnerable groups of people. 3 1.2 Present state of knowledge and identication of existing knowledge gaps a Climate plays one of the most important roles in shaping residential electricity consump- tion, but functional relationships between the two have not yet been established at high spatiotemporal resolution. Past studies conclude that climate is one of the most important drivers of residential electricity consump- tion. Generally, literature that has focused on or related to quantifying electricity-climate relationships can be classied into two groups according to their methodologies. Building energy simulation. Studies that fall into this category utilize building energy models to simulate and predict the energy demand of buildings. Most literature in this category aims to identify climate change's impact on building energy consumption. Due to dierences in study scales, this category can be further divided into two sub-classes. The rst is prototype building modeling. When modeling or predicting total energy demand, researchers often create multiple building prototypes that represent dierent types of buildings, e.g., single family house, small oce building, etc. Many of these studies compare energy demand under various climate scenarios across dierent building prototypes, which oers an opportunity to observe electricity-climate relationships based on building characteristics. By using this framework, [159] [50] [7] [166] [25] [111] [160] [4] [80] [72] [33] [50] [114] found that dierent types of buildings have heterogeneous responses to climate change; typically larger-size buildings of the same type had bigger energy demand response to climate change. However, due to dierences in study regions, models utilized, and assumptions in building prototypes, it is hard to draw a uniform set of conclusions from these studies. J. Huang and K. Gurney found that commercial buildings built after 2004 require less energy per square footage in the context of climate change, compared to ones built post-1980 and pre-1980 [72]. On the contrary, Y. J. Huang's modeling results showed newer homes will be more adversely aected by climate change than older homes in the US residential housing stock [73]. Chan et al., Lu et al., and Dirks et al.'s modeling results show residential buildings will experience a bigger increase in cooling energy demand than oce buildings 4 caused by climate change [33] [111] [41]. Huang et al. suggest the opposite results in a U.S.-based study [72]. This method has three main shortcomings: 1) Without real-world data, it is dicult to validate modeling results or quantify the modeling uncertainty. 2) A majority of studies using prototype building energy modeling can only acquire qualied eects of building characteristics on electricity-climate relationships. Simulation may show big oce buildings' energy demand are more sensitive to climate variation than small oce buildings. Without providing detailed information on each building prototype, it is impossible to quantify the role of square footage in this topic. 3) Due to dierent energy models utilized, dierent building prototypes created, and dierent assumptions made, it is hard to compare across various studies despite their similarities in scope. The second sub-class of studies use single building modeling. Some studies, e.g., [161] [32], simulate building energy consumption in certain climate scenarios for a single building located in one or more cities. The advantage of simulating only one building is to allow more detailed parameters and make the modeling results closer to reality. These studies also mainly aim at providing information to policy development related to building codes and energy eciency operations [161] [32]. Empirical studies. Compared to building energy simulation studies, fewer analyses utilize historical data to analyze electricity-climate relationships. Auhammer and Mansur completed a review of the liter- ature measuring climatic impacts on energy consumption in 2014, and found that there were few empirical studies assessing the eect of climate on buildings' energy consumption using high-spatiotemporal resolution data [10]. They also conclude that high temporal resolution and micro scale household and rm data are needed to address key knowledge gaps in the eld. Temporally, the vast majority of past empirical studies rely on aggregated electricity records ranging from daily to yearly [81] [125] [130] [131] [132] [122] [133] [64]. Highly temporally-resolved electricity data (e.g., hourly/subhourly) can be used to reveal detailed information such as daily patterns of peak demand and interpret behavioral factors through disaggregation techniques [52] [88] [168] [31]. However, no research to the author's knowledge has investigated how the temporal resolution of data used will impact the quantication of electricity-climate relationships. Most previous studies used relatively coarse data, in regards to its spatial 5 resolution, to represent climatic characteristics and aggregated electricity consumption at a relatively broad spatial extents. Such extents can span from city to country scale [131] [167] [20]. Using data at these low spatial resolutions may introduce inaccuracies and the loss of information since aggregating data averages out building-level variability including housing stocks, socioeconomic, behavior, etc. It is straightforward and intuitive to assume that using highly-resolved spatiotemporal electricity and climate data can give valuable information and insights on electricity-climate relationships through observing individual-level and intra- day variations as well as the major drivers behind the variations; however, increasing data resolution can also increase computational burdens to data processing and storage. Hence, analysis is needed to assess the trade-os in choosing data resolutions. To the author's knowledge, prior to this body of work, few studies analyzed how choices in data resolution and various indicators of data impacted building functional relationships between electricity consumption and climate. b Air conditioning is a major driver of electricity-climate relationships, but knowledge of how to quantify AC penetration rates at a high spatiotemporal resolution level remains lacking Since AC is a major player in driving the relationship between climate and residential electricity con- sumption, it is critical to identify the locations and scales of AC usage and penetration (dened as the percentage of homes having access to ACs in an area) in order to better understand any spatial and tempo- ral variations in electricity-climate relationships. Currently, knowledge of AC penetration rates mostly relies on survey-based studies such as appliance saturation surveys and residential energy consumption surveys conducted by federal and state governments. The spatial resolution of the previously available datasets span from climate zones (for example, in California, climate zones are about/bigger than the size of counties [100]) to groups of states in the US [153]. The resolution of these surveys is limited by sample sizes that are not big enough to be statistically representative at smaller spatial extents. It is dicult to incease sample sizes in surveys due to budget constraints [164] and time-consuming survey processes [116]. New methods need to be developed to estimate AC penetration rates at a higher resolution level to facilitate energy management and planning. 6 The emergence of smart meter records has increased the availability of high-resolution electricity data. Some studies have used smart meter data to disaggregate energy consumption from the building-level into unique end-use activities [52] [88] [168] [31], which can be applied to estimate the existence and usage of space cooling equipment. However, these methods typically require data at the minute-level or higher resolution, which is not currently supported by smart meter networks [24]. New methods need to be established to better utilize currently available smart meter datasets for quantifying AC penetration rates at high spatiotemporal resolutions. c Residential electricity-climate relationships likely depends on socioeconomic factors, but this has not been explicitly investigated Energy equity (i.e., referring to accessibility and aordability of energy resources in a region [59]) has been the focus of many energy experts, social scientists, and policy makers, because energy consumption is closely related to quality of life and human rights, as it enables access to clean water, clean forms of cooking, indoor heating, cooling, and ventilation (HVAC), other healthcare equipment, etc. Previous studies from multiple regions around the world have explored the impact of income or socioeconomic levels on residential electricity consumption, and a large fraction conclude that higher income households typically have residential electricity consumption [79]. Even in developed areas like the US and Europe, signicant dierences in residential electricity consumption have been observed across groups of varying household income [135] [165]. There are also studies that disagree with the conclusion above. For example, [81] does not observe any signicant correlation between residential electricity consumption and income level, but also admitted that this may be caused by the data sample, which did not include a diverse population in terms of socioeconomic status. Rising temperatures are expected to exacerbate existing challenges of energy equity and environmental justice. Past research has evaluated how vulnerable populations might be aected by heat exposure under current and/or future conditions [134] [140] [120] [89] [99] [104] [61] [69] [62] [84] [103] [70]. While space cooling can oset heat-related vulnerabilities, under the pressure of existing or emerging warming, less auent people might not be able to aord increasing electricity costs for air conditioning (AC), and therefore might be susceptible to heat-related health threats due to lack of AC. Many regions of the US are already 7 seeing the health ramications of such warming, which can include elevated risks of hospitalizations and increased heat-related mortality [86] [15]. To the author's knowledge, very few studies have directly investigated the impact of socioeconomic factors on the relationship between residential electricity consumption and ambient temperature. Although demographic data are readily available in the US at ne geospatial resolution, there are still large knowledge gaps related to 1) how patterns of electricity usage are in uenced by temperature at the household-scale; and 2) how these patterns vary across climate zones and demographic groups. Bridging these knowledge gaps have been hampered by the lack of publicly available, high-resolution (e.g., hourly) residential electricity data [34]. Previous studies in this area mainly utilize customer billing data, which are at the resolution of monthly or yearly, combined with socioeconomic data acquired by surveys [79]. There are two drawbacks of this method. Firstly, as to be illustrated by Chapter 2, coarse temporal resolution can cover up valuable information embedded in electricity-climate relationships. Secondly, survey data that acquire socioeconomic information are typically limited by small sample sizes, which might not statistically represent a metropolitan region. To overcome these problems, utilizing data at high spatiotemporal resolution and development of novel analysis methods are both needed. 1.3 Denition of research goals The aim of this research is to develop functional relationships between climatic parameters and residential electricity consumption based on highly resolved (spatially and temporally) data. Ultimately, we wish to address the overarching question: how does urban warming impact residential electricity consumption in time and space? To do so, we will address four major research questions through this eort: 1. How does the spatiotemporal resolution of selected datasets aect the calculated relationship between residential electricity consumption and climatic parameters, i.e., ambient temperature? How does the choice of temperature indicators aect the calculated relationship between residential electricity and ambient temperature? My work addresses this question in Chapter 2 by analyzing 1,245 households' electricity smart-meter data across California. 8 2. What role do penetration rates of air conditioners play in the functional relationship between electricity consumption and ambient temperature? Can we develop a method to quantify air conditioner penetra- tion rates using smart meter data? My research focuses on this question in Chapter 3 by developing a method to identify the use of air conditioners in residential households through the use of smart-meter and weather data. 3. How does residential electricity consumption across Southern California respond to changes in ambient temperature? To what extent is residential electricity behavior in uenced by spatial variations in baseline climate and socioeconomic factors? Can we better understand vulnerability to future increases in temperature by considering geospatial distributions in poverty level, AC penetration rates, and extreme heat events? My work focuses on these questions in Chapter 4 by analyzing 180,476 residential households' smart-meter data. This research targets Southern California area because it oers an opportunity to (a) observe how local variations in climate aect electricity use and (b) quantify how the UHI and climate change impact electricity use, across relatively small spatial extents with widely varying microclimates. Hence, it is an ideal test bed for developing a research methodology that can be applied to other cities around the world in future work. Moreover, Los Angeles has housing stocks varying from tiny apartments to huge villas, with a similarly wide variation in socioeconomic status. Currently only 65% of homes in the western US have AC, compared to 95% in the south [149]. Thus, future growth of AC energy use in the US will be in regions like Southern California that are growing in population, and likely to increase in AC market saturation, resulting in a non-linear sensitivity of electricity use to changes in local climate. Cities in Southern California also have an abundance of smart-meter data, enabling analysis to be completed with unprecedented resolution, thus providing a unique opportunity to address these research questions. 1.4 Structure of document and resulting publications to date This document is organized into ve chapters. Chapter 2 through 4 each corresponds to one research questions raised above. The work described in Chapter 2, 3, and 4 are published in the following peer-reviewed journals: 9 • Chapter 2: M. Chen, G.A. Ban-Weiss, and K.T. Sanders (2018). \The role of household level electricity data in improving estimates of the impacts of climate on building electricity use". Energy and Buildings, 180 (2018): 146-158. (2018 Impact Factor 4.495) • Chapter 3: M. Chen, K.T. Sanders, and G.A. Ban-Weiss (2019). \A new method utilizing smart meter data for identifying the existence of air conditioning in residential homes". Environmental Research Letters, 14: 094004. (2018 Impact Factor 6.192) • Chapter 4: M. Chen, G.A. Ban-Weiss, and K.T. Sanders (2020). \Utilizing smart-meter data to project impacts of urban warming on residential electricity use for vulnerable populations in Southern California". Environmental Research Letters, 15: 064001. (2018 Impact Factor 6.192) Finally, Chapter 5 concludes the main ndings and signicance of this dissertation. 10 Chapter 2 The role of household level electricity data in improving estimates of the impacts of climate on building electricity use This chapter re ects work published in Energy and Buildings in 2018. [34] 2.1 Motivation Household electricity demand increased by 16.5% [151] between 2001 and 2015 in the US and is projected to increase by 8% and 11% between 2015 and 2040 with and without the Clean Power Plan, respectively [150] [3]. Much of this increase is expected to come from increases in space cooling demand. Although a large number of factors impact residential electricity consumption, climate has been shown to play one of the most important roles in driving variations in residential electricity consumption [124] [81] [92] [16] [125]. Because of the diverse nature of the residential sector, analyzing the sensitivity of electricity demand to ambient temperature across the residential sector presents unique challenges compared to other sectors [131] [66]. Households tend to have larger spatiotemporal variations in electricity consumption compared to other sectors, driving more uncertainty in prediction [93], presumably due to factors such as highly variable housing stock characteristics, appliances and other energy consuming device selections, occupant behavioral patterns, heating and cooling sources, energy prices, demographic factors, and other socio-economic indicators, which can vary signicantly across regions [11] [135] [39] [14] [82]. Thus, to maximize our understanding of factors aecting residential sector electricity use, energy-climate sensitivity should be derived using data at the household level so that variability across residences can be observed and analyzed. 11 Coarse data resolution has been a limiting factor in the majority of prior research endeavors in this eld. Although there are a few studies that have used high resolution data, past studies typically rely on daily, monthly or annual electricity consumption data that might be insucient for resolving the relationship between climate and electricity consumption [81]. Additionally, most studies utilize relatively coarse climatic data to represent relatively broad spatial extents, often ranging from megacity- to country-wide in scale [131] [167] [20]. These spatial scales are not sucient for building a highly resolved understanding of climate- driven variations in energy consumption behavior. Using datasets that have coarse spatial and/or temporal resolution can average out important insights and cause loss of valuable information, especially for studies addressing the residential sector. Confounding analysis of the residential sector is the fact that residential homes have greater daily and seasonal variations in electricity use than other sectors, distinguishing the residential sector as the most dicult to analyze due to high amounts of variability and uncertainty [131] [66] [93] [82] [71] [47]. Relying on temporally aggregated data, in particular, diminishes the ability to gain insight on electricity consumption patterns, which can lead to uncertainties in quantifying electricity- temperature relationships. Although this issue has been partially addressed by several recent studies using either high temporal resolution data (e.g. hourly or sub-hourly) [16] [68] [42] [123] [119] [49] or high spatial resolution data (e.g. at building level) [68] [18] [6], knowledge gaps still exist. Our main insights based on a survey of existing literature in this eld (detailed below in Section 2.2) are that previous studies: 1) use datasets that vary widely in spatiotemporal resolution, spanning hourly to yearly resolution, across various spatial regions of interest; 2) rarely utilize electricity datasets that are both highly temporally and spatially resolved, and 3) utilize dierent types of electricity and temperature indicators to determine electricity-temperature sensitivities (e.g. hourly temperature, daily average temperature, daily minimum or maximum temperature, cooling degree days (CDD) or heating degree days (HDD), monthly average temperature, and some other derived indicators). Despite these large methodological dierences, no research to the authors' knowledge has investigated how electricity-temperature sensitivities vary according to the spatiotemporal resolution of electricity and climate data or choice in temperature indicators. While it is straightforward to assume that increasing data resolution is valuable to establishing rened and robust functional relationships between residential 12 electricity usage and climate parameters, these increases in data resolution can cause large increases in the computational resource requirements of analysis, so gaining insight into these tradeos oer merit. Thus, research questions addressed in this study are as follows: 1) How does the spatiotemporal resolution of selected datasets aect the calculated relationship between residential electricity consumption and climatic parameters, i.e., ambient temperature? 2) How does the choice of temperature indicators aect the calculated relationship between residential electricity and ambient temperature? Understanding and quantifying the functional relationships between residential electricity consumption and climatic parameters is crucial to developing eective energy conservation, peak energy management, and climate adaptation strategies, as well as informing meaningful and cost-eective power capacity investments in the future. Establishing robust electricity-temperature sensitivities is particularly important for future studies attempting to understand the role that phenomena such as climate change and the urban heat island eect might have on the power sector. 2.2 Literature review Previous studies conclude that climate plays one of the most important roles in driving variability in res- idential electricity consumption [124] [81] [92] [16] [125]. In an eort to improve estimates of electricity- temperature relationships (hereafter referred to as \electricity-temperature sensitivity"), we conducted a survey of existing literature on this topic. Table 2.1 summarizes 24 publications in the literature analyzing climate-related in uences on electricity consumption. These studies come from somewhat disparate elds including grid-scale electricity demand forecasting [16] [107] [21] [138] [132] [118], building-level energy use modeling [68] [18], and assessing the impact of climate change on electricity consumption [71] [42] [119] [49] [18] [6] [138] [9] [64] [127] [8] [95]. The studies investigate regions in more than 40 countries and dier signif- icantly according to research scope and objectives, data availability, researcher preferences on data metrics, and spatio-temporal resolution. Major modeling and data selection considerations across these studies are discussed in the sections below and resulting research objectives to be explored are then identied. 13 2.2.1 Models and observations Studies have used dierent methods to quantify energy-climate relationships, including statistical techniques (e.g., regression) that relate energy use and climate indicators, and physics-based building energy modeling [144]. Statistical analyses oer advantages over other methods that rely on model-simulated data since they generally make use of real historical energy use and climate data. Regression models describe the relation- ship between a dependent variable, usually electricity consumption, and a temporally aligned independent variable, such as ambient temperature. Other climatic parameters such as humidity, wind speed, and solar insolation, have also been used as independent variables in multivariable regression analyses. Twenty out of 24 studies summarized in Table 2.1, representing the vast majority of analyses in this space, use regression methods. Within these 20 studies, 13 applied linear regression models [71] [123] [119] [107] [138] [132] [118] [9] [127] [95] [139] [64] [163], four applied non-linear regression models [68] [6] [54] [37], and three applied a mixture of linear and non-linear models [49] [18] [53]. 14 Table 2.1 Literature review of studies that investigate the in uence of climate on electricity consumption Number Model type Temperature indicator Stationary point tempera- ture Form of elec- tricity data Data tempo- ral resolution Data spa- tial reso- lution Derived electricity- temperature sensitivity Region Time period Citation 1 Tobit model (quadratic to CDD) CDD 18.3°C Air con- ditioning elec- tricity load Three min interval a , hourly b Household (metered only at air condi- tioner) Not reported Pittsburgh 2010 (Horowitz, Mauch, and Sowell 2014)[68] 2 Recurrent neural net- work Humidex in- dex (derived from tem- perature and humidity) N/A Mean electric current intensity Hourly a;b Sub-city (a district in Italy) Not reported Italy 2002- 2003 (Beccali et al. 2008)[16] 3 Time-series econometric model Hourly tem- perature N/A Hourly elec- tricity demand Hourly a;b City 0.3-0.5% per 1% tempera- ture increase Singapore 2003- 2012 (Doshi et al. 2012)[42] 4 Linear re- gression Hourly tem- perature 18°C Hourly elec- tricity demand Hourly a;b Grid scale (simi- lar to county) 6%°C 1 Sacramento County (Califor- nia) 08-08- 2012 (one day) (Pomerantz et al. 2015)[123] 5 Single- variable linear regres- sion CDH c 24°C Hourly elec- tricity demand Hourly a;b Country Mean hourly demand: 2.4%-3.5% °C 1 , peak hourly demand: 2.8%-4.2% °C 1 Thailand 2004 (Parkpoom and Harrison 2008)[119] 15 Table 2.1 continued from previous page Number Model type Temperature indicator Stationary point tempera- ture Form of elec- tricity data Data tempo- ral resolution Data spa- tial reso- lution Derived electricity- temperature sensitivity Region Time period Citation 6 Cubic re- gression for daily de- mand and linear re- gression for hourly peak demand Average daily tem- perature & maxi- mum hourly temperature N/A Daily elec- tricity demand and hourly peak elec- tricity demand Hourly a , Daily b , & monthly b State Annual demand: 1.4%-4.4% °C 1 d , daily peak demand: 1.7%-5% °C 1 d California 2004- 2005 (Franco and Sanstad 2008)[49] 7 Non-linear regression (formulas not speci- ed) Daily aver- age temper- ature and CDD/HDD 22°C Daily energy demand Hourly a , daily b City Summer daily de- mand: 0.6% °C 1 d Greece 1993- 2001 (Gianna- kopoulos and Psiloglou 2006)[54] 8 Multivariable linear regres- sion (also a semi- parametric function) Daily aver- age tempera- ture 21°C Daily average or peak demand Hourly a , daily b County Hourly load: 1.6%°C 1 d , Daily peak demand 1.9%°C 1 d USA 2006- 2014 (Auhammer, Baylis, and Hausman 2017)[9] 9 Multivariable non-linear regression (cubic) Daily max temperature N/A Daily peak demand Daily a;b Regional Daily peak demand: 2.3%°C 1 d Canada 1991- 1995 (Colombo, Etkin, and Karney 1999)[37] 10 Linear re- gression Daily max temperature N/A Daily peak demand Hourly a , daily b State August peak demand: 5.6%-7% °C 1 d California 1960- 1990 (Sathaye et al. 2013)[138] 16 Table 2.1 continued from previous page Number Model type Temperature indicator Stationary point tempera- ture Form of elec- tricity data Data tempo- ral resolution Data spa- tial reso- lution Derived electricity- temperature sensitivity Region Time period Citation 11 Multivariable linear regres- sion CDD & HDD 18°C Daily elec- tricity demand Daily a;b Country Not reported Spain 1983- 1999 (Pardo, Meneu, and Valor 2002)[118] 12 Linear re- gression Temperature at 8:00 and 14:00 N/A Summer peak- hour elec- tricity load Daily a;b Country Daily peak demand: 2.6%-2.7% °C 1 b Israel 1987- 1988 (Segal et al. 1992)[139] 13 Linear re- gression Daily aver- age tempera- ture N/A Daily energy demand Daily a;b Country Daily aver- age demand: 0.5%°C 1 Netherlands 1970- 1999 (Hekkenberg et al. 2009)[64] 14 Multivariable linear regres- sion CDD & HDD 18.5°C Monthly elec- tricity demand Hourly a Daily b and monthly b Country Daily de- mand: 1.1%-1.9% °C 1 d Greece 1993- 2002 (Mirasgedis et al. 2006)[107] 15 Using both linear re- gression and physical models CDD & HDD Building specic (but only average is reported) Monthly energy demand for com- mercial buildings Monthly a;b Buildings sampled in 101 cities 14%°C 1 d USA 1989 (Belzer, Scott, and Sands 1996)[18] 16 log-linear specication model Number of days per year that the mean daily temperature falls in each temperature bin (every 5°F) N/A Monthly elec- tricity demand of house- holds Monthly a;b Household 9%-13% °C 1 d California 2003- 2006 (Aroonrueng- sawat and Auhammer 2011)[6] 17 Table 2.1 continued from previous page Number Model type Temperature indicator Stationary point tempera- ture Form of elec- tricity data Data tempo- ral resolution Data spa- tial reso- lution Derived electricity- temperature sensitivity Region Time period Citation 17 Multivariable linear regres- sion CDD Not spec- ied Monthly elec- tricity demand Monthly a;b City Monthly demand: 7.49% °C 1 Bangkok, Thailand 2002- 2006 (Wangpattara- pong et al. 2008)[163] 18 Both quadratic and linear regression Monthly average temperature N/A Monthly elec- tricity demand Monthly a;b City domestic: 8.9% °C 1 commercial: 3.0% °C 1 industrial: 2.0% °C 1 Hong Kong 1990- 2004 (Fung et al. 2006)[53] 19 Multivariable linear regres- sion CDD & HDD 18°C Summer Peak elec- tricity demand Daily a , Monthly b State Not reported California 1970- 2005 (Lebassi et al. 2010)[95] 20 Time series multivari- able linear regression CDD & HDD (pop- ulation weighted) State spe- cic Monthly elec- tricity demand Monthly a;b State 2.54% °C 1 d USA 2008- 2012 (Huang and Gurney 2016)[71] 21 Multivariable linear regres- sion CDD & HDD 53°F- 71°F (dierent across fuels and sectors) Monthly elec- tricity demand Monthly a;b State residential: 0.1% °F 1 commercial: 0.04%°F 1 Maryland 1977- 2001 (Ruth and Lin 2006)[127] 22 Multivariable linear regres- sion Monthly average temperature Region specic Monthly elec- tricity demand Monthly a;b State Summer monthly: 5.97-32.2 kWh per capita °C 1 month 1 8 states in USA 1984- 1993 (Sailor and Muoz 1997)[132] 18 Table 2.1 continued from previous page Number Model type Temperature indicator Stationary point tempera- ture Form of elec- tricity data Data tempo- ral resolution Data spa- tial reso- lution Derived electricity- temperature sensitivity Region Time period Citation 23 Panel analy- sis models Average seasonal temperature N/A Monthly energy demand (in- cluding electric- ity) Monthly a Seasonally b and yearly b Country Not reported 31 coun- tries around the world 1978- 2000 (Bigano, Bosello, and Marano 2006)[21] 24 Not specied Annual aver- age tempera- ture N/A Grid load at a spe- cic time Not speci- ed a;b City 1. Peak demand: FL: 6% °C 1 , AL: 3% °C 1 , West TX: 6% °C 1 , NM: 3% °C 1 , AZ: 1% °C 1 , Southern CA: 3% °C 1 , North- ern CA: 1.5%°C 1 2. Annual us- age: FL: 3% °C 1 , AL: 1.5% °C 1 , West TX: 3% °C 1 , NM: 0.5% °C 1 , AZ: 6% °C 1 , Southern CA: 1.5% °C 1 , North- ern CA: 0.5%°C 1 Dierent cities in US 1986 (Akbari 1992)[1] 19 Table 2.1 continued from previous page Number Model type Temperature indicator Stationary point tempera- ture Form of elec- tricity data Data tempo- ral resolution Data spa- tial reso- lution Derived electricity- temperature sensitivity Region Time period Citation a Source data resolution b Processed data resolution c CDH : Cooling Degree Hours, dened by the cited literature as: \a short-term version of CDD described by: P N h=1 (T h T b ) for T T b and 0 otherwise, where N is the number of hours in the period of interest,T is the air temperature, and T b is the cooling base temperature, commonly taken to be 24 °C in Thailand." d Value is calculated from percentage or absolute change in electricity consumption under dierent climate change scenarios versus a baseline period. 