University of Southern California Dissertations and Theses

Integration of topographic and bathymetric digital elevation model using ArcGIS interpolation methods: a case study of the Klamath River Estuary

(USC Thesis Other)

Integration of topographic and bathymetric digital elevation model using ArcGIS interpolation methods: a case study of the Klamath River Estuary

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content i

Integration of Topographic and Bathymetric Digital Elevation Model using ArcGIS
Interpolation Methods:

A Case Study of the Klamath River Estuary

by

Rachel R. Rodriguez

A Thesis Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
MASTER OF SCIENCE
(GEOGRAPHIC INFORMATION SCIENCE AND TECHNOLOGY)

May 2015

Copyright 2015 Rachel R. Rodriguez
ii

Dedication
I would like to dedicate this work to my wonderfully supportive and understanding
Dog, Hurley. For all the missed sunny days we spend indoors in front of the computer not
chasing Kong.

iii

Acknowledgements
I would like to sincerely thank Dr. Su Jin Lee and Mitra Parineh, for their continual
commitment to my thesis process. I would also like to thank Dr. Jennifer Swift and Dr.
Tarek Rashed for their excellent insight and guidance.
Credit also goes to Yurok Tribe’s GIS Program for their support of my educational
goals and the Bathymetry data collection.

iv

Table of Contents
Dedication ........................................................................................................................... ii
Acknowledgements ............................................................................................................ iii
List of Tables ..................................................................................................................... vi
List of Figures ................................................................................................................... vii
Special Terms and Acronyms .......................................................................................... xiii
Abstract ........................................................................................................................... xvii
Chapter 1: Introduction ....................................................................................................... 1
Chapter 2: Background ....................................................................................................... 4
2.1 Elevation Measurement ............................................................................................. 5
2.1.1 LiDAR ................................................................................................................ 5
2.1.2 Sonar ................................................................................................................... 6
2.2 Spatial Statistics Influence ........................................................................................ 7
2.2.1 Normal Distribution ............................................................................................ 7
2.2.2 Data Stationarity ................................................................................................. 8
2.2.3 Trends ................................................................................................................. 9
2.2.4 Spatial Auto-correlation ..................................................................................... 9
2.3 Surface Interpolation Algorithms ............................................................................ 10
2.3.1 Inverse Distance Weighting (IDW) .................................................................. 11
2.3.2 Natural Neighbor (NN) ..................................................................................... 11
2.3.3 Spline ................................................................................................................ 12
2.3.4 Kriging .............................................................................................................. 13
2.3.5 Empirical Bayesian Kriging (EBK) .................................................................. 13
2.3.6 Topo to Raster .................................................................................................. 14
2.4 ArcGIS Extensions for Interpolation....................................................................... 15
2.4.1 Spatial Analyst Extension ................................................................................. 17
2.4.2 Geostatistical Analyst Extension ...................................................................... 17

v

Chapter 3: Related Work .................................................................................................. 21
3.1 Spatial Statistics Importance ................................................................................... 21
3.2 Interpolation Comparison ........................................................................................ 23
3.3 Topographic/Bathymetric DEM integration ........................................................... 26
Chapter 4: Methodology ................................................................................................... 29
4.1 Klamath River Study Area ...................................................................................... 30
4.2 Datasets ................................................................................................................... 31
4.2.1 Statistics ............................................................................................................ 35
4.3 Interpolation ............................................................................................................ 47
4.4 Illustrating Results................................................................................................... 48
Chapter 5: Results ............................................................................................................. 51
5.1 Parameter Variation................................................................................................. 51
5.1.1 IDW .................................................................................................................. 52
5.1.2 Spline ................................................................................................................ 53
5.1.3 Kriging .............................................................................................................. 57
5.1.4 EBK .................................................................................................................. 62
5.1.5 Natural Neighbor & Topo to Raster ................................................................. 64
5.2 Results of Multiple Interpolation Methods ............................................................. 65
5.2.1 Project Area 1 ................................................................................................... 66
5.2.2 Project Area 2 ................................................................................................... 73
5.2.3 Project Area 3 ................................................................................................... 81
5.2.4 Project Area 4 ................................................................................................... 89
5.2.5 Project Area 5 ................................................................................................... 97
5.3 Results Summary................................................................................................... 104
Chapter 6: Discussion and Conclusions .......................................................................... 106
Bibliography ................................................................................................................... 115
Appendix A: .................................................................................................................... 119

vi

List of Tables
Table 1: ArcGIS 10.2 Extensions and Interpolation Comparison. ....................................19
Table 2: Statistical comparison of the entire Project Area to the subset dataset used to
create the interpolated surfaces. .........................................................................................36
Table 3: Table illustrating the differences between the IDW interpolation parameters
in the Tool, the Wizard's default and the optimized settings. ............................................53
Table 4: Table illustrating the differences between the Spline Regular interpolation
parameters in the Tool, the Wizard's default and the optimized settings...........................56
Table 5: Parameter variation of the Spline with Tension interpolation. ............................56
Table 6: Kriging parameter variation: O - ordinary, P - Prediction. ............................ 60-61
Table 7: Table of EBK interpolation parameters. ..............................................................63
Table 8: Statistical distribution of the raster surfaces created from the various
interpolation methods in Project Area 1. ...........................................................................67
Table 9: Table of the statistical distribution in Project Area 2. .........................................75
Table 10: Table of the statistical distribution in Project Area 3. .......................................82
Table 11: Table of the statistical distribution in Project Area 4. .......................................89
Table 12: Table of the statistical distribution in Project Area 5. .......................................97

vii

List of Figures
Figure 1: Map illustrating Study area. .................................................................................5
Figure 2: Screenshot of the Spatial Analyst toolbox (left) and the Geostatistical
Wizard (right). These two extensions provide an easy-to-use accessibility of point to
Raster Grid Surface. ...........................................................................................................18
Figure 3: Map of the Klamath River Estuary with the selected Project Areas
highlighted. ........................................................................................................................30
Figure 4: (Left) Individual LiDAR point measurements. (Right) Density map of the
LiDAR measurements averaging 2.28 points per square foot. ..........................................33
Figure 5: (Left) Individual Sonar point measurements. (Right) Density map of the
Sonar measurements averaging 9.68 points per square foot. .............................................33
Figure 6: Maps of Project Area(s) with LiDAR - Sonar Measurements with the Subset
data used in the creation of the surface profiles. ................................................................37
Figure 7: Graph illustrating the distribution of the merged LiDAR and Sonar points
used in the creation of the surface profiles for Project Area 1. ..........................................38
Figure 8: Graph illustrating the distribution of the merged LiDAR and Sonar points
used in the creation of the surface profiles for Project Area 2. ..........................................40
Figure 9: Graph illustrating the distribution of the merged LiDAR and Sonar points
used in the creation of the surface profiles for Project Area 3. ..........................................42
Figure 10: Graph illustrating the distribution of the merged LiDAR and Sonar points
used in the creation of the surface profiles for Project Area 4. ..........................................44
viii

Figure 11: Graph illustrating the distribution of the merged LiDAR and Sonar points
used in the creation of the surface profiles for Project Area 5. ..........................................46
Figure 12: IDW interpolation prompts, accessed via the Spatial Analyst Toolbox (left)
and the Geostatistical Wizard (right). ................................................................................52
Figure 13: Screen shot of the Spline interpolation prompt, accessed via the Spatial
Analyst Toolbox.................................................................................................................54
Figure 14: Screen shot of the Geostatistical Wizard prompts for the Spline Regular
(left) and the Spline Tension (right). ..................................................................................55
Figure 15: Screen shot of the Kriging interpolation window from the Spatial Analyst
toolbox. The sub-window of the advanced parameters has been open to illustrate the
further functionality in not used by default........................................................................58
Figure 16: Screen Shots of the Kriging interpolation, third step in using the
Geostatistical Wizard for this interpolation methodology. Semivariogram on the left
is the default Wizard values, on the right are the optimized setting. .................................58
Figure 17: Screen shot of the Geostatistical Wizard's fourth step for the Kriging
interpolation. The search neighborhood is the same for both the default Wizard and
the optimization setting. .....................................................................................................59
Figure 18: Screen capture of the EBK interpolation accessed via the Geostatistical
Wizard. ...............................................................................................................................62
Figure 19: Screen capture of the EBK interpolation accessed via the Geostatistical
Analyst Toolbox.................................................................................................................63
ix

Figure 20: Screen capture of the Topo to Raster tool available in the Spatial Analyst
Extension............................................................................................................................64
Figure 21: Surface raster maps of interpolation algorithms. Note the differences
between the different interpolation algorithms. .................................................................68
Figure 22: Profile graphs of the interpolation methods completed for Project area 1,
Transect A. All graphs represent the same geographic area. .............................................68
Figure 23: Profile graphs of transect A for Project Area 1. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project
Area. ...................................................................................................................................69
Figure 24: Profile graphs of transect B for Project Area 1. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project
Area. ...................................................................................................................................70
Figure 25: Profile graphs of transect C for Project Area 1. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project
Area. ...................................................................................................................................71
Figure 26: Difference maps, between the tool and the Wizard..........................................73
Figure 27: Surface raster maps of interpolation algorithms. Note the differences
between the different interpolation algorithms. .................................................................75
Figure 28: Profile graphs of transect A for Project Area 2. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project
Area. ...................................................................................................................................76
x

Figure 29: Profile graphs of transect B for Project Area 2. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project
Area. ...................................................................................................................................77
Figure 30: Profile graphs of transect C for Project Area 2. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project
Area. ...................................................................................................................................78
Figure 31: Maps illustrating the difference between the surfaces created by the Spatial
Analyst tool and the Geostatistical Analyst Extensions. ...................................................81
Figure 32: Surface raster maps of interpolation algorithms. Note the differences
between the different interpolation algorithms. .................................................................83
Figure 33: Profile graphs of transect A for Project Area 3. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project
Area. ...................................................................................................................................84
Figure 34: Profile graphs of transect B for Project Area 3. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project
Area. ...................................................................................................................................85
Figure 35: Profile graphs of transect C for Project Area 3. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project
Area ....................................................................................................................................86
Figure 36: Surface raster maps of interpolation algorithms. Note the differences
between the different interpolation algorithms. .................................................................90
xi

Figure 37: Profile graphs of transect A for Project Area 4. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project
Area ....................................................................................................................................93
Figure 38: Profile graphs of transect B for Project Area 4. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project
Area ....................................................................................................................................94
Figure 39: Profile graphs of transect C for Project Area 4. Each graph represents
interpolation algorithms, and all graphs represent the same transect within the Project
Area. ...................................................................................................................................95
Figure 40: Difference maps comparing the surfaces created from the Spatial Analyst
and the Geostatistical Analyst Extension. ..........................................................................96
Figure 41: Surface raster maps of interpolation algorithms. Note the differences
between the different interpolation algorithms. .................................................................98
Figure 42: Profile graphs of transect A for Project Area 5. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project
Area. ...................................................................................................................................99
Figure 43: Profile graphs of transect B for Project Area 5. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project
Area. .................................................................................................................................100
Figure 44: Profile graphs of transect C for Project Area 5. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project
Area ..................................................................................................................................101
xii

Figure 45: Difference maps of the Interpolation algorithms comparing the surfaces
created by the Spatial Analyst and Geostatistical analyst. ...............................................103
Figure 46: California Coastal Conservancy (CCC) LiDAR Bare Earth DEM
subtracted from the interpolated surface. The differences are repersented by the color
differences. .......................................................................................................................109
Figure 47: Two examples of the resulting Cross comparison. Taking the interpolation
method/extension combination and subtracting it to each interpolation - extension
interpolated raster surface created for Project Area 1 within the Study Area. Top
image shows the cross comparison of the IDW interpolation used with the Spatial
Analyst extension. Bottom image shows the cross comparison of the Natural NN
interpolation methodology. ..............................................................................................110
Figure 48: Elevation difference between the Measured points LiDAR and Sonar
Dataset to the Esri interpolated surfaces. As the difference illustrated by the X-Axis,
is closer to 0, the more similar the overall interpolated datasets fit the measured
points. ...............................................................................................................................112
Figure 49: Cross comparison. Taking the interpolation method/extension combination
and subtracting it to each interpolation - extension interpolated raster surface created
for Project Area 1 within the Study Area................................................................. 119-121

xiii

Special Terms and Acronyms
ArcGIS – Product developed by Esri, currently the leading GIS software internationally.
Bathymetry – The below water ground surface, usually measured as a depth value from
the reference water.
Covariance – A scaled version of correlation by illustrating the correlation between the
direction and distance of the pairs of values and their locations. In graph form, the
y-axis is representing the similarity of values and the x-axis represents the
distance relationship between them.
DEM (Digital Elevation Model) – A raster data model recorded as a regular grid
representing the height values of elevation for the (earth’s) surface.
Deterministic – In interpolation, this refers to the input data having a known influence
on the interpolation model algorithm based on the dataset’s statistical distribution.
EBK (Empirical Bayesian Kriging) – A deterministic interpolation method, which
automates the parameter values compared to traditional Kriging interpolation
methods.
ESDA (Exploratory Spatial Data Analysis) – Functionality available in Esri’s
Geostatistical Analyst Extension to investigate the data points prior to
interpolation, which is highly suggested.
Esri (Environmental Systems Research Institute) - The world’s leading GIS
software developer. Creator of ArcGIS software suite.
Extension – Additional software sold by Esri for use with the ArcGIS platform, includes
many additional tools and functionality not included in the standalone ArcGIS
software.
Geospatial Analyst – An Extension to ArcGIS software with additional functionality for
Geostatistical data review and analysis
GIS (Geographic Information System) – A computer based system to manage
geographic information for use in visualization, analysis and modeling.
GPS (Global Positioning Systems) - A network of space-based and ground-based
equipment used in the mathematical calculation of a physical location.
Inverse Distance Weighting (IDW) - A Deterministic interpolation method it which
uses the exact data variables when creating the surface. Available in both Spatial
Analyst and Geostatistical Analyst Extensions of ArcGIS.
xiv

Interpolation – Estimation of surface values at unsampled locations based on known
values of neighboring points.
Interpolation Algorithms - Using known data points to with a mathematical algorithm
in order to create a surface.
Kriging – An algorithm used in interpolation of surfaces, which using the surrounding
measured values are weighted to derive a predicted value for an unmeasured
location.
LiDAR (Light Detection and Ranging) – A remote sensing technique used in the detailed
collection of surface measurements.
Natural Neighbor (NN) - An algorithm used in interpolation of surfaces, which is based
on Voronoi (Thiessen) polygons of a neighboring points which weights them
based on proportionate areas to interpolate a value.
Normal distribution – Where the values of the dataset are clustered overwhelmingly
near the mean value, also called a bell-shaped curve. Usually visualized on a
histogram.
Normal QQ (Normal Quantile-Quantile) – Usually represented in a graph of the point
data to compare the distribution of the data to a standard normal distribution.
Optimize (Parameter Optimization) - Optimization process for each model is used to
change the parameter(s) in order to minimize the mean square error.
Parameter – Values or settings within a tool or Wizard which can be modified by the
user or maintain their default setting.
Radical Based Function (Spline) - A deterministic algorithm which interpolates to
ensure the estimates values are exact using a mathematical function that
minimizes overall surface curvature, resulting in a smooth surface that passes
exactly through the input points.
Raster – A cell based data model used to store information for faster processing and
analysis within a computerized environment. Sometimes referred to as a grid.
RMSE (Root Mean Square Error, RMS error) - a measure of cross validation in
which comparisons are performed from the predicted values to the observed
values for values.
Semivariogram – Usually represented in a graph or chart form illustrates and quantifies
the amount of auto-correlation.
xv

Sonar (Sound Navigation and Ranging) – a remote sensor used in the collection of
underwater elevation measurements. Common types include Multi-beam r,
Single-beam and Side Scan Sonar.
Spatial Analyst – An Extension to ArcGIS software with additional functionality for
analysis of spatial features, one such spatial feature is the interpolation from
points to a raster surface.
Spatial Auto-correlation (Tobler's First Law of Geography) – Principal within GIS
that features that are closer together have a greater relationship than feature
further apart.
Spatial Statistics (Geostatistics) – Using the location or value statistics in different types
of analyses, including pattern analysis, shape analysis, surface modeling and
surface prediction, spatial regression, statistical comparisons of spatial datasets,
statistical modeling and prediction of spatial interaction, and more
Spline Regular - A type of spline interpolation. Regularized Spline modifies the
minimization criteria leading to smooth, gradually changing surface with values
that may lie outside the sample data range
Spline Tension - A type of spline interpolation. Tension Spline modifies the
minimization criteria leading to a less smooth surface with values more closely
constrained by the sample data range.
Stationarity - the statistical relationship between two points which depend on the
distance between them. Usually viewed by plotting the points on Voronoi map to
illustrate the change.
Surface – A raster grid model consisting of values which are representative of the
elevation of the earth with either build environment or the bare elevation.
Tool – A user interface within ArcGIS, assessed by the ArcGIS Toolbox(s) to perform
complex procedures.
Topo to Raster - An algorithm used in interpolation of surfaces, which is designed for
the creation of hydrologically correct DEM. This interpolation can take a variety
of data models as inputs rather than just points to create the surface.
Topography – The ground surface in an arrangement of the natural and artificial
physical features of an area, usually represented in a raster surface.
Transects (Profile) - a straight line or narrow section through an across the earth's
surface, along which observations illustrated in profile graphs. Showing the 3D
representation of the elevation variation.
xvi

Trend(s) – Data being clustered in any form within a dataset. Procedures are available
for removing trends within a dataset such as Box-Cox, arcsine, and log
transformations which turn the data to a normal distribution.
USGS (U.S. Geological Survey) – An independent fact-finding agency within the
Department of the Interior.
Wizard - A user interface within ArcGIS, which walks the user through a multiple step
process in order to perform complex procedures.

xvii

Abstract
High quality topographic (land elevation) and bathymetric (water depth) data is targeted
by the USGS and other Federal agencies as a need for update and modernization,
particularly with the rapidly advancing technological innovations for use in modeling
hydrological and environmental changes. Esri’s ArcGIS provides advanced and various
options to interpolate surfaces using two ArcGIS Extensions: Spatial Analyst and
Geostatistical Analyst. These extensions provide access to advanced mathematical
algorithms used in the interpolation of measured points into an elevation surface, through
a user-friendly interface with pre-defined, yet highly technical input parameters.
Using Light detection and Ranging (LIDAR) elevation measurements and Single
Beam Sonar on the Klamath River Estuary, this project compares interpolation methods
provided by ArcGIS in the Spatial Analyst and Geostatistical Analyst Extensions, in
order to determine how varying the parameter settings affect the resulting surfaces. This
case employs seven commonly use interpolation algorithms: Inverse Distance Weighting,
Natural Neighbor, Spline Regular, Spline Tension, Kriging, Empirical Bayesian Kriging,
and Topo to Raster, all of which can be used in Digital Elevation Model (DEM) surface
creation. Understanding the differences between the two extensions and modifying
parameters in each interpolation algorithm results in statistically reliable elevation
surfaces. The results prove that modifying the default interpolation parameters to fit the
statistical variability, which is completed by the optimization of the Geostatistical
Analyst Wizard, improves the functional use of the study area raster surface.
xviii

Chapter 1 introduces the study’s necessity. Chapter 2 describes the technical
background, and provides all of the necessary information for understanding the results.
Chapter 3 provides the related works. Chapter 4 describes the methodology and explains
how to interpret the results. Chapter 5 presents the results from the case study on the
Klamath River. Chapter six provides the conclusion and discussion.
1

Chapter 1: Introduction
Elevation is a fundamental component in any geographic study, utilized as a concrete
measurement of the Earth’s surface above sea level. Thus for this study, Klamath River
Estuary which currently uses a DEM developed in the 1970’s requires an more timely
and reliable elevation surface (Stoker, et al. 2013). Teams of surveyors historically
collected elevation measurements, performing what is known as geodetic leveling.
Geodetic leveling measures the relative elevation from one area (usually the ocean or
benchmark) to the area of interest (Stoker, et al. 2013). Today, this time-intensive and
rather inaccurate process has been reduced to a series of fly-overs from remote, passive
or active sensors to acquire the measurements. The traditional definition of elevation is
considered the above water measurement of topographical features, but in order to
enhance the knowledge of global environmental factors, underwater elevation or depth
must be measured as well. Underwater elevation measurements, also known as
bathymetry, historically used line measurements to obtain depth values. Today passive or
active sensors are utilized by boats and even aircraft to measure elevation (Kearns and
Breman 2010). Integrating these two elevation datasets (topography and bathymetry) into
a seamless Topographic/Bathymetric Digital Elevation Model, has recently been targeted
as a spatial need with little guide for the procedure (NOAA Coastal Services Center
2007, Bernstein, et al. 2011, Evans 2013).
The sensors used to obtain the measurements of elevation or depth can vary, but are
generally standardized. Topography can be measured by either Synthetic Aperture Radar
2

(SAR) or LiDAR. Airborne LiDAR has become the industry standard due to its high
accuracy and precision as well as the wealth of information (Newkirk 2005). Bathymetry
collection technology has not changed as rapidly as elevation technology, but recently a
variation of the LiDAR technology is being utilized as well as multi and single beam
Sonar (Sound Navigation and Ranging) transducers (Kearns and Breman 2010). When
determining what technology to use for elevation measurement collection, factors such as
expense, availability, and quality of required data points are all integral. The most
common sensors utilized for obtaining topography and bathymetry is LiDAR and multi
beam sonar. These sensors provide point elevations at varying accuracies, level of detail
and measurement spacing. Using known values to estimate the values of areas without
measurements, these measured points are then interpolated into a raster grid surface for
use in modeling and mapping (Stoker, et al. 2013).
Digital Elevation Models (DEM) are used in many types of analytical studies related
to many topics. The importance of deriving the topography and bathymetry
measurements into seamless DEMs can be contextualized by hydrodynamics for tsunami,
flood or drought modeling, sea-level change, as well as the geophysical erosion,
subsidence and coseismic uplift (Goodchild 2000, NOAA Coastal Services Center 2007,
Stoker, et al. 2013). These processes and events rely immensely on accurate, timely and
precise elevation measurements of topography and bathymetry to perform such studies.
Traditional DEM’s only used topographic measurements to interpolate the earth’s
surface, leaving underwater elevation as a single constant value. Integrating both
3

topography and bathymetry into a single seamless dataset can enhance the level of
understanding of their environment for geospatial managers in both coastal and riverine
communities (Evans 2013, Stoker, et al. 2013). Although obtaining the most precise and
accurate DEM using any technology is difficult, utilizing the highest density elevation
measurement point spacing as well as the most recent data collection can mitigate other
potential errors arising from interpolation. The choice of interpolation algorithms used
when creating the Topographic/Bathymetric DEMs cause variation in the raster surface.
Technology and methodology development revolutionizes the geospatial field every
few years with version releases of Esri’s ArcGIS Geostatistical Analyst Extension. The
analytical ability to generate interpolated surfaces has increased, yet many professionals
are still relying on the Spatial Analyst Extension. This project examines the variation
between the interpolation methods provided by ArcGIS in its Spatial Analyst and
Geostatistical Analyst Extensions in order to determine how varying the parameter
settings affects the resulting surfaces. This Klamath River Estuary use case will employ
seven commonly use interpolation algorithms, including Inverse Distance Weighting,
Natural Neighbor, Spline Regular, Spline Tension, Kriging, Empirical Bayesian Kriging,
and Topo to Raster in DEM surface creation. Understanding the extensions differences
and modifying the parameters in each interpolation algorithm results in statistically
reliable elevation surfaces.

4

Chapter 2: Background
The Klamath River Estuary, the study area, is located in Del Norte County of Northern
California (Figure 1). The Klamath River flows from Oregon where it encounters a series
of dams and impediments. There is also climatic variability between the headwaters and
the estuary that can cause river flow rate to change day to day and year to year. This
variability in hydrological flow can influence the deposition of sediments, which affect
the impact of tsunamis, storm surges, and sea level changes on the surrounding
community. The ability for the community to prepare depends on pre-disaster elevation
modeling using the most recent information. The currently available DEM from the
USGS, specifically from the study area, was created using synthetic aperture radar in the
1970’s (Stoker, et al. 2013) with unknown river flows. This unknown timing of data
acquisition and the DEM’s continued influence over all environmental planning of the
region can be improved by incorporating all available elevation information. Agencies
are now looking into a “whole environment” approach to resource and land management,
incorporating topographic and bathymetric elevation that can improve environmental
planning.
5

Figure 1: Map illustrating Study area.
2.1 Elevation Measurement
The technology for measuring elevation has improved with the utilization of
Global Positioning Systems (GPS). GPS and computerized mathematics have simplified
the process of obtaining point measurements of greater accuracy than ever before.
Centimeter accuracy and the ability to validate the measurements using real-time base
stations allow for overall confidence in the DEM creation from remote sensors.
2.1.1 LiDAR
LIDAR is a remote sensing technology traditionally deployed from an aircraft
which uses light in the form of a pulsed laser to measure distances to Earth. The laser,
pointed at a targeted area on the ground, transmits a beam of light and as the light is
reflected by the surface, it receives the signal back (Raber and Cannistra 2005).
Calculating the transmitted and received signals as well as the position and orientation
6

data generated from integrated GPS and inertial measurement unit systems. The result is
dense point measurements, which can be used to visualize the surface of the earth and all
its features (Newkirk 2005). Specialized sensors can also penetrate water to obtain
bathymetric information by modifying the laser, but because of the high cost, this
technology was not utilized in this project (NOAA Coastal services center 2014).
2.1.2 Sonar
Sound Navigation and Ranging (Sonar) technology uses sound waves similar to
that of LiDAR, where it transmits a sound pulse and receives a signal once reached a
feature below the water (Breman 2010). The calculation of information from the position
and orientation data generated from integrated GPS and inertial measurement unit result
in high-density point measurements of depth. Sonar technology has several varieties
commonly used in bathymetric data collection such as Multi-beam Sonar, Single-beam
and Side Scan Sonar, each with their own applications for measurement collections.
Complicating the measurements is the level of reference of the water; most surveys
cannot be completed within the same tidal influences so manipulation and modification
of the data must be post-processed into the resulting dataset. For this project, the
bathymetry point dataset was collected using single beam sonar technology and was post-
processed to equalize the tidal discrepancies.
The point values from the LiDAR and single beam sonar sensors provide
individual elevation measurements. Both technologies are generally accepted as accurate
and commonly used in DEM surface creation. The surfaces are created by estimating the
7

values between the measured points, using the theory of spatial auto-correlation which
identifies that values closer together are more likely to be similar than those that are
farther apart (Stoker, et al. 2013). The estimation of the values in-between the
measurements is known as interpolation. The interpolation method can be based on a
mathematical formula or a series of formulas; all points are influenced by the measured
point elevations contained in the input datasets and their statistical variation.
2.2 Spatial Statistics Influence
Spatial Statistics is a concentration, of GIS, which is often overlooked (ESRI
2001) (Lynch and Krause 2014). The ability to analyze data for variability and integrity
should be required as part of the validation and verification of a dataset’s applicability.
The point data used in the interpolation of the surface feature, depending on the spatial or
statistical distribution, modifies the surface (Kravchenko 2003). Interpolation methods
can be accomplished with either sparse or dense datasets, although the more points
contained in an area, the more likely the surface is to mimic the true conditions. Datasets
need to be analyzed prior to interpolation in order to determine the suitability of the
algorithm or method. Factors such as normal distribution, Data Stationarity, Trends, and
spatial auto-correlation can sway the decision for an algorithm (Chiles and Delfiner 1999,
ESRI 2001, Bohling 2005, Kalkhan 2011).
2.2.1 Normal Distribution
In statistics, normal distribution is described as a case where the value points of
the dataset are clustered overwhelmingly near the mean value. This normal distribution is
8

also known as a bell-shaped curve, when visualized on a histogram (Berry n.d.). The
histogram must not contain values too distant from the mean and median values. Some
cases where the data is not normally distributed can be also seen in a Normal Quantile-
Quantile (Normal QQ) plot the existence of outliers, clusters or grouping of data can raise
an issue within some of the modeling algorithms by creating skewness in the distribution.
The skewness factor should be near zero, signifying the mean and median are near equal.
If the data is not normally distributed, then applying a transformation might resolve some
of the irregularities. Transformations modify the histogram by using either empirical or
Logarithm (Log) empirical in order to reduce the data’s skewness (Berry n.d., Kalkhan
2011, Lynch and Krause 2014).
2.2.2 Data Stationarity
If the data is stationarity the statistical relationship between two points depend on
the distance between them. If the data is not stationarity, transformations might stabilize
the variances or the utilization of certain interpolation algorithms, which do not require
stationarity. To visualize stationarity the Voronoi map is utilized, symbolized by entropy
or standard deviation. Entropy polygons are categorized using five classes based on a
natural grouping of data values (smart quantiles). The value assigned to a polygon is the
entropy that is calculated from the polygon and its neighbors—where entropy is the
proportion of polygons that are assigned to each class (Kalkhan 2011, Lynch and Krause
2014).

9

2.2.3 Trends
Identifying trends contained in a dataset is essential to understanding the changes
of the data values across the study area. The Trend analysis within ArcGIS can be
completed using Exploratory Spatial Data Analysis (ESDA) tool can be used to find
global trends within the dataset. Trends in the X, Y, or Z axis can be visualized by
manipulating the three dimensional (3D) graph. If the data shows signs of trends, there
might be problems with the dataset but they might be indistinguishable from auto-
correlation and anisotropy. Trend removal tools or a deterministic interpolation method
(for example, global or local polynomial), can reduce this global trend influence.
2.2.4 Spatial Auto-correlation
Spatial auto-correlation, also known as Tobler's First Law of Geography, is
visualized by the Semivariogram /Covariance Cloud tool in ArcMap. This tool examines
the local characteristics of spatial auto-correlation contained in a dataset. The
Semivariogram value is the squared difference between the values of two data points
making up a pair representing the y-axis and the distance between the two points on the
x-axis. With the Semivariogram the values variation should increase with distance. The
Covariance graph illustrates a scaled version of correlation by illustrating the correlation
between the direction and distance of the pairs of values and their locations, where the y-
axis is representing the similarity of values and the x-axis represents the distance
relationship. With the Covariance graph, the function should decreases with distance.
These two graphs are used in understanding the relationship the points have within their
geographic space (Kalkhan 2011, Lynch and Krause 2014). For this analysis, the
10

utilization of two datasets collected using two different methods and therefore two
different accuracy standards, processing requirements, and spatial constraints. Accuracy
is assumed from the data producer. If the interpolation method selected does not
generally conform to the parameters of the interpolation model, then the data will not be
as accurate. When analyzing the various point datasets, the identification of the spatial
trends, the spatial dependency, the point distribution of values and locations, and the
Stationarity of the dataset must be taken into consideration for use with the various
interpolation methods.
2.3 Surface Interpolation Algorithms
Interpolation is necessary due to the incomplete measurement of a feature within
an area. For the interpolation of ground surface, measurements are taken at a series of
locations, then using an interpolation algorithm; the elevation values between the points
are estimated. Interpolation of the ground surface is where the theory of spatial auto-
correlation, where points closer together are more similar to each other than points further
away. The basic two principals of interpolation algorithms based on 1) the definition of
neighborhood area and 2) the neighboring values to determine the mathematical
algorithm for the interpolation.
The choice of algorithm or formula used can influence the estimated elevation
surface, which is created from interpolation. Understanding the interpolation algorithms
and their assumptions can assist when creating any interpolated surface. Two
classifications of interpolation methodologies are Deterministic and Geostatistical.
11

Deterministic models estimate data relationships utilizing priori models, which are
directly based on the surrounding measured values. Geostatistical models utilize the
entire dataset to infer similar relationships for use in the neighborhood value
interpolation.
2.3.1 Inverse Distance Weighting (IDW)
Inverse Distance Weighting (IDW) is a type of Deterministic interpolation
method it uses the exact data variables when creating the surface. IDW initially created
by engineers for use in the petroleum field (Philip and Watson 1982, Watson and Philip
1985). The parameter setting for IDW is a power function, where the power function
illustrates the weight for the values that are proportional to the distance (Lu and Wong
2008). This method is widely used by Geoscientists, because of the ability to easily
explain the IDW methodology summarized by the implementation of spatial auto-
correlation (Kalkhan 2011). IDW is an appropriate interpolation methodology when
interpolating from point datasets that are non-clustered, spatially correlated, contain no
data outliers, and are non-directionally dependent. For elevation surfaces, IDW results in
flattening peaks and valleys unless these sites have a high point density.
2.3.2 Natural Neighbor (NN)
Natural Neighbor (NN) is a type of IWD deterministic function. NN can perform
both interpolation and extrapolation. It preserves localized minimum and maximum
values. This interpolation method was created by Robin Sibson as a simple interpolation
which aims to create a smooth elevation surface (Sibson 1981, D. Watson 1992). As
12

compared to traditional IWD, NN selects the closest points to create a convex hull around
the interpolation point, and weights their values by proportionate area which is also
known as Delauney Triangulation. Natural neighbor provides very little parameters for
modification, only the size of the raster cell. NN is not available in the Geostatistical
Analyst Extension due to its complexity of integrating various outsourced datasets, but it
can be accessed as a tool within the Spatial Analyst toolbox. Due to limited assumptions
made by the natural neighbor model, it generally works well with a variety of type of data
inputs including uneven points in distribution and density.
2.3.3 Spline
Spline, also known as a radical basis function, is another Deterministic
interpolation method that tries to ensure a smooth surface (Franke 1982, Mitas and
Mitasova 1988). This is achieved by finding a function that minimizes the tradeoff
between fitting the points and the smoothness of the surface. Both IDW and Spline rely
on spatial auto-correlation. In comparison to IDW, Spline is able to predict values above
and below the measured values, using the neighboring values as well as not requiring
quantification of spatial auto-correlation. Regular and Tension Spline are investigated in
this study. Regular Spline curve produces a flowing, sometimes irregular representation
of the measured points. Tension Spline curve is a flatter representation of the data,
staying closer to the measured values. The Kernel parameter controls the scales of
variation within the surface, where higher values equate to a smoother surface. Spline
functions produce smoother results when the measured points are not rapidly changing or
highly dense.
13