20 2.2.2 Electricity data metrics and resolution Studies utilizing real-world electricity data have used a variety of metrics or indicators to characterize elec- tricity use based on source datasets with widely varying spatiotemporal resolutions. While most studies surveyed use hourly, daily, monthly, seasonally, or yearly accumulated electricity usage data [71] [42] [123] [119] [49] [18] [6] [107] [21] [132] [118] [9] [127] [64] [163] [54] [53], six use peak electricity demand (i.e. elec- tricity load during time periods of highest demand and electricity prices) [49] [138] [9] [95] [139] [64], and one uses mean electric current intensity (Amperes) [16]. Ideally, the electricity indices utilized in a particular research study should re ect the research questions under investigation. For example, electricity usage data are most suitable for predicting future energy use trends and patterns (e.g. as a result of climate change or urban heat islands) [16] [71] [42] [119] [18] [6] [107] [118] [127] [163] [37] [53], while peak electricity demand data are valuable for informing grid reliability [49] [138] [9] [95] [139] [64]. The underlying resolution of these datasets is also an important driver of the accuracy in computed energy use-climate relationships [133] [74]. Coarse spatial resolution has been a major limitation across the majority of prior research endeavors in this eld. Past studies rely on electricity data at the sub-city [16], city [42] [163] [54] [53] [1], county [123] [9], state [71] [49] [138] [132] [127] [95], regional [37], or country levels [119] [107] [21] [118] [139] [64] since house-level data have been less commonly available. However, these spatial scales are not sucient for building a highly resolved understanding of climate-driven variations in energy consumption behavior, especially for regions with large climatic variations, such as those adjacent to mountains and coasts like the Los Angeles basin. The temporal resolution of datasets also varied considerably across the surveyed studies. For source data resolution, eight of the 24 studies use monthly aggregated electricity usage [71] [18] [6] [21] [132] [127] [163] [53], ve use daily aggregated electricity usage [118] [64] [95] [139] [37], nine use hourly electricity data [16] [42] [123] [119] [49] [107] [138] [9] [54], one uses sub-hourly meter data at the appliance level instead of the entire household [68], and one study does not specify data resolution[1]. The majority of studies have source data and processed data of the same resolution, but eight utilize processed data at a coarser resolution than source data [68] [49] [107] [21] [138] [9] [95] [54], meaning that the researchers chose to aggregate their datasets prior to analysis. 21 Although a few studies address the importance of high resolution data [9] [133] [74], typically no justica- tion is provided on why one resolution is chosen over other possible resolutions. It is assumed that temporal resolution re ects the availability of source data in most cases. Most studies rely on data being shared by utility companies or grid operators, so resolution is constrained by the data provided to researchers [16] [42] [123] [119] [49] [107] [138] [118] [9] [64] [139] [163] [54] [37] [53]. Relatively low temporal resolution data (e.g., monthly or yearly averages) have traditionally been most easily acquired from technical reports or bills [18] [6]; while the widespread dissemination of smart electricity meters has enabled the collection of hourly electricity data, few studies have had access to these data for analysis [68]. 2.2.3 Temperature data metrics Prior studies have used a variety of indicators for characterizing climate. For example, recent studies have utilized a range of temperature metrics, including Cooling/Heating Degree Days [71] [68] [119] [18] [107] [118] [127] [95] [163], hourly temperature [42] [123], daily average temperature [49] [9] [64] [54], daily max temperature [138] [37], monthly average temperature [132] [53], seasonal average temperature [21], histograms of daily temperature [6] [1], and other indices derived from temperature data [16]. Of the literature surveyed in Table 2.1, 10 out of 24 used CDD/HDD [71] [68] [119] [18] [107] [118] [127] [95] [163] [54] and six used daily average or max temperature [49] [138] [9] [63] [54] [37], suggesting that daily temperatures have been the most commonly utilized resolution in this body of literature. 2.2.4 Stationary point temperatures Studies utilizing CDD and HDD (see Section 3.2 and Eq. (2) for more details on CDD/HDD) as a tem- perature indicator need to choose a pre-dened, xed threshold temperature to calculate this metric. The threshold (also sometimes called a \stationary point temperature" or \base temperature") refers to the tem- perature below (above) which no cooling (heating) is needed (discussed in more detail in Section 3.2). In past studies, 18 °C is the most common threshold temperature, chosen by ve out of 13 studies that use CDD and/or HDD [68] [123] [107] [118] [95]. 60 °F (15.6 °C), 21 °C, 22 °C, and 24 °C are also used in past studies [119] [9] [127] [54]. Three studies assign specic stationary point temperatures to dierent buildings 22 or regions [71] [18] [132]. One study does not specify stationary point temperature [163]. Several methods have been applied to set a stationary point temperature, including: 1) choosing the temperature threshold arbitrarily; 2) referencing a previous study in the same or neighboring region; and 3) extracting it from a preliminary electricity-temperature plot. Only one study analyzes the impact of setting a region-specic stationary point temperature using a segmented regression model, but the study calculates this point at the state level only [71]. 2.3 Methods To address the research questions presented above, this study utilizes a dataset representing the hourly electricity consumption of 1245 households across California for a one-year period. We also utilize data from a network of 145 weather stations to assess hourly temperatures in locations adjacent to each home. A seg- mented linear regression model is applied to assess the electricity-temperature sensitivity of each household. The electricity data are spatially and temporally aggregated in various ways (i.e. both before and after computing electricity-temperature sensitivity) to assess how data resolution impacts electricity-temperature sensitivity. In addition, the dependence of chosen temperature indicators on computed sensitivities is as- sessed. The dataset used here includes only residential homes and thus varies from many previous studies using spatially aggregated datasets, which would also include commercial and industrial buildings. 2.3.1 Datasets Hourly smart meter data records of electricity usage at the household level from 1245 residential customers (after data cleaning and screening) across California were analyzed. These households re ect utility customers that voluntarily downloaded an energy-related smart phone app for tracking their electricity use. Since this sample is likely biased towards energy-conscious households, this paper focuses on comparing methods for computing energy-temperature sensitivity but does not claim that computed values are representative of the general population of California cities. Only zip code information for each household included in the dataset was provided to protect customer privacy. Several procedures for data cleaning and screening were carried out. First, to fully capture the year-round relationship between residential electricity consumption 23 and ambient temperature, only customers with one full year of electricity data (05/18/2015-05/17/2016) were included in the study. Second, households that could be identied as energy generators (e.g., with solar photovoltaic installations) were removed from the dataset to reduce the impacts of these households on load curves. We dened these generating households as those that had a negative value of electricity usage at any time; thus, households whose onsite generation never exceeded their own energy use through the period of study would not be agged and is a limitation of the study. The households included in the dataset spanned 41 counties, 549 zip codes, and 15 of 16 climate zones in California [29] [51]. These climate zones were established by the California Energy Commission (CEC) and specically characterize building energy use under various climate characteristics [29] [51]. Hourly ambient temperature data over the year under investigation were retrieved from the California Irrigation Management Information System (CIMIS) [27] [52], which includes a network of over 145 auto- mated weather stations in California covering most of the state's population centers. Geospatial analysis was executed with ArcGIS (10.4.1, ESRI, Redlands, CA, USA) to map each household with an electricity record to its nearest weather station. Figure 2.1 illustrates the number of homes per zip code and location of weather stations of this study. It is important to note that this dataset, which includes utility records averaging 30 households per county and less than three households per zip code, is not statistically representative of the population from a spatial perspective. Accordingly, the objective of this study is not to dene the eects of climate on electricity across regional boundaries (i.e. city, county, climate zone); rather, the goal is to assess the in uence of spatiotemporal electricty data resolution and climate indicators on derived electricity-temperature sensitivity (see research questions in the Introduction). 2.3.2 Statistical models The nonlinearity of relationships between building energy consumption and ambient temperature has been established in previous studies [49] [132] [54] [37] [53]. Nonlinear regression models (i.e., polynomial functions) have been developed that can achieve a good t among variables. However, sophisticated models may have issues with overtting and can thus fail to generalize trends under investigation and are also not applicable 24 Figure 2.1 Map showing locations of the 1245 residential electricity customers (shown as num- ber of households per zip code) and 145 CIMIS weather stations considered in this study. Each household was linked to a weather station based on shortest distance. 25 to other regions [65]. To address the nonlinear relationship between electricity use and ambient temperature in residential homes, while avoiding overtting, a segmented linear regression model proposed by [108] will be utilized in this analysis. The segmented linear regression reveals two important pieces of information. The rst is the stationary point temperature (SPT ), which sits at the stationary point of the piece-wise linear function and can be thought of as analogous to the base temperature in the CDD method. In other words, stationary point temperature is the temperature at which household electricity consumption reaches a minimum, with the assumption that no cooling or heating is needed at this temperature. In the segmented regression model, the stationary point is calculated iteratively to determine the best overall piece-wise linear t of the original dataset. The second is the slope of the linear regression to the right of the stationary point temperature (referred to in this analysis as the \electricity-temperature sensitivity"), representing the change in electricity consumption that corresponds to a change in ambient temperature of one degree Celsius. Electricity-temperature sensitivity can be aected by factors like house size, insulation, behavior, etc., since these factors also aect air-conditioning use. Figure 2.2 shows an example segmented regression for a home in Clovis, California. Daily aggregated elec- tricity usage is plotted against daily average temperature, and stationary point temperature and electricity- temperature sensitivity are illustrated. The plot in Figure 2.2 is thus divided into two regimes: (1) strong positive sensitivity between electricity use and temperature to the right of the stationary point tempera- ture, and (2) electricity use that is relatively insensitive to temperature change to the left of the stationary point temperature. In California, cooling energy demand from air conditioning is driven by electricity while heating is mainly supported by natural gas [46], which is why there is not a strong increase in electricity use as temperatures decrease below the stationary point. For the same reason, only one stationary point is identied in the segmented model, whereas in some regions there might be two (e.g., in the case of Israel described in [17]). 26 Figure 2.2 Example of a home in Clovis, CA illustrating the stationary point temperature (SPT) and electricity-temperature sensitivity through a segmented linear regression method. 27 To address research question 1, stationary point temperature and electricity-temperature sensitivity are computed for dierent spatial aggregation levels using a segmented regression dened as: E s;t t = 8 > > < > > : 1 + 1 T s;t 1 ; T s;t <SPT s;t 2 + 2 T s;t 2 ; T s;t SPT s;t (2.1) where E s;t is a vector of residential electricity consumption over a period of time t (vertical axis in Figure 2.2). Es;t t is expressed in units of electric power (kW). T s;t is a vector of near-surface ambient temperatures (in the units of°C) over the same period of time (horizontal axis in Figure 2.2). The rst row of Equation (2.1) describes the relationship between Es;t t andE s;t forT s;t < SPT s;t . The second row of Figure 2.2 describes the relationship between Es;t t andT s;t forT s;t SPT s;t . The electricity-temperature sensitivity, S s;t , is dened as the slope of the regression line above the SPT (i.e., 2 ), with units of kW °C 1 . SPT is calculated by iteratively locating the intersection of the two linear regions to maximize the model's overall coecients of determination (r 2 ). Thus, in Equation (2.1),E s;t andT s;t are inputs to the segmented regression and all other variables are outputs. (Note: 1 , 2 , and 1 are additional regression coecients and is the error term.) The spatial and temporal aggregations of data represented in vectors E s;t andT s;t , as well as scalars S s;t , and SPT s;t , t, are indicated by subscripts s and t, respectively. Values of subscript s in this study include household, city, county, and climate zone, andt can be hourly or daily. For example, ifs = household forT s;t , thenT s;t corresponds to the observed temperature at the nearest weather station for that home, while if s = city,T s;t corresponds to the population weighted spatial mean observed temperature for that city. By \population", we mean \number of homes", so \population-weighted spatial mean" means we take the average of temperature readings from multiple weather stations in the area, weighted according to how many homes are assigned to each weather station. (Note that we discuss various daily temperature metrics below when discussing research question 2.) For s = household (i.e., no spatial aggregation), segmented linear regression is conducted separately for each of 1245 homes using their hourly or daily aggregated electricity consumption (i.e., depending on t). The mean values of both stationary point temperatures (SPT s=household;t ) and sensitivities (S s=household;t ) 28 are computed by taking the mean over all 1245 households of computed stationary point temperatures and sensitivities. For s = city, county, or climate zone, the segmented linear regressions are carried out using spatially averaged electricity consumption over spatial extents, along with population-weighted temperature to mimic studies using more spatially aggregated data to compute sensitivities. The mean values S s;t and SPT s;t are then computed by taking the population-weighted average of all city, county, or climate zone level sensitivities and stationary point temperatures (i.e., depending on s), respectively. For example, to compute S s=city;t=daily , we rst compute spatially aggregated electricity use for each city, and then the city level (hourly) electricity data is accumulated to daily resolution. For temperature, rst the metric of choice (see the next paragraph) is computed for each weather station (i.e. daily minimum, average, or maximum), and then city population-weighted averages are computed. Then, segmented linear regressions are applied using the averaged data per city to compute city-level stationary point temperatures and electricity-temperature sensitivities. Lastly, S s=city;t=daily and SPT s;t are computed by taking the mean of city-level values. To address research question 2, the relationship between residential electricity use and various temper- ature indicators (i.e., hourly temperature, daily average temperature, daily maximum temperature, daily minimum temperature, and CDD) are explored. To quantify the eect of utilizing dierent temperature indicators on computed electricity-temperature sensitivities, we carry out the segmented linear regression using hourly temperature (i.e., t = hourly), daily maximum, daily minimum, and daily average temperature (i.e., for t = daily). This comparison is carried out for both a typical household in San Jose, and also for all households (within our dataset) in the City of San Jose. We also compute sensitivity using CDD assuming a uniform base temperature T b = 18°C for all homes. Since CDD calculations already include base temperatures, a standard linear regression model is applied rather than the segmented regression; the slopes of these linear regressions represent the electricity-CDD sensitivity. The coecient of determination (r 2 ) values of these regression models are compared to assess the quality of t. CDD is computed as: CDD = R day max[0;T s;t=hourly (h)T b ]dh 24 (2.2) T s;t=hourly (h) is the hourly ambient temperature for hourh expressed in°C andT b is the base temperature (i.e., 18.0°C in this study). The daily value of CDD can be obtained by integrating T s;t=hourly (h) over each 29 day as done in [12]. Physically, the base temperature is the ambient temperature at which a building's heat loss and heat gain reaches an equilibrium, such that cooling is not needed. The base temperature is often chosen based on previous studies that focus on a similar geographical zone or is set arbitrarily. Due to dierent climate zones, building characteristics, and occupant behavior patterns, the base temperature can vary signicantly among spatial areas [71] [132]. This issue has been identied by several previous studies [16] [71] [18] [132] [127] included in Table 2.1. In our study, CDD is calculated as the cumulative degrees beyond 18.0°C for each hour on a daily basis. We note that computing electricity-CDD sensitivities using linear regression is analogous to that of electricity-temperature sensitives using segmented linear regression with a xed stationary point temperature. 2.4 Results 2.4.1 Sensitivity and stationary point temperature distributions One of the biggest advantages to using household level electricity consumption data to derive electricity- temperature relationships is that these data illustrate home-to-home variability in terms of (a) the ambient temperatures at which homes start using increased electricity (i.e., stationary point temperature), and (b) the amount of additional electricity that homes use as temperatures increase beyond the stationary point. Probability density distributions of electricity-temperature sensitivities (top row) and stationary point tem- peratures (bottom row) for all households investigated here are plotted in Figure 2.3. Subplots a and c in this gure use daily average temperature, while subplots b and d use daily maximum temperature. Both distributions of sensitivities are skewed to the right. The long tails of the probability density distributions of electricity-temperature sensitivities represent homes that have large increases in electricity consumption as temperatures increase above the stationary point (Figure 2.3 (a) and (b)). When daily maximum temperature is used in the segmented regression, 47% of households in the dataset have a sensitivity less than 0.05kW °C 1 , while 24%, 24%, and 6% of households have a sensitivity value of 0.05 to 0.1, 0.1 to 0.2, and over 0.2kW °C 1 , respectively. For daily average temperature, 34% of households in the dataset have a sensitivity less than 0.05kW °C 1 , while 30 Figure 2.3 Probability density distributions of electricity-temperature sensitivities (a, b) and stationary point temperatures (c, d) of all 1245 homes in this study's dataset. Red dashed lines indicate the mean values of all 1245 homes. Blue dashed lines indicate the median values of all 1245 homes (Note that in panel d, the blue dashed line partially overlaps the red dashed line.) Sensitivity in units of kW °C 1 can be converted to kWh day 1 °C 1 by multiplying by a factor of 24. (For interpretation of the references to color in this gure legend, the reader is referred to the web version of this article.) 31 19%, 32%, and 15% of households have a sensitivity value of 0.05 to 0.1, 0.1 to 0.2, and over 0.2kW °C 1 , respectively. Both temperature indicators have similar distribution shapes, but daily average temperature leads to overall higher sensitivity values than daily maximum temperature. In other words, daily electricity consumption at the household level is generally more sensitive to daily average temperature than daily maximum temperature. If daily maximum temperature is used, the stationary point temperatures of the 1245 homes are dis- tributed within a range from about 10-35°C with a mean value of 23.1 4.9°C (73.6 8.9°F) (Figure 2.3(d)). For daily average temperature, the distribution of stationary point temperatures is almost normal within a range from about 5-25°C and more concentrated to the mean value, which is 17.1 3.9°C (62.8 7.0°F) (Figure 2.3(c)). It is interesting that a small percentage of homes have negative or zero sensitivity values in Figure 2.3(a) and (b). This can be attributed to a lack of cooling devices in these homes, or on-site energy generation (e.g., solar photovoltaics). Also, a small number of homes have stationary point temperatures less than 10°C in Figure 2.3(c) and (d), which appears anomalously low. One possible explanation is dier- ences between ambient and indoor air temperature due to solar heating; in this case, indoor temperatures may be higher than ambient, causing inhabitants to turn on air conditioners at lower ambient temperatures than expected. More information about building design is needed to further explore this possibility. Homes without cooling devices could also be the cause, with lower than expected stationary point temperatures being identied for reasons other than increasing cooling energy use. 2.4.2 Impact of temperature indicators on computed electricity-temperature sensitivities The impact of using various temperature indicators on computed electricity-temperature sensitivity is illus- trated using (a) electricity data for a typical household in San Jose, and (b) averaged electricity consumption for all households for which we have data in San Jose (Figure 2.4). The city of San Jose is chosen because our dataset includes a relatively large number of homes (n=80) compared to other cities. In Figure 2.4, the rst row shows hourly electricity consumption versus hourly temperature. The other rows show daily accumulated electricity consumption versus daily minimum temperature, daily average temperature, daily 32 maximum temperature, and cooling degree days at 18°C (CDD18C) both including and excluding days with CDD18C = 0. (Days with CDD18C = 0 occur when hourly temperatures remain below 18°C.) The temperature indicator utilized signicantly aects the coecient of determination (r 2 ) and the com- puted electricity-temperature sensitivity. Overall, using hourly electricity and temperature data shows weak coecients of determination relative to the daily metrics. Among the daily metrics (i.e., daily minimum, av- erage, and maximum temperature), daily average temperature is shown to lead to (a) the highest coecients of determination for both the typical home and all homes in San Jose, and (b) highest sensitivity. Daily maximum temperature and daily minimum temperature lead to the second and third highest sensitivities among the daily metrics. Linear regressions of daily aggregated electricity use versus CDD18C (including days withCDD18C = 0) show similar sensitivity as daily average temperatures for both the individual home and city of San Jose. The coecient of determination and sensitivity increases when days with CDD18C = 0 are removed from the regression. 2.4.3 Impact of spatial aggregation on computed electricity-temperature sensitivity Mean values of stationary points (SPT s;t=daily ) and sensitivities (S s;t=daily ) for California derived using data with dierent spatial aggregation levels (i.e., household, city, county, climate zone, and state) are displayed in Table 2.2. Sensitivity values are calculated using both daily maximum temperature and daily average temperature for comparison purposes. When daily maximum temperature is used, the mean value of sensitivities calculated using household level electricity data (i.e., no spatial aggregation) S s=household;t=daily is 0.079 kW °C 1 , about 19% higher than computing sensitivities where s is spatially aggregated to the city, county, climate zone or state-level, which range from 0.063 to 0.066 kW °C 1 , depending on the level of spatial aggregation (Table 2.2). A similar phenomenon is also observed using daily average temperature. The mean value of sensitivities computed for the 1245 homes using household level electricity data is 0.11 kW °C 1 , higher than that using aggregated data, which ranges from 0.097 to 0.10 kW °C 1 . 33 Figure 2.4 Segmented linear regression applied to a single household (left column) and the average of households in our dataset (n=80) within the City of San Jose, California (right column) using various temperature indicators: hourly temperature (a, b), daily minimum temperature (c, d), daily average temperature (e, f), daily maximum temperature (g, h), CDD18C including days with CDD18C = 0 (i, j), and CDD18C without days where CDD18C=0 (k, l). SPT corresponds to Stationary Point Temperature and S corresponds to electricity- temperature sensitivity. In panel (i-l), Base Temp corresponds to the base temperature, which can be seen as a prescribed stationary point temperature. 