2.3.4 Kriging
Kriging is a complex Geostatistical interpolation methodology, discussed in
numerous scholarly journals (Royle, Clausen and Frederiksen 1981, Burrough 1986,
Heine 1986, Press, et al. 1988, Oliver 1990). The primary intention of Kriging is to
produce the most accurate surface not a smooth surface. Kriging is able to provide the
prediction of uncertainty. Kriging assumes spatially correlated points, Normally
Distributed, Stationarity of the data and does not represent a trend. This interpolation type
measures distances between all possible pairs of sample points and uses the information
to model the spatial auto-correlation in order to derive the surface. The Kriging
interpolation has several types, which slight variation, each have different application.
For this research, we will focus on Ordinary and Universal Kriging types. Ordinary
Kriging uses several potential Semivariograms models and Universal Kriging uses linear
with linear drift or quadratic drift. There are also more advanced parameters for lag size
and modifying the variograms parameters for major range, partial still, and nugget. The
ability for the Kriging tool also provides for modifying the search radius for either fixed
or variable as well as transformation and trend removal options. Kriging provides a
wealth of variability in the parameter choices and methodologies, but the outputs from
this interpolation method can held to a higher standard due to the uncertainty value the
model provides.
2.3.5 Empirical Bayesian Kriging (EBK)
Empirical Bayesian Kriging (EBK) is a type of Kriging methodology, which
allows for accurate predictions of non- Stationarity data (Chiles and Delfiner 1999, Pilz
14

and Spock 2007). This algorithm uses repeated simulations to obtain the best-fit
Semivariogram where the parameters are implemented using a restricted maximum
likelihood. EBK uses local models to capture small-scale changes in the data; it also does
not assume one model fits the entire dataset. The ability to predict the standard error is
more accurate than with traditional Kriging methods. The EBK model provides many
parameters, just as ordinary Kriging due to the complexity of the interpolation and
prediction methods. The model parameters such as the maximum number of points in the
local model, the model area overlap, and the number of simulations can be changed. The
search area parameters can also be modified for the type of search neighborhood, the
radius, maximum or minimum neighbors, sector type and angle. EBK has the ability to
manage highly complex data rich datasets with ease. This method using both global and
local scales for use in the interpolation has fostered EBK’s use in non-traditional
interpolations.
2.3.6 Topo to Raster
Topo to Raster uses ANUDEM methodology from Australian National
University, Canberra that the latest version of ANUDEM is included within ArcGIS.
Inputs for topo to raster include point elevations, contours, and others with hydrological
applicability. Topo to raster is not available in the Geostatistical Analyst due to its
complexity of integrating various outsourced datasets, but it can be accessed as a tool in
the Spatial Analyst toolbox (Liu, Zhang and Peterson 2009). Topo to Raster provides the
functionality of incorporating other types of geographic features, which can assist in the
creation of a DEM. The parameters required for topo to raster are mostly optional to
15

change from the default. Spot elevation, contours, cliffs, lakes, coasts and other boundary
information can be inputted and can be utilized in creation of the final raster surface. The
drainage enforcement within the tool will ensure a mostly hydrological correct surface.
For this project, point elevation features were used (ESRI 1995-2013 ).
Overall, these interpolation methodologies provide the basic requirements of testing
the integration of the topographic and bathymetric point elevation values in order to
create a single raster elevation surface. Although the point datasets vary in their
distribution, spatial dependency, spatial trends, and the stationarity, the surfaces created
using the interpolation methods can visualize the basic differences between the surfaces
created. If the point datasets were to meet the assumptions of the interpolation method,
that surface would provide the most accurate prediction for the elevation within that area.
Even with the Spatial Statistics taken into account for use in the interpolation, the
parameters within the interpolation still influence the resulting surface (Childs 2004).
Within ArcGIS, several interpolation methods have various extensions of how they can
be accessed. The utilization of an interpolation from one ArcGIS Extension may have
slightly different parameters, which causes changes to the resulting surface models when
compare to the same points interpolated using a different extension.
2.4 ArcGIS Extensions for Interpolation
Esri’s ArcGIS is the driving force of GIS development and implementation globally.
Within their software suite, elevation data can be represented in several ways. Elevation
is known as a functional surface, where for every X, Y coordinate pair, a Z or elevation
16

value exists (X, Y, Z). Within ArcGIS Elevation can be represented in points, contours,
Triangular Irregular Networks (TIN), or Grids. Points are the measured point elevation,
where the only X, Y, Z coordinate to have elevation is the measured locations. Another
representation of elevation is contours, which are representative lines of equal elevations.
TINs represent an initial representation of a functional surface, where triangles are
connecting the closest measured points. Raster grids are regular spaced cells of equal
size, where the cell represents a surface of equal elevation. Grids are currently the most
utilized spatial data structure for elevation (Childs 2004).
The transition from measured point to the grid data structure depends on the data
density relative to the required cell size. If the points are dense and the cell size is large,
than a simple tool within ArcGIS can be utilized (Point to Raster) to create a functional
surface but if the points are sparse, then an estimation or interpolation of value must
occur. This estimation of the values is achieved by interpolation algorithms, which use
the measured points to derive the surface. Slight changes in the algorithms or
modification of the parameters within the algorithm drastically alter the resulting surface,
which sometimes occurs when accessing the interpolations via different software
extensions.
Esri provides several software extensions, which can perform the interpolation of the
gird or raster surface. Many GIScientists rely on the extensions available to them, which
they are comfortable using, but potentially unexplored extensions and tools might be able
to provide a more reliable surface. Contrasting the same tools accessed from two different
17

extensions and comparing the results might encourage the use of future investigations.
This study will focus on are ArcGIS’s Spatial Analyst and Geostatistical Analyst
comparing and contrasting their ability for interpolation. These two extensions can be
used for interpolation to a surface using many of the same algorithms, although some of
the resulting surfaces have slight variation between them (ESRI 1995-2013 , Childs 2004,
Krivoruchko 2012, a).
2.4.1 Spatial Analyst Extension
The Spatial Analyst Extension provides many interpolation algorithms in the form of
tools or scripts within the extension. Spatial Analyst is widely used for interpolation due
to the extensions other functions within GIS. Spatial Analyst allows for the manual
modification or tool parameters, which requires extensive knowledge of the Spatial
Statistics influence on the model. Without the Spatial Statistics knowledge regarding
which parameters to modify, ArcGIS does offer default parameter settings, which can be
used (ESRI 1995-2013 , Childs 2004).
2.4.2 Geostatistical Analyst Extension
The Geostatistical Analyst Extension provides access to the interpolation algorithms
by use in a dynamic Wizard. The Wizard walks the user through the various steps in
parameter assignment; some interpolation methods retain the generic default where others
require optimization for each dataset. The optimization can be done by default within the
Geostatistical Wizard or be optimized by selecting the setting. The optimization uses the
18

Spatial Statistics of the input dataset to assign the parameter values in order to reduce the
standard error of the resulting surface (ESRI 2001, Krivoruchko 2012, a).

Figure 2: Screenshot of the Spatial Analyst toolbox (left) and the Geostatistical Wizard
(right). These two extensions provide an easy-to-use accessibility of point to Raster Grid
Surface.
The ArcGIS Extensions of Spatial Analyst and Geostatistical Analyst contain several
comparable interpolation algorithms, but not all interpolation algorithms are available
within the Geostatistical Wizard. As well as some interpolations within the Geostatistical
Wizard do not have the optimization option, several are optimized by default.

19

Table 1: ArcGIS 10.2 Extensions and Interpolation Comparison.

Esri ArcGIS Desktop Extension

Spatial Analyst Geostatistical Analyst

Tool Wizard
Interpolation Method
Inverse Distance Weighting  
Natural Neighbor  
Spline
Regular  
Tension  
Kriging  
Empirical Bayesian
Kriging
  

Topo to Raster  


Available as a tool in the Geostatistical
Toolbox.


Needs selection of the parameter for
optimization.


Not Available in Geostatistical Analyst
Wizard.

Although EBK is not available within the Spatial Analyst Toolbox, in this document
it is referenced as such. This is due to the ability to access EBK through the toolbox for
the Geostatistical Analyst. This is an important distinction, and has been completed due
to the finding of different parameter choices from the Wizard and the toolbox specifically
within the Geostatistical Analyst Extension (Krivoruchko 2012, a & b).
The following section provides a general review of the previous work completed on
the general topics of the study. First, the review investigates the previous work completed
on the comparison of interpolation methods. Second, the review investigates the previous
cases of topographic and bathymetric integration to create a Topographic/Bathymetric
20

DEM. Special notes are be made for the areas where the study lacked the detail to truly
inform the reader regarding the extension utilized to access the interpolation algorithm.

21

Chapter 3: Related Work
The previous sections emphasized the technical documentation of elevation data
acquisition, Spatial Statistics, interpolation algorithms and the use of various ArcGIS
specific extensions; this section summarizes some of the existing case studies. Since there
is no functional way to compile the case studies related to this thesis, classifications of
the topics in this section is divided into Spatial Statistics, interpolation comparisons and
the Topographic/Bathymetric integration.
In summary, the Spatial Statistics of the point dataset can influence the study
depending on interpolation methodology. Without the studies mentioning the specific
extensions and software utilized, results of interpolation cannot be recreated. The point
dataset has various results of accuracy and suitability when comparing interpolation
algorithms surfaces. The use of pre-existing DEM surfaces with unknown interpolation
origins cannot be held as reliable without understanding the pre-interpolated point data or
without a large amount of measurement verifications.
3.1 Spatial Statistics Importance
Spatial Statistics, in particular Geostatistics is challenging because of the
complicated interaction of geography and parameter values. Greater consideration should
be assigned to Spatial Statistics and its roll in analysis (Scott 2010). The value taken into
consideration when interpolating is not necessarily important other than the requirement
of spatial auto-correlation. The study by A. N. Kravchenko (2003) details the
interpolation method using soil samples. It concluded that the spatial structure of the
22

dataset plays a greater role than the variability of the data points. Additionally Heritage et
al. (2009) reinforces A. Kravchenko’s research that the impact of the terrain and density
of the points have a greater influence than the interpolation models although it is
important in terms of representative accuracy. The two studies follow the technical
documentation that understanding the Spatial Statistics within the point dataset can foster
a better understanding of the interpolated surface. As mentioned in several accredited
journals the interpolation algorithm used in creating a surface directly affects the
accuracy of the DEM (Erdogan 2009, Aykut, Akpinar and Aydin 2012, Eakins and
Grothe 2014).
The spatial auto-correlation of a large riverbed effects the representation of the
correlation and trends within the dataset. This is illustrated by a series of research articles
writing by V. Merwade (2009) which discusses a methodology to reduce the impact by
converting the X, Y coordinates into a channel fitted coordinate system essentially
straightening the river meanders. The impact of this change within the datasets, better
accounts for the variability and potential issues that might arise from interpolating river
bathymetry datasets. These studies although very important were outside of the scope of
this project. It is important to note that without Merwade methodology the Trends and the
visualization of the Semivariogram /Covariance Clouds might not follow the traditionally
expected results.
23

The optimization of parameters is reliant on the spatial statistics for any number
of interpolation methods. Lu and Wong (2008) identified the importantce of modifying
the parameters in the IDW interpolation methodology to create an surface.
3.2 Interpolation Comparison
The selection of interpolation algorithms has led to numerous studies comparing
and contrasting the interpolation algorithms. No comprehensive study has been
completed on the suitability of one algorithm over another (Krivoruchko and Gotway
2002). Studies focus either on topographic interpolation, bathymetric or on some
combing the two (Liu 2008, Erdogan 2009, Heritage, et al. 2009, Merwade 2009, Aykut,
Akpinar and Aydin 2012).
Erdogan (2009) interpolates topographic elevation by using the interpolation
algorithms: IDW, ordinary Kriging, multiquadratic radical based function, and thin plate
Spline using the Geostatistical Analyst Extension within ArcMap, looks at the effect of
the density of the raw data, the interpolation techniques, the impact of interpolation on
accuracy on hilly terrain, and examines methods for uncertainty. The results of this study
concluded that IDW produces greater overall uncertainty, due to exacerbated results by
low density of point data. The use of the Geostatistical Analyst for interpolation and the
reduction in point density lead to further investigation on the Spatial Statistical influence
on the models and the parameters (Erdogan 2009).
Aykut, Akpinar and Aydin (2012) discuss the comparison of seven interpolation
methodologies, but is not explicit on the software utilized for interpolation or the
24

parameters implemented. The conclusion states that the interpolation is important for the
overall accuracy of the DEM and in general, Kriging methodology is most suitable for
low-density single beam sonar datasets. P.V. Arun (2013) discusses the comparison of
several interpolation methodologies, as with the previous study only mention of ArcGIS
was stated but no explicit mention of software extension (Arun 2013). The need to
identify not only the software but the extension us for interpolation is gravely important.
Looking into the interpolation methodology using the ArcGIS Geostatistical
Analyst Extension, Badea (2013) discusses the need in geomorphic modeling for
comparing models with the same data. This includes utilizing the same interpolation
methods, but modifying the parameters for optimization (Badea 2013). In summary, there
is no interpolation algorithm, which is best for all datasets. The statistical and data
variability when comparing two or more interpolation algorithms, some criteria must be
met to provide justification for over method to be chosen over another. This accuracy
assessment can be defined as representative against the true surface, or statistically valid
using Root Mean Square Error, or by simple visualization of what would seem to be a
functioning surface.
Badea (2013) concludes that there is no practical methodology for accuracy
assessment, but utilization of the Root Mean Square (RMS/ RMSE) Error prediction can
augment the understanding of accuracy using cross validation. RMSE is a measure of
cross validation in which comparisons are performed from the predicted values to the
observed values for all point values used in the creation of the surface raster. The
25

utilization of RMSE for the interpolation methods provides a valuation of the interpolated
error but many particularly within the Spatial Analyst Extension do not provide RMSE.
Therefore, it is important to identify and estimate errors enforcing the concept of
understanding the spatial variability and Spatial Statistics of the dataset (Erdogan 2009,
Liu, Zhang and Peterson, Evaluation of the preformance of DEM interpolation algorithms
for LiDAR data 2009, Badea 2013). The documentation for the Geostatistical Analyst
Extension does not mention the “best” interpolation or “most accurate” it simply states
that the reliability of a surface and ease for the user is important (Krivoruchko 2012, a).
Creation of Topographic/Bathymetric DEM is not a new; geographers have
placed the two type of information side-by-side for oceanic navigation purposes.
Integration of River Bathymetry has failed to develop do to the constantly changing
riverbed morphology. Ever complicating the advanced analysis of bathymetric data
collection, is that the measured points are not taken in a random sequence, rather a
method of measuring in straight lines called transects is the preferred method. This causes
obvious complications for data dependency and spatial Trends. Merwade (2009) resolves
this issue, which also accounts for the morphology of river channels. The factors of the
morphology will impact the data analysis of spatial auto-correlation for the different
Project Areas, but the solutions posed by Merwade (2009) is not preformed within this
study therefor the spatial variation of the dataset may seem untraditional.

26

3.3 Topographic/Bathymetric DEM integration
The USGS and NOAA lead the data integration process of
Topographic/Bathymetric DEMs using a variety of datasets from historical nautical
charts to advanced multi-color laser sensor LiDAR technologies. Their overall process is
to take previously created DEMs and preform a merging procedure (Gesch and Wilson
2002, Barnard and Hoover 2010, Medeiros, et al. 2011, Foxgrover and Barnard 2012,
Dewberry Consultants LLC 2013). It does not take the point values in which the surfaces
are create into account, creating a dataset with compounding errors.
According to Foxgrover and Barnard (2012) and Barnard and Hoover (2010),
when creating a Topographic/Bathymetric model for San Francisco Bay and Southern
California, details the systematic procedure to integrate various topographic and
bathymetric DEM datasets, these rely on the already created DEM’s from LiDAR and
Sonar point datasets and integrates. The implemented methodology as the standard within
the USGS lacks the integrity of an algorithmic interpolation method as well as inheriting
the interpolation uncertainties from the previous creation of the various DEMs. Although
the methodology creates a compilation dataset, issues such as data overlaps and data gaps
are , the research in this study is looking into the interpolation algorithms of point to
surface and not on surface to surface.
Gesch and Wilson (2002) uses a similar procedure with the existing DEM’s as
was done for San Francisco Bay, although the variation includes converting the existing
DEM’s into points contains the derived elevation. Then the points were processed using
27

the ArcGIS tool entitled TopoGRID (in ArcGIS 9.X+ this tool is called Topo to Raster),
which uses a Thin Plate Spline which is optimized for creating topographic surfaces. The
detailed methodology used in both Tampa Bay studies attempts to provide an overall
hydrological enforced interpolated surface. In this study, adding the Topo to Raster
toolset as an interpolation algorithm is an interesting comparison, although no version of
the tool exists within the Geostatistical Analyst Extension.
The NOAA requested a report be compiled detailing the steps in the creation of a
California Coastal Topographic/Bathymetric DEM, this report is entitled “Coastal
California Data Merge Project” (Dewberry Consultants LLC 2013). It discusses the
procedure utilized to create and integrated DEM by focusing on the data inputs and
additional post-processing of the dataset to classify the data into a hierarchy of suitable
datasets. The hierarchy is then used to create a Topographic/Bathymetric DEM and again
is preceded with smoothing algorithms, void identifications, and other post DEM creation
methodology. Most of NOAAs projects emphasize the ocean bathymetry, leaving the
near-shore and river bathymetry as data voids. The methodology is helpful in large-scale
project analysis, but community-planning level does not benefit for this type of large-
scale data integration. This project will tackle integrating some of the local data products
into a seamless datasets for future integration into larger compilation DEMs.
A hydrodynamic interpretation of combining topographic and bathymetric data
for Bogue Inlet, North Carolina is published in the U.S. HYDRO 2011 (Bernstein, et al.
2011). It discusses the challenges of integrating the various interpolation methods with
28

the different and ever-changing morphology of the coastal/river inlet. This study
reinforces and details some of the challenges of measuring river and tidal channels for
perceived accuracy, when conditions are ever changing. The conclusion of the study
focuses on the need for advanced field surveying techniques and flexible interpolation
methodology. It suggests selecting the interpolation methodology based on the measured
point density of the surveys, although it does not state which software or extensions were
used in the creation of their final dataset.
The government developed studies complete the Topographic/Bathymetric DEM
creation with an general understanding of accuracy as defined by “best available”
information as the input, but some of the “best available” is over twenty years old. The
changes in morphology and sedimentation occur on a shorter time scale. The
methodology presented in the government case studies seems appropriate when the raw
data cannot be acquired, but only used as a last option.
The integration of the Topographic/Bathymetric datasets focused on the
“seamless” data products from existing datasets, where a focused need for continual
mentoring of coastal waters to ensure the reliability of the datasets. The study in the
Bogue Inlet of North Carolina hints at the need for generating data products using more
recent data collection techniques, technology, and post-processing interpolation.

29

Chapter 4: Methodology
Many past studies have lacked investigation on the interpolation algorithm and the
extensions influence on the surface model. This framework offers that it is a useful
methodology on any study where interpolation is used to understand and select the most
suitable interpolation algorithm. In order to reproduce the exact results within this study,
the exact data requires identical configuration. Any slight variation in data inputs,
geography shifts, or the distribution of the data and the results will changes the data’s
reaction to the algorithm.
This study mainly implements Spatial Analyst Extension Toolbox or the
Geostatistical Analyst Geostatistical Wizard (Seen in Figure 2). For the Toolbox, a
particularly selected interpolation algorithm runs under the default settings and
parameters. The Geostatistical Wizard runs under the default settings and optimized
parameters. This creates up to three surfaces for a single Project Area for one
interpolation algorithm. Comparisons of these surfaces illustrate the variability within the
Spatial Analyst and Geostatistical Analyst parameters of the algorithm and its effect on
the surface. The algorithms used as part of the comparison include IDW, NN, Spline
Regular, Spline Tension, Kriging, EBK, and Topo to Raster. These interpolation methods
were most prevalent in the literature for interpolation of elevation surfaces.
The methodology process is simple although understanding the resulting surfaces
is complex and essential for making a determination for usability. The study area
30

geomorphology and the data utilized within this project have a direct impact on the
surfaces output.
4.1 Klamath River Study Area

The Klamath River, spans from north-west Oregon to the Pacific Ocean in
California. Many natural and man-made conditions affect the flow of the river and the
river’s ability to flush sediments into the Ocean. The Klamath River Estuary located in
the town of Klamath, CA develops the sediment buildup, which can modify the river’s
natural flow from day to day. The study area was selected due to the recent multi-agency
data collection efforts and future removal of dams upstream.
The Study Area of the Klamath River Estuary divided into 5 Project Areas, (See Figure
3) to investigate the river flow influencing on sedimentary layers. Each Project Area is
approximately 10 acres, spanning the entire river, and encompassing at least 200 ft. of
land on either side of the River’s banks. Although the Project Areas are referenced in
detail within this document, understanding the relationship to the larger study area is
important because their impact on the Spatial Statistics on the Topographic/Bathymetric
DEM surface creation.
31

Figure 3: Map of the Klamath River Estuary with the selected Project Areas highlighted.
4.2 Datasets
The LiDAR and Sonar datasets are vastly different in terms of measurement
technology and therefore have different accuracy standards, processing requirements, and
spatial constraints (NOAA Coastal Services Center 2007).
The topographical datasets were collected using a LiDAR sensor from aircraft
under the project title of the California Coastal LiDAR Dataset under the supervision of
NOAA, USGS, etc. The data was collected over the Klamath River on approximately
32

November 8 through December 7, 2010. The data producers verified all the accuracy
standards, performed post-processing to remove artifacts, and completed the final
reporting as stated within the contract.
The California Coastal LiDAR Dataset is free to download via the
http://coast.noaa.gov/ website. The dataset consists of several configuration options for
representing the LiDAR data by the classified points: “Ground,” “unclassified,” “water
surface,” “overlap”; as well as return types: “any,” “first,” and “last”. The selection for
modifying the coordinate information is also available. For this analysis, the classified
“ground” point features of “any” return type were selected from the download website
with associated X,Y,Z values in ASCII format in a projected coordinate system
California State Plane North American Datum of 1983 (NOAA Coastal Services Center
2007, NOAA Coastal services center 2014). No other post processing was completed on
the dataset. For the Study Area around the estuary of the Klamath River, the data
consisted of 6,394,830 measurements, with an average measurement density of 2.28
points per square foot (See Figure 4).
33

The bathymetry data, as part of a project by the Yurok Tribe between July 16 and 23,
2013, was collected using single-beam sonar or echo sounder system. The data collection,
processing, quality assurance was completed by a contractor specializing in hydrographic
surveys while training the Yurok Tribe staff in the technical requirements of collection,
processing, and quality assessment which were completed to the industry standards set by
the Army Corps of Engineers Hydrographic Survey Manual. The final data and report
products are not publicly available, although requests can be made to the Yurok Tribe.
The horizontal control is set by California State Plane North American Datum of 1983
and the vertical control is referenced to North American Vertical Datum of 1988
(NAVD88) as detailed in the unpublished survey report. The quality assessment process,
detailed in a report that was attached to the geographic dataset when requested. The
bathymetric dataset for the Klamath River estuary consists of 16,510 measurements, with
an average measurement density of 9.68 points per square feet (See Figure 5).
Figure 4: (Left) Individual LiDAR point measurements. (Right) Density map of the
LiDAR measurements averaging 2.28 points per square foot.
34

Due to the final product consisting on a single DEM, the points from the LiDAR
and Sonar technologies were merged geographically. Merging the point files prior to
interpolation reduces the propagation of errors seen in the government case studies,
although some slight overlapping of data points do occur due to the gap in time between
LiDAR and Sonar collections. This reduces the challenge in creating a seamless DEM as
well as accepting the point distribution within the datasets.
When analyzing the various point datasets the identification of the spatial Trends,
the spatial dependency, the point distribution of values and locations, and the stationarity
of the dataset must be taken into consideration for use with the various interpolation
methods.

Figure 5: (Left) Individual Sonar point measurements. (Right) Density map of the Sonar
measurements averaging 9.68 points per square foot
35

4.2.1 Statistics
The importance of the Spatial Statistics has been discussed within the background
section of this document. Understanding the data’s distribution, stationarity trends, and
spatial auto-correlation are factors, which influence the decision of choosing an
interpolation algorithm. For this project and review of the data used, the visualizations of
the factors will not be illustrated-- only the resulting conclusion derived from the
visualizations.
Interpolation algorithms do not have a limit on the number of points they can
interpolate, but due to the limitations of visualization and characterization of the various
methodologies, the selection of Project Areas are limited. From various Project Areas, the
LiDAR and Sonar points were randomly selected to create subset datasets with 250
points of LiDAR and Sonar respectively. This resulted in the five various Project Areas
to have each 500 elevation measurements equally derived from LiDAR and Sonar. These
subset datasets were used to create the interpolated surface elevations and the various
other results (Table 2, Figure 6).

36

Table 2: Statistical comparison of the entire Project Area to the subset dataset used to
create the interpolated surfaces.

Number of
Measurements Min Max Mean Median
Std.
Dev
Project
Area 1
Entire Area 13,107 -16.0 116.5 14.7 5.7 19.6
Subset 500 -16.0 105.9 3.6 0.9 18.3

Project
Area 2
Entire Area 4,976 -9.0 41.9 9.2 9.3 7.4
Subset 500 -9.0 39.7 2.5 0.7 8.7

Project
Area 3
Entire Area 10,765 -19.7 42.4 11.2 4.7 14.1
Subset 500 -19.7 42.2 1.2 0.4 14.7

Project
Area 4
Entire Area 9,984 -31.6 72.0 9.5 10.3 11.4
Subset 500 -31.6 70.4 -0.2 0.2 13.8

Project
Area 5
Entire Area 19,072 -17.9 90.6 7.2 5.3 7.3
Subset 500 -17.7 51.4 -2.6 0.4 11.2

37

Figure 6: Maps of Project Area(s) with LiDAR - Sonar Measurements with the Subset data
used in the creation of the surface profiles.

4.2.1.1 Project Area 1
Project Area 1 is characterized by a deep channelization of the river on the
southwest corner, as well as a gradual sloping riverbank on the opposite side of the river.
This area contains 13,108 measured points, 12,724 points resulting from LiDAR and 384
points from Sonar. The points used in the interpolation of the surface contain a
randomized sample of 250 LiDAR and 250 Sonar Points (Table 2).
38

Figure 7: Graph illustrating the distribution of the merged LiDAR and Sonar points used in
the creation of the surface profiles for Project Area 1.
Normal distribution within the point dataset is an important factor to account for
when choosing an interpolation method. Project Area 1, for the subset and Project Area
dataset is slightly skewed to the right with a skewness factor of 2.3 (Figure 7) and 1.9.
The statistical distribution of the full point datasets varies slightly in the minimum, mean,
median and max. Although the datasets do not equally reflect each other and are skewed,
they are considered normally distributed and do not require any transformation. The
representation of the Normal QQ plot illustrates the dataset as compared to a standard
normal (Gaussian) distribution, which is shown as a solid line. The variation of the
dataset is not enough to confirm the findings of the histogram and require any data
modifications. The data reduction has very little effect on the results as seen on the
histogram and Normal QQ plots.
The stationarity of the dataset is the statistical relationship between two points,
which depends on the distance between them. The Voronoi map are used illustrated the
39

stationarity for Project Area 1 the subset dataset, the data is stationary. Visualization of
trends within Project Area 1 can be seen functionally within a 3D environment, due to the
morphology.
The spatial auto-correlation as illustrated by the Semivariogram and Covariance
Cloud contains the spatial relationship between the individual points within the dataset.
Within Project Area 1, both do not represent what would normally be illustrated within
the Semivariogram and Covariance Cloud graphs. This issue is illustrated and accounted
for in research performed to account for the morphology of river channels (Merwade
2009) is not preformed within this study therefor the untraditional spatial auto-correlation
of the datasets.
These representations of the spatial and statistical variation within the dataset are
important to keep in mind when performing the interpolation. Understanding the
relationship the points have within their geographic space is the bases of the interpolation
algorithm methodology. From Project Area 1, the distribution of the dataset is normal,
with little stationarity and no trends, and the data is spatially auto-correlated. The effects
of Project Area 1’s spatial and statistical variation will be compared in the results section.
4.2.1.2 Project Area 2
Project Area two is characterized by a general straight flow of the river, with distinct
riverbanks. The area is notorious for containing underwater impedances due to old bridge
pilings. This Project Area contains 4976 number of measured points, with 4552 from
LiDAR measurements and 424 from the Sonar collection. The points used in the
40

interpolation of the surface contain a randomized sample of 250 LiDAR and 250 Sonar
Points (Table 2).

Figure 8: Graph illustrating the distribution of the merged LiDAR and Sonar points used in
the creation of the surface profiles for Project Area 2.
The statistical variation between the entire Project Area and the subset dataset can be
illustrated in the histogram with a skewness factor of 0.82 as seen in Table 2 and Figure
8. The point values on the histogram vary slightly from a traditional normal distribution,
but this variation is not large enough to require the use of a transformation. Normal QQ
plot illustrates that project are 2 points, follows relatively close to a standard normal
distribution.
Voronoi map are used to visualize the stationarity of the dataset and it shows very
little variation with the dataset for Project Area 2 the subset dataset. The trends shown in
the 3D environment illustrate the relationship for the data to their environment.
41

The Semivariogram and Covariance Cloud illustrate the spatial relationship between
the individual points within the dataset or the spatial auto-correlation. Within Project
Area 2, the clouds are not illustrating the expected results. The reasoning is due to the
theory discussed within Project Area 2, where the river morphology influences the spatial
relationship of the points.
Within Project Area 2, analysis of the point dataset reviles that the spatial and
statistical variation in the dataset can be seen that the data is normally distributed, has
little to no stationarity, no trends which need to be removed, and that the data is spatially
auto-correlated. From these illustrations of the spatial and statistical variation within the
dataset for Project Area 2, a majority of the interpolation methods can be utilized without
much concern for the applicability. The spatial and statistical variation within the dataset
for the entire Study Area is still important when comparing the interpolation
methodologies.
4.2.1.3 Project Area 3
Project Area 3 is the most complex Project Area, not with respect to river
morphology, but with respect to having a peninsula within the Project Area. This Project
Area is a representation of the data inconsistencies that might occur when there is a time
delay in LiDAR or Sonar data collections discussed in the USGS publication for the
Conceptual Prototype for the Next-Generation National Elevation Dataset (Stoker, et al.
2013). The peninsula during the LiDAR collection was visible, but through the
bathymetry collection, either the river level was higher or the morphology changed to
42

remove the sedimentation that had created the spit. This influences the Spatial Statistical
variation within this Project Area and is something to note and revisit within the results.
This Project Area contains 10,765 numbers of measured points, with 10,505 from LiDAR
measurements and 260 from the Sonar collection. The points used in the interpolation of
the surface contain a randomized sample of 250 LiDAR and 250 Sonar Points (Table 2).

Figure 9: Graph illustrating the distribution of the merged LiDAR and Sonar points used in
the creation of the surface profiles for Project Area 3.
The histogram and the Normal QQ plots both provide the visualization that the data is
normally distributed to such a degree that transformation is not required. The scattering
of the data points can be seen by the statistics contained within the histogram. The
histogram looks skewed with a value of 1.41 as seen on Figure 9. The data is slightly
clustered towards the lower data values, but extends out to the high values provides a sort
of equalization.
43

For Project Area 3 the subset dataset the variation seen in the Voronoi map is a
method to describe the stationarity of the dataset, shows that the data is stationary. The
trends shown in the 3D environment illustrate the relationship for the data to their
environment.
Visualizing the spatial auto-correlation of the datasets is performed by analyzing the
Semivariogram and Covariance Clouds. These graphs within Project Area 3 illustrate the
geographic relationships between the datasets and although they do not exactly fit the
expected result of increasing with distance and decreasing with distance, their general
trend leans toward this conclusion.
The visualizations for Project Area 3 illustrate that the point dataset is normally
distributed; stationarity with little trends and follows the spatial auto-correlations. These
factors can be used with selecting an interpolation method, as some are better-suited data
distribution types.
4.2.1.4 Project Area 4
Project Area 4 has a gradually sloping bankline with deep channelization of the river
due to the force of the water upstream and has a sharp incline into the land once again via
a levee feature and boat docks. This Project Area contains 9,984 of measured points, with
9,623 from LiDAR measurements and 361 from the Sonar collection. The points used in
the interpolation of the surface contain a randomized sample of 250 LiDAR and 250
Sonar Points (Table 2).
44

Figure 10: Graph illustrating the distribution of the merged LiDAR and Sonar points used
in the creation of the surface profiles for Project Area 4.
This point dataset plotted on a histogram can determine if that data is normally
distributed. This distribution although visually skewed with a value of 1.90 as seen in
Figure 10, offer enough distributed variation to not require any transformation processes.
The Normal QQ plots for Project Area 4 generally follow the Gaussian distribution,
which is shown as a solid line. There are a few outliers, but their impacts on the overall
dataset are very little.
The Voronoi map for Project Area 4 illustrate that the dataset is stationarity when
classified by entropy. The darker colors represent the higher rates of change of the values
within the dataset for neighboring point values. Project Area 4 has very little obvious
trends within the dataset, aside from the river channelizations.
The Semivariogram and Covariance Clouds illustrate the level of spatial auto-
correlation contained within the dataset. The representations of the geographic
45

relationships are not within the expected range, but the relationship tends to follow the
traditional variation.
The point dataset for Project Area 4 demonstrates that the point dataset is normally
distributed; that the dataset have little to no trends and that the data is known to be spatial
auto-correlations. These features should be used when wanting to create a surface feature,
due to influence that these features have on the interpolation parameters.
4.2.1.5 Project Area 5
This project Area 5 contains a low elevation wetland island, with a deep channel river
outlet as the water flows near the ocean to a manmade levee and boat ramp. This Project
Area is most susceptible to tidal influence, although we are not accounting for tidal flow
into the measurements. This Project Area contains 19,072 of measured points, with
18,791 from LiDAR measurements and 281 from the Sonar collection. The points used in
the interpolation of the surface contain a randomized sample of 250 LiDAR and 250
Sonar Points (Table 2).
46