34 Table 2.2 Electricity-temperature sensitivities and stationary point temperatures computed with various spatial aggregations using electricity consumption data for 1245 California house- holds. Resolution of electricity use data, E s;t , used in the segmented regression analysis Mean value of electricity- temperature sensitivity S s;t=daily (kW °C 1 ) Standard deviation of electricity- temperature sensitivity (kW °C 1 ) Mean value of sta- tionary point temperature SPT s;t=daily in °C (°F) Standard deviation of stationary point temperature in °C (°F) Using Daily Maximum Temperature s = household a 0.079 f 0.18 23.1 (73.6) 4.93 (8.87) s = city b 0.066 0.058 22.4 (72.3) 4.46 (8.03) s = county c 0.063 0.04 22.2 (71.9) 3.21 (5.78) s = climate zone d 0.063 0.036 22.3 (72.1) 3.30 (5.93) s = state e 0.064 N/A 22.1 (71.8) N/A Using Daily Average Temperature s = household 0.11 f 0.1 17.1 (62.8) 3.86 (6.95) s = city 0.098 0.082 17.1 (62.8) 3.50 (6.30) s = county 0.098 0.058 17.2 (63.0) 2.08 (3.74) s = climate zone 0.097 0.04 17.2 (63.0) 2.64 (4.75) s = state 0.1 N/A 17.0 (62.6) N/A a Segmented linear regression was performed for each household, and then sensitivity and stationary points per home were averaged for all homes in California. b Electricity data were averaged by city, segmented regression was performed for each city, and then sensitivity and stationary points were population-weighted averaged for all cities. c Electricity data were averaged by city, segmented regression was performed for each city, and then sensitivity and stationary points were population-weighted averaged for all counties. d Electricity data were averaged by city, segmented regression was performed for each city, and then sensitivity and stationary points were population-weighted averaged for all climate zones. e Electricity data were averaged for entire state of California and then segmented regression was performed for state-averaged data f Equivalent to 10.5% change in electricity consumption °C 1 (% change means the relative change in electricity consumption per °C increase using the consumption at the stationary point temperature as a baseline). g Equivalent to 15.3% change in electricity consumption °C 1 The level of spatial aggregation aects electricity-temperature sensitivity more than stationary point temperature (up to 19% for sensitivity vs. 4% for stationary point temperature by using daily maximum temperature, and up to 6% vs. 1% by using daily average temperature). Using daily maximum temperature, the mean value of computed stationary point temperature SPT s=household;t=daily for all 1245 households is 23.1 °C (73.6 °F). Using electricity data that are spatially aggregated, stationary point temperatures SPT s=city=county=climatezone=state;t=daily are slightly lower, ranging from 22.1 to 22.4 °C (71.8 to 72.3 °F). 35 Using daily average temperature, the stationary point temperature is 17.0-17.2 °C (62.6-63.0°F), regardless of level of aggregation. 2.5 Discussion 2.5.1 Advantages of utilizing high spatiotemporal resolution data Using high spatiotemporal resolution electricity and climate data to investigate the eects of climate variabil- ity on energy consumption oer advantages over using aggregated data. From a research perspective, having access to household-level data enables the ability to investigate how data resolution in uences computed electricity-temperature interactions. Table 2.2 indicates that computed electricity-temperature sensitivity is dependent on the level of spatial aggregation of the data used in the segmented linear regressions. For example, our research suggests that computing the electricity-temperature sensitivity using household data and then averaging all households in a state results in a dierent sensitivity value than computing the sensi- tivity using state-mean electricity data, as illustrated in Table 2.2. In addition, using electricity data at the household level is ideal for most accurately calculating electricity-temperature sensitivities given that more representative temperature data for each household can be used in the analysis. This is especially important for cities like Southern California that have strong spatial variability in climate. Two case studies are presented here to explicitly illustrate how dierent electricity-temperature sensitivi- ties can arise using household level versus aggregated data (see Figure 2.5). In Case I, Household A (zip code 94583, San Ramon) has large daily non-cooling loads (i.e., electricity use to the left of the stationary point) and a large sensitivity. Household B (zip code 90504, Torrance) has relatively small daily non-cooling loads and a small corresponding sensitivity. A smaller sensitivity value is calculated if we take the mean value of sensitivities computed per household (i.e., S s=household;t=daily ) compared to performing the segmented re- gressions after aggregating the electricity consumption of two households (i.e., S s=city;t=daily ). In the latter case, the average sensitivity will be weighted towards Household A because of its higher electricity use and thusS s=household;t=daily >S s=city;t=daily . In Case II, Household C (zip code 92571, Perris) has overall high daily electricity use with relatively small sensitivity while Household D (zip code 92069, San Marcos) has 36 Figure 2.5 Two case studies illustrating two households with dierent non-cooling electricity usages and sensitivities. In Case I (top row), Household A has higher daily electricity use and a larger electricity-temperature sensitivity value than Household B. In Case II (bottom row), Household C has a higher daily electricity use, but a smaller sensitivity than Household D. In either case, the average sensitivity for the two households, if calculated based on aggregated electricity use, will more heavily weight the larger electricity consumer. small daily electricity consumption with relatively high sensitivity. In this case, a higher sensitivity value is calculated if we take the mean value of sensitivities computed per household compared to performing the segmented regression after aggregating the electricity consumption of two households (S s=household;t=daily < S s=city;t=daily ). In either case, computing the mean sensitivity after spatial aggregation will weight big electricity consumers more heavily, and give less weight to smaller consumers regardless of their sensitivity values. A second advantage to using high spatiotemporal resolution data is that they oer the ability to investigate the distribution of energy use patterns among dierent households, which in this study is re ected by stationary point temperatures and sensitivities. Figure 2.3 illustrates that households in this sample have a wide distribution of sensitivities and stationary point temperatures. Variability in sensitivities are likely a result of variations in occupant behavior patterns, building and HVAC system characteristics, and climate zone. More information about building characteristics at the household level is needed to further quantify the relative importance of these causal factors of variability in sensitivity. We hypothesize that a large number of households show small sensitivity to ambient temperature change due to lack of air conditioning equipment presumably concentrated in coastal locations, and possibly also due to homes with relatively low 37 square footage and/or occupants that cannot aord air conditioning. These hypotheses should be validated with additional datasets in future analyses. Figure 2.3 also illustrates that dierent households have unique stationary point temperatures, which is an important distinction between the method used in this study and previous studies that assume xed base temperatures that are not necessarily computed based on the dataset (e.g. CDD18C). Spatial variations in stationary point temperatures re ect building characteristics, occupant behavior, and climate variability [71]. If aggregated electricity data are used, only one stationary point temperature for the entire region can be computed and used in regressions. 2.5.2 Roles of temperature indicators As indicated in Figure 2.4 and Table 2.2, electricity-temperature sensitivities are dependent on the temper- ature indicator used in the regression. We suggest that the following considerations be used to help decide which temperature indicator is of interest. First, electricity and temperature show the strongest relationships when computed using data at daily temporal resolution. Regressions between hourly household electricity consumption and hourly temperature result in relatively low coecients of determination (r 2 ). This can be partially explained by the daily electricity use patterns of residential homes, which can be heavily aected by household energy consumption behavior. For example, energy use patterns in many cases will not directly follow hourly temperatures since occupants that go to work during normal business hours may peak in their electricity usage in the evening when ambient temperatures are not at their daily peak. There can also be a timing lag between ambient outside temperature rise and its impact on indoor temperature (and thus, air conditioning usage). This reasoning can also be partially observed by Figure 2.6, which presents a histogram of the hour of day at which summertime (dened as July, August, September) peak electricity consumption occurs for each household. (In other words, the height of each bar represents the total number of households that have summertime peak electricity consumption at that time of day.) The hour of day corresponding to peak electricity consumption per household represents the most frequently occurring daily peak time over the summertime period. We observe in this study that the timing of most households' peak electricity use does not correspond to daily peak ambient temperature (usually during mid afternoon). By contrast, a 38 Figure 2.6 Histogram of time of day corresponding to peak energy for each household during summer (July, August, September). Peak hourly electricity use occurs in the late afternoon to early evening for the majority of households in this study. large number of households have peak energy use in the late afternoon through early evening. Although we currently lack data to calculate how much of this peak energy use is driven by air conditioners, it is reasonable to assume that air conditioners are a major driver of evening electricity consumption since a large fraction of occupants are home from work during this period and might choose to cool their homes for occupant comfort. Second, the choice of whether to use daily average or maximum temperature depends on the research questions under investigation. For example, most research analyses assessing the impacts of climate change on electricity utilize daily average temperature, since this indicator is what is estimated most commonly in global climate modeling studies [71] [49] [18] [21] [64] [127] [8] [95]. On the other hand, daily maximum 39 temperature is often used to predict future peak electricity demand, which is driven instantaneously by extreme heat during the day [49] [138] [37]. While it should be noted that total electricity usage is dependent on many factors, in this study, daily average temperature shows the best segmented linear relationship with electricity use relative to other temperature indicators (i.e., hourly, and daily minimum and maximum temperature). One of the driving reasons for this trend is likely due to nature of temperature uctuations across diering climates, which can change the need for cooling throughout the day. For example, while a coastal home may experience similar daily average temperature (e.g., 30°C) with an inland home in a dry desert region, the diurnal temperature range that each home experiences can be vastly dierent (e.g., coastal daily temperature range: 28-32°C vs inland: 20-40°C). In this example, the maximum daily (or minimum daily) temperature is vastly dierent in each region, even when the daily average temperature is the similar. While one might assume that total daily electricity consumption might scale with maximum temperature, the inland home would experience a great deal more nighttime cooling than the coastal home; this nighttime cooling might attenuate the need for some daytime air-conditioning use since it experiences pre-cooling. On the other hand, while the coastal home might not be subjected to extreme maximum temperatures, it also experiences less cooling relief during the evening in this example. Third, setting a uniform, pre-dened base temperature as is done in the CDD calculation is not as good as computing household level stationary point temperatures. Using pre-dened base temperatures can lead to inaccuracies in regressions when occupant behaviors lead to the AC turning on at ambient temperatures below the threshold. This eect can be observed in Figure 2.4(i) and (j), illustrated by data points with CDD18C = 0. Including these zero values aects the regression slope (i.e., sensitivity) relative to excluding the zeros (see Figure 2.4(k) and (l)). In addition, using CDD18C as the indicator (with linear regression) leads to coecients of determination that are smaller than when using daily average temperature (with segmented regression). Thus, using daily average temperatures with segmented regression may be best for studies that investigate sensitivities of daily electricity use (as opposed to peak energy use) rather than CDD18C. 40 2.5.3 Comparing computed electricity-temperature sensitivities to previous studies Using the dataset described in this study, the computed electricity-temperature sensitivityS s=household;t=daily is 0.079 kW °C 1 using daily maximum temperature and 0.11 kW °C 1 using daily average temperature. However, previous studies commonly present electricity-temperature sensitivity in units of percentage change in electricity consumption per °C increase in ambient temperature (% °C 1 ). Thus, to be comparable with sensitivity values from past studies, we also computed electricity-temperature sensitivity in units of percent- age change in electricity consumption per °C increase in ambient temperature (%°C 1 ). These sensitivities (S s=household;t=daily ) were 10.5%°C 1 using daily maximum temperature and 15.3%°C 1 using daily aver- age temperature in this dataset. (All of these values are computed by calculating the sensitivities in percent units for each household and then averaging over all households.) Among previous studies surveyed, three re- port electricity-temperature sensitivity values computed using hourly or monthly average temperature data, presented as 6% °C 1 [123], 8.9% °C 1 [53], and 9-13% °C 1 [6], which are similar in magnitude to those computed in this study. Calculating electricity-temperature sensitivity in units of kW °C 1 versus % °C 1 presents tradeos in terms of insights gained. Sensitivities in units of kW °C 1 will be highest for households with high cooling loads regardless of the magnitude of noncooling loads, while reporting in units of %°C 1 is dependent on the magnitude of cooling loads versus non-cooling loads. Thus, a household with small non-cooling loads would have a higher percentage increase in cooling load per unit temperature rise relative to a household with high non-cooling loads, even if the cooling load increase in kW °C 1 are equal; yet, reporting the percentage of cooling load increase is insightful for understanding trends such as the relative increases in electricity costs for dierent socioeconomic populations. In addition to these considerations regarding selected units, several caveats of such comparisons in sen- sitivities between this and prior studies should be noted: 1) the sample size of this study is not statistically representative of electricity users in the state of California; and 2) electricity-temperature sensitivities can be driven by numerous factors, e.g., occupant behavior patterns, climate zones, housing characteristics, etc. 41 Neither this study nor previous studies have revealed enough detailed information to explain these dierences in sensitivities, but will be the focus of future research. 2.6 Conclusion Despite a growing body of literature utilizing various types of electricity usage and temperature source data across a wide range of spatiotemporal resolutions, no research to our knowledge has focused on assessing the impacts of data resolution and choice of temperature metrics on computed functional relationships between electricity usage and ambient temperature. To address this, we use hourly energy use records from 1245 customers across California along with corresponding hourly ambient temperature data to investigate the dependence of spatiotemporal data resolution and temperature metrics on computed electricity-temperature sensitivities. Sensitivities are computed using a segmented linear regression model. We use this regression model with input data at various resolutions to emulate source data of spatial resolutions including household, city, county, and climate zone, and temporal resolutions including hourly and daily. In addition, we compare the impacts on computed electricity-temperature sensitivity of using hourly, daily minimum, daily mean, daily maximum, or cooling degree days as temperature indicators in the regression model. Results indicate that the strongest relationships between electricity consumption and temperature, as indicated using the coecients of determination, are computed when using data at daily temporal resolution (i.e., daily accumulated electricity consumption and daily average temperature), even when compared to those relationships computed using more resolved hourly electricity consumption and temperature data. This nding indicates that increasing the temporal resolution of electricity data to increments smaller than daily do not translate to higher regression model performance. By contrast, increasing the spatial resolution of electricity data improved the accuracy of computed electricity-temperature sensitivity (i.e., since ambient temperatures experienced by the house can be more accurately determined), and elucidated new trends masked by using spatially aggregated data as well. The choice of temperature indicator can also impact the computed electricity-temperature sensitivities and stationary point temperature values. In this study, regression models utilizing daily average temperature oered higher coecients of determination than those using daily minimum temperature, daily maximum 42 temperature, or cooling degree days, thus showing the best segmented linear relationship with electricity usage. Moreover, we nd that computing a unique household level stationary point temperature is superior to setting a uniform, pre-dened base temperature as is done in standard CDD calculations. Having access to household level data enables the calculation of a household specic base temperature (assuming that base temperature is eectively the equivalent of stationary point temperature), which can enhance the accuracy of computed sensitivities. In addition, having household level electricity data allows for determining more representative temperatures for that household, which can improve the accuracy of computed sensitivities especially in places like Southern California where temperature variations are signicant. However, it should be noted that the choice of using daily average or maximum temperature depends on the research questions under investigation, and there are cases where daily average temperature might not be the indicator of choice. To summarize, the take-away points of this study are: • Sensitivities between residential electricity consumption and ambient temperature are best computed using daily data, as indicated using the coecient of determination from a segmented linear regression model. Daily data led to improved regressions even when compared to hourly electricity consumption and temperature data. Daily data refers to daily accumulated electricity usage or daily average power consumption. The choice of whether to use daily average versus maximum temperature depends on the research question under investigation (see next bullet). • Daily average temperature is the best choice for exploring general relationships between residential electricity consumption and ambient temperature. While use of daily average temperatures led to the highest coecients of determination, use of daily maximum temperature can be more appropriate for investigating certain research questions (e.g., relationships between peak electricity use versus temperature). • Having access to household level data can enhance the accuracy of computed sensitivities, and elucidate new trends (e.g., household-to-household variations in electricity-temperature sensitivity and stationary point temperature, which can be thought of as the ambient temperature at which the AC is turned on) masked by using spatially aggregated data. 43 For future research, a more statistically representative dataset of electricity consumption records is needed for better quantifying and understanding the relationship between residential electricity consumption and cli- matic parameters, which is essential for investigating eective energy conservation, peak energy management, and climate adaptation strategies. 44 Chapter 3 A new method utilizing smart meter data for identifying the existence of air conditioning in residential homes This chapter re ects work published in Environmental Research Letters in 2019. [36] 3.1 Motivation Globally, the use of air conditioning (AC) is expected to increase signicantly over the coming decades, particularly across the developing world. Although 87% of US households currently have air conditioners (ACs), the US still led the world in new sales of ACs in terms of cooling output capacity (about 315 GW) in 2016 [76]. Worldwide, AC penetration is estimated to grow dramatically from 1.6 billion units currently to 5.6 billion units by mid-century [83], most of which will be driven by developing countries like China, India, and Indonesia. Furthermore, future increases in daily average ambient temperatures and more frequent extreme heat waves resulting from global climate change will increase the usage of current AC installations [9]. Thus, global AC growth will re ect both increases in adoption [13] and increases in the use of existing AC capacity. This growth can exacerbate warming at both global (i.e. from increased greenhouse gas emissions) and urban scales (i.e. from increased anthropogenic heat [162]), resulting in a positive feedback loop that accelerates the need for more cooling. Despite the importance and urgency of understanding future trends in AC growth, current research quantifying AC penetration with high spatial and temporal resolution is lacking. Estimating AC penetration rates with high spatial resolution would be particularly valuable for identifying communities that might be 45 vulnerable to extreme heat events due to the lack of AC or rising energy costs. Developing knowledge of these AC penetration patterns would also be useful for future grid planning, as it would enable the identication of peak energy hotspots, as well as areas that might be prone to large increases in AC installations in the future. Developing an understanding of the temporal changes in AC penetration rates is also important, particularly in areas where AC usage is currently low and its growth potential is high. Past studies estimate AC penetration rates at low spatial resolution due to a lack of available data. Most reported AC penetration rates come from appliance saturation surveys or residential energy consumption surveys carried out by federal and state governments, which are often time-intensive undertakings (e.g., multi- year eorts) [116] and constrained by budgets [164]. The spatial resolutions of existing datasets typically range from climate zones (e.g., in the case of California [100], which tends to have more available data than other regions of the US) to larger geographic regions (e.g., groups of multiple states [153]). The latest 2015 Residential Energy Consumption Survey (RECS) released by the US Energy Information Administration in 2018 showed that the Pacic region (comprised of Alaska, California, Hawaii, Oregon, and Washington) of the US had an overall AC penetration rate of 66% [45]. In California, the Advanced Residential Energy and Behavior Analysis Project conducted by the California Energy Commission (CEC) revealed that over 60% of all California homes had AC units [100]. More detailed data at the utility level were reported through the 2009 California Residential Appliance Saturation Study, which indicated that 75% of customers in the Southern California Edison utility territory had AC equipment in 2009 [116]. These reports have major shortcomings because of the long time between updates and their low spatial resolution. Other studies (e.g., [106] [105] [137]) have used economic and demographic parameters to model the diusion of residential appliances including AC, but these studies have only achieved results at the country-scale (i.e., they report one value for entire countries). New methods are needed to derive higher resolution estimates of AC penetration rates to facilitate better planning from an energy management perspective. Smart meter data have enabled a diversity of analyses in the building energy space. A subset of these analyses, which include peer-reviewed studies [48] [90] [57] [85] [2] [117] [56], [121] and open source platforms [22], [115], estimate cooling and/or heating loads from whole building energy consumption data. Other analyses have used smart meter data to calculate potential energy savings from demand response (DR) 46 programs [91] [43] [23] and disaggregate energy consumption according to end-use activity from whole- building data [52] [88] [168] [31]. (While methods have been established to disaggregate the electricity consumption signatures of individual appliances (e.g., an AC unit) [52] [88] [168] [31], they typically require data at the minute-level resolution or higher, which is higher resolution than is supported by most smart meter infrastructure [24]. Despite the growing body of literature utilizing these datasets, studies that use smart meter data for determining highly-resolved AC penetration rates are lacking. In this study we present a new method to compute AC penetration rates with high spatial resolution (i.e., the census tract level) utilizing the electricity consumption records of 180476 households in the Greater Los Angeles Area along with local site weather data. Our proposed method is capable of dierentiating households that currently use space cooling from those that do not. Trends in AC penetration are then analyzed in the context of regional variations in climate. Given that only 65% of homes in the western US currently have AC (compared to 95% in the South [45]), future increases in AC use in the US are likely to expand in the West, particularly in dense regions such as Los Angeles that are growing in population and will experience increasing temperatures [143] [158]. Moreover, Los Angeles is diverse in terms of its socioeconomic demographics, enabling us to observe dierences in AC penetration across disparate groups. Furthermore, the Greater Los Angeles Area is observed and projected to have uneven spatial distributions of warming due to urban heat islands and future climate change, respectively [157] [58] [143] [158] [97]. Hence, the Los Angeles region represents an important area to assess AC penetration rates at high spatiotemporal resolution. Such knowledge can help enable (a) tracking AC adoption over time and (b) projecting the spatiotemporal trends in future AC growth, which is important for managing peak energy demand, guiding electricity asset investments, and building energy eciency and demand-side management incentives. There is also an abundance of available smart-meter data in the Greater Los Angeles Area, enabling analysis with unprecedented spatiotemporal resolution. 47 3.2 Methods 3.2.1 Datasets We obtained hourly residential electricity data records from the Investor Owned Utility (IOU), Southern California Edison, for the years 2015 and 2016. The dataset contains more than 200000 randomly chosen households across the Greater Los Angeles Area. The sample size of 200000 was calculated to be representa- tive of the region's 4.5 million residential households at a 99% condence level. We screened out customers that had less than an entire year of electricity records. We also did not consider homes that were mostly uninhabited, determined as those whose annual electricity consumption was lower than 20 kWh, which is the amount of electricity an average home in California consumes per day [46]. Although no information about onsite generation installation (e.g., solar panels) was provided by the utility, a heuristic ltering approach was applied to remove homes with solar generation to avoid distorting electricitytemperature relationships. Customers who had no electricity consumption for at least one hour between 10:00 and 16:00 and positive electricity consumption between 17:00 and 23:00 for more than 36 d (5% time of the 2 year span) were considered to have solar generation. Although this method would be insucient to detect homes that had enough battery storage to oset night-time electricity generation, we expect these instances to be very small in number. (Note that zeros, rather than negative values for electricity use, were reported in the dataset for hours when self-generation exceeded consumption.) After all ltering was complete, 180 476 households were included for analysis in this study (see statistics of analyzed data records in Table B.1). Household street addresses were also provided to enable geospatial analysis. All data were stored and processed on USC's Center for High-Performance Computing (HPC) with a highly secure HPC Secure Data Account, allowing us to perform computations that met the strict data security requirements of the IOU. Two sources were used to retrieve daily average ambient near-surface air temperature (hereafter referred to as \daily average ambient temperature") data for the years under investigation (2015 and 2016): the Cal- ifornia Irrigation Management Information System (CIMIS) [27] and the National Oceanic and Atmospheric Administration (NOAA)'s National Centers for Environmental Information (NCEI) [109]. Both networks are made up of land-based, automated, quality-controlled weather stations that cover most population centers 48 in Southern California. This study utilizes 36 CIMIS stations and 43 NCEI stations, which were selected from all available stations by choosing the nearest weather monitors to each household with electricity data records. Census tract boundary shapeles were acquired from the US Census Bureau website [148]. Building climate zone boundaries established by CEC, based on building energy consumption characteristics, were also retrieved to characterize climate in the investigated region [29]. Sample records that fell within each census tract were counted and compared against the sample sizes needed to statistically represent the population of that census tract. Detailed methods can be found in the supporting information (available online at stacks.iop.org/ERL/14/094004/mmedia). 