Figure 11: Graph illustrating the distribution of the merged LiDAR and Sonar points used
in the creation of the surface profiles for Project Area 5.
Histograms and the Normal QQ Plots can visualize the distribution of the point
dataset within Project Area 5. The histograms illustrate the datasets distribution around a
normal distribution curve, the skewness of the dataset falls within the limits of not
requiring any transformations at a value of 0.54 in Figure 11. The Normal QQ Plots
which create a linear feature to attempt to best fir the dataset.
The Voronoi map for Project Area 5 illustrate that the dataset is stationarity when
classified by entropy. The darker colors represent the higher rates of change of the values
within the dataset for neighboring point values. Project Area 5 has very little trends
within the dataset. The 3D view of the point dataset
The Semivariogram and Covariance Cloud in Project Area 5, the clouds are not
illustrating the expected results. The clouds show the representation of the data’s
distribution of the points for the area verse the distance between the points themselves.
47

Project Area 5, has an interesting point distribution which reviles the spatial and
statistical variation in the dataset can be seen that the data is relatively normally
distributed. This dataset has little to no stationarity, no trends which need to be removed,
and that the data is spatially auto-correlated. The illustrations of the spatial and statistical
variation within the dataset for Project Area 5, shows that a majority of the interpolation
methods can be utilized for this dataset.
For the 5 Project Areas datasets, none of the Project Areas follows exactly the
expected results of the data analysis. Even with this variation, the irregularities are not
great enough to require modification of the datasets themselves. The datasets, with the
traditional distribution and variation which are used in the creation of the various raster
surfaces.
4.3 Interpolation
For processing, the point datasets for the various Project Areas Esri software was
utilized, particularly Esri’s ArcGIS Desktop 10.2.1 software access to the Spatial Analyst
and Geostatistical Analyst Extensions. The Spatial Analyst tools were accessed through
the ArcToolbox window, Spatial Analyst toolbox. The Geostatistical Analyst accessed
through the Geostatistical toolbar and performed the interpolations the Geostatistical
Wizard. The Geostatistical Analyst Extension and Wizard requires users to interface with
the Wizard prompts having an overall greater functionally.
The Geostatistical Wizards use of the optimization settings within the Wizard has
been proven in the research (Krivoruchko 2012, Badea 2013). This optimization modifies
48

the parameters within the interpolation methods to better fit the point dataset. Although
this optimization can only enhance the resulting interpolated surface, it is not always
defaulted parameter within the Wizard. Manual selection is required in Geostatistical
interpolation methods IDW and Kriging. Not all interpolation algorithms are available
within the Geostatistical Wizard.
This project will investigate the variation between the Spatial Analyst Extension’s
default parameters and the default and optimized parameters of the Geostatistical Analyst
Extensions Wizard, see Figure 3 for summary of interpolation methods available in
which extension. The IDW and Kriging are the only interpolation algorithms to require
the three surfaces creation; this is due to the Geostatistical Wizard not automatically
optimizing the parameters. Spline Regular and Spline Tension have results with the
Spatial Analyst tool and the optimized by default Geostatistical Wizard. The algorithms
of Topo to Raster and Natural Neighbor are not accessible via the Geostatistical Wizard
as well as EBK not being a part of the Spatial Analyst toolset, but are important
interpolation algorithms to the creation of Topographic/Bathymetric surfaces in many
studies.
4.4 Illustrating Results
The results from the interpolation into a Topographic/Bathymetric DEM can be
visualized in several such as statistical tables, profile graphs and maps for each raster
Project Area.
49

The non-geographic tables provide a way to analyze statistically, the entire
interpolated raster surface as a set of values. The variation provides a broad
understanding of the underlying raster surfaces, particularly important within
Geostatistical interpolation models. The local and global variation of the measured values
can assist in understanding how might the data distribution’s effect on the interpolated
surface. The comparisons of the tables are for each individual Project Area. Illustrating
the non-geographic variation in the elevation values within the interpolated raster
surfaces of the subset Project Area. The raster surface created with the Spatial Analyst
tools does not have a valuation for the root means square error (RMSE). As discussed in
chapter 2, not all interpolation algorithms are available within the Extensions of Spatial
or Geostatistical Analyst, nor have all the parameters within the Geostatistical Wizard
have been optimized.
The profile views provide the comparison of the various interpolated surfaces to make
a determination of the suitability and continuity of the dataset for the topography and
bathymetric datasets integration. The entire selected Project Areas contain Land-Water-
Land surfaces and are visualized by three transects of the Land-Water-Land integration.
The graphs, labeled A, B, and C respectively, depending on the interpolation
methodology can have between one and three raster datasets represented, which illustrate
the tool, the default Wizard and the Wizard optimized. The measured points are also
illustrated on the graphs represent a 50ft area around the interpolated transect as potential
50

measurements used to create the surface, although the points may not have been
accounted for within that surface.
The general visualization or maps of the entire Project Areas surface raster’s is
helpful to see the overall elevation variation. Although comparing the distinct elevations
is challenging in map visualizations, the visualization of the difference between the tool
and the interpolation with the least RMSE was also completed. Each dataset is classified
in order to exaggerate the elevation variation between the merged datasets, at the water
surface by classifying the data values as follows: -30, -25, -20, -15, -10, -5, -4, -3, -2, -1,
0, 1, 2, 3, 4, 5, 10, 15, 20, 25, 30, 35, 40, 50, 60, 70, 80, 90, and100. This visualization
lacks the ability to compare the interpolation’s differences.

51

Chapter 5: Results
The results of this study provide difference between the Spatial Analyst and the
Geostatistical Analyst Extensions. Those can also help GIS Professionals to make an
informed choice when tasked with creating an interpolated surface. The results chapter
consists of two subsection; 1) parameter variation section illustrate the software
variability of the seven different interpolation algorithms, and 2) interpolation section
provides statistical and visual (Surface Raster’s and Transects) examples of the impact on
the resulting surface with the variability on the interpolated surfaces by use of case study
of the Klamath River estuary. These comparisons are conceptually interesting, and should
increase awareness of the complicated role of the data’s statistics influence on the
interpolated surface, which can be completed by using Geostatistical Analyst Extension’s
optimization setting, but it also provides the proof that documenting the interpolation
algorithms used should be part of the metadata of all raster surfaces.
5.1 Parameter Variation
ArcGIS is known for having multiple ways to access a function within their software.
The implication of this is that many GIS Professionals assume standardization between
the two functions with the same name. With interpolation, this is not the case.
Standardization between the interpolation methodologies of the Spatial Analyst
Extension and the Geostatistical Analyst Extension vary, resulting in surface features
with large variability. The visualization of the Spatial Analyst tool and Geostatistical
Wizard are shown to identify parameter differences in naming as well as having a
52

reference for the differences illustrated in the tables (Table 3, 4, 5, 6,and 7) . The tables
compare the parameter values from the tool defaults, Wizard defaults, and Wizard
optimization for each Project Area. The comparisons should broaden the understanding
of the functional differences of interpolation within the Geospatial Community.
5.1.1 IDW
The variability in the IDW interpolation tool in the Statistical Analyst Extension and
the IDW interpolation in the Geostatistical Analyst have noticeable differences (Figure
12), particularly when it comes to the selection of the points used in the interpolation
process. The IDW tool has a default setting of a variable search type with the selection of
12 measured points. By default, the distance parameter is left blank. The Geostatistical
Wizard has slightly different terminology as the tool, where the search type has become
the neighborhood type. The Geostatistical Wizard’s neighborhood type is Smooth with a
maximum neighbors of 15, minimum neighbors of 10 and a 1-sector search type.

Figure 12: IDW interpolation prompts, accessed via the Spatial Analyst Toolbox (left) and
the Geostatistical Wizard (right).
53

Detailed differences are seen in Table 3. The power setting for the IDW, which is the
weighted measure, in both the Statistical Analyst tool and the Geostatistical Wizard
default parameters has a power value of 2. When the Geostatistical Wizard is optimized
for Project Area 1, the power becomes 3.01. For Project Area 2-5, the optimized power
values are as follows 3.32, 2.00, 2.98, and 9.87 respectively. IDW Wizard offers an
estimation of this variability can later be seen in the raster surface and the transects of the
various Project Areas. The power setting should be influenced by the statistical
variability within the point dataset used as the input in the interpolated surface.
Table 3: Table illustrating the differences between the IDW interpolation parameters in the
Tool, the Wizard's default and the optimized settings.
5.1.2 Spline
The Spline interpolation tool within the Spatial Analyst Extension (Figure 13) and the
Geostatistical Wizard (Figure 14) vary, within the Geostatistical Wizard Spline is termed
IDW
Tool Wizard Wizard Optimized per Project Area
Default Default *PA 1 *PA 2 *PA 3 *PA 4 *PA 5
Power 2 2 3.01 3.32 2 2.9 9.87
Search Neighborhood
type
Std. Std. Std. Std. Std. Std.
Maximum Neighbors 15 15 15 15 15 15
Minimum neighbors 10 10 10 10 10 10
Search Radius Variable
Search Radius #
points
12
Search Radius Max
Distance
Blank
Sector type 1 1 1 1 1 1
Angle 0 0 0 0 0 0
Major Semiaxis Opt Opt Opt Opt Opt Opt
Minor semiaxis Opt Opt Opt Opt Opt Opt
Anisotropy factor 1 1 1 1 1 1
*PA – Project Area
54

as a Radical Based Function. Changes in the terms used to describe the parameters of the
interpolation algorithm also occur. For the Tool, Spline Type options are Regular or
Tension, where in the Geostatistical Analyst the term Kernel Function is used to describe
the same parameters as Completely Regularized Spline and Spline with Tension as seen
in Figure 10 and 11. As with the IDW comparison, the method for selecting the
interpolating points varies for the weight given to the points and the number of points
used in the interpolation. The Spatial Analyst tool has a weight of 0.1 and number of
points as 12. The Geostatistical Wizard terms the weight as Kernel Parameter that is
optimized by default, with a standard neighborhood, maximum neighbors of 15,
minimum neighbors of 10 and a 1-sector search type.

Figure 13: Screen shot of the Spline interpolation prompt, accessed via the Spatial Analyst
Toolbox.
55

Figure 14: Screen shot of the Geostatistical Wizard prompts for the Spline Regular (left)
and the Spline Tension (right).
The modification of the parameters for Spline interpolation algorithm is visualized in
Table 4 and 5. The Kernel Parameter is optimized within the Geostatistical Analyst
Extension by selecting the value, which produces the lowest RMSE for the selected
Kernel Function type. The Kernel Parameter for the Completely Regularized Spline
Project Areas 1-5 are: 0.14, 0.31, 0.19, 0.12, and 0.18 respectively. The Kernel Parameter
for the Spline with Tension Project Areas 1-5 are: 0.003, 0.007, 0.006, 0.009, and 0.004
respectively. Comparing these values within the Geostatistical Analyst Wizard with the
automatic selection of 0.1 within the Spatial Analyst Tool, the variability seems slight
with the Completely Regularized Spline, but with the Spline with Tension, the variation
can cause large manipulations within the raster surface.
56

Table 4: Table illustrating the differences between the Spline Regular interpolation
parameters in the Tool, the Wizard's default and the optimized settings.
Spline Regular
Tool Wizard Default which is Optimized per Project Area
Default *PA 1 *PA 2 *PA 3 *PA 4 *PA 5
Kernel
Function/Spline Type
Regular
Kernel Parameter 0.14 0.31 0.19 0.12 0.004
Search Neighborhood
type
Std. Std. Std. Std. Std.
Weight 0.1
Num Points 12
Maximum Neighbors 15 15 15 15 15
Minimum neighbors 10 10 10 10 10
Sector type 1 1 1 1 1
Angle 0 0 0 0 0
Major Semiaxis 5,909 5,909 5,909 5,909 5,909
Minor semiaxis 5,909 5,909 5,909 5,909 5,909
Anisotropy factor 1 1 1 1 1
* Project Area
Table 5: Parameter variation of the Spline with Tension interpolation.
Spline Tension
Tool Wizard Default which is Optimized per Project Area
Default *PA 1 *PA 2 *PA 3 *PA 4 *PA 5
Kernel Function/Spline
Type
Regular
Kernel Parameter 0.003 0.007 0.006 0.009 0.004
Search Neighborhood
type
Std.
Std. Std. Std. Std.
Weight 0.1
Num Points 12
Maximum Neighbors 15 15 15 15 15
Minimum neighbors 10 10 10 10 10
Sector type 1 1 1 1 1
Angle 0 0 0 0 0
Major Semiaxis 5,909 5,909 5,909 5,909 5,909
Minor semiaxis 5,909 5,909 5,909 5,909 5,909
Anisotropy factor 1 1 1 1 1
57

* Project Area
5.1.3 Kriging
The Kriging interpolation algorithm is the most complicated of the interpolation
algorithms within this study due to the sheer number of potentially modified parameters.
The variation from the Spatial Analyst tool and the Geostatistical Wizard are intensive
and within the Geostatistical Wizard, the selection of optimizing the parameters is
required. The windows for the Spatial Analyst Tool for Kriging can be seen in Figure 15,
16 and 17. The Semivariogram properties determine types of Kriging algorithm. The
Kriging method and the Semivariogram model, which are defaulted to Ordinary and
Spherical respectively, is the second step of the Geostatistical Wizards Kriging
methodology. There is an option for selecting the advanced parameters for the Kriging
algorithm, but by default, these are blank but options include Lag Size, and the
variograms parameters of Major Range, Partial Still and Nugget. The search Parameter
within the Tool is identical to IDW and Spline, with the Variable search radius, number
of points 12 and a blank selection for a maximum distance.
58

Figure 15: Screen shot of the Kriging interpolation window from the Spatial Analyst
toolbox. The sub-window of the advanced parameters has been open to illustrate the further
functionality in not used by default.

Figure 16: Screen Shots of the Kriging interpolation, third step in using the Geostatistical
Wizard for this interpolation methodology. Semivariogram on the left is the default Wizard
values, on the right are the optimized setting.
59

Figure 17: Screen shot of the Geostatistical Wizard's fourth step for the Kriging
interpolation. The search neighborhood is the same for both the default Wizard and the
optimization setting.
The Geostatistical Analyst Extension has a large selection of parameters for the
Kriging algorithm, which requires multiple steps within the Wizard. The first step of the
Wizard is the selection of the Kriging type, which by defaults is Simple, but for
comparison purposes, Ordinary Kriging was selected, there are options for
Transformation Type and Trend Removal, but both have None selected. The next step in
the Wizard has specific details of the Kriging algorithm; the variation for each study area
can be seen in Table 6. The point selection for the Geostatistical Analyst has a Search
Neighborhood Type of Standard with a Maximum of 5, Minimum of 2, and a Sector
Type of 4 with a 45 degree off-set. The Wizard offers the ability to see the surface
Raster’s predicted error given in terms of a Regression Function, Root-Mean-Square,
Mean Standardized, Root-Mean-Square Standardized, and Average Standard Error.
60

Kriging
Tool
Wizard per Project Area
Project Area 1 Project Area 2 Project Area 3 Project Area 4 Proj. Area 5
Default Default Optimized Default Optimized Default Optimized Default Optimized Default Opt.
Kriging Type Ordinary O O O O O O O O O O
Output Surface Type Prediction P P P P P P P P P P
Semivariogram Model Spherical
Transformation Type None None None None None None None None None None
Order of trend removal None None None None None None None None None None
Variable Semivariogram Semivariogram Semivariogram Semivariogram Semivariogram Semivariogram Semivariogram Semivariogram Semivariogram
Semivariogra
m
Model Nugget enabled TRUE TRUE TRUE TRUE TRUE TRUE TRUE TRUE TRUE TRUE
Calculate Nugget TRUE FALSE TRUE FALSE TRUE FALSE TRUE FALSE TRUE FALSE
Nuggets Size Blank 0.45 0 0 0.29 0.21 0.07 0 0 0 0.03
Measure Error 100% 100% 100% 100% 100% 100% 100% 100% 100% 1000%
Model Type Stable Stable Stable Stable Stable Stable Stable Stable Stable Stable
Parameter 2 1.52 1.35 1.95 2 2 1.5 1.96 1.86 2
Major Range Blank 465.38 940.44 735.53 267.76 320.57 60.32 531.45 323.35 663.38 129.33
Anisotropy FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
Calculate Partial Sill Blank TRUE FALSE TRUE FALSE TRUE FALSE TRUE FALSE TRUE FALSE
Partial Sill Size 455.97 671.03 117.07 65.87 207.51 6.04 257.62 200.46 205.64 30.28
Lag Size Blank 74.42 117.55 88.84 33.47 51.09 7.54 78.46 40.41 76.74 16.2
Number of Lags 12 12 12 12 12 12 12 12 12 12
Search Radius # points 12
Search Radius Max
Distance
Blank
Table 6: Kriging parameter variation: O - ordinary, P - Prediction.
61

Search Neighborhood
type
Variable Standard Standard Standard Standard Standard Standard Standard Standard Standard Standard
Maximum Neighbors 5 5 5 5 5 5 5 5 5 5
Minimum neighbors 2 2 2 2 2 2 2 2 2 2
Sector type 4 w/ 45deg 4 w/ 45deg 4 w/ 45deg 4 w/ 45deg 4 w/ 45deg 4 w/ 45deg 4 w/ 45deg 41 w/ 45deg 4 w/ 45deg 4 w/ 45deg
Angle 0 0 0 0 0 0 0 0 0 0
Major Semiaxis 465.38 940.44 735.53 267.77 320.57 60.32 531.45 323.35 663.38 129.33
Minor semiaxis 465.38 940.44 735.53 267.77 320.57 60.32 531.45 323.35 663.38 129.33
Anisotropy factor 1 1 1 1 1 1 1 1 1 1

62

5.1.4 EBK
EBK, being a recently created Geostatistical interpolation method has many
parameters when are optimized by default, the parameters can be seen of the screen
capture of the Geostatistical Wizard in Figure 18. Although EBK is not available within
the Spatial Analyst toolbox, it is available within the Geostatistical Analyst toolbox with
only slightly different parameters when those available within the Geostatistical Analyst
Wizard (Figure 19). The only parameter that was identified as different between the
methods was the search radius of the neighborhood (Table 7). Although this variation is
slight, it seems to have an effect in the resulting surface. The Wizard offers the ability to
see the surface Raster’s predicted error given in terms of a Regression Function, Root-
Mean-Square, Mean Standardized, Root-Mean-Square Standardized, and Average
Standard Error.

Figure 18: Screen capture of the EBK interpolation accessed via the Geostatistical Wizard.
63

Figure 19: Screen capture of the EBK interpolation accessed via the Geostatistical Analyst
Toolbox.
Table 7: Table of EBK interpolation parameters.

Tool Wizard
Subset Size 100 100
Overlap Factor 1 1
Number of Simulations 100 100
Output surface Type Prediction Prediction
Transformation None None
Semivariogram Type Power Power
Search Neighborhood
type
Standard
Circular
Standard
Circular
Maximum Neighbors 15 15
Minimum neighbors 10 10
Sector type 1 1
Angle 0 0
Search Radius Blank Opt

64

5.1.5 Natural Neighbor & Topo to Raster
Both Natural Neighbor and Topo to Raster interpolation methodologies are not
available within the Geostatistical Analyst Wizard. The surfaces models are a result of
using the default setting with only the point dataset as input. Screen shots of the Topo to
Raster tool can be seen in Figure 20, note the additional vector capabilities. These
methods are used within the interpolation literature.

Figure 20: Screen capture of the Topo to Raster tool available in the Spatial Analyst
Extension.
65

The search neighborhood, weight of proximity, and algorithm differences account for
the variation within the same geographic area. This variation can cause large
discrepancies in the volumetric calculation used in flood and sea level change modeling
due to the raster surface differences. The comparison of the raster surfaces are completed
below and illustrate the impact of the discrepancies of the Spatial Analyst Extension and
the Geostatistical Analyst Extension was well as the variation between the interpolation
algorithms available in each.
5.2 Results of Multiple Interpolation Methods
The results provide an overall picture of the surface created from the seven
interpolation algorithms and parameter options. In each Project Area, tables and multiple
figures are provides for an overall understanding of the interpolation’s influence on that
area.
A table is provided with minimum, maximum, mean, standard deviation and for some
interpolation methods an RMSE, this table should be compared to the entire Project Area
and the subset Project Area statistics to see it’s comparison of the interpolation’s
processing of the point data into a surface.
The project map, organized in the same order than the table, shows the actual
interpolated surface (instead of the statistics of the surface as in the table), these surfaces
were created using the same data, the variation lies with the interpolation algorithm and
the ArcGIS Extension utilized. The colorization is the same for each surface, where blue
66

represents color in the below water and green representing above water. This colorization
is applied in each Project Area in the study.
Difference maps provide a method to contrast the Spatial Analyst Tool surface with
the Geostatistical Analyst Wizards optimized setting surface for the interpolation
algorithms, the difference maps which are colorized to exaggerate the differences, darker
red the color the greater the difference in either positive or negative elevation
measurement between the surfaces.
From the interpolated surfaces, 3 transects were created for each Project Area, these
transects can are represented in profile graphs of the surface. Transects illustrate the
smoothness of the data integration better than the color scales are able to. Each transect,
labeled A,B, and C, are representing the same ground surface the only variation lies in the
interpolation algorithm and extension used to create the surface, where on the graphs, the
various lines within the graph represent the extensions used.
Together these representations of the interpolated surfaces facilitate an overall
understanding of the need to optimize the parameters of the interpolation algorithm used
5.2.1 Project Area 1
Project Area 1 is characterized by a deep channelization of the river on the southwest
corner, as well as a gradual sloping riverbank on the opposite side of the river. The three
profiles encompass changes in the bank morphology.
67

Table 8: Statistical distribution of the raster surfaces created from the various interpolation
methods in Project Area 1.

Use Table 8 to compare the statistical variation within Project Area one subset data.
The Kriging methodology, using the Wizard default settings, caused a drastic
minimization of the mean values within the area when compared to the tool and the
Wizard optimized setting which provides less dramatic variation in the surface created.
Spline Regular identifies drastic changes from the distribution of the measured points.
The natural neighbor tool skews the mean of the subset data from 2.3 to 5.1. These
variations are interesting within the Project Area, but can only foster a limited
understanding of the entire Project Area. The maps of the raster surface, seen in Figure
21, provide a visual comparison of the interpolation and parameter variation results.
68

Figure 21: Surface raster maps of interpolation algorithms. Note the differences between
the different interpolation algorithms.

69

Figure 23: Profile graphs of transect A for Project Area 1. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project Area:
a. IDW, b. NN, c. Spline Regular, d. Spline Tension, e. Kriging, f. EBK, g. Topo to Raster.
70

Figure 24: Profile graphs of transect B for Project Area 1. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project Area:
a. IDW, b. NN, c. Spline Regular, d. Spline Tension, e. Kriging, f. EBK, g. Topo to Raster.
71

Figure 25: Profile graphs of transect C for Project Area 1. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project Area:
a. IDW, b. NN, c. Spline Regular, d. Spline Tension, e. Kriging, f. EBK, g. Topo to Raster.

72

Transect A (Figure 23) produces variation within interpolation algorithms based on
the use of the tool, the Wizard’s default parameters and the optimized version of the
Wizard parameters. Spline Regular, Spline Tension and Kriging can be seen with drastic
variation. The profile seems to generate a step at a distance of about 740 ft.; this can be
seen by the interpolation methods of IDW, Spline Regular, Kriging and EBK.
Transect B’s (Figure 24) variation of the interpolation algorithms can be seen in all
the profiles with Spline Regular and Kriging being the most dramatic at a profile distance
of about 50ft to 250ft. When comparing the methods, natural neighbor seem to cause
increasing elevation at the beginning of the profile, where most of the other interpolations
restrict the surface to create a level surface.
Transect C (Figure 25) IDW, Spline Regular and Tension as well as Kriging
illustrates the differences when using the tool or the Geostatistical Wizard as seen from a
distance of about 50ft to 250ft. Spline Regular causes a wave within the data, where
Spline Tension and Kriging interpolations create an erratic change, even while using the
Wizards’ optimization options.
The variations mentioned within the profile graphs can be visually seen on the maps
for each transects (Figure 21). The bank lines have slight differences as well as the depth
of the water within the Project Area. Spline Regular has the most visually unsettling
surface, where islands and lakes area created. The remaining interpolation
methodologies, the differences are challenging to visualize, the use of a difference maps
73

(Figure 26). The difference within the Kriging interpolation is rather great within the
topographic areas.

Figure 26: Difference maps, between the tool and the Wizard.
When comparing the interpolations overall the Wizard default and optimized
parameters create generally smoother surfaces where the tools can create generally
unpredictable results. This can be seen in most interpolation algorithms used, with the
exception of Spline Regular, where both the tool and the Wizard generate unnecessary
variations.
5.2.2 Project Area 2
Project Area 2 is characterized by a general straight flow of the river, with distinct
riverbanks. The area is notorious for containing underwater impedances due to old bridge
pilings.
74

Project Area 2 has a wider variety mean and standard deviations seen in Table 9.
Within the Spline Regular used within the Spatial Analyst tool or Wizard and the natural
neighbor tool has the greater dissimilarity. The Kriging interpolation using the Wizard
default setting and the optimized setting produced similar results when looking at the
RMSE, but the max elevation has a difference of around 14 ft. EBK Wizard default
produces a lower minimum value than the rest of the interpolation methods with one of
the lowest RMSE values. Even with the statistics of the interpolated surfaces, matching to
the measured point statistics are still skewed, where the measured points had a mean of
2.5, the average interpolated mean was 0.2. Visualization of the maps Figure 27,
transects Figures 28 – 30, and is necessary to identify the anomalies.
75

Table 9: Table of the statistical distribution in Project Area 2.

Figure 27: Surface raster maps of interpolation algorithms. Note the differences between
the different interpolation algorithms.
76

Figure 28: Profile graphs of transect A for Project Area 2. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project Area:
a. IDW, b. NN, c. Spline Regular, d. Spline Tension, e. Kriging, f. EBK, g. Topo to Raster.
77

Figure 29: Profile graphs of transect B for Project Area 2. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project Area:
a. IDW, b. NN, c. Spline Regular, d. Spline Tension, e. Kriging, f. EBK, g. Topo to Raster.
78

Figure 30: Profile graphs of transect C for Project Area 2. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project Area:
a. IDW, b. NN, c. Spline Regular, d. Spline Tension, e. Kriging, f. EBK, g. Topo to Raster.

79

Transect A has similar trends between the surfaces created within the Tool, Wizard
Default parameters and the Wizard optimized, except in the case of the Spline Regular as
seen in Figure 28. The Spline Regular tool interpolated surface contained great variation
of the below water surface, creating peaks of over 20 ft., where the Wizard Default
parameters created a smoother surface comparable to the other interpolated surfaces.
Spline Tension tool, created not so drastic anomalies as the Regular Spline tool method,
although the Wizard Default settings also minimized these variations. The Kriging
interpolation contained some variation within the various uses methods, but generally
followed the same path.
Transect B (Figure 29), has similar results between the tool, Wizard default or Wizard
optimized except in the case of the Spline Regular then using the tool. The slight
variation can be seen within the Kriging interpolated surfaces, where the Wizard Default
and optimized settings were slightly above the tool’s surface at around 200ft, then at
1500ft shot above the tools estimation, this is caused by the distance between the known
points. The IDW interpolation methodology in created an interesting bench at around
1250ft where the Spline Regular Wizard default and the Topo to Raster tool also creates
some sort of benched feature. This benched feature is actually a retaining wall for an old
bridge and decommissioned road.
Transect C contain interesting variation of the interpolated surfaces at a distance on
the transect of 200 to 300 ft. illustrated in Figure 30. This area is about where the land
and water measured points join, although on the other end of transect at between 1000 to
80

1100ft the variations can also be seen. Kriging compared to the IDW interpolation
methods are most interesting between the 200 and 300ft area the Kriging Wizard default
seems similar to the IDW interpolation, but the Kriging Wizard optimized setting the
smooth surface disappears into sharp peaks where the only other interpolation method
which dose this is the EBK Wizard default. The Spline Regular tool again produces
undulating anomalies within the interpolate surfaces, but when using the Wizard default
setting seem to produce respectable results.
The stark differences can be seen between the Spline Regular tool when compared to
the Wizard default setting as seen in Figure 29. The Spline Regular default Wizard is
comparable to the EBK Wizard outputs when compared visually, where the Kriging
interpolation surfaces create a choppy representation of elevation for the Project Area.
The Topo to Raster tool appears to encompass more of the morphological channelization
of the stream channels entering the river. The differences between the interpolated tools
and Wizards are moderately tame within this Project Area as seen in Figure 31, with the
exception of the Spline Regular interpolation method, using the Wizard default setting,
Spline Regular produces near similar results than other interpolation methodologies.
81

Figure 31: Maps illustrating the difference between the surfaces created by the Spatial
Analyst tool and the Geostatistical Analyst Extensions.
Overall Project Area 2 several interpolated surfaces seem most appropriate to
ascertain the smooth elevation surfaces of the Project Area, such as all the IDW
interpolations, Kriging Wizard default, Spline Regular and Tension Wizard defaults,
Natural Neighbor and Topo to Raster interpolation tools.
5.2.3 Project Area 3
Project Area 3 is the most complex, not with respect to river morphology, but with
respect to having a peninsula within the Project Area. This causes slightly varying results
then if the study was completed with the entire dataset and the interpolated surfaces were
clipped to the Project Areas. Project Area 3 is a great representation of the data
inconsistencies, which might occur when there is a time delay in LiDAR or Sonar data
collections. The peninsula at the LiDAR collection was visible, but during the bathymetry
82

collection, either the river level was higher or the morphology changed to remove the
sedimentation, which had created the spit.
Table 10: Table of the statistical distribution in Project Area 3.

Project Area 3’s statistics is summarized in Table 10. This Project Area had two
interpolation methodologies with the smallest RMSE, which include Kriging and Spline
Regular when using the Wizard optimized and default setting. Although a small RMSE is
important, following the trends of the measured points is also a factor in determining
suitability of a surfaces interpolation. The subset data for Project Area 3 has a man of 1.2
and a wide range of standard deviation with a value of 14.7. The interpolated surfaces
followed similar trends away from the measured values, although natural neighbor and
Spline Regular tools had results not following the other surfaces. The surfaces of Kriging
produce statistically interesting results, the tool had a mean of 0.4 and an standard
deviation of 14.4, but when utilizing the Geostatistical Wizard in the default parameters,
the mean drops to -0.5 and a standard deviation of 15.1 and even further to the Wizard’s
optimized settings, where the mean is 0.2 and the standard deviation is at 14.1. These
83

statistical changes can also be seen within the map outputs (Figure 32) and the profiles
(Figures 33-35).

Figure 32: Surface raster maps of interpolation algorithms. Note the differences between
the different interpolation algorithms.
84

Figure 33: Profile graphs of transect A for Project Area 3. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project Area: a.
IDW, b. NN, c. Spline Regular, d. Spline Tension, e. Kriging, f. EBK, g. Topo to Raster.
85

Figure 34: Profile graphs of transect B for Project Area 3. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project Area:
a. IDW, b. NN, c. Spline Regular, d. Spline Tension, e. Kriging, f. EBK, g. Topo to Raster.
86

Figure 35: Profile graphs of transect C for Project Area 3. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project Area:
a. IDW, b. NN, c. Spline Regular, d. Spline Tension, e. Kriging, f. EBK, g. Topo to Raster.

87

Transect A for Project Area 3 has dramatic variation in the surfaces created from
interpolated methods as well as the tool, Wizard default and Wizard optimized settings.
(Figure 33) The IDW methods area extremely similar that only slight variations are
visible in the wide measurements gaps within the surface. The surfaces created from the
Kriging interpolations, particularly within the instance along the transect of 700 to 900ft
are drastic, where the tool surface is similar to the interpolated surfaces such as IDW,
EBK, and natural neighbors. Spline Regular using the tool seems to produce an
uncharacteristic smooth result but in this case does not follow the data trends. The Topo
to Raster tool seems to smooth some of the rapid variation in the surface when compared
to the other interpolation methods.
Transect B has one feature which has in interesting influence on the surfaces created
as seen in Figure 34. This transect has the least amount of measured points surrounding
the profile which causes the surfaces to rely on point measurements further away, some
of which are within the discrepancy areas. Both the Spline Regular and Tension tools
return to the normal form of creating exaggerated mounds within the surface, which some
might consider the Tension Spline tool to be favorable when compared to the erratic steps
created by the Wizard default surface.
The peninsula creates interesting results within transect C as seen in Figure 35,
although the transect does not exactly include the peninsula, the effects of the data
variability can be visualized in the interpolated surfaces. The Kriging interpolation just as
the rest of the Project Area 3 shows the variation with the tool, Wizard default and
88

Wizard optimized settings. The peninsula influence can be visualized within the 300 and
350ft area. The influence can be seen as the Kriging, Spline Regular and Tension, with
the Kriging creating different surfaces for the discrepancies. The variation in most
surfaces can be seen as well in the distance range of 600 to 900ft, due to a large data gap.
The influences of the surrounding measured points at a further distance creates
manipulations within the interpolated surfaces, where most interpolated surfaces when
using the Wizard default of optimized methodology follow a sharp curve at about a
distance of 825ft.
The visualizations of the surfaces are starkly different then the visualized profiles or
statistical overview of the surfaces. The representations of the peninsula vary, the surface
created from the Spline Regular, Spline Tension and EBK all produce multiple
island/peninsula visualization. Although the visualizations of this Project Area visually
seem skewed, the representations made by the various interpolations represent the
influencing factors of the interpolation methodology. Where the measured point values
are densely spaced, the values are all within a reasonable variation and where
discrepancies exist such as in Transect 3, the measured points directly on the raster grid
cell prevails. The differences between the Tool, Wizard default or optimized settings can
be seen within this Project Area more than others. Not only within the peninsula area, but
with the bank on the right hand side of the page, the variations are distinct in all
interpolation methodologies.
89

The lack of consistency within the overall Project Area might be attributed to the
sparseness of the overall measured points. The timing of the data collections, the choice
of LiDAR and Sonar point integration methodology and the lack of verification on the
ground all attributed to the irregularities within Project Area 3. Although the data
succeeded in illustrating the interpolated surfaces methods for either incorporating, the
surrounding measured points to varying degrees.
5.2.4 Project Area 4
Project Area 4 is has a gradually sloping bankline with deep channelization of the
river due to the force of the water upstream and has a sharp incline into the land once
again via a levee feature and boat docks.
Table 11: Table of the statistical distribution in Project Area 4.