3.2.2 Statistical model To describe the nonlinear relationship between residential electricity consumption and ambient temperature, we implemented the segmented linear regression model described in our previous study [34]. In this model, a stationary point is located by iteration to achieve the overall best two-piece linear t to the dataset. Our previous study [34] shows that using daily accumulated electricity consumption data and daily mean temperature yields the best model performance. Hence, we aggregated hourly electricity data records to daily accumulated electricity consumption (kWh day 1 ) for analysis. Three examples of this segmented model showing daily electricity consumption versus daily average ambient temperature for households in the Greater Los Angeles Area are illustrated in Figure 3.1. Two physically-relevant metrics can be retrieved from the segmented linear regression: (1) the station- ary point temperature (SPT ), which represents the stationary point identied by the regression model, corresponding to the threshold ambient temperature beyond which households are likely to use AC (see distribution of SPT in Figure B.3, and (2) electricitytemperature sensitivity (ET sensitivity), the slope of the least squares linear regression for temperature values SPT (Figure 3.1(a)). Physically, this metric represents the change in electricity consumption for a household corresponding to a unit increase in ambient temperature. 49 a b c Figure 3.1 Daily cumulative electricity consumption versus daily average ambient tempera- ture for three example homes in Southern California. Stationary point temperature (SPT), electricity-temperature sensitivity (i.e., the slope of the least squares best t for temperature values SPT, labeled as \slope right"), and the slope for temperature values < SPT (labeled as \slope left") are identied in panel (a). Penel (a) illustrates a home identied by our method as having AC, whereas (b) and (c) are identied as not having AC. To determine whether a household has AC, our method compares the slopes of least squares regressions for temperature values falling below SPT (referred to as slope left versus that for temperature values greater than or equal to SPT (referred to as slope right) (see Figure 3.1(a)). We dene households with AC as those that meet both of the following criteria: (1) slope right greater than zero, and (2) the sum of slope left and slope right greater than zero. Electricity consumption versus temperature for a typical household with AC is shown in Figure 3.1(a). In this case, electricity consumption increases when the daily average ambient temperature is greater than or equal toSPT , presumably due to increased cooling demand, thus meeting criteria (1). Criteria (2) is set mainly to rule out households that have negligible positive slope right caused by noise or other appliances that might consume more electricity on hot days like refrigerators [129]; it removes the need to set an arbi- trary non-zero cut-o value for slope right to account for a possible positive but negligible slope right. According to data from the 2015 Residential Energy Consumption Survey, space heating in Southern Cal- ifornia is mainly supported by natural gas, and thus, would not impact analysis of electricitytemperature sensitivity [152]; meanwhile, households in the investigated region generally consume more electricity for space cooling than heating [28]. Hence, slope left is typically near-zero, with an absolute value much 50 less than slope right. (See supporting information Figure B.2 showing slope right versus slope left values). Households that do not meet the two criteria above are deemed as having no AC. Electricity consumption versus temperature for two example households identied as having no AC are shown in Figure 3.1(b) and (c). The household in Figure 3.1(b) does not fulll the condition slope left + slope right> 0, while that shown in Figure 3.1(c) does not fulll the condition slope right> 0. Visually we see no signicant energy use increase, even on hot days when daily average ambient temperature exceeds 30°C (86°F), suggesting no AC usage. It is important to note that these criteria may not be sucient to identify AC penetration in all regions throughout the US and rest of the world. Applicability to other locations is discussed in the discussion section. For more discussion about shortcomings of this method, please see the Supporting Information section B.2. We applied this classication framework to partition the full dataset (180,476 households) into two categories, i.e., households with and without AC, for the two years investigated. We then computed AC penetration rates, discussed below in Section 3.3.1, at the census tract level by dividing the number of households identied as having ACs by the number of households in our dataset falling within each respective census tract. 3.3 Results and discussion 3.3.1 AC penetration rates One of the major advantages of the method proposed in this study is the ability to quantify AC penetration rates at high spatial resolution. Figure 3.2 displays the spatial distribution of AC penetration rates in the Greater Los Angeles Area at the census tract level. Since our source data was provided by Southern California Edison, penetration rates are not shown for regions served by other utilities (e.g., Los Angeles Department of Water and Power). Census tracts containing too few records to statistically represent the population are designated with cross-hatching. (Note that the entire dataset is representative of the overall population at a 51 99% Condence level, while the data shown at the census tract level in Figure 3.2 are representative at the 90% Condence level, except where noted with cross-hatching.) Since AC penetration rates are calculated using household level data, AC penetration rates can be esti- mated at any spatial resolution so long as the sample size per aggregation area is statistically representative of the region. For reasons of privacy protection, data were aggregated to the census tract level. (Census tracts, established by the U.S Census Bureau, are small and relatively stable geographical units in terms of area and population [147].) To the authors' knowledge, this is the rst study to calculate AC penetration rates with such high spatial resolution. The overall AC penetration rate in the investigated area is computed as 69%. Figure 3.3 shows proba- bility density distributions of electricitytemperature sensitivities for all households included in our dataset (Figure 3.3(a)) and separately for homes classied with and without AC (Figure 3.3(b)). The mean (median) electricitytemperature sensitivity for all households is 0.068 (0.054) kW °C 1 (Figure 3.3(a)). 52 Ai r Condi t i oner Penet r at i on Rat e Summer mean t emper at ur e: Cl i mat e Zone 6: 21° C Cl i mat e Zone 8: 22° C Cl i mat e Zone 9: 23° C Cl i mat e Zone 10: 24° C Cl i mat e Zone 14: 28° C Cl i mat e Zone 15: 34° C Cl i mat e Zone 16: 21° C 6 9 1 6 1 4 1 6 1 5 1 0 8 Figure 3.2 A choropleth map of AC penetration rates (i.e., the ratio of total number of homes identied with AC versus the total number of homes per aggregation area in our dataset) for all census tracts in Southern California that are powered by Southern California Edison. Generally, coastal and mountainous areas tend to have lower AC penetration rates relative to inland and desert areas. The dark grey boundaries indicate building climate zones, with average summer temperatures per climate zone indicated at the bottom of the gure. White indicates that no data were available. Cross-hatched census tracts do not have sucient records to statistically represent the population with a 90% condence level and margin of error of 10%. These tracts account for less than 30% of analyzed census tracts. 53 a b Figure 3.3 Probability density distributions of E-T sensitivities for (a) all 180,476 homes in- vestigated in this study, and (b) shown separately for homes identied as having AC (in light blue) and those identied as not having AC (in red). In (a), the red dashed line indicates the mean E-T sensitivity value (0.068 kW °C 1 ) for all homes, while the blue dashed line indicates the median (0.054 kW °C 1 ). Though there is some overlap between the probability density distributions for homes identied with and without AC, we see that most households with near-zero and negative sensitivities are grouped into those without AC, as we would expect. On the contrary, households with AC have a wider range of ET sensitivities. One of the advantages of this classication method for determining whether or not a household has an AC unit is that there is no need to dene an arbitrary cut-o based on electricitytemperature sensitivity. On the other hand, the lack of a \cut-o" also results in the region overlap in ET sensitivities between the two groups in Figure 3.3(b). Generally, AC penetration rates are higher in hotter climate zones (note the average summertime tem- peratures per climate zone indicated in Figure 3.2). Along the coast, low AC penetration rates are observed, with the exception of a few census tracts that are generally in more wealthy regions. Compared to other lo- cations within the domain, coastal areas, mainly included in Climate Zone 6, have the lowest summer mean temperature, resulting in a relatively low demand for cooling. Climate Zone 16, an inland, mountainous region, has lower AC penetration rates relative to surrounding areas on the east side of the basin, likely due 54 to its relatively low summer mean temperature. Also, Climate Zone 16 includes Big Bear Lake, a resort area utilized for skiing and winter vacationing, leading to a portion of seasonal residents who may not install ACs. Future work can explore other causes of spatiotemporal variation in AC penetration rates, such as socioeconomic status and building characteristics. 3.3.2 Comparison of computed AC penetration rates to other studies Here we compare aggregated AC penetration rates presented in this investigation to existing survey data for the studied region (Table 3.1). The overall calculated AC penetration rate for all 180 476 households (69%) is slightly less than the California Residential Appliance Saturation Study value for the SCE territory (75%). It should be noted that our classication method might not identify evaporative cooling devices as they consume much less energy than central or room ACs [101], and therefore, are unlikely to cause a signicant increase in electricity consumption per unit temperature increase when used. If we consider that the California Residential Appliance Saturation Study estimates that 6% of cooling systems in Southern California are evaporative, our results are consistent with its computed penetration rate. 55 Table 3.1 Comparison of AC penetration rates acquired from this study and publicly available survey data Source AC penetration rate AC types included Investigated year Investigated area Citation This study 68% All types (other than possibly evaporative cooling devices) 2016 Southern California Edison territory Residential Energy Consumption Survey (RECS) 66% All types 2015 Pacic region of US (Residential Energy Con- sumption Survey 2015)[45] Advanced Residential Energy and Behavior Analysis Project >60% All types 2017 California (Lutzenhiser et al 2016)[100] California Residential Appliance Saturation Study 75% All types (including 6% evaporative coolers) 2009 Southern California Edison territory (Palmgren et al 2010)[116] Borgeson PhD Dissertation 60% or 68%-70% a All types 2008-2011 Pacic Gas and Electric territory (Borgeson 2013)[24] A more recent report, the Advanced Residential Energy and Behavior Analysis Project, released by California Energy Commission, suggests that California's overall AC penetration rate is > 60%, which is consistent with our value though is representative of a larger region. In his PhD dissertation, Borgeson (2013) calculated the AC penetration rate for Pacic Gas and Electric's territory in 2008 through 2011, which covers large portions of Northern and Central California, as 60% or 68%70%, depending on which of two dierent methods was employed [24]. 3.3.3 Applicability to other regions We believe that this method can be applied to other regions around the world where household level smart meter data are available. However, some modications to the method might be required in some cases, depending on the region under investigation. Three factors should be taken into consideration to determine a Value varies by method employed. 56 whether modifying the method is needed: climate characteristics (energy demand for space cooling and heating), the typical fuel(s) utilized for space cooling and heating, and most common AC and heating tech- nologies utilized. These three factors together determine the quantitative relationships between household electricity consumption and ambient temperature. Although US regions have diverse space cooling and heating needs [141] as well as fuel choices [152], we expect the methods dened in this study to be applicable for much of the country under current climatic and fuel usage trends. For example, US regions with cold winters, including the east coast, the mid-west, and the west, typically use natural gas as the dominant space heating fuel [152]. Thus, we would not expect a strong electricitytemperature sensitivity on days when temperatures are cold enough to require heating, meeting the criteria to use the method dened in this study. In the Southern US, which typically has a hot- humid climate, electricity is the main space heating energy source during their mild and short winters [152]. Considering the intense use of AC in the summertime in hot-humid regions, even in this case slope right would be anticipated to be larger than the absolute value of slope left; thus, in these regions our method would likely be viable. This reasoning is supported by a study in Shanghai [98], which has a comparable hot-humid climate, signicant use of electric heating, and similar AC penetration rate (near-100%) to the Southern US [153], [98], [142]. In this study, they nd that cooling drives residential electricity usage when compared to heating, suggesting that our method could be eective even in warm regions where electric heating dominates. The U.S. is expected to electrify over time [40](Deason et al 2018). The 2015 RECS observes more electric heating in the residential sector than 2009 RECS [154](Energy Information Agency 2018). The methods presented here would likely need to be modied, for example, in cases where the change in energy usage per change in temperature during electric heating (i.e., slope left) would exceed the rate of change during electric cooling (i.e., slope right). Our method might also fail in regions with year-round hot climates, such as those near the equator, as it might be dicult to identify a stationary point temperature in relationships between household electricity consumption and ambient temperature in cases where cooling is required all year. These households would likely have a linear positive slope in electricity-temperature plots without an identiable stationary point. 57 Lastly, AC type should be checked, particularly in regions with high usage of evaporative cooling, since this cooling type requires less energy and is therefore more dicult to identify from electricity use records. Consequently, this method would likely be less eective in identifying evaporative cooling through the use of electricitytemperature plots. Above all, such methods cannot be applied without an abundance of household level smart meter data, as well as reliable, high-resolution weather data. To conclude, a new classication method to identify whether homes have AC at the household level was developed and presented here. The method relies on having access to household daily aggregated electricity and daily average ambient temperature data. The methods were applied to estimate AC penetration rates at the Census tract level in Southern California (within the Southern California Edison service territory). The mean penetration rate computed for the service territory was also compared to estimates from previous studies. Using our study as a baseline for AC penetration, repeating the study in future years could allow for quantifying trends in AC penetration over time. 58 Chapter 4 Utilizing smart-meter data to project impacts of urban warming on residential electricity use for vulnerable populations in Southern California This chapter re ects work published in Environmental Research Letters in 2020. [35] 4.1 Motivation Today over half of the global population lives in urban areas, and by 2050, this fraction is expected to grow to nearly 70% [146]. Accordingly, increases in urban warming, in uenced by urbanization, population densication, and the local impacts of global climate change, represent key drivers for future changes in energy usage in the United States (US) due to increased cooling needs [5], [77], [136]. This warming is likely to have disproportionate eects on poor communities that might be sensitive to increasing electricity bills or lack access to air conditioning (AC) altogether [60] [26]. As heat currently kills more people each year in the US than storms, oods, and lightning combined [110], identifying those communities most vulnerable to rising residential energy costs is critical [79]. Similarly, while exploding global demand for cooling is expected to bring AC to billions of people in the coming decades [38], [83], the spatio-temporal distributions of these future energy needs for space cooling in the US and abroad are not well understood, and analytical techniques to anticipate these trends with available data are limited. 59 The residential sector is the biggest electricity consumer in the US, accounting for 37% of total US electricity consumption [155]. Space cooling accounts for 15% of electricity consumption within US homes, ranking as the highest single energy-consuming end use activity in the residential sector [154]. Cities are currently warming due to the local impacts of global climate change and the urbanization-induced inten- sication of the urban heat island eect. Urban warming is expected to increase residential sector energy usage in the US due to growing cooling and comfort needs [5] [136], [78], both in terms of increased AC usage in homes with existing AC and new AC installations in homes without existing AC. Intensifying AC usage will lead to additional greenhouse gas emissions, and its impact on energy demand is expected to pose challenges to current electricity infrastructure and peak electricity management [3], [9], [94], [102]. In order to anticipate challenges associated with future cooling needs, it is important to quantify how urban warming is expected to increase cooling energy use with high spatial delity. However, this task is dicult given spatially heterogeneous distributions in future warming [87] and building characteristics across the built environment [136], both of which aect the sensitivity of cooling energy use to temperature increases. Developing a quantitative understanding of how residential electricity use by disparate populations will be impacted in the future by factors such as climate change and urban heat islands requires high resolution energy and climate data to (a) identify the geospatial distribution of warming and (b) quantify how electricity consumption changes according to those temperature and climatic variations [34]. To the authors' knowledge, few empirical studies have been performed to understand the functional relationships between electricity usage and ambient temperature at ne spatiotemporal scales. This research gap is largely due to a lack of publicly available, high-resolution (e.g., household level and hourly) residential electricity data [34]. However, the recent availability of highly resolved electricity use data, collected by smart meters at the household level, enable such analysis. Additionally, few studies have directly investigated the correlation between socioeconomic factors and how a household's residential electricity consumption varies as a function of ambient temperature (hereafter referred to as the `electricity-temperature relationship'). Based on these knowledge gaps, this study addresses the following research questions: 1. How does residential electricity consumption across Southern California respond to changes in ambient temperature? 60 2. To what extent is residential electricity behavior in uenced by spatial variations in baseline climate and socioeconomic factors? 3. Can we better understand vulnerability to future increases in temperature by considering geospatial distributions of poverty level, AC penetration rates (i.e., the fraction of homes with AC), and extreme heat events? To answer these questions, we utilize two years (i.e., 20152016) of hourly electricity consumption records for 180476 households in Southern California, as well as local site weather data, to compute household- level relationships between electricity consumption and ambient temperature, using the segmented linear regression model detailed in our previous study [34]. We also assess the prevalence of AC usage across the investigated area at the household level, in the context of climate characteristics, as well as socio-economic demographics, using a methodology presented in our previous study [36]. In this study, the investigated area within Southern California is dened as the territory serviced by Southern California Edison, which includes Los Angeles County, Orange County, Riverside County, and San Bernardino County. (However, the City of Los Angeles, within Los Angeles County, is excluded since it is serviced by Los Angeles Department of Water and Power.) Southern California was selected as a study region because it has widely varying microclimates across relatively small spatial extents, spanning coastal, mountainous, and desert climates. It is also comprised of densely urban through sparsely populated rural communities. Southern California is projected to have uneven spatial distributions of warming due to the combined signatures of urban heat islands and climate change [157] [58]. Moreover, Southern California has a diverse housing stock ranging from tiny apartments to huge villas, re ecting wide variations in socioeconomic status and housing preferences. This analysis was enabled by the abundance of Advanced Metering Infrastructure smart-meter data across Southern California, which facilitated results to be generated with unprecedented spatio-temporal resolution. In this study, we characterize the relationship between residential electricity consumption and ambient temperature using two indicators, including: 61 (1) Electricity-temperature sensitivity (referred to as `ET sensitivity', in kW °C 1 ), dened as the change in instantaneous electricity consumption of a household corresponding to a unit change in ambient temperature, i.e., one degree Celsius. (2) Stationary point temperature (referred to as `SPT ', in°C), dened as the ambient temperature thresh- old beyond which households are likely to turn on AC if equipped. For example, a household having an electricity-temperature sensitivity of 0.25 kW°C 1 and a stationary point temperature (SPT ) of 20°C indicates that the household's electricity consumption is observed to increase at a rate of 0.25 kW per degree C of temperature increase (or 6 kWh °C 1 day 1 ) on days with daily mean ambient temperatures higher than 20°C. 4.2 Methods Electricity dataset. Hourly residential electricity data records for more than 200000 randomly chosen households in Southern California were acquired. The spatial distribution of households at the census tract level can be found in Appendix C.4 in supplementary information (available online at stacks.iop.org/ERL/15/064001/mmedia). These data were provided by the Investor Owned Utility (IOU), Southern California Edison, for the year 2015 and 2016. The sample size of 200 000 was calculated such that it would be representative of the region's 4.5 million residential households (all equipped with smart meters) at a 99% condence level using Equation (4.1), allowing for additional degrees of freedom for analysis n = Z 2 p(1p) e 2 N (N 1) + Z 2 p(1p) e 2 (4.1) In Equation (4.1), n is the required sample size; Z is the Z-score, determined as 2.576 by the condence level chosen (99%); p is the expected prevalence, chosen as 50%; and e is the margin of error, set as 0.5% in this study;N is the total number of residential customers in the studied area (4.5 million). (See Appendix C.6 in supplementary information for detail on how we determined these parameters.) To fully capture seasonal variations in the relationship between electricity consumption and ambient temperature, we screened out customers with less than 365 days of electricity records. The data provider did not give information about 62 onsite generation installations (e.g., solar panels). In addition zero values, rather than negative values, were recorded in the dataset when generation exceeded consumption, further complicating the identication of these self-generation. Accordingly, we used a heuristic ltering approach to remove homes with potential solar generation since their electricity-temperature relationships were likely to be distorted. Homes that had at least one hour of zero consumption between 10:00 and 16:00 and positive consumption between 17:00 and 23:00 for more than 36 days (i.e., 5% of the 2-year period of study) were deemed to have solar generation. Once all ltering was complete, 180 476 households were analyzed from the original dataset. Data included hourly household-level electricity consumption and street address. Storage and processing of all data were performed using a highly secure HPC Secure Data Account (HSDA) on USC's Center for High-Performance Computing (HPC). This was to fulll the data security and privacy protection requirements of the IOU. Site weather dataset. Two weather data sources were used to retrieve daily ambient near-surface air temperatures for 2015 and 2016: the California Irrigation Management Information System (CIMIS) [27] and National Oceanic and Atmospheric Administration (NOAA)'s National Centers for Environmental In- formation (NCEI) [109]. The two networks are comprised of land-based weather stations that are automated, quality-controlled, and covering most population centers in the investigated area. We utilized 36 CIMIS and 43 NCEI stations in total, which were selected by measuring the shortest distance between a weather station and each household with an electricity data record. Figure C.2 shows the number of households per census tract and location of weather stations utilized in this dataset. Model. This study applies a model that has been utilized in our previous study [34]. Brie y, to describe the nonlinear relationship between residential electricity consumption and ambient temperature, a segmented linear regression model is used [108]. In this model, a stationary point is located by iteratively calculating to achieve the overall best two-piece linear t to the data points. Two variables can be retrieved from such segmented linear regression: (1) the SPT , which is the temperature corresponding to the `stationary point' captured by the regression model, and physically represents the threshold ambient temperature beyond which a household starts to use an air conditioner; (2) E-T sensitivity, the slope of the linear segment to the right ofSPT , which represents the change in electricity consumption of a household corresponding to a unit increase in ambient temperature, i.e., one degree Celsius. The segmented linear regression was chosen in this 63 study because space cooling is mainly supported by electricity while the majority of households in California utilize natural gas as a heating energy source [46]. Hence, there is not a signicant rise in electricity use as temperature drops below the SPT . Applicability and limitations of the segmented linear regression model are discussed in our previous study [36]. In this study, we chose daily accumulated electricity consumption (in kWh) and daily average temperature at the household level as the indicators of electricity consumption and weather to be regressed against each other (i.e. for computing ET sensitivity). This is motivated by the ndings of our previous study [34], which indicates that for the same household or region, regression between daily accumulated electricity consumption and daily average temperature shows the highest coecient of determination (r 2 ) values compared to other indicators (e.g., hourly electricity consumption, daily maximum/minimum temperature). Note that in this study we also use other weather indicators (e.g., daily maximum temperature, diurnal temperature range) in multivariable regression analysis while investigating relationships between E-T sensitivity and baseline climate. To answer research question 2, we developed maps to visualize the spatial distribution of E-T sensitivities andSPTs in Southern California (see Figure 4.1(a), (b)). For the purposes of protecting the privacy of data at the household level, we calculated the mean value of sensitivities and SPTs at the census tract level to create the resulting choropleth maps. We acquired census tract boundary shapeles from the US Census Bureau [148]. For characterizing climate zones in the investigated region, we used boundaries established by California Energy Commission (CEC) [29]. Socioeconomic data were retrieved from CalEnviroScreen 3.0, which is a mapping tool provided by the Oce of Environmental Health Hazard Assessment (OEHHA) within California's Environmental Protection Agency (CalEPA) to help identify vulnerable communities most at risk of being exposed to pollution [113]. These data were resolved at the census tract level. CalEnviroScreen reports multiple socioeconomic metrics. We used the poverty percentile index, dened as the percentile rank of each census tract in California according to the percent of population living below two times the federal poverty level within each tract, to represent auence level. Thus, each household is assigned a poverty percentile index corresponding to its census tract, regardless of its individual household-level income. 