The statically comparison in Table 11 of Project Area 4 is not as large as other Project
Areas, but the impact on the resulting interpolated surface raster is visible as seen in
Figure 36. The variation in standard deviation from the Kriging tool and the Kriging
Wizard optimized setting is drastic, as well as compared to the variation seen in the
Spline Regular tool. The subset measured points for the area has a mean of -0.2, where
90

the interpolated surfaces caused reeducation in this value ranging from -2.2 to -3.5. The
Kriging Wizard optimized setting also produced the lowest RMSE, and the maximum
and minimum just fall outside the measured point’s values.

Figure 36: Surface raster maps of interpolation algorithms. Note the differences between
the different interpolation algorithms.
Transect A majority of the variation can be seen at a profile distance of about 950 to
1350ft in Figure 37. This variation can be seen within the methodologies of the Kriging
tool, Wizard default and Wizard optimized. The Kriging Wizard optimized parameters
cause the depth to an exaggerated amount compared to the other interpolated surfaces,
91

another interesting note is that the tool for Kriging produces a bench at the end of the
transect where none of the other interpolation methods do so. Within this Project Area,
the topo to raster tool interpolates an interesting series of dips at a profile distance of 900
to 1500ft. where Spline Regular also has sever dip at this location, the influence of distant
measurements on the surface profiles.
The variation with the Wizard default and the optimized setting is small within
Transect B, but the differences in the tool parameters are visible within the profiles seen
in Figure 38. Kriging interpolation using the tool creates a stair-step surface around the
profile distance of 100 to 350ft and again at 1500 to 1250ft, this discrepancy within the
surface of at one point nearly 15ft. Spline Regular and Tension tools create rolling hills
within the surface to illustrate the value distribution, where a slight shelf follows the data
values more functionally. The IDW interpolation between the Wizard and the tool, a
slight peak is created where there is a lack of data values.
Transect C contains several sections of clustered data along the transect profile on
Figure 39. This clusterization and sparse distribution of other data causes models such as
Spline Regular tool to create steep peaks within the surface, where when using the
Wizard default smooth these features. The Kriging tool creates a visual stair-step
approach simulate to that of the IDW interpolated surfaces, but when using the optimized
setting of Kriging is generally smoothed with the exception of rugged variation near a
profile distance of 11000 to 1200ft. EBK tool created a seemingly smooth surface
92

compared to the optimized version, but within out complete visualization of the surface
the judgment cannot be made.
93

Figure 37: Profile graphs of transect A for Project Area 4. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project Area:
a. IDW, b. NN, c. Spline Regular, d. Spline Tension, e. Kriging, f. EBK, g. Topo to Raster.

94

Figure 38: Profile graphs of transect B for Project Area 4. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project Area:
a. IDW, b. NN, c. Spline Regular, d. Spline Tension, e. Kriging, f. EBK, g. Topo to Raster.

95

Figure 39: Profile graphs of transect C for Project Area 4. Each graph represents
interpolation algorithms, and all graphs represent the same transect within the Project
Area: a. IDW, b. NN, c. Spline Regular, d. Spline Tension, e. Kriging, f. EBK, g. Topo to
Raster.

96

The visual representation of the elevation profiles all generally follow a seemingly
smooth transition from topography to the bathymetry and back, with the exception once
more with the Spline Regular tool seen in Figure 36 and 40. The Spline Regular tool
surface has islands and anomalies where there are no measured points. The Topo to
Raster tool produces a visually pleasing output of their surface concentrating on the
gradients of elevation. The Kriging interpolation and the Spline Regular tool have the
greatest differences between the tool and the Wizard default or optimized settings.
Although, all surfaces have some high degree of variation

Figure 40: Difference maps comparing the surfaces created from the Spatial Analyst and
the Geostatistical Analyst Extension.
Project Area 4, which is complicated by deep channelization of the river flow as well
as manmade features of a levee within close proximity, illustrates the need for
understanding the statistical variation of the dataset and the mathematical interpolation
the data is undergoing when interpolating the result. The small jitters of surfaces such as
97

those created by using the Wizard optimized setting in Kriging and EBK can affect the
surface characteristics.
5.2.5 Project Area 5
This Project Area 5 contains a low elevation wetland island, with a deep channel river
outlet as the water flows near the ocean to a manmade levee and boat ramp. This Project
Area is most susceptible to tidal influence, although we are not accounting for tidal flow
into the measurements.
Table 12: Table of the statistical distribution in Project Area 5.

Project Area 5 has a measured mean elevation value of -2.6 and standard deviation of
11.2, although all the interpolated surfaces underestimated the mean elevation at around -
4.6, the standard deviation was just slightly off. (Table 12) The Spline Regular tool
produces most variation of the surface with a standard deviation of 23.5, where the Spline
Regular Wizard default is able to smooth the changes to an acceptable value of 11.2. The
Kriging Wizard default properties did appear to both lower and increase the minimum
and maximum value, the mean and standard deviation were within the normal variation,
where the RMSE was one of the lowest within this Project Area. The maps illustrating
the statistical variability can be seen in Figure 41.
98

Figure 41: Surface raster maps of interpolation algorithms. Note the differences between
the different interpolation algorithms.
99

Figure 42: Profile graphs of transect A for Project Area 5. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project Area:
a. IDW, b. NN, c. Spline Regular, d. Spline Tension, e. Kriging, f. EBK, g. Topo to Raster.
100

Figure 43: Profile graphs of transect B for Project Area 5. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project Area:
a. IDW, b. NN, c. Spline Regular, d. Spline Tension, e. Kriging, f. EBK, g. Topo to Raster.

101

Figure 44: Profile graphs of transect C for Project Area 5. Each graph represents an
interpolation algorithm, and all graphs represent the same transect within the Project
Area: a. IDW, b. NN, c. Spline Regular, d. Spline Tension, e. Kriging, f. EBK, g. Topo
to Raster.
102

Transect A has several large discrepancies within the transect profile see Figure 42.
The variations between the tool and the Wizard default or optimized can be visualized in
nearly all the interpolated surfaces. The IDW surface, although the surfaces created
seems irrational along the profile distance between 1100 to 1300ft the elevation behind
the levee actually dips to a certain degree, where other interpolation methods attempted to
smooth or reduce this on-the-ground data anomaly. The Kriging and spline tension
Wizard surfaces created sharp transitions of the bathymetric surface, where topo to raster,
EBK and natural neighbor seem to adjust these variations into a smooth surface.
A road features is contained within profile B, which should result in a bench
representation of the data in Figure 43. Topo to raster, EBK and natural neighbor do not
seem to account for the temporally consistent drop of elevation from the hillside below
and above the road as seen on the profile distance 1250 to 1325ft. Other variation
noticeable particularly within the Kriging surfaces occur within the profile distance of
250 and 400ft. where the variation between the tool the Wizard default and the Wizard
optimized have a gap of at least 3 feet between the surfaces. At the other end of the
Kriging profile at distance of 1100 to 1325 ft. the variations are also erratic.
Transect C has the similar features as transect B with the levee and road, but the
distribution of the clustered measured pointes is further, causing smoother but still
different interpolations within that area. (Figure 44) The spline tension interpolated
surfaces at a distance of 1250ft can be seen diverging, with the tool having a less gradual
slope increase than the Wizard default. The surfaces of IDW the variations between the
103

tool, Wizard default and Wizard optimized can be seen, where the tool and Wizard
default follow the similar pattern, but when the surface has the Wizards optimized setting
accounting for the Spatial Statistics of the measured points the surface changes greatly to
encompass the slight slopes and bank features.

Figure 45: Difference maps of the Interpolation algorithms comparing the surfaces created
by the Spatial Analyst and Geostatistical analyst.
When visualizing the elevation of these surfaces, a smooth transition of gradient is
necessary when looking in the surface maps in Figure 41. When viewing the surfaces
created by the Kriging interpolation methods both the default and optimized Wizard
create jittery appearance. IDW, Natural Neighbor and Topo to Raster produce smooth
results, where EBK Wizard’s surface has some disjointed colorization, the overall effect
is correct of incorporating the levee and road features. The differences are visible within
all surface profiles, with EBK tool and Wizard optimized having the least amount of
variation, which only occur near the manmade features as seen in Figure 45.
104

Project Area 5 has the most varying interpolated surface of any of the previous
Project Areas, not only between interpolations, but when comparing the tool to the
Wizard default or optimized settings. As expected from the previous interpolations,
Spline Regular tool created outrageously varied results. The clustered data and the
sparseness of some of the dataset areas cause such erratic variation within the
interpolated surfaces.
5.3 Results Summary
The anticipated result of comparing the parameters from the Spatial Analyst and the
Geostatistical Analyst Extensions was that there was some slight variance, but the degree
of discrepancy illustrated in the results was not anticipated. The largely held concept of
tools ArcGIS having multiple ways to access an interpolation algorithm cannot be held,
the parameter differences make any of these algorithms preform drastically different. The
change of the parameters had a great impact in all the resulting surfaces.
When analyzing the tables exclusively, one raster surface has noticeable different
results than the rest. The Spline Regular tool surface contains values greatly exaggerated
from any other surface or the measured points. Although when the Spline Regular
method of interpolation is utilized within the Geostatistical Analyst, the values return to a
reasonable degree of variation. The interpolation method of the Geostatistical Analyst
Wizard Spline Tension, Kriging and EBK, shows generally a lower overall value of
RMSE. These methods follow closely the measured points for both the subset surface in
105

which they were created from as well as the entire Project Area. This phenomenon is
visualized in the transect profiles of the Project Areas.
The profile view provides the comparison of the various interpolated surfaces to make
a determination judgment on the suitability and continuity of the dataset for the
topography and bathymetric datasets integration. The surface profiles when utilizing the
Spatial Analyst tools, provided variation compared to the Geostatistical Analyst Wizard,
when the optimization setting was utilized. The variation of the surfaces between the tool
and the Wizard are overall greater than expected.
The raster surfaced visualized as a map output, then classified in a color schema
exaggerating the land/water integrations provided overall characteristics of the surface
created by the various interpolation methods that cannot be ascertained by the transects
alone.

106

Chapter 6: Discussion and Conclusions
Providing information is the not the same as providing an understanding. The developing
field of GIS is tasked with not only collecting and illustrating information; it has to
provide an understanding of the world in a narrative. The challenge of bridging the gap
between the information collection and the storyline is usually lost in translation by a
data-rich GIS community. GIS professionals rely heavily on the derived datasets from the
nation’s leading government agencies such as the USGS, without consideration for the
background development of the information. Elevation, which is the most fundamental
geospatial dataset available from the USGS in the form of DEMs, lacks the attention and
scrutiny it deserves. The measurement of elevation is currently set by industry standards
for LiDAR collections, but there is no standard on how to convert the LiDAR
measurements into a raster surface for use in geospatial analysis. ArcGIS, the world’s
leading geospatial software company that is used by all US government agencies,
provides multiple pathways (extensions) for the interpolation of points into a raster
surface this allows for uncertainty. Documenting the exact software, version, extension,
and parameter selections can help the users to become scientists and better storytellers by
building the platform to develop the geospatial story.
Creating a representative surface of the Earth’s elevation, interpolating from point
datasets to a raster, is taught in most introductory GIS courses with the latest Esri ArcGIS
software but attempting to create the most reliable and accurate surface is more
complicated than relying on the default parameters of the Spatial Analysis Extension’s
107

interpolation algorithms. Esri provides a Geostatistical approach to point to raster surface
creation in the Geostatistical Analyst Extension, which has many of the same
interpolation algorithms as the Spatial Analyst Extension, yet it produces different
results. The variation, usually unbeknownst to the GIS professional, can have
implications on the resulting analysis. The variation can be seen in the selection of which
points to use as the cornerstones of the interpolation, also known as the search
neighborhood. The power functions in many of the weighted interpolation algorithms can
also differ between the Spatial Analyst Extension and the Geostatistical Analyst
Extension, with the Geostatistical Wizard providing optimization (sometimes by default).
The Geostatistical Wizard takes the statistical variability of the point dataset into
consideration when interpolating, this influences the Wizard’s parameter choices and
provides an illustrative level of confidence in the elevation surface created from the
parameters in the forms of a RMSE. The change in parameters when using the
Geostatistical Wizard optimized parameter values creates a more reliable interpolated
surface than using the Spatial Analyst interpolation tools’ parameters alone.
The incorporation of two diverse datasets in density, distribution and accuracy, such
as Topographic/Bathymetric measurements, provide an illustration of the need to
statistically understand the dataset prior to and during interpolation. The use of the
interpolation algorithms: Inverse Distance Weighting, Natural Neighbor, Spline Regular,
Spline Tension, Kriging, Empirical Bayesian Kriging, and Topo to Raster,- can be
accessed through Spatial Analyst and/or Geostatistical Analyst. The interpolation
108

algorithms have their own requirements of Normal Distribution, Data Stationarity,
Trends, and spatial auto-correlation, can and should influence the choice of interpolation
based on the statistical distribution of the point dataset. Although each interpolation
algorithm’s method is essentially correct based on the parameters selected, their
reliability fluctuates. With Topographic/Bathymetric surface field validation, the
interpolated surface is not feasible due to the high rate of surface variability. The ability
to validate the surface is attempted by three methods; comparing the interpolated surfaces
to an existing DEM, comparing the differences between the measured points and the
interpolated surfaces, and using a type of cross validation across all surfaces. These
attempts to determine a reliable measure of implied correctness of a surface is performed
on Project Area 1 and is achieved with both visual inspection and statistical checking.
The California Coastal Conservancy Coastal LiDAR project developed a digital
elevation model from the bare earth LiDAR. This attempt to validate the raster surfaces
will compare the bare earth DEM to the Interpolated surfaces created from that same
LiDAR point dataset and the single beam sonar dataset. One would expect that the
differences would only show within the Bathymetric area, but variation can be visualized
throughout the surfaces as seen in Figure 46.
109

Figure 46: California Coastal Conservancy (CCC) LiDAR Bare Earth DEM subtracted
from the interpolated surface. The differences are repersented by the color differences.
Within the results section, spatial comparison of the Tool and the Geostatistical
Wizard within the same interpolation algorithm but the overall comparison between
interpolation algorithms did not occur. This complete comparison visible within appendix
a., provides a general overview of the differences between the surfaces. Taking each
interpolation algorithm created from each Esri extension, using raster calculator to
compare it to every other interpolation algorithm and extension. An example of the cross
comparison can be seen in figure 47, gives a better understanding of the how the various
interpolation algorithms create different results.
110

Figure 47: Two examples of the resulting Cross comparison. Taking the interpolation
method/extension combination and subtracting it to each interpolation - extension
interpolated raster surface created for Project Area 1 within the Study Area. Top image
shows the cross comparison of the IDW interpolation used with the Spatial Analyst
extension. Bottom image shows the cross comparison of the Natural NN interpolation
methodology.
111

As both the LiDAR and Sonar datasets have undergone post processioning and
additional quality assessments, this data is considered the true values of the elevation and
depth. Using the 500 subset points used to create the interpolated surfaces is then
compared to the raster value of the interpolated surfaces. This illustrates two concepts,
first if the interpolation is an exact or inexact interpolator and that when comparing the
surfaces, using the Wizard’s optimized settings improves the overall measured accuracy.
This is visualized by the histograms in Figure 48. The difference between the measured
points and the interpolated surfaces should be near zero an example of this is IDW. IDW
is an exact interpolator, where when using the Spatial Analyst tool surface 98% falls
within 0.1 foot difference, with the Geostatistical Wizard surface it is also 98%, but with
the optimized Wizard interpolated surface 100% of the points fall on the surface exactly.
Another example using the Kriging interpolation, where in Esri help documentation states
that if the Kriging Parameter is set to zero then Kriging is an exact interpolator, but
within the Spatial Analyst there is no parameter setting for this, and for the Geostatistical
Analyst is default set to 100, therefor the kriging is assumed to not be an exact
interpolator. When using the Spatial Analyst tool surface 25% falls within 0.1 foot
difference, with the Geostatistical Wizard surface it is 21%, but with the optimized
Wizard interpolated surface 88% of the points fall on the surface exactly. This is an
interesting concept for the overall justification for using the Geostatistical Wizard’s
optimized setting.
112

Figure 48: Elevation difference between the Measured points LiDAR and Sonar Dataset to
the Esri interpolated surfaces. As the difference illustrated by the X-Axis, is closer to 0, the
more similar the overall interpolated datasets fit the measured points.
113

For the entire study area’s interpolated surfaces, the measures discussed can only
provide an implied correctness, but some interpolation algorithms require special note;
Regularized Spline and EBK.
The surface generated by the Completely Regularized Spline interpolation algorithm
has several very different results when comparing the usage of the Spatial Analyst and
the Geostatistical Analyst Extension. This interpolation method attempts to ensure a
smooth surface by finding a function that minimizes the tradeoff between fitting the
points and the smoothness of the surface. When using the Spatial Analyst default
parameters, this function is static and cannot vary based on the statistical variation of the
dataset. This results in erratic undulations of the raster surface. The Geostatistical Analyst
Extension accomplished a slight reduction in the drastic variations, but when the
Geostatistical Wizard is set to optimize the dataset the parameters of the variation, the
surface produces both statistically and visually becomes an expected result.
Empirical Bayesian Kriging (EBK) model is the newest interpolation methodology
incorporated into ArcGIS Desktop. EBK is “a particularly robust method of data
interpolation because it accounts for the error introduced by estimating the
Semivariogram model” (Krivoruchko 2012, b). EBK is only available in Esri’s
Geostatistical Analyst Extension, although when investigating the Geostatistical Analyst
Extension and the Geostatistical Analyst toolbox, a change in search radius does occur.
This variation between two seemingly exact interpolation algorithms modifies the surface
114

elevations and is important to note for usage of the Geostatistical Analyst with automated
geospatial processes such as Esri’s ModelBuilder, which requires the usage of toolboxes
rather than Wizards.
Although the general finding within this study concludes that the Geostatistical
Wizard’s optimized settings will produce greater reliability, and a mathematical
explanation for the parameters selected, this can also be achieved manually within the
Spatial Analyst Extension, but requires a more advanced mathematical background than
that provided by most University bachelor’s programs (D. Watson 1992). The
Geostatistical Wizard provides the flexibility to walk the user through the parameters
selection with the opportunity to on-the-fly modify the parameters and see the effect on
the resulting surface without committing to the changes, while providing the ability to use
advanced statistical modeling to select the parameters with a click of the optimization
setting.
As GIS continues to advance as a field, the creation of additional data models will
need to be developed, as the data models of Vector, TIN, and Raster cannot adequately
describe all conditions of the earth’s surface, which GIS professionals need to model.
GIS Professionals can move towards to not only provide information but provide an
understanding, remembering that not everything that counts can be counted, not
everything that can be counted should be and never forget the Metadata.

115

Bibliography
Ali, T, S. C. Medeiros, S. C. Hagen, and J. P. Ralford. 2011. Development of a seamless
topographic / bathymetric digital terrain model for Tampa Bay, Florida.
Photogrammetric Engineering and Remote Sensing 77(12): 1249-1256.
Arun, P. V. 2013. A comparative analysis of different DEM interpolation methods.
Geodesy and Cartography 39(4): 171-177.
Aykut, N. O., B. Akpinar, and O. Aydin. 2012. Hydrographic data modeling methods for
developing precise seafloor topography. Computer Geoscience (Springer) 17:
661-669.
Badea, A.-C. Comparative Study between Geostatistical Models applied to analyze the
phenomena caused by geomorphic processes. Recent Advances in Geodesy and
Geomatics Engineering. GENG’13 Conference, 2013.
Barnard, P. L., and D. Hoover. 2010. A seamless, high-resolution, coastal digital
elevation model (DEM) for Southern California. Data series 487, U.S. Geological
Survey.
Bernstein, D. J., S. W. Benjamin, C. W. Freeman, and H. Mitasova. 2011. Modern
techniques for improved topo/bathy elevation modeling of tidal inlets. U.S.
HYDRO.
Berry, J. K. 1996-2013. Topic 8: Investigating spatial dependency. In Beyond Mapping
III : Compilation of Beyond Mapping columns appearing in GeoWorld magazine.
Bohling, G. October 2005. Introduction to geostatistics and variogram analysis.
Resources for C&PE 940. Lawrence, Kansas: University of Kansas.
Breman, J., ed. 2010 Ocean Globe. Redlands: ESRI.
Burrough, P. A. 1986 Principales of Geographical Information Systems for Land
Resources. New York: Oxford University Press.
Childs, C. July-September 2004. Interpolating surfaces in ArcGIS spatial analyst. Arc
User, 32-35.
Chiles, J-P, and P. Delfiner. 1999. Chapter 4: Geostatistics: modeling spatial uncertainty.
New York: John Wiley & Sons.
Dewberry Consultants LLC. 2013.Coastal california data merge project. NOAA
Contract: EA 133C-11-CQ-0007, Tampa, FL: Dewberry/ National oceanic and
atmospheric administration (NOAA).
116

Eakins, B. W., and P. R. Grothe. September 2014. Challenges in building coastal digital
elevation models. Journal of Coastal Research 30, 5: 942-953.
Erdogan, S. 2009 A comparision of interpolation methods for producing digital elevation
models at the field scale. Earth Surface Processes and Landforms 34: 366-376.
ESRI. ArcGIS 10.2.1 Help. Redlands, California, 1995-2013 .
—. Summer 2001. Spatial data exploration extension for ArcGIS: introducing
geostatistical analyst. ArcNews.
ESRI. 2007. GIS for Ocean Conservation. GIS Best Practices, Redlands: ESRI,.
Evans, G. 2013. National Elevation Dataset. Release Notes, U.S. Geological Survey.
Foxgrover, A. C., and P. Barnard. 2012. A seamless, high-resolution digitial elevation
model (DEM) of the north-central California coast. Data Series 684, U.S.
Geological Survey.
Franke, R. 1982. Smooth interpolation of scattered data by local thin plate splines.
Computer and Mathmatics with Applications 8(4): 273-281.
Gesch, D., and R. Wilson. 2002. Development of a seamless multisource topographic /
bathymetric elevation model of Tampa Bay. Marine Technology Society Journal
35:4.
Goodchild, M. F. 2000. Part 1: Spatial analysts and GIS practitioners: The current status
of GIS and spatial analysis. Journal of Geographical Systems 2: 5-10.
Heine, G. W. 1986. A controlled study of some two-dimensional interpolation methods.
COGS Computer Contributions 3(2): 60-72.
Heritage, G. L., D. J. Milan, A. R.G Lsrge, and I. C. Fuller. 2009. Influence of survey
strategy and interpolation model on DEM quality. Geomorphology 112: 334-344.
Kalkhan, M. A. 2011. Spatial Statistics: Geospatial Information Modeling and Thematic
Mapping. Boca Raton, Florida: CRC Press.
Kearns, T. A., and J. Breman. 2010. Bathymetry - the art and science of seafloor
modeling for modern applications. In Ocean globe, by J. (Ed.) Breman, edited by
J. Breman, 1-36.
Kravchenko, A. N. 2003. Influence of spatial structure on accuracy of interpolation
methods. Soil Science Socuety of America Journal 67(5): 1564-1571.
117

Krivoruchko, K. Fall 2012. Empirical bayesian kriging: implemented in ArcGIS
Goestatistical Analyst. Arc User.
—. Fall 2012. Modeling contamination using empirical bayesian kriging. ArcUser.
Krivoruchko, K., and C. Gotway. 2002. Expanding the "s" in GIS: incoperating spatial
statistics in GIS. CSISS Special Meeting on Spatial Data Analysis Software Tools.
Santa Barbara.
Liu, X. 2008. Airborne LiDAR for DEM generation: some critical issues. Progress in
Physical Geography. SAGE Publications. 32(1): 31-49.
Liu, X., Z. Zhang, and J. Peterson. 2009. Evaluation of the preformance of DEM
interpolation algorithms for LiDAR data. Proceedings of the Surveying & Spatial
Sciences Institute Biennial International Conference. Surveying & Spatial
Sciences Institute. 771-780.
Lu, G. Y., and D. W. Wong. 2008. An adaptive inverse-distance weighting spatial
interpolation technique. Computers and Geoscience 34: 1044-1055.
Lynch, S., and E. Krause. Geostatistical Analyst - Getting Started. 2014 Esri
International User Conference. San Diego: ESRI, 2014.
McBratney, A. B, and R. Webster. 1986. Choosing functions for semi-variograms of soil
properties and fitting them to sampling esimates. Journal of Soil Science 37: 617-
639.
Medeiros, S. C., T. Ali, S. C. Hagen, and J. P. Ralford. 2011. Development of a Seamless
Topographic / Bathymetric Digital Terrain Model for Tampa Bay Florida.
Photogrammetric Engineering and Remote Sensing (American Society for
Photogrammetry and Remote Sensing) 77(12): 11249 - 1256.
Merwade, V. 2009. Effect of spatial trends on interpolation of river bathymetry. Journal
of Hydrology 371: 169-181.
Mitas, L., and H. Mitasova. 1988. General variational approach to the interpolation
problem. Computers and Mathematics with Applications 16(12): 983-922.
Newkirk, G. 2005. LIDAR Guidebook: Concepts, Project Design, and Practical
Applications. 2. Vol. 17. City of Vacouver, Washinton: Urban and Regional
Information Systems Association (URISA).
NOAA Coastal services center. 2014. Guidance: Lidar provisioning guidance for the
digital coast data access viewer. NOAA.
118

NOAA Coastal Services Center. 2007. Topographic and bathymeteric data
considerations: datums, datum conversion techniques, and data intergration.
Technical report NOAA/CAC/20718-PUB, Washington, DC: National Oceanic
and Atmospheric Administration (NOAA).
Oliver, M. A. 1990. Kriging: a method of interpolation for geographical information
systems. International Journal of Geographic Information Systems 4: 313-332.
Philip, G. M., and D. F. Watson. 1982. A precise method for determining contoured
surfaces. Austrailian Petroleum Exploration Association Journal 22: 205-212.
Pilz, J., and G. Spock. 2007. Why do we need and how should we implement bayesian
krigiing methods. Stochastic Environemental Research and Risk Assessment. 621-
632.
Press, W. H., S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery. 1988. Numerical
Recipes in C: The Art of Scientific Computing. New York: Cambridge University
Press.
Raber, B. R., and J. Cannistra. 2005. LiDAR guidebook: concepts, project design and
practical applications. Park Ridge, Il: URISA.
Royle, A. G., F. L. Clausen, and P. Frederiksen. 1981. Practical Universal Kriging and
Automatic contouring. Geoprocesing, 377-394.
Scott, L. April 2010. Understanding Spatial Statistics and Geostatistics. ESRI ArcWatch.
Sibson, R. 1981. Chapter 2 : A brief description of natural neighbor interpolation. In
Interpolating Multivariate Data, 21-36. Ney Yourk: John Wiley & Sons.
Stoker, J. M., H. K. Heidemann, G. A. Evans, and S. K. Greenlee. 2013. A conceptual
prototype for the next-generation national elevation dataset. Open-File report
2013-1023, U.S. Geological Survey.
Watson, D. 1992. Contouring: A Guide to the Analysys and Display of Spatial Data.
London: Pergammon Press.
Watson, D. F., and G. M. Philip. 1985. A Refinement of inverse distance weighted
Interpolation. Geoprocessing 2: 315-327.
Wechsler, S. P. 2007. Uncertainties associated with digital elevation models for
hydrologic applications: a review. Hydrology and Earth System Sciences 11: 148-
1500.

119

Appendix A:
Comparing surfaces visually is an important component to understanding the variation
between interpolation method’s and extensions. Taking each interpolation algorithm
created from Spatial Analyst and Geostatistical Analyst extension, using raster calculator
to compare it to every other interpolation algorithm and extension. This cross comparison
gives a better understanding of the how the various interpolation algorithms create
different results.

120

121

Figure 49: Cross comparison. Taking the interpolation method/extension combination and
subtracting it to each interpolation - extension interpolated raster surface created for
Project Area 1 within the Study Area. a) IDW interpolation algorithm using the Spatial Analyst
extension is used as the base for comparison. b) IDW interpolation algorithm using the
Geostatistical Analyst extension default parameters is used as the base for comparison. c) IDW
interpolation algorithm using the Geostatistical Analyst extension optimized parameters is used as
the base for comparison. d) Natural Neighbor interpolation algorithm using the Spatial Analyst
extension. e) Spline Regular interpolation algorithm using the Spatial Analyst extension. f) Spline
Regular interpolation algorithm using the Geostatistical Analyst extension. g) Spline Tension
interpolation algorithm using the Spatial Analyst extension. h) Spline Tension interpolation
algorithm using the Geostatistical Analyst extension. i) Kriging interpolation algorithm using the
Spatial Analyst extension. j) Kriging interpolation algorithm using the Geostatistical Analyst
extension default parameters. k) Kriging interpolation algorithm using the Geostatistical Analyst
extension optimized parameters. l) EBK interpolation algorithm using the Geostatistical Analyst
toolset. m) EBK interpolation algorithm using the Geostatistical Analyst wizard default
parameters. n) EBK interpolation algorithm using the Spatial Analyst extension.

Abstract (if available)

Abstract High quality topographic (land elevation) and bathymetric (water depth) data is targeted by the USGS and other Federal agencies as a need for update and modernization, particularly with the rapidly advancing technological innovations for use in modeling hydrological and environmental changes. Esri’s ArcGIS provides advanced and various options to interpolate surfaces using two ArcGIS Extensions: Spatial Analyst and Geostatistical Analyst. These extensions provide access to advanced mathematical algorithms used in the interpolation of measured points into an elevation surface, through a user-friendly interface with pre-defined, yet highly technical input parameters. ❧ Using Light detection and Ranging (LIDAR) elevation measurements and Single Beam Sonar on the Klamath River Estuary, this project compares interpolation methods provided by ArcGIS in the Spatial Analyst and Geostatistical Analyst Extensions, in order to determine how varying the parameter settings affect the resulting surfaces. This case employs seven commonly use interpolation algorithms: Inverse Distance Weighting, Natural Neighbor, Spline Regular, Spline Tension, Kriging, Empirical Bayesian Kriging, and Topo to Raster, all of which can be used in Digital Elevation Model (DEM) surface creation. Understanding the differences between the two extensions and modifying parameters in each interpolation algorithm results in statistically reliable elevation surfaces. The results prove that modifying the default interpolation parameters to fit the statistical variability, which is completed by the optimization of the Geostatistical Analyst Wizard, improves the functional use of the study area raster surface. ❧ Chapter 1 introduces the study’s necessity. Chapter 2 describes the technical background, and provides all of the necessary information for understanding the results. Chapter 3 provides the related works. Chapter 4 describes the methodology and explains how to interpret the results. Chapter 5 presents the results from the case study on the Klamath River. Chapter six provides the conclusion and discussion.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Geological modeling in GIS for petroleum reservoir characterization and engineering: a 3D GIS-assisted geostatistics approach

PDF

Effect of spatial patterns on sampling design performance in a vegetation map accuracy assessment

PDF

Implementing spatial thinking with Web GIS in the non-profit sector: a case study of ArcGIS Online in the Pacific Symphony

PDF

Using Landsat and a Bayesian hard classifier to study forest change in the Salmon Creek Watershed area from 1972–2013

PDF

Building a spatial database of biochar research and practice with Web-GIS

PDF

Preparing for immigration reform: a spatial analysis of unauthorized immigrants

PDF

Walkability study for school accessibility: case study of the San Juan, Puerto Rico elementary schools

PDF

Evaluating surface casing depths of oil & gas operations in an effort to protect local groundwater: a GIS enabled process

PDF

The role of amenities in measuring park accessibility: a case study of Downey, California

PDF

Trojan Food Finder: a web-based GIS campus food sharing application

PDF

Assessing the impact of a web-based GIS application to promote earthquake preparation on the University of Southern California University Park Campus

PDF

Analyzing earthquake casualty risk at census block level: a case study in the Lexington Central Business District, Kentucky

PDF

Beach morphodynamic change detection using LiDAR during El Niño periods in Southern California

PDF

GIS-aided production of equipment locator maps for Metro rail maintenance and support

PDF

Wrong way driving in San Antonio, Texas: a transportation route study using Network Analyst

PDF

Links, tempos, and mass balances of cyclic deformation and magmatism in arcs: a case study on the Mesozoic Sierra Nevada arc integrating geological mapping, geochronology, geobarometry, strain an...

PDF

Spatial and temporal patterns of long-term temperature change in Southern California from 1935 to 2014

PDF

A Maxent-based model for identifying local-scale tree species richness patch boundaries in the Lake Tahoe Basin of California and Nevada

PDF

Investigation and analysis of land use/land cover to determine the impact of policy developments in cities

PDF

Increase in surface temperature and deep layer nitrate in the California Current: a spatiotemporal analysis of four-dimensional hydrographic data

Asset Metadata

Creator Rodriguez, Rachel R. (author)

Core Title Integration of topographic and bathymetric digital elevation model using ArcGIS interpolation methods: a case study of the Klamath River Estuary

School College of Letters, Arts and Sciences

Degree Master of Science

Degree Program Geographic Information Science and Technology

Publication Date 02/23/2015

Defense Date 12/05/2014

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

Tag ArcGIS,bathymetric,DEM,interpolation,Klamath River Estuary,OAI-PMH Harvest,topographic

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Lee, Su Jin (committee chair), Rashed, Tarek (committee member), Swift, Jennifer N. (committee member)

Creator Email rachel.r.rodriguez@gmail.com,rachelrr@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-536196

Unique identifier UC11297933

Identifier etd-RodriguezR-3207.pdf (filename),usctheses-c3-536196 (legacy record id)

Legacy Identifier etd-RodriguezR-3207.pdf

Dmrecord 536196

Document Type Thesis

Format application/pdf (imt)

Rights Rodriguez, Rachel R.

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