64 Identifying households with AC. We utilized a method developed and detailed in our previous study [36] to identify whether a household has AC using household electricity and ambient temperature data. Air conditioner penetration rates at the census tract level were computed as the ratio of homes with AC to total homes in our dataset. We determined whether a home has AC by comparing the slopes to the left and right of the SPT . For simplicity, hereafter we refer slope left as the slope of the linear segment to the left of the SPT , and slope right as the slope to the right of the stationary point. We treat households with (1) slope right > 0 and slope left + slope right; > 0 as households that have air conditioners. In California, space heating is mainly supported by natural gas [100]. Hence slope left is typically near-zero and has an absolute value much less than slope right. Using this method, we calculated the overall AC penetration rate of Southern California as 69%, which matches well with previously reported survey-based data in the same or similar geographical area (60% 75%) [116] [24] [45], [100]. Projections of future climate change. Historical observations and projections of the number of extreme heat days per year in Southern California were derived from data provided by Cal-Adapt, a web- based tool funded and managed by California Energy Commission [30]. An extreme heat day was dened to have a daily maximum temperature of over 35°C (95°F). Historical observations were from a gridded dataset at a spatial resolution of 1/16°derived from land-based daily temperature observations from about 20 000 NOAA Cooperative Observer (COOP) stations. In this study, we calculated the mean value of the annual number of extreme heat days from 1961 through 1990 to represent the historical average of observed events for each census tract analyzed. Projections of future extreme heat days were calculated based on data statistically downscaled using the LOCA technique with four climate models (HadGEM2-ES (warmer/drier), CNRM-CM5 (cooler/wetter), CanESM2 (average), and MIROC5 (complement)) at a 1/16°spatial resolution and daily temporal resolution. In this study, for each census tract analyzed we take the mean value of the yearly number of extreme heat days across the four climate models under the RCP8.5 scenario from 2070 through 2099 to project extreme heat events at the end of century. We choose RCP8.5 because it is the RCP scenario with the largest temperature increase to bound the analysis. 65 4.3 Results and discussion 4.3.1 Spatial trends in the electricity use response to increases in ambient temperature Figure 4.1(a) illustrates spatial variability in ET sensitivities across Southern California. The mean and median value of ET sensitivity is 0.068 and 0.054 kW°C 1 , respectively. Details about ET sensitivities and SPTs can be found in Appendix C.1 in supplementary information. In general, ET sensitivities are larger for census tracts within climate zones with higher summer mean temperatures. Similarly, the electricity consumption of homes in coastal census tracts is less sensitive to ambient temperature change relative to those in inland areas. The meanSPT for all households in this study is 18.8°C 3.7°C (65.8 6.7°F), which is similar to the widely used base temperature 18°C (65°F) for calculating Cooling/Heating Degree Days. The majority of households haveSPTs around 18°C (65°F). There are some areas, mainly located in Climate Zone 16, showing abnormally lowSPT values. We discuss these results in further detail in Appendix C.1 and Appendix C.3 in supplementary information. Climate characteristics across Southern California are captured by California Building Climate Zones (hereafter referred to as Climate Zones) established by California Energy Commission [29]. Homes contained within Climate Zone 6 (i.e., the only coastal zone) have a mean ET sensitivity of 0.04 kW°C 1 , which falls below the average sensitivities observed for inland climate zones (0.06 to 0.11 kW °C 1 ). This coastal climate zone also has the lowest summer mean temperature (20°C), resulting in a low demand for cooling compared to other places in the investigated region. Accordingly, the average AC penetration rate computed for the coastal census tracts within Climate Zone 6 is 46%, which is lower than other climate zones (59% 80%) (see Figure 4.1(c)). Within inland regions, hotter climate zones (i.e., with higher summer mean temperature) generally have larger mean ET sensitivities, compared to other regions. Quantitatively, Climate Zones 8, 9, 10, 14, and 15 have summer mean temperatures of 22°C, 23°C, 25°C, 28°C, and 32°C, and mean ET sensitivities of 0.06, 0.08, 0.07, 0.08, and 0.11 kW °C 1 , respectively. This trend can be partially explained by higher AC penetration rates in hotter areas since the aforementioned average ET sensitivities include all households (i.e., those with and without AC). For example, Climate Zone 8 has a mean AC penetration rate of 64%, 66 while low desert-like Climate Zone 15 has a mean AC penetration rate of 80%. We hypothesize that another contributing factor is that inland regions generally contain larger homes, compared to areas closer to coast, which would require more energy for cooling. 67 a b Cl i mat e Zone Summer Mean Temp: 6: 20° C 8: 22° C 9: 23° C 10: 25° C 14: 28° C 15: 32° C 16: 24° C d c Figure 4.1 Maps showing (a) average electricity-temperature sensitivity values per census tract in Southern California (including all households in the dataset), (b) average stationary point temperatures per census tract (including only households identied as having AC), (c) AC penetration rates per census tract, and (d) poverty indices per census tract. In (b) the diverging color scheme is centered around 18°C, which is commonly used as the reference for Cooling Degree Days. Dark gray outlines indicate boundaries of building climate zones. Average summer temperatures are shown for each climate zone in (b). In (d), 0%10% refers to the top 10% auence level within the state of California, while 90%100% refers to the poorest 10%. White indicates that no data are available. 68 4.3.2 Impact of auence on the electricity use response to increases in ambient temperature With the exception of Climate Zone 16 (i.e., mountainous areas), households in more auent census tracts generally have higher ET sensitivities than those in less auent census tracts (see Figure 4.1(a) and (d)), indicating that electricity usage for more auent households is typically more sensitive to ambient tempera- ture changes. This trend is conrmed in Figure 4.2, which presents box-and-whisker plots of household-level ET sensitivity, for all households identied as having AC, dierentiated by census tract poverty percentile and Climate Zone. AC penetration rates also re ect this trend, as illustrated in Figure 4.3, which shows that more auent census tracts have higher AC penetration rates in almost all climate zones, except for the coastal and mountainous climate zones, i.e., Climate Zone 6 and 16, respectively. (Coastal regions had the lowest overall AC penetration rates in the investigated region, due to the relatively mild climate. More discussion regarding the mountainous regions contained within Climate Zone 16, which contains relatively large stocks of seasonal vacation homes, can be found in Appendix C.3 of supplementary information.) Although gure 2 only includes homes identied with ACs, we still observe a decline in ET sensitivity as fractions of poverty increase within climate zones, indicating that auence level aects ET sensitivities, even across homes with space cooling. Thus, auent populations tend to use AC more intensely, on average, presumably due to: (1) more luxurious lifestyle; and/or (2) larger house sizes. For example, the auent census populations of Malibu, Rancho Palos Verdes, and Laguna Beach, are comprised of larger homes and their energy use is more sensitive to changes in temperature compared to other communities along the coast. (Note: since census tracts are dened to represent similar population (from 1200 to 8000, with an average of 4000), larger land areas indicate less population density and larger homes. However, no data about lifestyle and home sizes were utilized in this study and should be considered in future research.) 4.3.3 Response of electricity use to increases in ambient temperature in Southern California: combined eects of climate and auence levels In this section, we investigate the combined eects of climate and auence levels on ET sensitivities and AC penetration. Hotter regions tend to have larger ET sensitivities than cooler regions, on average, while 69 Summer mean t emp: Cl i mat e Zone 6: 20° C Cl i mat e Zone 8: 22° C Cl i mat e Zone 9: 23° C Cl i mat e Zone 10: 25° C Cl i mat e Zone 14: 28° C Cl i mat e Zone 15: 32° C Cl i mat e Zone 15: 32° C Cl i mat e Zone 16: 24° C poor r i ch Figure 4.2 Box-and-whisker plots of household level electricity-temperature sensitivities for households identied with AC, grouped by climate zone. In each subplot, the distribution of electricity-temperature sensitivities is displayed across 10 bins, each representing a dierent poverty classication. Integer values inside each box represents the number of households per bin. See methods for details on the calculation of poverty percentile index. Each home plotted in the gure is assigned the poverty percentile index for its respective census tract as reported by CalEnviroScreen3.0. 70 T=20°C n=29760 T=22°C n=49117 T=23°C n=35971 T=25°C n=40289 T=28°C n=12660 T=32°C n=6090 T=24°C n=3582 0% 25% 50% 75% 6 8 9 10 14 15 16 Climate Zones Air Conditioning Penetration Rate Poverty Percentile Index 10% (rich) 20% 30% 40% 50% 60% 70% 80% 90% 100% (poor) Figure 4.3 Bar chart of air conditioning penetration rate versus poverty percentile index, grouped by climate zone, in Southern California. AC penetration rate is computed as the number of households identied as having AC divided by the total number of households in our dataset. The total number of households and summer mean temperature per climate zone are shown above each group of bars. 71 richer communities tend to have larger ET sensitivities than poorer populations within similar climate zones. The dierences in ET sensitivities and AC penetration rates between the most auent and least auent communities are generally smaller in hotter climate zones. For example, in the hottest Climate Zone 15, dierences observed among socioeconomic groups are smaller than other climate zones. Since cooling demand is higher in hotter climate zones, there appears to be a threshold above which AC is used similarly across populations for comfort. We conducted single and multiple variable linear regressions at the census tract level among a variety of variables to assess the relative importance of climate and socioeconomic factors on ET sensitivities and AC penetration rates. Explanatory variables associated with larger coecients of determination (r 2 ) perform better in explaining variance in response variables. The values of slopes cannot be directly compared to assess the relative importance across variables because they have dierent units, so we compare the sign of the slopes to indicate the direction of correlation between the response and explanatory variables. Statistics are presented in Table C.1 in supplementary information. Daily mean temperature, daily minimum temperature, daily maximum temperature, and diurnal tem- perature range (i.e., the dierence between daily maximum and minimum temperature) averaged over sum- mertime (July, August, September) were selected to represent the climate characteristics of each census tract in the regression analysis. Among all climatic variables, the diurnal temperature range best explained the variations in single variable regression in both census tract mean ET sensitivity (r 2 = 0.18) and AC penetration rate (r 2 = 0.42), both with positive slopes. In other words, census tracts located in places with large diurnal temperature ranges tend to have higher AC penetration rates and ET sensitivities. In South- ern California, inland regions typically had larger diurnal temperature ranges and daily mean temperatures compared to coastal areas. The positive slopes of ET sensitivity versus diurnal temperature range and daily mean temperature are consistent with observations presented in Figure 4.1(a) and Figure 4.2, illustrating that electricity use for homes in hotter (inland) climate zones tends to be more sensitive to temperature than cooler (coastal) climate zones. Single variable regressions reveal that, compared to climate variables, auence level was stronger (weaker) in explaining ET sensitivities (AC penetration rates). Auence level alone can explain 12% of variations in 72 ET sensitivities while it can only explain 6% of that in AC penetration rates. Climate variables, particularly diurnal temperature range, can explain 18% of variations in ET sensitivities but 42% of that in AC pene- tration rates. Multivariable regressions show that variability in summertime daily maximum temperature, auence level, and summertime diurnal temperature range together explain 60% of the variability in AC penetration rates but only 38% of that in ET sensitivities. When summertime average diurnal temperature range and auence level are used in multivariable regressions, 58% of the variability in AC penetration rates can be explained, compared to 42% when only summertime diurnal temperature is used and 6% when only auence level is used. This result indicates that a community's AC penetration rate is associated with both its climate characteristics and auence level. Other factors likely aect variability in ET sensitivities such as occupant behavior, home size, and building characteristics, but these factors were not considered in this study due to data limitations. Future research should focus on these variables. 4.3.4 Locating the most temperature-sensitive hotspots and most heat-vulnerable communities Although the entire region is likely to experience the impacts of increasing extreme heat events, there are two categories of hotspots that will be particularly important to watch for future grid planning. The rst category includes census tracts across the studied region that currently have relatively low AC penetration rates, and thus their overall electricity consumption is not currently very sensitive to changes in temperature. Many places in this category are located in coastal (Climate Zone 6) and mountainous areas (Climate Zone 16). The majority of census tracts in these two climate zones are not in extreme poverty, so as temperatures warm, populations are likely to install AC units and/or intensify their usage (see Figure 4.1(d)). Given projections of anticipated future warming, these might be the regions that experience the largest magnitude of change in residential cooling energy use due to the need for more cooling. Thus, from a grid management perspective, these are the regions that are critical to identify in order to ensure adequate energy services in the future. The second category includes census tracts with populations that currently have high ET sensitivity values because their electricity consumption could increase more than others given the same temperature rise. (These census tracts are shown in Figure 4.1(a) as census tracts with the highest ET sensitivities, >0.25 73 kW °C 1 .) Some of these census tracts are located in low desert, generally within rich communities (e.g. Palm Springs, Temecula), while others are simply very auent communities with large houses (e.g. Beverly Hills). The high ET sensitivities in these places are driven by hot climate, large housing square footages, and/or lifestyle factors (e.g. lack of energy saving behaviors), all of which require more cooling loads. It is important to note that these are regions that currently have high AC loads, so while these loads might intensify, larger growth in cooling-related energy usage might occur in areas with lower AC penetration rates today. However, these highly ET sensitive places can potentially serve as reference hotspots for future grid planning as well as targets for heat mitigation plans. 4.3.5 Identifying communities that may be most vulnerable to ambient temperature increases Using the aforementioned insights, census tract-level AC penetration estimates along with socioeconomic data were used to identify communities that might be most vulnerable to future warming and increases in extreme heat events. Vulnerable census tracts were dened as those that currently have both low AC penetration rates and high poverty rates (i.e., are more sensitive to future energy cost increases); more specically, vulnerable census tracts were dened as those falling into the top 10% of the studied region when the (1) percentage of population living in poverty and (2) percentage of households not having AC were summed for each respective census tract and ranked across all tracts in the region. A full list of identied top 10% most vulnerable census tracts can be found in Appendix C.7 in supplementary information. Some of the census tracts identied as potentially vulnerable using these criteria are already experiencing large numbers of extreme heat days (e.g. >16 d per year), as illustrated in Figure 4.4(a). However, by the end of the century, 100% of the census tracts identied as most vulnerable to future extreme heat events are expected to have more than four extreme heat days per year based on the latest climate change projections for California [145]. Moreover, 80%, 55%, and 30% of these potentially vulnerable communities are expected to experience over 8, 16, and 32 extreme heat days per year, respectively. These extreme events can pose dire health-related impacts on those populations that cannot aord access to sucient cooling in the future. Accordingly, this map can serve as a reference point to develop targeted climate change adaptation and 74 energy management policies to help vulnerable populations prepare for a future with more extreme heat events. 4.3.6 Further considerations It should be noted that the results in this study are based on present day data and climate change projections for the year 2100 based on the RCP 8.5 scenario. There are several factors other than climate that may change in the future, adding additional complexity to how electricity consumption patterns and vulnerability to extreme heat may evolve in the future. For example, buildings are expected to be more energy ecient as a result of more stringent building energy codes [126], which will change residents' electricity-temperature relationships. The increasing utilization of onsite generation (e.g., solar photovoltaic panels) and storage technologies might also markedly aect net energy consumption proles. While this study investigates factors in uencing electricity consumption patterns in the context of future climate change, another important consideration is whether or not existing infrastructure can meet projected increases in electricity demand in general. Similar to other regions, Southern California is expected to experience increases in cooling capacities as well as AC penetration [51]. However, if targeted policies are not put in place properly, such increases in AC penetration could add additional nancial burden to vulnerable populations. Furthermore, increasing AC penetration does not relieve constraints associated with operating AC units. More research needs to be done to quantify these uncertainties. Though assessing heat exposure based on home address is usually used for practical reasons, negative health eects can involve many more factors (e.g., the time people spend commuting within the region, the time people spend in an air-conditioned environment, and more) [67]. These broader activities should be considered in future studies on the topic of heat-related vulnerability assessment. 75 a b Figure 4.4 Map of (a) historically observed and (b) end-of-century projected number of extreme heat days (dened as daily maximum temperature >35°C (95°F)) per year at the census tract level in Southern California. Census tracts that we identied as vulnerable to increases in extreme heat are highlighted using yellow borders. 76 4.4 Conclusion and future work In this research we nd large spatial variability in sensitivity of electricity to ambient temperature for residential homes across Southern California. Variabilities in these ET sensitivities are associated with spatial patterns of climate characteristics, as well as socioeconomic distributions. Specically, homes in coastal (cooler) climate zones show electricity consumption that is generally less sensitive to ambient temperature change relative to inland (hotter) communities. Coastal communities also have lower AC penetration rates (i.e., the fraction of homes with AC) relative to inland communities. In addition, more auent communities generally have ET sensitivities that are higher than their less auent counterparts and have higher AC penetration rates in most climate zones. We nd that, compared to climate, auence level was stronger in explaining spatial variability in ET sensitivities but weaker in explaining AC penetration rates. We identied communities that are likely to be most vulnerable to increases in extreme heat events as those that have the lowest levels of both AC penetration and auence. 55% and 30% of the census tracts identied as most vulnerable are expected to have more than 16 and 32 extreme heat days per year by the end of the century, respectively. Future work should focus on assessing the impact of housing stock characteristics (e.g., square footage, year of built, insulation levels), as well as behavior related factors, on the relationships between residential electricity consumption and climate. 77 Chapter 5 Conclusion The research described throughout this dissertation focuses on developing quantitative methods for assessing the impacts of climate on residential electricity consumption. The residential sector is by far the biggest con- sumer in electricity market in the US [44]. Due to the diverse nature of the residential sector, it is especially challenging to model or predict the response of electricity consumption in households to climate uctuations compared to other sectors [131] [66]. To address the highly variable contributing factors, like housing stock characteristics, socio-economic factors, heating and cooling energy sources, that can signicantly vary across regions or even neighborhoods, it is necessary to utilize data of high spatiotemporal resolution [34]. Prior to this work, methods to quantify the functional relationship between residential electricity consumption and climate variables using highly spatiotemporal resolved data, as well as how the resolution itself impacts such quantication, have not been well studied. This body of work addressed several of these knowledge gaps and methodological gaps in the energy-climate nexus literature. Chapter 2 conducted a thorough literature review in this eld and located several major knowledge gaps. Generally speaking, previous studies used datasets that vary widely in spatiotemporal resolution. However, they rarely utilize electricity datasets that have both high spatial and temporal resolution. Despite the variety of data resolution and types used, no research to the author's knowledge has investigated how choices in dataset resolution and data types in uence the quantication of electricity-temperature sensitivity. Chapter 2 addressed this gap by establishing a segmented linear regression model describing the relationship between residential electricity consumption and ambient temperature. Model performance was tested using data at various spatiotemporal resolution and concluded that data at the individual household level and daily 78 level yield the best model performance. In addition, results show that the strongest relationships between electricity consumption and temperature are computed using daily accumulated electricity consumption and daily mean temperature, among all types of data. These ndings oered insights in the importance of utilizing high spatiotemporal resolution data and how to choose among dierent temperature indicators in quantifying E-T relationships. Chapter 3 applied the regression model established in Chapter 2 to a much larger and statistically repre- sentative smart-meter dataset, which includes 180,476 households in Southern California region. Chapter 3 focuses on developing a method to identify whether homes use AC at the household level, by only observing their daily accumulative electricity use and daily average temperature data. The results align with previ- ously published survey-based data across a similar regional extent. To the author's knowledge, the study presented in Chapter 3 is the rst to allow the quantication of AC accessibility at high spaiotemporal res- olution. Repeating the study in future years can give us the ability to track AC penetration over time. The method developed in this study can also be applied to other regions around the world, with modication, where household-level smart meter data are available. Since AC is an important player in dening E-T relationships, knowledge gleaned from Chapter 3 can help the scientic community better understand how energy use as a function of temperature varies over space and evolves over time. Chapter 4 applied the model developed in Chapter 2 and the method developed in Chapter 3 to map out the E-T sensitivities in the Southern California region. Variability in E-T sensitivities are associated with spatial patterns of baseline climate, as well as socioeconomic distributions. In general, coastal homes show electricity consumption that is less sensitive to ambient temperature change relative to inland homes. In addition, more auent communities have higher E-T sensitivities than their less auent counterparts. Both trends align with AC penetration patterns in these areas/communities. Using this information, many communities, according to their AC penetration and auence levels, that may be the most vulnerable to future extreme heat events are also expected to experience large amount of extreme heat days (e.g., 8 or 16 days per year) by the end of century. Collectively, this body of work oers researchers electricity grid and city planners, and pertinent policy makers data and methods to better understand and assess the impact of climate on electricity consumption 79 in the residential sector. Such information is important especially with the fact that many major cities in the world are facing urbanization, global climate change, and urban heat island eects. Meanwhile, our energy systems are evolving at a fast pace as more and more renewable sources are replacing fossil fuel generators, adding some complexity to ensuring electricity reliability in the future. Thus, having a holistic and quantitative understanding of how climate plays a role in impacting electricity demand provides useful information to energy eciency policy-making, climate mitigation strategies, peak energy management, as well as addressing related energy/environmental justice issues. As more and more abundant energy and environmental data become available around the world, we believe that the methods and frameworks developed in this work can be applied to other regions and extended to study changes in E-T relationships over time. 80 Reference List [1] Hashem Akbari. Cooling our Communities: A Guidebook On Tree Planting And Light-Colored Sur- facing, 1992. [2] Muhammad Tauha Ali, Marwan Mokhtar, Matteo Chiesa, and Peter Armstrong. A cooling change- point model of community-aggregate electrical load. Energy and Buildings, 43(1):28{37, 2011. [3] Melissa R. Allen, Steven J. Fernandez, Joshua S. Fu, and Mohammed M. Olama. Impacts of climate change on sub-regional electricity demand and distribution in the southern United States. Nature Energy, 1(8):16103, 2016. [4] I. Andri c, Andr e Pina, Paulo Ferr~ ao, J er emy Fournier, Bruno Lacarri ere, and Olivier Le Corre. The impact of climate change on building heat demand in dierent climate types. Energy and Buildings, 149:225{234, 2017. [5] B. W. Ang, T. N. Goh, and X. Q. Liu. Residential electricity demand in Singapore. Energy, 17(1):37{ 46, 1992. [6] Anin Aroonruengsawat and Maximilian Auhammer. Impacts of Climate Change on Residential Elec- tricity Consumption Evidence from Billing Data. In The Economics of Climate Change: Adaptations Past and Present, number May, chapter 11, pages 311{342. University of Chicago Press, 2011. [7] D. A. Asimakopoulos, M. Santamouris, I. Farrou, M. Laskari, M. Saliari, G. Zanis, G. Giannakidis, K. Tigas, J. Kapsomenakis, C. Douvis, S. C. Zerefos, T. Antonakaki, and C. Giannakopoulos. Mod- elling the energy demand projection of the building sector in Greece in the 21st century. Energy and Buildings, 49:488{498, 2012. [8] Maximilian Auhammer and Anin Aroonruengsawat. Simulating the impacts of climate change, prices and population on California's residential electricity consumption. Climatic Change, 109(SUPPL. 1):191{210, 2011. [9] Maximilian Auhammer, Patrick Baylis, and Catherine H Hausman. Climate change is projected to have severe impacts on the frequency and intensity of peak electricity demand across the United States. Proceedings of the National Academy of Sciences, 114(8), 2017. [10] Maximilian Auhammer and Erin T. Mansur. Measuring climatic impacts on energy consumption: A review of the empirical literature. Energy Economics, 46:522{530, 2014. [11] Masood A. Badri. Analysis of demand for electricity in the United States. Energy, 17(7):725{733, 1992. [12] George Ban-Weiss, Craig Wray, Woody Delp, Peter Ly, Hashem Akbari, and Ronnen Levinson. Elec- tricity production and cooling energy savings from installation of a building-integrated photovoltaic roof on an oce building. Energy and Buildings, 56:210{220, 2013. [13] Matthew Bartos, Mikhail Chester, Nathan Johnson, Brandon Gorman, Daniel Eisenberg, Igor Linkov, and Matthew Bates. Impacts of rising air temperatures on electric transmission ampacity and peak electricity load in the United States. Environmental Research Letters, 11(11):114008, 2016. 81 [14] Cajsa Bartusch, Monica Odlare, Fredrik Wallin, and Lars Wester. Exploring variance in residential electricity consumption: Household features and building properties. Applied Energy, 92:637{643, 2012. [15] Rupa Basu and Bart D. Ostro. A multicounty analysis identifying the populations vulnerable to mortality associated with high ambient temperature in California. American Journal of Epidemiology, 168(6):632{637, 2008. [16] M. Beccali, M. Cellura, V. Lo Brano, and A. Marvuglia. Short-term prediction of household electric- ity consumption: Assessing weather sensitivity in a Mediterranean area. Renewable and Sustainable Energy Reviews, 12(8):2040{2065, 2008. [17] Michael Beenstock, Ephraim Goldin, and Dan Nabot. The demand for electricity in Israel. Energy Economics, 21(2):168{183, 1999. [18] David B. Belzer, Michael J. Scott, and Ronald D. Sands. Climate Change Impacts on U.S. Commercial Building Energy Consumption: An Analysis Using Sample Survey Data. Energy Sources, 18(February 2015):177{201, 1996. [19] Mark L. Berenson, David M. Levine, and Timothy C. Krehbiel. Chapter 8.7: Estimation and Sam- ple Size Determination for Finite Populations. Basic business statistics: concepts and applications. Pearson/Prentice Hall, 10th edition, 2006. [20] Marie Bessec and Julien Fouquau. The non-linear link between electricity consumption and tempera- ture in Europe: A threshold panel approach. Energy Economics, 30(5):2705{2721, 2008. [21] Andrea Bigano, Francesco Bosello, and Giuseppe Marano. Energy Demand and Temperature: A Dynamic Panel Analysis. Fondazione Eni Enrico Mattei, (112.06), 2006. [22] Sam Borgeson. ConvergenceDA/visdom: Toolkit for analyzing small or large samples of Smart Meter data and estimating attributes of customers, 2016. [23] Sam Borgeson, June A. Flora, Jungsuk Kwac, Chin-Woo Tan, and Ram Rajagopal. Learning from Hourly Household Energy Consumption: Extracting, Visualizing and Interpreting Household Smart Meter Data. Lecture Notes in Computer Science (including subseries Lecture Notes in Articial Intel- ligence and Lecture Notes in Bioinformatics), 9188:337{345, 2015. [24] Samuel Dalton Borgeson. Targeted Eciency: Using Customer Meter Data to Improve Eciency Program Outcomes. PhD thesis, Ph.D. dissertation, University of California Berkeley, 2013. [25] Marilyn A Brown, Matt Cox, Ben Staver, and Paul Baer. Climate Change and Energy Demand in Buildings. Proceedings of the American Council for an Energy Ecient Economy (ACEEE) Summer Study on Energy Eciency in Buildings. Pacic Grove, pages 26{38, 2014. [26] Edward Byers, Matthew Gidden, David Leclere, Juraj Balkovic, Peter Burek, Kristie Ebi, Peter Greve, David Grey, Petr Havlik, Astrid Hillers, Nils Johnson, Taher Kahil, Volker Krey, Simon Langan, Nebjosa Nakicenovic, Robert Novak, Michael Obersteiner, Shonali Pachauri, Amanda Palazzo, Simon Parkinson, Narasimha D. Rao, Joeri Rogelj, Yusuke Satoh, Yoshihide Wada, Barbara Willaarts, and Keywan Riahi. Global exposure and vulnerability to multi-sector development and climate change hotspots. Environmental Research Letters, 13(5), 2018. [27] California Department of Water Resources. The California Irrigation Management Information System, 2017. [28] California Energy Commission. Developing Advanced Energy Storage Technology Solutions to Lower Costs and Achieve Policy Goals (PON-13-302) ATTACHMENT 12 References for Energy End-Use, Electricity Demand and GHG Emissions Reference and Calculations. 2014. [29] California Energy Commission. California Building Climate Zone Areas, 2015. 82 [30] California Energy Commission. Cal-Adapt, 2019. [31] K. Carrie Armel, Abhay Gupta, Gireesh Shrimali, and Adrian Albert. Is disaggregation the holy grail of energy eciency? The case of electricity. Energy Policy, 52:213{234, 2013. [32] Maurizio Cellura, Francesco Guarino, Sonia Longo, and Giovanni Tumminia. Climate change and the building sector: Modelling and energy implications to an oce building in southern Europe. Energy for Sustainable Development, 45(2018):46{65, 2018. [33] A. L.S. Chan. Developing future hourly weather les for studying the impact of climate change on building energy performance in Hong Kong. Energy and Buildings, 43(10):2860{2868, 2011. [34] M. Chen, G.A. Ban-Weiss, and K.T. Sanders. The role of household level electricity data in improving estimates of the impacts of climate on building electricity use. Energy and Buildings, 180, 2018. [35] Mo Chen, George Alexander Ban-Weiss, and Kelly T Sanders. Utilizing smart-meter data to project impacts of urban warming on residential electricity use for vulnerable populations in Southern Cali- fornia. Environmental Research Letters, 2020. [36] Mo Chen, Kelly T Sanders, and George Alexander Ban-Weiss. A new method utilizing smart meter data for identifying the existence of air conditioning in residential homes. Environmental Research Letters, 14(9):94004, 2019. [37] Andrew B Colombo, David Etkin, and Bryan W Karney. Climate Variability and the Frequency of Extreme Temperature Events for Nine Sites across Canada : Implications for Power Usage. Journal of Climate, (IPCC 1996):2490{2503, 1999. [38] Lucas W. Davis and Paul J. Gertler. Contribution of air conditioning adoption to future energy use under global warming. Proceedings of the National Academy of Sciences, 112(19):5962{5967, 2015. [39] An bal De Almeida, Paula Fonseca, Barbara Schlomann, and Nicolai Feilberg. Characterization of the household electricity consumption in the EU, potential energy savings and specic policy recommen- dations. Energy and Buildings, 43(8):1884{1894, 2011. [40] Je Deason, Max Wei, Greg Leventis, Sarah Smith, and Lisa Schwartz. Electrication of buildings and industry in the United States. Technical Report March, Electricity Markets and Policy Group, Lawrence Berkeley National Lab, 2018. [41] James A. Dirks, Willy J. Gorrissen, John H. Hathaway, Daniel C. Skorski, Michael J. Scott, Tren- ton C. Pulsipher, Maoyi Huang, Ying Liu, and Jennie S. Rice. Impacts of climate change on energy consumption and peak demand in buildings: A detailed regional approach. Energy, 79(C):20{32, 2015. [42] Tilak K Doshi, Principal Fellow, Abhishek Rohatgi, Nahim Bin Zahur, Energy Analyst, and Yuen Kah Hung. Impact of climate change on electricity demand of singapore. pages 1{15, 2012. [43] Mark E.H. Dyson, Samuel D. Borgeson, Michaelangelo D. Tabone, and Duncan S. Callaway. Using smart meter data to estimate demand response potential, with application to solar energy integration. Energy Policy, 73:607{619, 2014. [44] EIA. U.S. Energy Information Administration (EIA), 2015. [45] Energy Information Administration. About the Residential Energy Consumption Survey (RECS): Fuels used and end uses in U.S. homes by housing unit type, 2015, 2015. [46] Energy Information Agency. Household Energy Use in California. Technical report, Energy Informa- tion Agency, 2009. [47] Energy Information Agency (EIA). Homes show greatest seasonal variation in electricity use - Today in Energy - U.S. Energy Information Administration (EIA), 2013. [48] Margaret F Fels. PRISM : An Introduction. Energy and Buildings, 9:5{18, 1986. 83 [49] Guido Franco and Alan H Sanstad. Climate change and electricity demand in California. Climatic Change, 87(August 2006):139{151, 2008. [50] Th Frank. Climate change impacts on building heating and cooling energy demand in Switzerland. Energy and Buildings, 37(11 SPEC. ISS.):1175{1185, 2005. [51] Andrew M. Fraser, Mikhail V. Chester, David Eisenman, David M. Hondula, Stephanie S. Pincetl, Paul English, and Emily Bondank. Household accessibility to heat refuges: Residential air conditioning, public cooled space, and walkability. Environment and Planning B: Urban Analytics and City Science, 44(6):1036{1055, 2017. [52] Jon Froehlich, Eric Larson, Sidhant Gupta, Gabe Cohn, Matthew Reynolds, and Shwetak Patel. Disaggregated End-Use Energy Sensing for the Smart Grid. IEEE Pervasive Computing, pages 28{39, 2010. [53] W. Y. Fung, K. S. Lam, W. T. Hung, S. W. Pang, and Y. L. Lee. Impact of urban temperature on energy consumption of Hong Kong. Energy, 31(14):2287{2301, 2006. [54] Christos Giannakopoulos and Basil E. Psiloglou. Trends in energy load demand for Athens, Greece: Weather and non-weather related factors. Climate Research, 31(1):97{108, 2006. [55] GISTEMP Team. GISS Surface Temperature Analysis (GISTEMP), version 4, 2020. [56] Jo~ ao Pedro Gouveia, J ulia Seixas, and Ana Mestre. Daily electricity consumption proles from smart meters - Proxies of behavior for space heating and cooling. Energy, 141:108{122, 2017. [57] J.S. Haberl and S. Thamilseran. The Great Energy Predictor Shootout II: Measuring Retrot Savings. ASHRAE Journal, 40(1), 1998. [58] Alex Hall, Fengpeng Sun, Daniel Walton, Neil Berg, Alexandre Jousse, Scott Capps, Maria Schwartz, Xin Qu, Mark Nakamura, Hsin-Yuan Huang, and Ruth Cerezo-Mota. Mid-Century Warming in the Los Angeles Region. Climate Change in the Los Angeles Region, 1, 2012. [59] Sarah Marie Hall, Sarah Hards, and Harriet Bulkeley. New approaches to energy: Equity, justice and vulnerability. Introduction to the special issue. Local Environment, 18(4):413{421, 2013. [60] Stephane Hallegatte and Julie Rozenberg. Climate change through a poverty lens. Nature Climate Change, 7(4):250{256, 2017. [61] Sharon L. Harlan, Anthony J. Brazel, Lela Prashad, William L. Stefanov, and Larissa Larsen. Neigh- borhood microclimates and vulnerability to heat stress. Social Science and Medicine, 63(11):2847{2863, 2006. [62] Sharon L. Harlan and Darren M. Ruddell. Climate change and health in cities: Impacts of heat and air pollution and potential co-benets from mitigation and adaptation. Current Opinion in Environmental Sustainability, 3(3):126{134, 2011. [63] M. Hekkenberg, R. M J Benders, H. C. Moll, and A. J M Schoot Uiterkamp. Indications for a changing electricity demand pattern: The temperature dependence of electricity demand in the Netherlands. Energy Policy, 37(4):1542{1551, 2009. [64] M. Hekkenberg, H. C. Moll, and A. J M Schoot Uiterkamp. Dynamic temperature dependence patterns in future energy demand models in the context of climate change. Energy, 34(11):1797{1806, 2009. [65] H. S. Hippert, D. W. Bunn, and R. C. Souza. Large neural networks for electricity load forecasting: Are they overtted? International Journal of Forecasting, 21(3):425{434, 2005. [66] Y. Hirano and T. Fujita. Evaluation of the impact of the urban heat island on residential and com- mercial energy consumption in Tokyo. Energy, 37(1):371{383, 2012. 84 [67] Christopher G. Hoehne, David M. Hondula, Mikhail V. Chester, David P. Eisenman, Ariane Middel, Andrew M. Fraser, Lance Watkins, and Katrina Gerster. Heat exposure during outdoor activities in the US varies signicantly by city, demography, and activity. Health and Place, 54(March):1{10, 2018. [68] Shira Horowitz, Brandon Mauch, and Fallaw Sowell. Forecasting residential air conditioning loads. Applied Energy, 132:47{55, 2014. [69] Frumkin Howard, Hess Jeremy, Luber George, Malilay Josephine, and McGeehin Michael. Climate Change: The Public Health Response. American Journal of Public Health, 98(3):435{445, 2011. [70] Cunrui Huang, Gerard Barnett Adrian, Wang Xiaoming, Vaneckova Pavla, Gerard FitzGerald, and Shilu Tong. Review Projecting Future Heat-Related Mortality under Climate Change Scenarios :. Environmental Health Perspectives, 119(12):1681{1690, 2011. [71] Jianhua Huang and Kevin Robert Gurney. Impact of climate change on U.S. building energy demand: sensitivity to spatiotemporal scales, balance point temperature, and population distribution. Climatic Change, pages 171{185, 2016. [72] Jianhua Huang and Kevin Robert Gurney. The variation of climate change impact on building energy consumption to building type and spatiotemporal scale. Energy, 111:137{153, 2016. [73] YJ Huang. The impact of climate change on the energy use of the us residential and commercial building sectors. Lawrence Berkeley National Laboratory, Report No. LBNL-60754, 2006. [74] T. Ihara, Y. Genchi, T. Sato, K. Yamaguchi, and Y. Endo. City-block-scale sensitivity of electricity consumption to air temperature and air humidity in business districts of Tokyo, Japan. Energy, 33(11):1634{1645, 2008. [75] International Energy Agency. Electricity generation by source worldwide, 2019. [76] International Energy Agency/Organisation for Economic Co-operation and Development. The Future of Cooling Opportunities for energy- ecient air conditioning. Technical report, International Energy Angency/Ognisation for Economic Cooperation and Development, 2018. [77] Morna Isaac and Detlef P. van Vuuren. Modeling global residential sector energy demand for heating and air conditioning in the context of climate change. Energy Policy, 37(2):507{521, 2009. [78] Anthony C. Janetos. Energy infrastructure: Mapping future electricity demand. Nature Energy, 1(8), 2016. [79] Rory V. Jones, Alba Fuertes, and Kevin J. Lomas. The socio-economic, dwelling and appliance related factors aecting electricity consumption in domestic buildings. Renewable and Sustainable Energy Reviews, 43:901{917, 2015. [80] Kelly Kalvelage, Ulrike Passe, Shannon Rabideau, and Eugene S. Takle. Changing climate: The eects on energy demand and human comfort. Energy and Buildings, 76:373{380, 2014. [81] Amir Kavousian, Ram Rajagopal, and Martin Fischer. Determinants of residential electricity con- sumption: Using smart meter data to examine the eect of climate, building characteristics, appliance stock, and occupants' behavior. Energy, 55:184{194, 2013. [82] Scott Kelly, Michelle Shipworth, David Shipworth, Michael Gentry, Andrew Wright, Michael Pollitt, Doug Crawford-Brown, and Kevin Lomas. Predicting the diversity of internal temperatures from the English residential sector using panel methods. Applied Energy, 102:601{621, 2013. [83] Kendra Pierre-Louis. The World Wants Air-Conditioning. That Could Warm the World, 2018. [84] Patrick L. Kinney, Marie S. O'Neill, Michelle L. Bell, and Joel Schwartz. Approaches for estimating eects of climate change on heat-related deaths: challenges and opportunities. Environmental Science and Policy, 11(1):87{96, 2008. 85 [85] John Kelly Kissock, Je S. Haberi, and David E. Claridge. Inverse Modeling Toolkit: Numerical Algorithms for Best-Fit Variable-Base Degree Day and Change Point Models (RP-1050). Energy Systems Lab., 109:1{9, 2003. [86] Kim Knowlton, Mirian Rotkin-Ellman, Galatea King, Helene G. Margolis, Daniel Smith, Gina Solomon, Roger Trent, and Paul English. The 2006 California heat wave: Impacts on hospitaliza- tions and emergency department visits. Environmental Health Perspectives, 117(1):61{67, 2009. [87] Reto Knutti, Robert L. Wilby, Markus G. Donat, Andy J. Pitman, and Sonia I. Seneviratne. Allowable CO2 emissions based on regional and impact-related climate targets. Nature, 529(7587):477{483, 2016. [88] J. Zico Kolter, Siddarth Batra, and Andrew Y. Ng. Energy Disaggregation via Discriminative Sparse Coding. Advances in Neural Information Processing Systems, 4(8):1153{1161, 2010. [89] R. Sari Kovats and Shakoor Hajat. Heat Stress and Public Health: A Critical Review. Annual Review of Public Health, 29(1):41{55, 2007. [90] J.F. Kreider and J.S. Haberl. Predicting hourly building energy use: The great energy predictor shootout { Overview and discussion of results. In American Society of Heating, Refrigerating, and Air Conditioning Engineers (ASHRAE) annual meeting, Orlando, FL, 1994. [91] Jungsuk Kwac and Ram Rajagopal. Demand response targeting using big data analytics. Proceedings - 2013 IEEE International Conference on Big Data, pages 683{690, 2013. [92] Joseph C. Lam. Climatic and economic in uences on residential electricity consumption. Energy Conversion and Management, 39(7):623{629, 1998. [93] Joseph C. Lam, H. L. Tang, and D. H W Li. Seasonal variations in residential and commercial sector electricity consumption in Hong Kong. Energy, 33(3):513{523, 2008. [94] Shaun Larcom, Po Wen She, and Terry van Gevelt. The UK summer heatwave of 2018 and public concern over energy security. Nature Climate Change, 9(5):370{373, 2019. [95] B. Lebassi, J. E. Gonzalez, D. Fabris, and R. Bornstein. Impacts of Climate Change in Degree Days and Energy Demand in Coastal California. Journal of Solar Energy Engineering, 132(3):031005, 2010. [96] Nathan J. L. Lenssen, Gavin A. Schmidt, James E. Hansen, Matthew J. Menne, Avraham Persin, Reto Ruedy, and Daniel Zyss. Improvements in the GISTEMP Uncertainty Model. Journal of Geophysical Research: Atmospheres, 124(12):6307{6326, jun 2019. [97] Xiaoma Li, Yuyu Zhou, Sha Yu, Gensuo Jia, Huidong Li, and Wenliang Li. Urban heat island impacts on building energy consumption: a review of approaches and ndings. Energy, 2019. [98] Yating Li, William A. Pizer, and Libo Wu. Climate change and residential electricity consumption in the Yangtze River Delta, China. Proceedings of the National Academy of Sciences, (14):201804667, 2018. [99] George Luber and Michael McGeehin. Climate Change and Extreme Heat Events. American Journal of Preventive Medicine, 35(5):429{435, 2008. [100] Loren Lutzenhiser, Mithra Moezzi, Aaron Ingle, and James Woods. Final Project Report for Advanced Residential Energy and Behavior Analysis Project by California Energy Commission. Technical Report September, Portland State University, California Energy Commision, 2016. [101] G. P. Maheshwari, F. Al-Ragom, and R. K. Suri. Energy-saving potential of an indirect evaporative cooler. Applied Energy, 69(1):69{76, 2001. [102] Jeremy Martinich and Allison Crimmins. Climate damages and adaptation potential across diverse sectors of the United States. Nature Climate Change, 9(5):397{404, 2019. 86 [103] James J. McCarthy and Intergovernmental Panel on Climate Change. Working Group II. Climate change 2001: impacts, adaptation, and vulnerability: contribution of Working Group II to the third assessment report of the Intergovernmental Panel on Climate Change. Cambridge University Press, 2001. [104] Michael A. McGeehin and Maria Mirabelli. The potential impacts of climate variability and change on temperature-related morbidity and mortality in the United States. Environmental Health Perspectives, 109(SUPPL. 2):185{189, 2001. [105] Michael A. McNeil and Ve Letschert. Future air conditioning energy consumption in developing coun- tries and what can be done about it: the potential of eciency in the residential sector. Lawrence Berkeley National Laboratory, page 14, 2008. [106] Michael A Mcneil and Virginie E Letschert. Modeling diusion of electrical appliances in the residential sector. Energy and Buildings, 42(August):783{790, 2010. [107] S. Mirasgedis, Y. Saradis, E. Georgopoulou, D. P. Lalas, M. Moschovits, F. Karagiannis, and D. Pa- pakonstantinou. Models for mid-term electricity demand forecasting incorporating weather in uences. Energy, 31(2-3):208{227, 2006. [108] Vito M R Muggeo. Regression Models with Breakpoints/Changepoints Estimation, 2015. [109] National Centers for Environmental Information (NCEI). Land-Based Station Data, 2016. [110] National Weather Services. Natural Hazard Statistics, 2018. [111] Ning Lu, Wei Jiang, J. Correia, Chunlian Jin, T. Taylor, Pak Chung Wong, and L.R. Leung. Climate Change Impacts on Residential and Commercial Loads in the Western U.S. Grid. Technical report, Pacic Northwest National Lavoratory, 2008. [112] NOAA. Climate Prediction Center - Monitoring & Data: Weekly & Monthly Degree Day Summaries Explanation, 2005. [113] California Environmental Protection Agency Oce of Environmental Health Hazard Assessment. CalEnviroScreen 3.0, 2018. [114] Mady Olonscheck, Anne Holsten, and J urgen P. Kropp. Heating and cooling energy demand and related emissions of the German residential building stock under climate change. Energy Policy, 39(9):4795{4806, 2011. [115] OpenEE Inc. EEmeter: tools for calculating metered energy savings, 2018. [116] C Palmgren, N Stevens, M Goldberg, R Bames, and K Rothkin. 2009 California residential appliance saturation survey. Technical report, California Energy Commission, 2010. [117] Luca Pampuri, Nerio Cereghetti, Davide Strepparava, and Paola Caputo. Analysis of the electricity consumptions: A rst step to develop a district cooling system. Sustainable Cities and Society, 23:23{ 36, 2016. [118] Angel Pardo, Vicente Meneu, and Enric Valor. Temperature and seasonality in uences on Spanish electricity load. Energy Economics, 24(1):55{70, 2002. [119] S.J. Parkpoom and G.P. Harrison. Analyzing the Impact of Climate Change on Future Electricity Demand in Thailand. IEEE Transactions on Power Systems, 23(3):1441{1448, 2008. [120] Jonathan A Patz, Michael A McGeehin, Susan M Bernard, Kristie L Ebi, Paul R Epstein, Anne Grambsch, Duane J Gubler, Paul Reiter, Isabelle Romieu, Joan B Rose, Jonathan M Samet, Juli Trtanf, Research Program Director, JD JC Scheraga We thank Beier, B Boutin, R Calderon, WR Daley, D Dockery, D Driscoll, D Easterling, D Engelberg, Da Focks, G Greenough, M Habib, W Jakubowski, L Kalkstein, T Karl, E Lipp, M Lipsett, M Mirabelli, R Nasci, E Noji, D Paxman, W Reisen, J Riad, 87 J Schwartz, J Selanikio, BH Sherman, R Shope, A Spielman, M Wilson, and W H Yap We also thank. Potential Health Impacts of Climate Variability and Change for the US Executive Summary. Environmental Health Perspectives *, 108(4):367, 2000. [121] Krystian X. Perez, Kristen Cetin, Michael Baldea, and Thomas F. Edgar. Development and analysis of residential change-point models from smart meter data. Energy and Buildings, 139:351{359, 2017. [122] Karoliina Pilli-Sihvola, Piia Aatola, Markku Ollikainen, and Heikki Tuomenvirta. Climate change and electricity consumption-Witnessing increasing or decreasing use and costs? Energy Policy, 38(5):2409{ 2419, 2010. [123] Melvin Pomerantz, Pablo J. Rosado, and Ronnen Levinson. A simple tool for estimating city-wide annual electrical energy savings from cooler surfaces. Urban Climate, 14:315{325, 2015. [124] B. E. Psiloglou, C. Giannakopoulos, S. Majithia, and M. Petrakis. Factors aecting electricity demand in Athens, Greece and London, UK: A comparative assessment. Energy, 34(11):1855{1863, 2009. [125] Manish Ranjan and V K Jain. Modelling of electrical energy consumption in Delhi. Energy, 24(De- cember 1997):351{361, 1999. [126] Janet L. Reyna and Mikhail V. Chester. Energy eciency to reduce residential electricity and natural gas use under climate change. Nature Communications, 8(May 2017):1{12, 2017. [127] Matthias Ruth and Ai Chen Lin. Regional energy demand and adaptations to climate change: Method- ology and application to the state of Maryland, USA. Energy Policy, 34(17):2820{2833, 2006. [128] D. A. Sachindra, A. W.M. Ng, S. Muthukumaran, and B. J.C. Perera. Impact of climate change on urban heat island eect and extreme temperatures: A case-study. Quarterly Journal of the Royal Meteorological Society, 142(694):172{186, 2016. [129] R Saidur, H H Masjuki, and I A Choudhury. Role of ambient temperature , door opening , ther- mostat setting position and their combined eect on refrigerator-freezer energy consumption. Energy Conversion and Management, 43:845{854, 2002. [130] D. J. Sailor and A. A. Pavlova. Air conditioning market saturation and long-term response of residential cooling energy demand to climate change. Energy, 28(9):941{951, 2003. [131] David J. Sailor. Relating residential and commercial sector electricity loads to climate - Evaluating state level sensitivities and vulnerabilities. Energy, 26(7):645{657, 2001. [132] David J. Sailor and J. Ricardo Mu~ noz. Sensitivity of electricity and natural gas consumption to climate in the U.S.A. - Methodology and results for eight states. Energy, 22(10):987{998, 1997. [133] David J. Sailor and Chittaranjan Vasireddy. Correcting aggregate energy consumption data to account for variability in local weather. Environmental Modelling and Software, 21(5):733{738, 2006. [134] Natalie R. Sampson, Carina J. Gronlund, Miatta A. Buxton, Linda Catalano, Jalonne L. White- Newsome, Kathryn C. Conlon, Marie S. O'Neill, Sabrina McCormick, and Edith A. Parker. Staying cool in a changing climate: Reaching vulnerable populations during heat events. Global Environmental Change, 23(2):475{484, 2013. [135] Thomas F. Sanquist, Heather Orr, Bin Shui, and Alvah C. Bittner. Lifestyle factors in U.S. residential electricity consumption. Energy Policy, 42:354{364, 2012. [136] M. Santamouris, C. Cartalis, A. Synnefa, and D. Kolokotsa. On the impact of urban heat island and global warming on the power demand and electricity consumption of buildings - A review. Energy and Buildings, 98:119{124, 2015. [137] Hajime Sasaki, Ichiro Sakata, and Weerin Wangjiraniran. Appliance Diusion Model for Energy Eciency Standards and Labeling Evaluation in the Capital of Lao PDR. 3(3):269{281, 2015. 88 [138] Jayant A Sathaye, Larry L Dale, Peter H Larsen, Gary A Fitts, Kevin Koy, and Sarah M Lewis. Estimating impacts of warming temperatures on California ' s electricity system. Global Environmental Change, 23:499{511, 2013. [139] M Segal, H Shar, M Mandel, P Alpert, and Y Balmor. Climatic-related evaluations of the summer peak-hours' electric load in Israel.pdf. Journal of Applied Meteorology, 31(December):1492{1498, 1992. [140] Scott C Sheridan, Michael J Allen, Cameron C Lee, and Laurence S Kalkstein. Future heat vulnera- bility in California , Part II : projecting future heat-related mortality. Climatic Change, 2012. [141] Michael Sivak. Air conditioning versus heating: Climate control is more energy demanding in Min- neapolis than in Miami. Environmental Research Letters, 8(1), 2013. [142] Weather Spark. Weather comparison between Shanghai and Altlanta, 2019. [143] Fengpeng Sun, Daniel B. Walton, and Alex Hall. A hybrid dynamical-statistical downscaling technique. Part II: End-of-Century warming projections predict a new climate state in the Los Angeles region. Journal of Climate, 28(12):4618{4636, 2015. [144] Lukas G Swan and V Ismet Ugursal. Modeling of end-use energy consumption in the residential sector : A review of modeling techniques. Renewable and Sustainable Energy Reviews, 13:1819{1835, 2009. [145] Nancy Thomas, Shruti Mukhtyar, Brian Galey, and Maggi Kelly. Cal-Adapt: Linking Climate Science With Energy Sector Resilience and Practitioner Need. Technical report, Geospatial Innovation Facility - University of California, Berkeley, California's Fourth Climate Change Assessment, California Energy Commission. Publication Number: CCCA4-CEC-2018-015, 2018. [146] United Nations, Department of Economic and Social Aairs, Population Division. World Urbanization Prospects: The 2018 Revision (ST/ESA/SER.A/420). Technical report, United Nations, New York, 2018. [147] U.S. Census Bureau. Geography Atlas - Census Tracts. [148] U.S. Census Bureau. Cartographic Boundary Shapeles - Census Tracts, 2017. [149] U.S. Energy Information Administration. 2009 RECES Survey Data - Residential Energy Consumption Survey (RECS), 2009. [150] U.S. Energy Information Administration. Annual Energy Outlook 2016. Technical report, U.S. Energy Information Administration, Washitong D.C., 2016. [151] U.S. Energy Information Administration. Electricity data browser - Retail sales of electricity, 2016. [152] U.S. Energy Information Administration. U.S. households' heating equipment choices are diverse and vary by climate region - Today in Energy - U.S. Energy Information Administration (EIA), 2017. [153] U.S. Energy Information Administration. 2015 Residential Energy Consumption Survey: Overview, 2018. [154] U.S. Energy Information Administration. How is electricity used in U.S. homes? - FAQ - U.S. Energy Information Administration (EIA), 2018. [155] U.S. Energy Information Administration. Sales of Electricity to Ultimate Customers, may 2020. [156] U.S. Energy Information Administration. What is U.S. electricity generation by energy source?, feb 2020. [157] P. Vahmani and G. Ban-Weiss. Climatic consequences of adopting drought-tolerant vegetation over Los Angeles as a response to California drought. Geophysical Research Letters, 43(15):8240{8249, 2016. 89 [158] Daniel B. Walton, Fengpeng Sun, Alex Hall, and Scott Capps. A hybrid dynamical-statistical downscal- ing technique. Part I: Development and validation of the technique. Journal of Climate, 28(12):4597{ 4617, 2015. [159] Kevin K.W. Wan, Danny H.W. Li, Wenyan Pan, and Joseph C. Lam. Impact of climate change on building energy use in dierent climate zones and mitigation and adaptation implications. Applied Energy, 97:274{282, 2012. [160] Haojie Wang and Qingyan Chen. Impact of climate change heating and cooling energy use in buildings in the United States. Energy and Buildings, 82(2014):428{436, 2014. [161] Liping Wang, Xiaohong Liu, and Hunter Brown. Prediction of the impacts of climate change on energy consumption for a medium-size oce building with two climate models. Energy and Buildings, 157(2017):218{226, 2017. [162] Y Wang, Y Li, S Di Sabatino, A Martilli, and P W Chan. Eects of anthropogenic heat due to air- conditioning systems on an extreme high temperature event in Hong Kong. Environmental Research Letters, 13(3):034015, 2018. [163] Kiattiporn Wangpattarapong, Somchai Maneewan, Nipon Ketjoy, and Wattanapong Rakwichian. The impacts of climatic and economic factors on residential electricity consumption of Bangkok Metropolis. Energy and Buildings, 40(8):1419{1425, 2008. [164] Michael Witt, Justin Holtzman, Madhur Lamsal, and D N V Gl. Conducting A Web-Based, Residential Appliance Saturation Survey (Rass) Across Multiple Electric Cooperatives: Methodology, Challenges, Results, And Next Steps. In 2017 International Energy Program Evaluation Conference, Baltimore, MD, 2017. [165] Peter Wyatt. A dwelling-level investigation into the physical and socio-economic drivers of domestic energy consumption in England. Energy Policy, 60:540{549, 2013. [166] Peng Xu, Yu Joe Huang, Norman Miller, Nicole Schlegel, and Pengyuan Shen. Impacts of climate change on building heating and cooling energy patterns in California. Energy, 44(1):792{804, 2012. [167] Yuk Yee Yan. Climate and residential electricity consumption in Hong Kong. Energy, 23(1):17{20, 1998. [168] Michael Zeifman and Kurt Roth. Nonintrusive appliance load monitoring: Review and outlook. IEEE Transactions on Consumer Electronics, 57(1):76{84, 2011. 90 Appendix A Supplementary material for Chapter 2: the role of household level electricity data in improving estimates of the impacts of climate on building electricity use Information on the correlation between electricity-temperature sensitivity and statistics graph of the dataset in this study are provided. A.1 Correlation between electricity-temperature sensitivity and stationary point temperature The correlation between electricity-temperature sensitivity and stationary point temperature is presented in Figure A.1. A positive correlation is observed between these two indicators. There are a few possible explanations for the positive correlation, related to occupants' behavior and home characteristics. For example, high stationary point temperature and high electricity-temperature sensitivity may coexist in newly built large houses due to better insulation (relative to older homes) and larger cooling devices used (relative to smaller homes). However, further analysis of such correlation should be assessed in future studies since the dataset used here is not statistically representative of California and lacks further relevant household and occupant details. 91 Figure A.1 correlation between electricity-temperature sensitivity (kW °C 1 ) and stationary point temperature (°C) at dierent spatial aggregation levels: a) individual household, b) cities, c) counties, d) climate zones. p-value and Pearson correlation coecient r are shown in each subplot. 92 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 Hour of the Day -1 0 1 2 3 4 5 Electric Power Consumption [kW] Figure A.2 Electric power throughout 09/10/2015 (hottest day of 2015 in California) averaged for 1,245 households (solid line) and standard deviation (shaded area). This gives a general description of statistics of the dataset in this study. 93 Appendix B Supplementary information for Chapter 3: a new method utilizing smart meter data for identifying the existence of air conditioning in residential homes Information on the method uses to calculate sample sizes needed to statistically represent the population in a spatial area is provided. B.1 Method used to calculate statistically representative sample sizes n = Z 2 p(1p) e 2 N (N 1) + Z 2 p(1p) e 2 (B.1) Equation (B.1)[19] is used to calculate the sample size needed to statistically represent the population in a spatial area, which is census tract in this study. In Equation (B.1), n is the required sample size. Z is the Z-score, determined by the condence level chosen. In this study, a condence level of 90% is chosen, corresponding to a Z-score of 1.645. p is the expected prevalence or proportion. 68% is used for p, as suggested by the overall AC penetration rate in the investigated area in Section 3.3.1. e is the margin of error, chosen as 10% for this study. N is the population in the area, which is the total number of households in each census tract. B.2 Discussion of potential method shortcomings Assuming a piecewise linear t with one or two \breakpoints" (equivalent to Stationary Point Temperature in this study), ASHRAEs inverse model toolkit [85] details ve cases to illustrate a household's electricity consumption response to ambient temperature, as displayed in Figure B.1. The method used in this study treats a household as having AC if 1) the household has a positive slope right and 2) the household's slope right is bigger than slope left. For each case illustrated in Figure B.1, we explain the applicability of the method in this study: • Case 1, Households with both electric heating and cooling (one stationary point temperature): Our method can capture most households in this category. Since cooling demand in Southern California is larger than heating demand [28], a majority of households in this case, if equipped with AC, should have a bigger slope right than slope left. However, if for some uncommon reason a household has AC and uses electric space heating more intensively than space cooling, our method will not identify the AC unit's existence. It should be noted that 80% of households in Southern California do not use electricity for space heating [152] so we know that the fraction of households that fall into this case is relatively small. • Case 2, Households with electric heating and cooling (two stationary point temperatures): Our method cannot capture households in this category because only one stationary point temperature is identied 94 Figure B.1 Schema of relationships between energy use and temperature dened in ASHRAE's inverse model toolkit (adapted from [24], [85]) in the model utilized in this study. Households in this category will be treated erroneously by our statistical model as Case 1. However, one stationary point temperature is enough to describe the relationship between electricity consumption and ambient temperature for most households in Southern California because space heating is mainly supported by natural gas, while cooling energy demand is driven by electricity [46]. (A more detailed discussion of this justication is provided in our previous study [34].) Thus, we believe the proportion of households falling into this category is relatively small. • Case 3, Households with electric cooling only: We can fully capture households in this category. We expect that most homes that have AC in our study region would fall into this category since heating is dominated by natural gas. In this case, households have a positive slope right and near-zero slope left, which our method ags as having AC. • Case 4, Households with electric heating only: We can fully capture households in this category. In this case, households have a negative slope right and slope left, which our method ags as not having AC. • Case 5, Households with neither electric cooling, nor heating: We can capture most households in this category. In this case, households have near-zero slope right and slope left, which our method ags as not having AC. Our method may erroneously identify some households in this category if slope left is small and negative, and slope right is small but positive, both of which are due to noise. It should be noted that since we are identifying AC existence based on smart meter data, only households having and using AC can be identied. Our method will not detect households that have AC but do not use it. 95 Figure B.2 (a) Scatter plot of slope left and slope right of all data records analyzed in Chapter 3 (n=180,476); (b) scatter plot of slope left and slope right of data records within -2 to 2 kW °C 1 range, where most of data points are located B.3 Additional study statistics Figure B.2 shows the relationships of slope left andslope right for all data records analyzed in this study (n=180,476). Data point positions in Figure B.2 include households that re ect dierent cases illustrated in Figure B.1 (ASHRAE Inverse Model Toolkit). For example, data points along the negative part of x-axis fall within Case 4 (electric heating only) due to their negative slope left and near-zero slope right. Data points along the positive part of the y-axis belong to Case 3 (electric cooling only). Data points in the upper left quadrant refers to Case 1, which include houses with both electric heating and cooling. Points around the origin point often fall within Case 5 (neither electric heating or cooling). Most data points fall into one of these ve categories. Figure B.3 shows the distribution of stationary point temperatures calculated with the dataset utilized in this study. Variability in SPT can be caused in part by dierences in occupant behavior and thermal properties of the building. Note that our method does not use stationary point temperature to identify existence of AC. 96 Figure B.3 Probability density distributions of stationary point temperatures (SPT) of all data records analyzed in Chapter 3 (n=180,476). The red dashed line (partially overlapped with blue dashed line) indicates the mean value (18.8°C 3.7°C) of SPT. The blue dashed line indicates the median value (18.6°C) 97 Table B.1 Statistics of annual-averaged electricity consumption for data records in this study by climate zones Climate Zone Summer mean tempera- ture (°C) Mean annual electricity consumption Number of house- hold records Maximum annual electricity consumption a (kWh) Minimum annual electricity consump- tion b (kWh) Standard deviation in annual electricity consumption (kWh) 6 21 4552 29878 986140 20 7660 8 22 4847 49185 403460 20 4527 9 23 5916 36003 750893 20 7887 10 24 6277 40353 809344 20 6999 14 28 6151 12681 177888 20 5271 15 34 8456 6149 1361641 20 22909 16 21 5190 3660 1460973 23 24418 a The maximum annual electricity consumption refers to the largest annual electricity consumption record observed in each climate zone within the dataset. b The minimum annual electricity consumption falls around 20 kWh because this value was set as a screening criterion to lter our unoccupied houses (i.e., any households with an annual electricity consumption lower than 20 kWh is not considered in the analysis). 20 kWh is the average daily electricity consumption of a California home [46]. 98 Appendix C Supplementary information for Chapter 4: utilizing smart-meter data to project impacts of urban warming on residential electricity use for vulnerable populations in Southern California C.1 Probability density distributions of electricity-temperature sensitivities and stationary point temperatures Figure C.1 Probability density distributions of electricity-temperature sensitivities (a) and sta- tionary point temperatures (b) for all 180,476 homes investigated in this study (i.e., including homes with and without space cooling). Red (blue) dashed lines indicate the mean (median) value for all homes. These gures are adapted from Figure 3.3 and Figure B.3 in [36]. Household-level electricity consumption data and high-resolution weather data were used to establish electricity-temperature relationships to enable the quantication of household-to-household variability across the two indicators of interest. Figure 1 displays a probability density distribution of E-T sensitivities (subplot a) and SPTs (subplot b) for all 180,476 customers (i.e., the number of households investigated after data culling was performed). The mean and median value of E-T sensitivity was computed as 0.068 and 0.054 kW °C 1 , respectively. In Figure C.1(a), the distribution of E-T sensitivities is skewed to the right. The peak in the probability 99 density distribution corresponds to sensitivities near zero kW °C 1 , representing households that are not sensitive to ambient temperature changes. (Near zero E-T sensitivities most likely imply that there are no air conditioning devices available at the home.) On the contrary, households represented within the high tail in the distribution in Figure C.1(a) have electricity consumption that is relatively sensitive to ambient temperature change. 48% of households have an electricity-temperature sensitivity of less than 0.05 kW °C 1 while 23%, 23%, 6% of households have electricity-temperature sensitivity values of 0.05 to 0.1, 0.1 to 0.2, and over 0.2 kW °C 1 , respectively. The distribution of SPTs across the studied households demonstrates a near normal distribution with a mean value of 18.8 3.7°C (65.8 6.7°F). This mean value is similar to the widely adopted base temperature of 18°C (65°F) that is used to calculate cooling degree days (CDD) and heating degree days (HDD), which are well established proxies to estimate the energy demand for indoor climate control purposes (e.g., utilized by NOAA's National Weather Center [112]). Though the meanSPT is consistent with the base temperature for CDD and HDD calculations, gaining an understanding of the distribution of household-specic SPT oers more insight into how dierent populations respond to changes in temperature. It is interesting to note that a small portion of homes have SPTs less than 14°C, which is abnormally low. These homes are likely lacking air conditioning devices, causing the model to identify SPT at lower than expected value for reasons other than increasing energy use. 100 C.2 Single variable and multivariable linear regression Table C.1 Linear regression between E-T sensitivity and AC penetration rates versus climatic and socioeconomic variables at the census track level. Values in columns under \Explanatory variables" refer to regression coecients of explanatory variables. Rows with only one value under \Explanatory variables" columns refer to single variable linear regression. Rows with multiple values under \Explanatory variables" columns refer to multivariable linear regressions including the corresponding variables in the columns with values. Values under column \r 2 " refer to coecients of determination of linear regressions in corresponding rows. Response variables Explanatory variables r 2 Summer T min (°C) b Summer T ave (°C) c Summer T max (°C) d Summer4T (°C) e Auence level f E-T sensitivity (kW °C 1 ) a -3.810 3 0.03 3.510 3 0.03 5.310 3 0.13 5.610 3 0.18 -5.910 4 0.12 4.910 4 -6.510 4 0.18 7.210 3 -6.210 4 0.34 6.710 3 -5.310 4 0.37 1.810 3 5.210 3 -6.110 4 0.31 3.110 3 4.410 4 -5.610 4 0.38 AC penetration rate (%) a -1.710 2 0.04 2.610 2 0.11 3.310 2 0.35 3.310 2 0.42 -1.510 3 0.06 3.010 2 -1.910 3 0.19 4.010 2 -1.710 3 0.51 3.710 2 -1.210 3 0.58 1.210 2 3.010 2 -1.710 3 0.49 2.610 2 1.510 2 -1.410 2 0.60 a mean value of census tracts b daily mean temperature averaged over summertime (Jul, Aug, Sep) at census track level c daily minimum temperature averaged over summertime (Jul, Aug, Sep) at census track level d daily maximum temperature averaged over summertime (Jul, Aug, Sep) at census track level e diurnal temperature range averaged over summertime (Jul, Aug, Sep) at census track level f rank of census track's percentage of population living twice under federal poverty level. Small number (higher rank) means more auent 101 C.3 Discussion about Climate Zone 16 (Big Bear Lake area) Climate Zone 16 has a lower average summer temperature (24°C) than other climate zones due to its mountainous landscape and high altitude. Hence, relatively low cooling needs lead to low AC penetration rates and E-T sensitivities, as displayed in Figure 4.1(a) and Figure 4.1(c). As for SPTs, it is observed in Figure 4.1(b) that the Big Bear Lake area has homes with abnormally low SPTs (<14°C). We hypothesize two possible explanations for these behaviors. First, the Big Bear Lake area contains a large fraction of vacation homes compared to most other regions. Vacation homes are often occupied part-time or during specic seasons, which could distort the year-round electricity-temperature relationship and lead to erroneous SPTs. Second, some households, especially vacation homes, may utilize electric rather than gas heating devices. Electric heating will lead to a dierent electricity-temperature relationship and the currently used segmented linear regression model may fail to characterize the home's behavior. Climate Zone 16 also has relatively low E-T sensitivities. The unusually low AC penetration rates coupled with low E-T sensitivities in auent communities (See Figure 4.2 and Figure 4.3) suggest that more vacation houses are located in rich neighborhoods in the Big Bear Lake area. Their E-T relationships are likely distorted by the fact that there is a relatively large number of homes that are not utilized all year round (e.g., used primarily during the ski season) and/or utilize electric heating. 102 C.4 Spatial span and distribution of data included in this study Figure C.2 Map showing locations of the 180,476 residential electricity customers (shown as number of households per census tract) and weather stations considered in this study (red dots). Each household was linked to a weather station based on shortest distance. 103 C.5 Map showing average electricity-temperature sensitivity values per census tract in Southern California (including only households identied as having AC) No Dat a <0. 05 0. 05- 0. 1 0. 1- 0. 15 0. 15- 0. 2 0. 2- 0. 25 >0. 25 CEC Cl i mat e Zone Boundar y Cl i mat e Zone Summer Mean Temper at ur e: 6: 20° C 8: 22° C 9: 23° C 10: 25° C 14: 28° C 15: 32° C 16: 24° C Sensi t i v i t y ( kW ° C - 1 ) 6 8 9 1 6 1 4 1 6 1 5 1 0 Figure C.3 Map showing average electricity-temperature sensitivity values per census tract Southern California (including only households identied as having AC). Dark grey outlines indicate boundaries of building climate zones. Average summer temperatures are shown for each climate zone. White means no data available. 104 C.6 Determination of expected prevalence p in sample size calculation We decide p to be 50% based on an experiment of how the determined sample size n changes with p value using equation: n = Z 2 p(1p) e 2 N (N 1) + Z 2 p(1p) e 2 (C.1) Table C.2 Determined sample size n to be statistically representative as a function of param- eters N Z e p n 4,500,000 2.576 0.50% 0.01 2626.2 4,500,000 2.576 0.50% 0.05 12572.8 4,500,000 2.576 0.50% 0.1 23762.6 4,500,000 2.576 0.50% 0.5 65393.5 4,500,000 2.576 0.50% 0.9 23762.6 4,500,000 2.576 0.50% 0.95 12572.8 4,500,000 2.576 0.50% 0.99 2626.2 As shown in the table above, the maximum sample size is needed to statistically represent the entire population in the area (4.5 million) whenp = 50%, which meansp = 50% sets the strictest rule of statistical representativeness here. Hence it is chosen to make sure that our sample size (180,476 after screening) is big enough to represent the entire population in the Southern California region. 105 C.7 List of identied vulnerable census tracts in Chapter 4 Vulnerable census tracts were dened as those that currently have both low AC penetration rates and high poverty rates (i.e., are more sensitive to future energy cost increases); more specically, vulnerable census tracts were dened as those falling into the top 10% of the studied region when the (1) percentage of population living in poverty and (2) percentage of households not having AC were summed for each respective census track and ranked across all tracts in the region. Baseline and projected number of extreme heat days are based on results of Cal-Adapt models, described in Section 4.2: Projections of future climate change in the main body text. Table C.3 List of identied vulnerable census tracts in Chapter 4 census tract id city Baseline number of extreme heat days (1961-1990) Projected number of extreme heat days (2070-2099) 06037402303 Pomona 24.6 93.8 06037402801 Pomona 29.0 100.6 06037402803 Pomona 29.0 100.6 06037408800 Pomona 24.6 93.8 06037433302 El Monte 27.2 111.9 06037433402 South El Monte 27.2 111.9 06037433801 South El Monte 19.8 91.7 06037434004 South El Monte 19.8 91.7 06037482502 Rosemead 24.8 98.5 06037501400 Whittier 13.9 72.3 06037503000 Whittier 11.3 58.8 06037530301 Los Angeles 14.6 76.3 06037530500 Los Angeles 14.6 76.3 06037530601 Los Angeles 14.6 76.3 06037530700 Los Angeles 8.7 44.6 06037530902 Los Angeles 14.6 76.3 06037531201 Los Angeles 8.7 44.6 06037531202 Los Angeles 8.7 44.6 06037531301 Los Angeles 8.7 44.6 06037531302 Los Angeles 8.7 44.6 06037531502 Los Angeles 14.6 76.3 06037531504 Los Angeles 14.6 76.3 06037531604 Los Angeles 14.6 76.3 06037532302 Los Angeles 8.7 44.6 06037532603 Huntington Park 5.7 31.0 06037532606 Huntington Park 5.7 31.0 06037532700 Los Angeles 5.7 31.0 06037532800 Los Angeles 3.1 12.9 06037532900 Los Angeles 3.1 12.9 06037533001 Los Angeles 5.7 31.0 06037533002 Los Angeles 5.7 31.0 06037533103 Huntington Park 5.7 31.0 06037533104 Huntington Park 5.7 31.0 06037533105 Huntington Park 5.7 31.0 06037533106 Huntington Park 5.7 31.0 06037533107 Huntington Park 5.7 31.0 06037533201 Huntington Park 5.7 31.0 06037533300 Maywood 5.7 31.0 06037533401 Maywood 5.7 31.0 06037533502 Huntington Park 5.7 31.0 106 Table C.3 continued from previous page census tract id city Baseline number of extreme heat days (1961-1990) Projected number of extreme heat days (2070-2099) 06037533503 Huntington Park 5.7 31.0 06037533601 Bell 5.7 31.0 06037533602 Bell 5.7 31.0 06037533603 Bell 5.7 31.0 06037533701 Maywood 13.7 66.5 06037533702 Maywood 13.7 66.5 06037533703 Maywood 13.7 66.5 06037533803 Bell 5.7 31.0 06037533901 Bell 13.7 66.5 06037534102 Bell 13.7 66.5 06037534201 Bell 13.7 66.5 06037534202 Bell 13.7 66.5 06037534203 Bell 13.7 66.5 06037534301 Bell 5.7 31.0 06037534404 Bell 5.7 31.0 06037534405 Bell 5.7 31.0 06037534406 Bell 13.7 66.5 06037534501 Huntington Park 5.7 31.0 06037534802 Huntington Park 5.7 31.0 06037534803 Huntington Park 5.7 31.0 06037534804 Huntington Park 5.7 31.0 06037534900 Los Angeles 5.7 31.0 06037535001 Los Angeles 3.1 12.9 06037535002 Los Angeles 5.7 31.0 06037535101 Los Angeles 3.1 12.9 06037535102 Los Angeles 3.1 12.9 06037535200 Los Angeles 3.1 12.9 06037535300 Los Angeles 5.7 31.0 06037535400 Los Angeles 5.7 31.0 06037535501 South Gate 5.7 31.0 06037535503 South Gate 5.7 31.0 06037535604 South Gate 5.7 31.0 06037535606 South Gate 5.7 31.0 06037535702 South Gate 5.7 31.0 06037535803 South Gate 5.7 31.0 06037535804 South Gate 5.7 31.0 06037536104 South Gate 13.7 66.5 06037540202 Lynwood 3.6 16.5 06037540203 Lynwood 3.6 16.5 06037540300 Lynwood 5.7 31.0 06037540400 Los Angeles 3.6 16.5 06037540501 Lynwood 3.6 16.5 06037540502 Lynwood 3.6 16.5 06037540600 Los Angeles 3.6 16.5 06037540700 Los Angeles 2.7 8.2 06037540901 Los Angeles 2.7 8.2 06037541001 Gardena 2.7 8.2 06037541100 Compton 2.7 8.2 06037541300 Compton 3.6 16.5 06037541400 Compton 3.6 16.5 107 Table C.3 continued from previous page census tract id city Baseline number of extreme heat days (1961-1990) Projected number of extreme heat days (2070-2099) 06037541500 Compton 3.6 16.5 06037541603 Compton 3.6 16.5 06037541604 Compton 3.6 16.5 06037541605 Compton 3.6 16.5 06037541606 Compton 3.6 16.5 06037541700 Lynwood 3.6 16.5 06037541801 Lynwood 3.6 16.5 06037542000 Compton 3.6 16.5 06037542103 Compton 3.6 16.5 06037542105 Compton 3.6 16.5 06037542106 Compton 3.6 16.5 06037542401 Compton 3.6 16.5 06037542502 Compton 3.6 16.5 06037542601 Compton 3.6 16.5 06037542602 Compton 3.6 16.5 06037542700 Compton 3.6 16.5 06037542900 Compton 2.7 8.2 06037543201 Compton 3.6 16.5 06037543202 Compton 3.6 16.5 06037553701 Paramount 3.6 16.5 06037553702 Paramount 3.6 16.5 06037554204 Bell ower 5.9 29.5 06037570301 Long Beach 3.6 16.5 06037570303 Long Beach 3.9 16.5 06037570304 Long Beach 3.9 16.5 06037570603 Long Beach 6.0 26.0 06037571600 Long Beach 6.0 26.0 06037572301 Long Beach 3.9 16.5 06037572500 Long Beach 3.9 16.5 06037572800 Long Beach 4.2 17.4 06037572900 Long Beach 4.2 17.4 06037573002 Long Beach 4.2 17.4 06037573004 Long Beach 4.2 17.4 06037573201 Long Beach 4.2 17.4 06037573202 Long Beach 4.2 17.4 06037573300 Long Beach 6.8 28.1 06037575102 Long Beach 6.8 28.1 06037575201 Long Beach 6.8 28.1 06037575202 Long Beach 6.8 28.1 06037575300 Long Beach 4.2 17.4 06037575401 Long Beach 4.2 17.4 06037575402 Long Beach 4.2 17.4 06037575801 Long Beach 4.2 17.4 06037575802 Long Beach 4.2 17.4 06037575803 Long Beach 4.2 17.4 06037576200 Long Beach 4.2 17.4 06037576301 Long Beach 4.2 17.4 06037576302 Long Beach 6.8 28.1 06037576401 Long Beach 6.8 28.1 06037576402 Long Beach 6.8 28.1 108 Table C.3 continued from previous page census tract id city Baseline number of extreme heat days (1961-1990) Projected number of extreme heat days (2070-2099) 06037576403 Long Beach 6.8 28.1 06037576901 Long Beach 6.8 28.1 06037576903 Long Beach 6.8 28.1 06037600100 Los Angeles 3.1 12.9 06037600201 Los Angeles 3.1 12.9 06037600202 Los Angeles 3.1 12.9 06037600303 Los Angeles 3.1 12.9 06037600304 Los Angeles 3.1 12.9 06037600602 Inglewood 1.8 5.6 06037600902 Inglewood 1.8 5.6 06037600911 Inglewood 1.8 5.6 06037600912 Inglewood 1.8 5.6 06037601001 Inglewood 1.8 5.6 06037601100 Inglewood 1.8 5.6 06037601202 Inglewood 1.8 5.6 06037601211 Inglewood 1.8 5.6 06037601302 Inglewood 1.8 5.6 06037601303 Inglewood 1.8 5.6 06037601401 Inglewood 1.9 5.6 06037601501 Inglewood 1.8 5.6 06037601502 Inglewood 1.8 5.6 06037601801 Inglewood 1.8 5.6 06037601900 Inglewood 1.8 5.6 06037602002 Hawthorne 1.7 5.5 06037602003 Inglewood 1.8 5.6 06037602004 Inglewood 1.8 5.6 06037602103 Hawthorne 1.7 5.5 06037602105 Hawthorne 1.7 5.5 06037602504 Hawthorne 1.7 5.5 06037602509 Hawthorne 1.7 5.5 06037602801 Los Angeles 2.7 8.2 06037603001 Gardena 2.7 8.2 06037603004 Gardena 2.7 8.2 06037603006 Gardena 2.7 8.2 06037603704 Hawthorne 1.7 5.5 06037603802 Lawndale 1.7 5.5 06037603900 Lawndale 1.7 5.5 06037604001 Lawndale 1.7 5.5 06037604002 Lawndale 1.7 5.5 06037604100 Lawndale 1.7 5.5 06037650604 Torrance 2.5 8.3 06037701100 Los Angeles 1.4 6.8 06037900102 Lancaster 57.2 126.7 06037910501 Palmdale 58.2 132.0 06037920011 Santa Clarita 21.1 88.4 06059062622 Laguna Woods 3.5 19.8 06059062625 Laguna Hills 1.1 5.4 06059062646 Laguna Woods 1.1 5.4 06059063604 Costa Mesa 1.7 9.4 06059063605 Costa Mesa 1.3 6.6 109 Table C.3 continued from previous page census tract id city Baseline number of extreme heat days (1961-1990) Projected number of extreme heat days (2070-2099) 06059063701 Costa Mesa 1.3 6.6 06059063808 Costa Mesa 1.3 6.6 06059074300 Santa Ana 6.3 34.3 06059074406 Santa Ana 5.9 36.4 06059074501 Santa Ana 6.3 34.3 06059074701 Santa Ana 5.9 31.3 06059074702 Santa Ana 5.9 31.3 06059074805 Santa Ana 5.9 31.3 06059074806 Santa Ana 5.9 31.3 06059074901 Santa Ana 5.9 31.3 06059074902 Santa Ana 5.9 31.3 06059075003 Santa Ana 6.0 35.5 06059075004 Santa Ana 5.9 36.4 06059075201 Santa Ana 6.0 35.5 06059099402 Huntington Beach 3.9 21.7 06059099510 Seal Beach 7.0 30.6 06065042405 Moreno Valley 25.3 97.4 06065043401 Hemet 85.8 148.1 06065044404 Idyllwild 9.5 70.8 06071003700 Rialto 59.3 129.6 06071004103 San Bernardino 51.8 122.3 06071004104 San Bernardino 51.8 122.3 06071004302 San Bernardino 59.3 129.6 06071004700 San Bernardino 51.8 122.3 06071004800 San Bernardino 59.3 129.6 06071004900 San Bernardino 59.3 129.6 06071005400 San Bernardino 57.8 127.6 06071005500 San Bernardino 57.8 127.6 06071005600 San Bernardino 65.7 135.4 06071005701 San Bernardino 65.7 135.4 06071005800 San Bernardino 65.7 135.4 06071006401 San Bernardino 65.7 135.4 06071006402 San Bernardino 65.7 135.4 06071006500 San Bernardino 65.7 135.4 06071007407 San Bernardino 40.4 110.3 06071007601 San Bernardino 63.0 131.7 06071009116 Adelanto 58.6 127.4 06071009201 Wrightwood 1.0 14.6 06071009400 Barstow 87.3 153.1 06071009800 Victorville 58.6 127.3 06071010423 Landers 55.7 131.6 06037300100 La Crescenta 3.4 39.6 110

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Developing high-resolution spatiotemporal methods to model and quantify water use for energy

PDF

Residential electricity demand in the context of urban warming: leveraging high resolution smart meter data to quantify spatial and temporal patterns…

PDF

Developing frameworks to quantify the operational and environmental performance of energy systems within the context of climate change

PDF

Evaluating energy consuming behaviors and the sufficiency of urban systems in the context of extreme heat hazards

PDF

Identifying and mitigating the effects of urban heat islands in California

PDF

Beyond greenhouse gases and towards urban-scale climate mitigation: understanding the roles of black carbon aerosols and the urban heat island effect as local to regional radiative forcing agents

PDF

Using demand-side management for decarbonization: developing methods to quantify the impact of altering electricity consumption patterns

PDF

From tugboats to trees: investigating the coupled systems of urban air pollution and meteorology

PDF

Evaluating the role of energy system decarbonization and land cover properties on urban air quality in southern California

Asset Metadata

Creator Chen, Mo (author)

Core Title Investigating the role of climate in affecting residential electricity consumption through high spatiotemporal resolution observations

School Viterbi School of Engineering

Degree Doctor of Philosophy

Degree Program Engineering (Environmental Engineering)

Publication Date 08/05/2020

Defense Date 06/04/2020

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag air conditioning,Climate,climate change,electricity-temperature sensitivity,energy-climate nexus,extreme heat events,OAI-PMH Harvest,residential electricity consumption,smart meter,Southern California,spatiotemporal resolution,urban warming,vulnerable population

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Ban-Weiss, George Alexander (committee chair), Sanders, Kelly Twomey (committee chair), Dilkina, Bistra (committee member)

Creator Email anthonychen118@gmail.com,chen921@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-360779

Unique identifier UC11665693

Identifier etd-ChenMo-8868.pdf (filename),usctheses-c89-360779 (legacy record id)

Legacy Identifier etd-ChenMo-8868.pdf

Dmrecord 360779

Document Type Dissertation

Rights Chen, Mo

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA