|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
USING REMOTE SENSING AND GEOGRAPHIC INFORMATION SYSTEMS FOR
FLOOD VULNERABILITY MAPPING OF THE C-111 BASIN IN SOUTH MIAMI-
WILLIAM ANDREW WEBB
A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF
FLORIDA INT PARTIAL FULFILLMENT FOR THE REQUIREMENTS FOR THE
DEGREE OF MASTER OF SCIENCE
UNIVERSITY OF FLORIDA
William Andrew Webb
This paper is dedicated to my parents Frank R. Webb and Brenda Y. Webb.
I would like to acknowledge my department professors Dr. Wendy Graham, Dr.
Carol Lehtola and Dr. Jack Jordan for their guidance and hard work in this proj ect. I
would like to thank Dr. Clint Slatton for providing his expertise in Airborne Laser Swath
Mapping. I would like to thank Don Pybas for his contribution and cooperation. I would
like to thank Charles Brown for his assistance during the development process.
TABLE OF CONTENTS
ACKNOWLEDGMENT S .............. .................... iv
LI ST OF T ABLE S ................. .............. vii...___.....
LIST OF FIGURES .............. .................... ix
GLOSSARY OF TERMS ............ _...... ._ ..............xii...
AB S TRAC T ......_ ................. ............_........x
1 INTRODUCTION ................. ...............1.......... ......
Background ................. ...............1.................
FI ood Management ................. ...............1.......... ......
Obj ectives ................. ...............3.......... ......
Proj ect Area ................. ...............4................
2 LITERATURE REVIEW .............. ...............7.....
Active and Passive Remote Sensing ................. ...............7............ ...
Spectral Si nature of Water............... ...............8.
Sensor Perform ance ................... ............ ............ ...............9.
Normalized Differential Vegetation Index (NDVI) .............. ....................1
Water Detection ................. ...............11.................
Cloud Detection ................. .......... ...............12.......
Airborne Laser Swath Mapping ................. ...............13................
ALSM Accuracy ........._..... ...._... ...............15.....
ALSM Point Removal .............. ...............16....
AL SM Applications ............._. ...._... ...............18....
Geographic Information Systems .............. ...............20....
Spatial M odeling.................... .............2
Inundation Mapping with GIS .............. ...............24....
3 DATA RESOURCES AND METHODOLOGY .............. ...............26....
Introducti on ................. ...............26___ .......
Surface Water Data ............... ... ........... ...............26...
Digital Elevation Model Construction................ .............2
Landsat 7 Enhanced Thematic Mapper .............. ...............28....
Vegetative Index Methodology .............. ...............28....
Unsupervised Classification .............. ...............29....
Bare Earth Modeling............... ...............30
Aerial Color Infrared Analysis .............. ...............31....
Ground Control Point Analysis .............. ...............32....
Topographic Spatial Modeling .............. ...............34....
Surface Water Elevation Map Methodology .............. ...............35....
Surface Water Elevation Map Interpolation ......._ ......... ___ ........._ ......37
Surface Water Inundation Map Methodology .............. ...............39....
4 RE SULT S AND DI SCU SSION ............... ...............4
Cloud Detection ............... ... .....___ ..... ....__ .............4
Vegetation Index Two and Vegetation Index Three ....._____ ........___ ..............51
NDVI .............. .. ...............55..
Topographic Analysis............... ...............60
Classified ALSM DEM .............. .. ...............61..
Surface Water Elevation Map Analysis............... ...............68
Surface Water Inundation Map ................. ...............93................
5 CONCLU SION................ .............11
Image Analysis ................. ...............114......... ......
Bare Earth Modeling ................. ...............114................
SW EM ................. ...............115......... ......
SW IM ................. ...............115......... ......
Conclusion ................ ...............116................
6 RECOMMENDATIONS FOR FUTURE STUDIES ................. ............ .........117
LIST OF REFERENCES ............__........_ ...............119....
BIOGRAPHICAL SKETCH .........._.... ...............123..__.........
LIST OF TABLES
2. 1 Specifications of a commercial Lidar system. .......................... ........13
3.1 Area 2 static GPS points. ............. ...............39.....
3.2 Inverse distance weighting search parameters for topography ................. ...............40
3.3 Global polynomial search parameters for topography ................. .......................41
3.4 Local polynomial search parameters for topography ................. ........................41
3.5 Radial based function search parameters for topography. ................... ...............4
3.6 Kriging search parameters for topography. ............. ...............42.....
3.7 Proj section Parameters for AL SM ................. ...............42........... .
3.8 Universal kriging search parameters for SWEM. .............. ...............43....
3.9 Simple kriging search parameters for SWEM. ............. ...............44.....
3.10 Ordinary Kriging search parameters for SWEM. ............. ...............45.....
3.11 Disjunctive kriging search parameters for SWEM. .................. ................4
3.12 Universal Kriging for SWEM ................. ...............48......____..
4.1 Radial based functions statistics for topography. ............. ...............106....
4.2 Inverse distance weighting statistics for topography. ................... ...............10
4.3 Global polynomial statistics for topography. ........_................. .............. ....107
4.4 Local polynomial statistics for topography ................. ...............107..............
4.5 Kriging statistics for topography. ............. ...............108....
4.6 Universal Kriging statistics for SWEM. .............. ...............108....
4.7 Universal kriging statistics for SWEM, 10/12-22/1999. ............. .....................0
4.8 Simple kriging statistics for SWEM, 10/12-22/1999. .............. ......__............109
4.9 Ordinary kriging statistics for SWEM. ............. ...............109....
4. 10 Disjunctive Kriging statistics for SWEM. ................ ...............111.............
4. 11 SWEM surface water values for October 12-22, 1999. ................ ............... .....112
4. 12 Vulnerability index classes used for SWIM. ................ .............................113
4. 13 Calculated inundation statistics for the study area ................. ................ ...._.113
LIST OF FIGURES
1.1 Map of Miami-Dade County and the proj ect area. ......___ ..... ...__ ........._.......4
1.2 M ap of the study area. ..............._ ...............5......... ...
2. 1 Illustration of a Lidar infrared beam. ......._...._ ... ........ ......_...........1
3.1 Color infrared aerial photos of the study area. ...._.._.._ ..... .._._. ....._.... .....3
3.2 Map of measurement sites. ........._.._.. ....__. ...._.._ ....._.. ........._......36
4. 1 Map of the Frog Pond with Band 8. ................ ...._.._ ...............50. .
4.2 Vegetation index two map of south Florida, October 16, 1999. ........._.._... ...............52
4.3 Vegetation index two map of the study area, October 16, 1999. .............. ... ............54
4.4. October 16, 1999, NDVI map of the south Florida. ............. .....................5
4.5. October 16, 1999, NDVI map of the study area. ............. ...... ............... 5
4.6 Planar view of the NAD 27 study area DEM. ................ ...............62.............
4.7 Graph of surface water elevation values.. ............. ...............70.....
4.8 SW EM ................. ...............71................
4.9 Prediction error for SWEM ................. ...............82...............
4. 10 SW IM ................. ...............95._ ___.....
F ro m
GLOSSARY OF TERMS
Determini sti c
ACIR is aerial color infrared imagery that is not
referenced with a coordinate system.
ALSM is a mapping technology that uses a laser to
map land or bathymetric topography.
A bare earth model is a DEM with artifact or
unwanted points removed.
A DEM is a 3D representation of a surface than may
be represented with raster cells or a TIN.
Deterministic interpolation uses deterministic
functions to predict values of a spatially distributed
field at unmeasured locations.
A DOQQ is similar to an aerial photograph except it
is referenced with a coordinate system and is used for
general GIS mapping applications.
A DSM is a 3D representation of a surface with
objects and man made features removed.
A DTM is a 3D representation of a surface that uses a
TIN to connect points.
FEMA is the disaster management and relief agency
of the federal government.
GIS is software that captures, stores, retrieves,
manipulates and displays geographically referenced
spatial tabular data.
Glossary of terms continued
GPS is a constellation of 24 satellites that provides
latitudinal and longitudinal data collected by a
Kriging is geostatistical interpolation technique that
uses the spatial correlation of a distributed field to
predict its value of unmeasured locations.
Landsat 7 ETM + is the seventh USGS satellite in a
series of satellites designed to capture environmental
data with visible, near infrared, mid-infrared, low and
high gain thermal sensor bands.
Light Detection and Ranging is the enabling laser
technology used for ALSM flight operations.
NAD 83 is the current horizontal datum used by the
National Geodetic Survey.
NAD 27 is the predecessor horizontal datum to NAD
NDVI is a vegetative index that is calculated as the
difference between the red and near infrared bands
divided by the sum of the red and near infrared
NGVD 29 is the predecessor vertical datum to
NGVD 88 is the current vertical datum used by the
National Geodetic Survey.
The National Park Service is controlled by the U.S.
Department of Interior and is responsible for the
management of all national parks.
A raster is a thematic map layer represented with a
Remote sensing refers to the capture of data without a
physical collection of the data.
NGVD 29 National
NGVD 88 National
NPS National Park
Glossary of terms continued
SCDS South Dade
SFWMD South Florida
Di stri ct
SWEM Surface Water
SWIM Surface Water
USGS U.S. Geological
The SDCS is the southern extension of the Central
and Southern Flood Control Proj ect and is located in
south Miami-Dade County.
The SFWMD is one of five water management
districts in Florida, and its district authority covers all
of southeast Florida.
The Surface Water Elevation Map is a representation
of the surface water for elevation over proj ect areas
The Surface Water Inundation Map is a
representation of the surface water elevation
measured in elevation above mean sea level with
respect to NGVD 88. The Surface Water Inundation
Map is the result of subtracting land surface elevation
grids from surface water elevation, and represents
depth of water on the land surface.
A TIN is a three dimensional representation of a
surface created by using triangles to link points.
The USGS is a multi-disciplinary science
organization that studies biology, geography,
geology, geospatial information, and water.
Vector is a thematic map layer represented by points,
lines and polygons.
Vegetative Index Two is calculated as the product of
the green band and low gain thermal band divided by
the high gain thermal band.
Vegetative Index Three is calculated as the low gain
thermal band divided by the sum of the mid-infrared
and red bands.
Abstract of Thesis Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Master of Science
USING REMOTE SENSING AND GEOGRAPHIC INFORMATION SYSTEMS FOR
FLOOD VULNERABILITY MAPPING OF THE C-111 BASIN IN SOUTH MIAMI-
William Andrew Webb
Chair: Wendy D. Graham
Maj or Department: Agricultural and Biological Engineering
The hydrologic cycle of south Florida frequently produces rain events that include
thunderstorms, tropical depressions and hurricanes. During 1999-2000, south Miami-
Dade was struck by two intense rain events that severely inundated local agricultural
operations for over a week. In the final assessment, agricultural losses sustained from
these storms totaled to nearly $430 million.
Flood hazard mapping has traditionally relied on paper maps that display the flood
extent with only polygon boundaries. Unfortunately, paper maps are greatly limited in
use, because they fail to show the extent, magnitude and duration of flooding. Recent
advances in airborne laser swath mapping, ALSM, and satellite sensor technology have
provided alternative types of data needed to more accurately map flood vulnerability. The
general scope of this proj ect is to improve mapping flood vulnerability in the southern C-
1 11 basin by combining a variety of remotely sensed data sets.
The procedure for mapping a severe flood condition following Hurricane Irene
involved the combination of ALSM, Landsat7 ETM+ and Geographic Information
Systems (GIS). Band 8, vegetation index two and vegetation index three derived from the
Landsat 7 ETM+ image were useful for mapping cloud cover, and the normalized
differential vegetation index (NDVI) was useful for mapping inundation produced by
Hurricane Irene. The primary limitations of vegetation index maps include the 30 meter
spatial resolution, and the obstruction of the spectral signature of water caused by
vegetation and clouds. Proj ect inundation maps created with regional surface water and
airborne laser swath mapped (ALSM) data displayed the flood duration, magnitude and
extent of the flood condition resulting from Hurricane Irene.
For nearly a century, south Miami-Dade's subtropical climate has provided a
suitable environment for consistent annual production of agricultural commodities.
Agricultural production heavily depends upon the regional climate that is characterized
by a high mean annual rainfall, warm temperatures and extremely mild winters.
Hurricanes and tropical storms often produce flood conditions that can remain for weeks.
During 1999-2000, south Miami-Dade was struck by two intense rain events. The
first event, Hurricane Irene, passed over South Florida on October 15, 1999 and the
second event, the October 2000 No Name Event (NNE), struck almost one year later on
October 4, 2000. The impact of both storms on the agricultural economy of south Miami-
Dade resulted in losses of nearly $430 million.
Flood control for south Florida became a federal priority in 1947 after back-to-
back hurricanes left most local communities and the newly created Everglades National
Park (ENP) inundated for weeks. In 1948, Congress authorized construction of the
Central and Southern Florida Flood Control Proj ect (CS&F) to regulate flooding and
mitigate damage. The current system contains 1,800 miles of canals, 25 maj or pumping
stations and other conveyance structures that stretch from Orlando to south Miami-Dade.
The South Dade Conveyance System (SDCS) is the Miami-Dade County extension
of the CS&F and is governed in a three party agreement between ENP, the United States
Army Corps of Engineers, and the South Florida Water Management District (South
Florida Water Management District 2000). Canals C-111 and L31W provide flood relief
for agricultural lands and discharge water into Taylor Slough and Florida Bay.
The frequency and magnitude of flood events in South Miami-Dade have increased
the demand for high-resolution flood maps that are capable of displaying the extent,
magnitude and duration of a specific flood event. In 1997, a University of Florida
Hydrologic Sciences Task Force (HSTF) addressed the maj or issues surrounding flood
management for agricultural areas in south Miami-Dade (Graham et al., 1997, pp.34),
Flooding in the agricultural area has intensified in frequency, duration and depth ..
.the lack of documentation concerning the negative impact of the experimental
water deliveries has hindered progress by the USACOE and SFWMD to address
The hydrologic and geographic databases in the agricultural area east of the C- 111
canal should be enhanced. Installation of additional monitoring stations,
development of new geographic information, and further historical and statistical
evaluations of the existing data bases is necessary to accurately assess the impact of
canal operations on groundwater levels in the agricultural area.
A local-scale, event based hydrologic model is needed to define the risk of flooding
to the agricultural community associated with alternative structural and operational
plans for the C- 111 proj ect...such a model could be used to produce maps of
flooding probability in the agricultural area associated with alternative structural
and operational plans for the C- 111 proj ect, which would allow local producers to
better plan for the future."
The development of a multi-hazard database currently is the highest priority for the
Department of Homeland Security and the Federal Emergency Management Agency,
FEMA (Lowe 2002). FEMA' s Multi-Hazard Flood Map Modernization initiative
involves the expansion of the current geo spatial hazard data base including the Multi-
Hazard Flood Map Modernization. The modernization proj ect is designed to produce a
more accurate geospatial flood vulnerability database that is accessible to the general
public. Furthermore, the National Flood Insurance Program has charged FEMA to head
the Coordination of Surveying, Mapping, and Related Spatial Data Activities. The action
mandates that flood hazard mapping becomes FEMA' s top priority among all natural
disasters. Part of the effort additionally includes the initiative to acquire geo-referenced
spatial data and micro-topographic Airborne Laser Swath Mapped data for advanced
hydrologic models and maps. In 1999, FEMA and NASA sponsored ALSM flight
operations of the C- 111 basin and Everglades National Park, ENP, for future flood
mapping proj ects. Modern FEMA flood maps are required to meet the standard of a 5
meter spatial resolution (Maune 2001).
Recently, FEMA and the Harris County Flood Control District developed the
Tropical Storm Allison Recovery Proj ect to assess flood vulnerability in response to the
aftermath of tropical storm Allison (www.hcfcd.org/tsarp.asp, April 2006). The project
methodology featured the integration of ALSM and GIS for creating digital flood
insurance rate maps, DFIRMS.
The research obj ective of this proj ect was to develop a method for assessing flood
vulnerability in the C-111 basin by integrating Landsat7 sensory data, regional surface
water elevation data and airborne laser scanned topographic data. Secondary objectives
included the creation of a bare earth model for agricultural fields, modeling regional
surface water elevation prediction grids and detecting clouds with vegetative indices. The
featured map product is the surface water inundation map (SWIM). SWIM is an
inundation map capable of displaying the magnitude, duration and extent of flooding with
a 3 meter spatial resolution. October 12-22, 1999, is the study period used for generating
surface water grids and inundation maps.
Proj ect Area
The study area lies in the southern area of the C- 111 basin and is comprised of
agricultural fields and protected wetland areas in ENP, Figurel.1. The Frog Pond is a
fertile tract of land in the C-111 partitioned into twenty-two parcels and leased by the
Federal government; however flood protection is not guaranteed for the Frog Pond.
Parcels 14, 15, 16, 17, 18 and 19 are located within the study area; however only 16 and
18 are completely displayed, (see Figure 1-2).
The three maj or land cover types that dominate the study area are wetland forest,
wetland marsh and agricultural row crop fields. Fiducial land features are permanent geo-
morphological features in the study area and they include the soil mound and the L-31W
canal, Figure 1.2. The S175 culvert is also located on the L31W canal inside the study
] Project Area
g Prog Pond
)'1 ~Streams and Canalsin
[ ] County Boundary
10 0 10 20 Miles
Figure 1.1 Map of Miami-Dade County and the project area.
,:II ~ ~ ~ i Wetland Forest~eln as
I I S175
0.6 0 0.6 1.2 Miles
Figure 1.2 Map of the study area. Leased parcels are numbered 14-21. The S175 culvert
is represented with a red circle on the L31W canal.
The sub-surface hydrology of south Florida, including Miami-Dade, is
characterized by an unconfined, highly permeable system called the Biscayne Aquifer
(Fetter 1998). The Biscayne aquifer is recharged by precipitation, and water table levels
fluctuate with the amount of precipitation. Below the Biscayne is a plastic semi-confining
unit, the gray limestone aquifer and a lower plastic unit (Graham et al. 1997). Canals
penetrate the most permeable part of the aquifer. The thickness and hydraulic
conductivity of the Biscayne in the southern C- 111 basin are approximately 46ft. and
25,000 ft. /day, respectively (Graham et al. 1997). Water levels in the Biscayne conform
to the land surface with the highest levels occurring in the high elevation areas, and
lowest levels in the low elevation areas.
Photogrammetry is the science of analyzing photographs and images to determine
the size, shape, and spatial attributes of the features in an image acquired with remote
sensing (Bethel and Cheng 1995). Remote sensing refers to the inferring of target or
media characteristics by the reception of energy from the target or media. The energy
may be electromagnetic, acoustic, subatomic particles, scattered energy originally
transmitted from an active system sensor or originating from the sun.
Active and Passive Remote Sensing
All remote sensing applications use either active or passive sensors. Passive remote
sensing is usually dependent on reflected solar illumination or the emission or
transmission of black body radiation. Active remote sensing involves sending a signal at
a specific wavelength to the earth surface, detecting a return signal and assigning a pixel
value to the received signal.
Emitted energy, an earth surface feature, is optimally sensed in the near infrared to
the far infrared bands and reflectance properties are optimally sensed in the visible
through the mid-infrared bands. For this reason, most passive sensor studies of planetary
surfaces are conducted in the visible and infrared regions. Madden et al. (1998) used
1994 color infrared imagery to identify wetland vegetation in Everglades National Park.
Doren, Rutchey and Welch (1998) used color infrared imagery to classify vegetation in
the southern Everglades. Welch, Madden and Doren (1998) used color infrared imagery
as ground control to classify vegetation in the Everglades.
The maj or applications of visible and infrared remote sensing include detecting
surface chemical compositions, vegetative cover and biological processes. Although,
visible and color infrared sensory data are useful for environmental studies, wave matter
interactions produce noise in the received signal.
Spectral Signature of Water
Variations in emitted and reflected radiation are used to measure, classify and
verify the spectral signature characteristics of the land surface. Similar surfaces will share
similar signature values within the electromagnetic spectrum for a specific wavelength,
and different surfaces typically possess different spectral signatures. Scatter, emittance,
reflectance and absorption of specific bands produce a unique "spectral signature" or
curve that is characteristic for a particular surface property. The remotely sensed spectral
signature is related to an associated curve obtained from laboratory measurements of
wavelength versus reflectance for the visible and infrared regions of the electromagnetic
spectrum for a library of materials.
Water has a low spectral signature reflectance in the visible and infrared region
compared to all other major land cover types. Water, vegetation and exposed ground are
the main ground cover types in the C-111, and the ability to recognize these ground cover
types with remotely sensed images is dependent on separating and distinguishing their
spectral characteristics. Water with sediment and debris will produce a higher reflectance
spectral signature than that of pure water.
Albedo is the reflectivity of a surface, and water possesses a low albedo in the near
infrared band. Vegetation possesses a high reflectance in the infrared spectrum due to
plant microstructure. Vegetation has a relatively low reflectance in the red band
compared with soil and turbid water, while wet soil and water have similar reflectance in
the red band. The ratio of red and infrared bands is used to distinguish between water
(pure and turbid) and land (vegetation and soil).
Because of sediment and debris, flood water will produce a maximum reflectance
peak in the red band. This signature is particularly useful for flood detection; however,
the presence of dense clouds may interfere with the signature of the land surface. No
verifiable method can be expected to eliminate cloud contamination to obtain visible and
near infrared based flood information under thick cloud formation, (Sheng, Su and Xiao
Active and passive signal errors are primarily attributed to absorption or scatter of
atmospheric noise components. Absorption is caused by the presence of water vapor and
gases, while scatter is caused by the presence of vapor, gases, dust and atmospheric
Four measures of sensor performance are used for determining the quality of the
resolution of an image. These measures include spectral resolution, spatial resolution,
radiometric resolution and temporal resolution.
Spectral resolution refers to the specific wavelength intervals in the electromagnetic
spectrum that the sensor records. A decrease in the wavelength interval results in an
increase in the resolution of the image.
Spatial resolution is the measure of the smallest feature that a sensor can detect or
the area on the ground represented by each pixel for a nadir view. Nadir is the point
diametrically opposed to the zenith, which is the point in the sky directly overhead. An
azimuth is an arc from the horizon to the zenith. Nadir can also be taken to mean "lowest
point" in the sense that zenith can be taken to mean "highest point."
Spatial resolution can also be described in the form of the instantaneous field of
view (IFOV) or the measure of the cone angle (radius) viewed by a single detector at a
specific point in time. The scale at which an image is captured provides useful
information about spatial resolution, and spatial resolution may vary for different sensory
bands in an image. For example, the panchromatic band of Landsat 7 possesses a 15
meter spatial resolution while the other bands possess a 30 meter resolution. Obj ects that
are smaller than the IFOV can be detected if they contrast strongly against the
background of surrounding pixels. Conversely, obj ects larger than the pixel may not be
detected if their reflectance does not dominate the surrounding pixels.
Radiometric resolution or dynamic range is the number of possible data values in
each band or the number of bits into which the remotely sensed energy is divided. For
example, when the Landsat7 ETM+ sensor records the electromagnetic radiation in its
IFOV, the total intensity of the energy is divided into 256 brightness values for 8 bit data.
Data fie values or digital numbers for 8 bit data range from 0 to 255 for each pixel.
Temporal resolution is a measure of how often the sensor records imagery for a
particular area, and for satellites, this is generally defined by its path or orbital cycle.
Normalized Differential Vegetation Index (NDVI)
Vegetation indices are created by combining data from sensor bands into a
specified algorithm. They are particularly useful for identifying features by enhancing
certain reflectance properties. NDVI is commonly used to visualize properties of land
cover that are elusive with only raw band imagery.
NDVI is most useful for mapping land cover including urban areas, water, soils,
dying and healthy vegetation. A value near 1 represents high near infrared reflectivity and
a value near -1 indicates strong near infrared absorption. NDVI is calculated as the
difference between near infrared and red divided by the sum of red and near infrared.
Melesse and Jordan (2003) calculated NDVI as
Band 4 + Band3 (1)
Todd and Hoffer (1998) used mid-infrared with Landsat 5 near infrared data to map
land surface moisture. NDVI increased with an increase in healthy vegetative cover.
Inundated surfaces possess extremely low NDVI values, because of high infrared
absorption and low infrared reflectance properties. The study investigated NDVI for
targets with specific vegetation cover amounts and varying soil backgrounds. Although
vegetation indices were less sensitive to soil background, they were effective for
determining vegetation biomass and vegetation cover for small areas. The relationship
between NDVI and vegetative land cover showed that NDVI was higher for moist soils
than the drier soils at the same percent vegetation. NDVI increased substantially as
moisture increased for the same vegetation cover.
Lunetta and Balog (1999) used multi temporal Landsat 5 data for identifying
wetland land cover including water bodies. The results showed that sensor data in the
mid-infrared, Band 5, best discriminated between dry and wet areas. Frazier and Page
(2000) successfully used visible and infrared bands from Landsat 5 to detect water bodies
in the floodplain of the Murrumbidgee River in central Australia.
Song, Duanjun and Wesely (2003) researched the short wave spectral signature of
water bodies. The signature of water is unique among signatures for most natural
surfaces, because of its low reflectance throughout the electromagnetic spectrum. The
reflectance of water bodies showed a decrease in reflectance with an increase in
wavelength. The signature of water produced a negative value with NDVI, and pixels in
satellite images with the most negative NDVI values were correlated with water bodies.
The shape of the spectral surface reflectance and its value in the red band greatly
depended on the relative amounts of suspended minerals, chlorophyll and dissolved
organic matter in the water. Under clear water conditions, the reflectance was found to
decrease linearly with wavelength.
Cloud contamination alters or sometimes completely obstructs the spectral
signature of the land surface. The significant difference in spectral reflectance between
clouds and the earth makes the process of distinguishing clouds from the Earth's surface
difficult due to the high variability in cloud expression. Sheng Su and Xiao (1998) used
thermal, infrared and visible channels of Advanced Very High Resolution Radiometer
(AVHRR) to distinguish cloud cover from land cover. The spatial variance of cloud top
temperature was noted to be greater than that of the Earth' s surface, and the contextual
feature of surface temperature was also used for cloud screening. Image analysis showed
that cloud shadow caused a reduction in solar irradiance, and cloud shadow and water
bodies were difficult to distinguish in the near infrared channel.
Melesse and Jordan (2003) used visible, short-wave infrared and thermal infrared
bands from Landsat 5 to develop two vegetative indices for detecting clouds, cloud
buildup and water in the Econ Basin, Florida. Clouds were detected and classified by
using the simplified Plank constant to convert Band 6 digital number values to
temperature. Image data was used to enhance dense clouds and urban features for visual
analysis. For Landsat 7 ETM+, vegetative index two and vegetative index three are
BBBBBBBBBBBBBBBBBBBBan 2 B and 6
VI 2 = (2)
B a n d7B~~~~BBBBB~~~~BBBB
VI3 = (3)
Band 5~~~~BBBBB~~~~BBBB + Band3
where Band 2 is the green band, Band 6 is the thermal infrared band, Band 7 is the
middle infrared band and Band 3 is the red band. These indices were found to be
effective for mapping dense cloud cover and partial clouds.
Airborne Laser Swath Mapping
Airborne Laser Swath Mapping (ALSM), or light detection and ranging, Lidar,
remote sensing utilizes a laser, detector, scanning system and Global Positioning Systems
for topographic mapping. The complete process involves planning, collection,
processing, filtering and editing echo points from the return signal data. Elevation post
spacing is a function of flying height, speed, pulse rate and scan angle. Specifications of a
commercial ALSM system generally describe laser, scanning, GPS, INS and flying
operations and information concerning error and delivery (Table 2.2).
Table 2. 1 Specifications of a commercial Lidar system.
Specification Typical Value
Wavelength 1,064 Clm
Pulse Repetition Rate 5 33 k
Pulse Width 10 ns
Beam Divergence 0.25 2 mrad
Scan Angle 40 (75 o Maximum)
Scan Rate 25 40 Hz
Scan Pattemn Zig-Zag, Parallel, Elliptical, Sinusoidal
GPS Frequency 1 2 times per second
INS Frequency 50 (200 maximum)
Operating Altitudes 100 1,000 m (6,000 m max)
Footprint 0.25 2 m (from 1,000 m)
Multiple Elevation Capture 1 5
Grid Spacing 0.5 2m
Vertical RMSE 15+ cm
Table 2. 1 Continued
Horizontal RMSE 10 100 cm
Post-Processing Software Proprietary
Topographic ALSM lasers use an infrared light beam (1064 nm) that is invisible,
absorbed by water and strongly reflected by healthy vegetation and concrete. The laser is
sent at a narrow dispersion angle (0.3 Cprad), and laser spot size or footprint is determined
by flying height. The infrared beam reflects strongly off healthy vegetation, concrete and
dry soils, however any presence of water will absorb and warp the beam path. It is
important to note that birds and other airborne obj ects will reflect the infrared beam and
produce an exceptionally high elevation value. Figure 2.1 shows how the beam reflects
off of and penetrates a tree canopy to produce an elevation point.
figure z. 1 Illustration or a Liaar intrarea D~eam. Ine actual D~eam diameter Is smaller than
what i s shown in the figure. Source http://earthob servatory.nasa.gov
Intensity is the measure of the energy reflected from an obj ect. Detecting return
intensity involves recording the reflected or return beam energy from the earth surface.
Obj ects possessing high reflectivity properties show a higher return energy than objects
possessing low reflectivity properties. Obj ects such as metal roofs and sand possess high
reflectivity values, while water and black tar pavement possess low reflectivity values.
Different sensors have been developed to record multiple returns and reflected
intensity. Multiple return signals occur when part of a distended beam strikes an above
ground obj ect and the remaining portion strikes the ground. When this occurs, the
recorded signal will then display multiple elevation values from a single pulse. The above
ground signal is the "first return" and the ground signal is the "last return". Multiple
returns are found in high, dense canopy areas, because the first and middle returns
provide elevations for the top and intermittent growth. The last return usually reflects
from the ground, however extremely dense canopy will prevent full penetration.
Kraus and Pfeifer (1998) noted if the beam strikes canopy or branches then the
measured ground elevation value might be overestimated. This can lead to an asymmetric
distribution error of laser scanner points. The research results emphasized the necessity to
remove vegetation without deleting ground points for areas possessing low penetration
Post processing of ALSM data is performed to satisfy two requirements for product
delivery. The first requirement derives accurate results based on GPS stations to provide
a frame of reference for the airborne operation (Maune 2001). The second requirement is
to solve a bare earth condition by removing irrelevant points, and this is accomplished
with automatic or manual post processing methods. Automatic processing uses software
algorithms to view neighborhood points and weigh them before removal. Manual
processing is necessary, because automated algorithms may produce anomalies not
characteristic of the bare earth condition. Occasionally, some apparent data anomalies
appear in files and the analyst may review aerial photography, digital imagery or
videotape to identify anomalies (Maune 2001).
Few empirical studies exist for assessing the accuracy of digital elevation models
created with ALSM. Under ideal conditions, absolute vertical accuracy for grass and
pavement may be within 15 centimeters, but vertical accuracy cannot be obtained within
10 centimeters (Maune 2001). Daniels (2001) evaluated datum conversion issues and
accuracy of ALSM by comparison of real time kinetic GPS sample points and lidar spot
elevations. Base station, local orthometric height and regional offset corrections used to
isolate potential datum offsets in lidar were necessary for mapping dynamic geo-
Hodgson et al. (2003) found elevation root mean square error with ALSM was 33
centimeters for low grass and 153 centimeters for shrub/scrub land cover. In general,
vertical errors with low grass and high grass were much smaller than in areas of heavy
vegetation canopies. Hodgson and Bresnahan (2004) also noted that variation in land
surface elevation was strongly correlated with a change in vegetation. Root mean square
error values ranged from a low of 17-19 centimeters for low grass and pavement.
Shrestha et al. (2000) performed an accuracy assessment for surveying and
mapping applications with ALSM. The results showed that elevation values for bare earth
ground were accurate to within +/- 5 10 centimeters. The authors noted that ALSM
technology was an innovative approach for high resolution flood plain and drainage
ALSM Point Removal
Automated post processing of ALSM data attempts to model a bare ground
condition by using software to identify and remove artifacts. Automated methods for
point removal are based on neighborhood operators that iteratively identify the lowest
points within a defined search neighborhood. The operator then adds them to a candidate
set of ground returns. Subsequent iterations select the candidate set by adding returns that
are low or exhibit some angular deflection from a surface modeled by the current
candidate set of points.
The details of search neighborhood operators and parameters vary by lidar mapping
vendor. Generally, the analyst will examine a candidate set of ground returns to further
improve the accuracy of labeling features. The procedure also requires an analysis of
small areas as a three dimensional cloud of ALSM points overlain on available digital
orthophotography. Thus, the process of point removal may contain errors, because
removal is both adaptive and subj ective.
Krabill et al. (2000) used ALSM to study changes in beach morphology. The
research showed that post processing remained problematic for removing artifacts
including near ground vegetation. Okagawa (2002) assessed multiple automatic fi1ters to
extract artifacts from digital surface models. The author concluded that image
information was indispensable for identifying artifacts during post processing. Kampa
and Slatton (2004) used a multiscale filter to segment bare ground from artifact points in
ALSM data. To compute the mean square error for performance, the adaptive fi1ter was
initially applied to simulated ground data. The ground surface was distinguishable from
artifact points for a point density of twelve points per 25 square meter grids.
Raber et al. (2002) used an adaptive fi1ter to minimize the overall error by applying
different vegetation point removal parameters based on vegetation type. The study
involved extracting vegetation land cover type information using only ALSM multiple
return data. The study showed that land cover information could be used adaptively in
ALSM vegetation point removal for the production of accurate elevation models. Land
cover observations involved analysis of color infrared imagery, ground control points and
vegetation land cover. Histogram analysis showed that monoculture canopies were
characterized by a dampened bimodal histogram. A statistical analysis further showed
that among all land cover types, low and high grass possessed the lowest mean absolute
Huising and Pereira (1998) studied bare earth modeling and found that separating
dense vegetation from bare ground was a protracted process. The authors observed that
manual filtering may be better than automated, however more time is required for post
processing large areas. The manual method was found to be ideal for filtering vegetation
and other artifacts in small areas. The authors concluded that using only topography data
compounded the problem, and aerial photography was determined essential for
classifying land features. The accuracy of elevation measurements was related to the laser
system and terrain geometry, and flat terrain and low grass areas were used to estimate
Persson, Holmgren and Soderman (2002) used ALSM to detect individual trees by
estimating height crown closure and stem volume. The study used the lowest laser
reflection points to derive bare earth DTMs. The study further noted that return intensity
and type return pulse data provided more information about tree structure. Hodgson et al.
(2003) also used orthophotography and ALSM surface cover height to map impervious
Multiple remotely sensed data sets may be used to separate vegetative height from
a theoretical bare earth condition. Popescu and Wynne (2004) combined lidar and
multispectral data to accurately estimate plot level tree height by focusing on the
individual tree level. Combining small footprint airborne lidar data in conjunction with
spatially coincident optical data was found to help accurately predict tree heights of
interest for forest inventory and assessment. The study recommended that proj ect
methodology can be applied to process lidar data for vegetation removal, and individual
Popescu and Wynne (2003) developed analysis and processing techniques to
facilitate the use of small foot print ALSM for estimating plot level tree height. This was
accomplished by measuring individual trees identifiable on a three dimensional ALSM
elevation model. The study used the combination of ALSM and multi-spectral optical
data fusion to differentiate between forest types and improve the estimation of average
plot heights for pines. The research demonstrated that small foot print ALSM, used in
conjunction with spatially coincident optical data, was accurately able to predict the tree
heights of interest for forest inventory and assessment.
Hopkinson et al. (2004) used ALSM to map snowpack depth under forested
canopies. Snow pack distribution patterns were mapped by subtracting a bare earth DEM
grid from a peak snowpack DEM grid. Snow pack depth was used to predict water
availability and flood levels during the warming period. The study also found that a high
proportion of last pulse returns led to an overestimation of ground elevation. The study
recommended a further assessment of type dependent elevation offsets for improving
elevation and snow depth estimation.
Brock et al. (2001) used ALSM to recognize and map surfaces that provide
accurate low variability topographic measurements. These features were termed fiducial
and were used as reference base line features for mapping morphology. Fiducial features
are naturally occurring bald earth features such as beaches, bare dunes and ice sheets. The
process for separating dense plants of less than 10 centimeters was difficult based solely
on passive spectral signatures of ALSM. The presence of vegetation increases the
difference between ALSM and ground survey elevations from a minimum of 0.26 meters
over bare sand to values near 0.40 meters for all vegetation classes. Of the four defined
vegetation classes consisting of mono, sparse, medium and dense, sparse vegetation
possessed the highest variance between ALSM and coordinate survey elevations.
Evans et al. (2001) used sampling theory to map individual trees and estimate tree
height. Small foot print lidar failed to yield ground returns in areas dominated by dense
Renslow and Gibson (2002) developed bare earth models from ALSM and high
resolution aerial photography to assist the decision making process for increasing
services for utility companies. This was accomplished by mapping fast track utility
corridors using bare earth models. Heinzer et.al (2002) used ALSM and aerial images to
model inundation, velocity and steady state flow of water. Interestingly, bare earth
models were interpolated from group points; however buildings were reinserted to
display realistic structural definitions.
Geographic Information Systems
Geographic Information Systems, GIS, is software designed to create maps by
capturing, storing, retrieving, manipulating and displaying geographically referenced
spatial tabular data (www.usgs.gov, April 2006). The three types of spatial data common
in GIS are points, arcs (lines) and polygons (areas).
GIS thematic map layers can display topological relationships between mapped
features. Topology refers to recording the spatial relationship between points, arcs and
polygons. A coverage is a GIS data Hile that display topology, however some GIS data
files such as shape files do not display topological relationships.
Metadata files list important parameters describing attributes of remotely sensed
and GIS data products. Metadata typically includes the coordinate system, period of data
capture and ancillary information pertinent for mapping applications with other data sets.
There are two classes of interpolation, deterministic and geo-statistical.
Deterministic methods such as inverse distance weighting, splines, and radial based
functions are directly based on an interpolator that uses the surrounding measured values
or mathematical formulas applied to those values. Geostatistical models, such as kriging,
predict values by accounting for the probabilistic spatial relationship among neighboring
points. Kriging is able to predict estimation errors and is often preferred over
The surface calculated using inverse distance weighting depends on the selection of
a power value and the neighborhood search strategy. For inverse distance weighting the
maximum and minimum values in the interpolated surface can only occur at sample
points. The output surface is sensitive to clustering and the presence of outliers. Inverse
distance weighting assumes that the surface is being influenced by the local variation,
which can be captured throughout the neighborhood.
The local polynomial method is a moderately quick and smooth deterministic
interpolator. It is more flexible that the global polynomial method; however there are
more parameter decisions. There is no assessment of prediction errors; however the
method provides prediction surfaces that are comparable to kriging with measurement
errors. Local polynomial methods do not allow any analysis of the spatial autocorrelation
of the data, thus it is less flexible and more automatic than kriging.
The global polynomial method is also a quick and smooth deterministic
interpolator. There are fewer decisions to make regarding model parameters than for the
local polynomial method. It is best used for surfaces that change slowly and gradually.
There is no assessment of the predictions errors and this method may produce a surface
that may be too smooth. Values at the edge of the data can have a significant impact on
the interpolated surface.
Radial based functions are moderately quick deterministic interpolators that are
exact, and they are considerably more flexible than inverse distance weighting, however
there are more parameter decisions, and there is no assessment of prediction errors. The
method provides prediction surfaces that are comparable to the exact form of kriging.
Radial based functions do not allow for analysis of the autocorrelation of the data, thus
making it less flexible and more automatic than kriging. Radial based functions are used
for calculating smooth surfaces from a large number of data points, and are preferred for
gently varying surfaces such as elevation. The radial based function is inappropriate
when there are large changes in the surface values within a short horizontal distance
and/or when the sample data is prone to error or uncertainty.
Ordinary kriging produces interpolated values by assuming a constant but unknown
mean value, allowing a local influence from nearby neighboring values. Because the
mean is unknown, there are few assumptions about the data. This makes ordinary kriging
flexible but less powerful.
Simple kriging produces interpolated values by assuming a constant but known
mean value, allowing local influences due to nearby neighboring values. Because the
mean is known it is slightly more powerful than ordinary kriging but in some cases the
selection of a mean value is not well known.
Universal kriging produces interpolated values by assuming a trend surface with
unknown coefficients in the model; however it allows local influences from nearby
neighboring values. It is possible to overfit the trend surface, which fails to leave enough
variation in the random errors to properly reflect uncertainty in the model. It can be more
powerful than ordinary kriging because it explains much of the variation in the data
through a non-random trend surface.
Disjunctive kriging considers functions of the data, rather than just the original data
values themselves, and stronger assumptions are required. Disjunctive kriging assumes
all data pairs come from a bivariate normal distribution and the validity of these
assumptions should be checked. A bivariate normal distribution describes relative
frequencies of occurrence in the population of pairs of values. When this assumption is
met, the functions of the data are indicator variables that transform the continuous data
values to binary values based on a decision threshold value.
Doucette and Beard (2000) evaluated inverse distance weighting, splines and
universal kriging as interpolators to fill gaps left by occlusions in digital elevation data.
The results favored splines as a surface interpolator, especially as terrain roughness
increased. The study additionally found that altering the search radius parameter
significantly impacts interpolation error statistic values.
Selecting a best fit model depends on the assessment of several modeling statistics.
In general, the best fit model is one that has the standardized mean error closest to zero,
the lowest root mean squared prediction error, the average standard error nearest to the
root mean squared prediction error and the standardized root mean squared prediction
error closest to one (ESRI 2001).
Inundation Mapping with GIS
Previous flood mapping efforts have used remote sensing and GIS to map the
extent, duration and magnitude of flooding. Ball and Schaffranek (2000) used
topographic and surface water grids to map water depth in the southern Everglades.
Temporal inundation patterns were mapped and compared to historical and current water
depths. A comparison to other hydroperiods was conducted to isolate temporal changes
affected by anthropogenic influences of water management policy. To estimate water
depth accuracy, computed depths were subtracted from depths measured in the wetlands
adjacent to the C-111 canal and in Taylor Slough in 1997 and 1999.
Ball and Schaffranek (2000) employed a similar method to study water surface
elevation and water depth for Taylor Slough in the southern Everglades. A GIS program
was used to subtract topographic elevation grids from surface water elevation grids. The
extremely low topographic relief of the southern Everglades produced significant spatial
variability in surface water gradients. Furthermore, the land surface elevation grid was
calculated from interpolating global positioning systems (GPS) topographic data
sponsored by the USGS, National Mapping Division. Daily surface water data was
obtained from the SFWMD, and National Park Service (NPS) Everglades National Park.
The research concluded that water depth and topographic accuracy were directly
correlated to the spatial resolution and accuracy of input data. Proj ect inundation grids
were calculated with the same method; however an ALSM topographic grid was used
instead of a GPS topographic grid. ALSM topographic grids produce inundation maps
with a finer spatial resolution than GPS topographic grids.
DATA RESOURCES AND METHODOLOGY
Proj ect data sources include aerial color infrared imagery, Landsat7 EMT+ sensor
data, surface water elevation data and ALSM topographic data. All tables that are
referenced can be found at the end of this chapter.
Color infrared imagery was acquired from Land Boundary Information Systems
(www.1abins.com April 2006), Labins, and each image possessed a 1 meter spatial
resolution. Color infrared imagery for the study area is found in quadrangle Royal Palm
Ranger Station or Quadrangle 1205 S.W. and was obtained for 1994 and 1999. The
primary use of color infrared imagery was to identify vegetation points in the NAD 27
and NAD 83 ALSM point data sets.
Surface Water Data
Surface water elevation values for regional canal stations and well monitoring sites
were obtained from the South Florida Water Management District (SFWMD), U.S.
Geological Survey (USGS) and National Park Service (NPS). Surface water elevation
data covers the period during October 12-22, 1999. SFWMD canal stage elevations were
recorded for both head and tail stage, and the mean between head and tail was used to
create surface water grids. Surface water values were recorded in feet, and NGVD29 was
used as the reference vertical datum for hydrologic and topographic data sets.
Digital Elevation Model Construction
In 1999, FEMA and NASA sponsored 30010, a consulting firm, to conduct ALSM
flight operations for the C-111 basin. ALSM Hiles are arranged by flight area, and each
flight area includes a cache of Eiles that provide a variety of elevation data products.
ALSM data was captured by single pulse return data, and the products included text Hiles
of x,y,z coordinates and processed DTM files. All coordinate elevation values were
recorded in feet, and text Hiles were prepared for all first return raw data and automated
filtered bare earth data. NAD 27 raw data values were recorded to either one-hundredth
or one thousandth of a foot. All NAD 83 bare earth fie data were recorded to one one-
hundred thousandth of a foot. Only accuracies of hundredths of a foot should be
considered for DEM analysis, because real time kinematic GPS values are only accurate
to one one-hundredth of a foot. The post spacing for points was 10 feet along the track
direction and 23 feet across the track direction. Bare earth contour line DTMs were
included for each area and elevation lines were categorized by one-foot intervals. These
DTMs were not used in this study, because of their low vertical resolution.
Data quality reports were prepared for all areas and they included coordinates for
flight area ground control points in addition to the methodology used to create the bare
earth digital terrain models. These reports are commonly used to inform the client about
the accuracy of ALSM data by a statistical comparison between ground surveyed GPS
points to associated ALSM points. Data quality reports for the study area listed a vertical
accuracy of 15 centimeters, and bare earth DTM files were created using Delaunay
triangulation (3001 1999). Bare earth Hiles were created using proximal analysis to filter
unwanted points according to the report; however no additional information was provided
about the procedure.
Landsat 7 Enhanced Thematic Mapper
Landsat 7 ETM + image data were used to detect inundated surfaces and dense
clouds within the study area after Hurricane Irene. Landsat7 scenes for October 16, 1999,
and April 9, 2000, were obtained from the USGS, and all data deliverables were stored on
CD-ROM media and delivered as Geotiff files. For image processing, all Geotiff files
required both importation in ERDAS Imagine and exportation as an ERDAS Imagine
image file. The first scene was captured on October 16, 1999, nine hours after Hurricane
Irene passed over the C-111, and the second was captured on April 9, 2000 during the
peak of the dry season.
A bend in the L31W canal was used to detect an offset between the October 16,
1999, and April 9, 2000, Landsat 7 ETM+ scenes. Band 8 was used from both scenes to
locate the x and y values for associated pixels, and the offset was measured. The offset
between associated pixels was 48 ft. north and 1 ft. east.
Vegetative Index Methodology
ERDAS Imagine is software that is specifically designed to work with large geo-
referenced image data sets. NDVI map methodology was initiated by creating a layer
stack of red and infrared bands for both Landsat7 scenes. The ERDAS Imagine layer
stack function combined sensor data, and the NDVI function automatically created an
NDVI image by separating Band 3 and Band 4 from the stack and substituting them into
the NDVI equation.
The ERDAS Imagine Spatial modeler extension selected the appropriate individual
band layers in a composite layer stack and substituted them into their designated
vegetative index equations. The Spatial Modeler tool was used to create vegetative index
two and vegetative index three indices described in Chapter 2.
An unsupervised classification divides pixels into classes based on their digital
number value. ERDAS Imagine unsupervised classification was performed on the three
vegetative index images, and 30 classes were created for all vegetative index maps. All
vegetative index unsupervised images were scaled with Class 1 representing the lowest
reflectance grouped values and class 30 representing the highest reflectance grouped
Topographic grids were created from NAD 83 / NGVD 29 and NAD 27 / NGVD
88 ALSM data. The procedure for creating a point map theme from text data required a
list of vertical, horizontal and elevation values. All text files were space delimited and
consequently, no files could be opened by GIS software. Only tab and comma delimited
formats are recognized by GIS software for importation. Furthermore, all text files were
too large to fit the 65,536 spreadsheet row entry maximum. In response to this constraint,
a quick and effective procedure was developed to convert text files into shape files.
The conversion of text files into database tables was required for importation into
GIS software. This initial step involved opening each text file with Microsoft Wordpad,
and converting the native .xyz format to an ASCII text file. The ASCII text file was
opened in a spreadsheet and saved as a database file; however spreadsheet row entries
were limited to only 65,536 displayed values. The Find and Replace tool in Microsoft
WordPad located the 65,536th value in the text file, and all values listed above the
65,536th value were selected and deleted. The altered file was saved under the original
text file name, and displayed the 65,536th value as the first value in the spreadsheet. All
cells were converted to a number format with six decimal places and the column
containing coordinate values were assigned X, Y, and Z field headings respective to their
measurements. The process was applied to all files until the original text file was reduced
to an acceptable size for one spreadsheet. The end result produced a series of
spreadsheets with each representing one sub-area.
Arcview 3.2 was used to develop a point shape file from the data base file. All
database tables were imported into Arcview, and the Add Event function was used to
display the table coordinate values as points for all data base tables imported into
Arcview. Areview' s Geoprocessing tool was used to merge and convert sub area database
files into a point shape file. The geoprocessing tool designates the tables to be merged
and then exports the resultant thematic shape file to a known file directory. The offset
coordinate value between NAD 83 and NAD 27 was 156235.73 ft. false east and 159.86
ft. false north. These values were later used to proj ect surface water data points from
NAD 27 to NAD 83.
Bare Earth Modeling
The procedure for modeling a bare earth condition for the study area involved the
manual removal of ALSM points that represent vegetation, fiducial features and
structures. Point removal was based on the assumption that the study area possessed a flat
topography and low elevation characteristic of the C- 111 basin. The steps utilized in the
process included a cross-comparison between color infrared imagery, NAD 27 ALSM
DEM, and ground control point elevations in the study area. Multiple interpolators and
search parameters were tested for predicting grid elevation values.
ALSM data was collected in NAD 83 and NGVD 88 datum; however the unfiltered
data was placed in NAD 27 and NGVD 29. The measured difference between NGVD 29
and NGVD 88 first order benchmark elevation values is 1.5 125 ft. Proj ect inundation
maps were placed in NGVD 88, because this was the vertical datum used in the data
The proximal analysis method was used by 3001 C to model the bare earth
condition for NAD 83 data, however this method is not defined in the data quality report
for Area 2. Furthermore, this method was not sufficient for flood mapping, because
vegetation points were found in the bare earth model for the study area. The success of
topographic bare earth models relies on the accuracy of the estimated maximum bare
earth elevation threshold used to remove points.
Aerial Color Infrared Analysis
The available Land Boundary Information System (Labins) aerial imagery covered
the dates of December 27, 1994, Figure 3.1(a) and February 21, 1999, Figure 3.1(b).
Aerial color infrared imagery was useful for identifying vegetation patterns and
associated land features in ALSM maps. Healthy vegetation in images possessed a strong
reflectivity in the infrared region of the electromagnetic spectrum, and was displayed as
red. Although row crop vegetation reflected strongly as red, the individual boundaries
varied between both images. In both images, dense tree canopies reflected the strongest
and were easily distinguished from the surrounding land cover. In the 1999 image,
wetland forest was characterized by variable red reflectivity values; however only high
dense tree canopies were consistently reflected as red in the 1994 image. The soil mound
seen in Figure 1.2, reflected as white in the 1994 image when leaf canopy was reduced,
and was difficult to distinguish from the surrounding land cover. In the 1999 aerial
image, the soil mound was easily distinguished from the heterogeneous cover of healthy
vegetation and exposed bare soil.
In the 1999 aerial image row crop vegetation conformed to Hield boundaries,
however in the 1994 aerial image not all row crop vegetation conformed to Hield
boundaries. This dissimilarity was most noticeable in canopy patterns found in leased
parcel 19. A semi-circular arc of vegetation can be seen in the southeast quadrant of the
1994 image, Figure 3.1(a). This was useful for identifying suspect vegetation patterns
found in the raw ALSM topographic DEMs.
Ground Control Point Analysis
Although ground control points, GCP, were not in the study area, they were
analyzed for determining the threshold value for estimating the bare earth condition,
Table 3.1. The process for determining vegetation points was subjective. This depended
on visual analysis and analysis of nearby ground control points in the NAD 27 DEM. The
obj ective of the approach was the removal of vegetation points, while preserving points
that represented roads and bare earth. Based on this methodology, the value of 4.80 ft.
was determined to be the maximum bare earth elevation value for the study area for NAD
27. Consequently, all points in the associated NAD 83 DEM that exceeded the value of
3.29 ft. were also identified as vegetation and removed using the clip tool in Arcview.
Recall that the difference between NAD 83 and NAD 27 maximum bare earth elevation
values is 1.51 ft., and this is equal to the difference between NGVD 29 and NGVD 88
elevation values. Areview's query filter was used to remove elevation points that
exceeded the designated maximum elevation threshold value. The clip tool in Arcview
was used to create a separate shape Eile consisting only of points that were not deleted.
Figure 3.1 Color infrared aerial photos of the study area. (A) 1994 color infrared aerial
image (B) 1999 color infrared aerial image. The study area is outlined in
0 68 Miles
Topographic Spatial Modeling
Multiple interpolators were used to develop prediction grids for NAD 27 and NAD
83 point shape files. These interpolators included inverse distance weighting, local
polynomial, global polynomial, radial based functions, universal kriging, ordinary
kriging, simple kriging and disjunctive kriging. The root mean square error, RMSE,
statistic was calculated by sequential dropping of each observed elevation point and
estimating it using the appropriate interpolation procedure. The RMSE was used to select
the optimum interpolation method for surface water and topographic grids. If two tests
possessed an equal RMSE statistic, then the mean absolute error, was used as the next
All prediction grids, including surface water, were exported as raster surfaces to be
later used for calculating inundation grids. All interpolation methods were used to
generate z prediction values for a test x and y location, the results of the different
prediction methods showed that the predicted values at the test location ranged from
2.9217 ft. to 2.9483 ft. The difference in z prediction values indicates that a small
variation exists between predicted topographic grids using the various methods.
Table 3.2 lists the search parameters for the inverse distance weighting method that
were not set to a default value. The neighborhood method was used for all tests. The
search ellipse used for the neighborhood search had maj or and minor semi-axes of
2, 134.6 ft., and the anistropy factor was set to a value of 1. The x and y test prediction
locations were 797,555.55 ft. and 394,614.36 ft. respectively. Table 3.3 lists the search
parameters for the global polynomial method. Table 3.4 lists the search parameters for
local polynomial method. Table 3.5 lists the search parameters for radial based function
Table 3.6 lists the search parameters used for kriging methods. For all kriging
methods, no transformation was applied, and no trend was assumed. The angle direction
and tolerance were 150 and 450 and the band width was 6. The number of lags was 12,
the search shape angle were 3 and 150 respectively. The major and minor semi-axes were
6,306.9 ft. and 5,546.8 ft., and the anistropy factor was 1.137. The test x and y locations
were 797,555.95 ft. and 394,614.36 ft., and twenty neighbors were used for the test
Surface Water Elevation Map Methodology
The surface water elevation map (SWEM) was created to show the change in
surface water elevation values over the study period, and was used to calculated
inundation grids. All surface water data was acquired from the SFWMD, USGS, and the
NPS. All surface water proj ect data values were referenced with NAD 27 horizontal
datum and NGVD 29 vertical datum. NGVD 29 surface water elevation values were
converted to NGVD 88 by subtracting 1.51 ft from NGVD 29 values.
REMO is the SFWMD internet data retrieval program that provided surface water
elevation data. REMO hydrologic data was delivered in text format and all data was
converted to data base file format. All water elevation data was recorded in feet and
referenced to NGVD29. Canal elevation measurements included head and tail
measurements; however the mean value between head and tail was used to create surface
water grids. SFWMD data sites included the S175, S177, S178, FP, FPl, FP2 and S332D.
FP, FP1 and FP2 are wells and the S175, S177, S178 and S332D are water control
structures. Figure 3.2 shows in situ measurement sites used to create SWEM.
USGS provides maximum daily ground water elevations for monitoring stations in
Miami-Dade. USGS maximum water elevation data for G3355 was acquired through the
2 01 2 4 Miles
USGS water resources internet link with the SFWMD, and USGS water resources for
Miami-Dade. G33 55 is located in the southeast corner of Figure 3.2.
Figure 3.2 Map of measurement sites.
USGS sponsors Tides and Inflows in the Mangroves in the Everglades, TIME
(time.er.usgs.gov). TIME provides telemetric surface water elevation measurements in
daily, hourly and fifteen minute intervals. TIME water monitoring wells NPl l2 and
NPl158 were used to create surface water grids. The locations NP ll2 and NPl158 are
shown in Figure 3.2.
The procedure for developing the surface water site point shape file began with
transforming the Latitudinal and Longitudinal coordinates from Degrees-Minutes-
Seconds to Data-Decimal-Degrees, DDD,
,,=,,,,( Minutes Seconds
60 3600 1
The DMS coordinates were divided into separate spreadsheet fields, and the
conversion for each coordinate was performed using cell formulas. Longitudinal
coordinates were assigned negative values. The resultant spread sheet file was exported
as a tab delimited text file and assigned a .dat extension.
The ERDAS Imagine vector tool was used to export the .dat files as Arcinfo
coverages. The coverage file was opened in Arcview, and the view properties were set to
match the proj section parameters of ALSM as defined in the meta-data report, Table 3.7.
Finally, the coverage file was converted to a shape file with the same coordinates as the
Surface Water Elevation Map Interpolation
Tables 3.8, 3.9, 3.10 and 3.11 list the initial search parameters for kriging tests used
to estimate the surface water elevation. A description of these parameters is discussed
For universal kriging no trend removal and no transformation were performed. A
bandwidth of 6 ft. and the lag size and lag number were 2,721 ft. and 12. The maj or semi-
axis and minor semi-axis for the neighborhood search ellipse were 30,000 ft. and 24,000
ft. The anisotropy factor was set to a default value of 1.25 for all tests. The x and y test
prediction locations were 804,799 ft. and 389,908 ft. The software's default value for
search neighbors was used, and the number of search neighbors was set to five for
For disjunctive kriging, no transformation or trend removal was conducted. The
direct method was used, and the maj or and minor ranges were 30,023 ft. and 12,607 ft.
The search direction, partial sill and nugget were 64.90, 0.64504 ft. and 0.0221170 ft.,
respectively. The lag size and number were 2,721 and 12. The major and minor semi-
axes were 30,023 ft. and 12,607 ft. The anisotropy factor was set to 2.3815. The x and y
test location values were 804,799 ft. and 389,908 ft., and the bandwidth was set to 6 ft.
For ordinary kriging, no transformation or trend removal was conducted. The maj or
and minor ranges were 30,705 ft. and 26,007 ft. The angle direction, partial sill and
nugget were 14.80, 1.1645 ft. and 0 ft., respectively. The lag size and number were 2,721
ft. and 12, and the bandwidth was set to 6 ft. The maj or and minor semi-axes were 30,705
ft. and 26,007 ft. The anisotropy factor was 1.1806, and the x and y test location values
were 804,799 ft. and 389,908 ft.
For simple kriging, no transformation was applied, and the mean threshold value
not to be exceeded was 3.857 ft. The bandwidth was set to 6 ft., and the maj or and minor
ranges were 30,091 ft. and 21,698 ft. The anisotropy factor was activated for all tests, and
the nugget was 0.423 52 ft. The lag size and number were 2,721 ft. and 12 respectively.
The search angle direction and partial sill were 200 and 0.61003 ft. The maj or and minor
semi-axes were 30,705 ft. and 26,007 ft. The test x and y prediction locations were
804,799 ft. and 389,908 ft.
Table 3.12 lists the search parameters for universal kriging for SWEM. The
universal kriging maj or range was set to 30,668 ft. and the minor range was set to 25,892
ft. The maj or semi-axis and minor semi-axis were set to 30,000 ft. and 24,000 ft. The
anisotropy factor was set to 1.25. The test prediction location was 804,799 ft. and
389,908 ft. The lag size and number were set to 2,721 ft. and 12 respectively. Eight
neighbors were used for the search parameters. No trend removal and or transformation
were conducted. The global influence, local influence, angle direction, angle tolerance,
search direction and were equal to 65%, 35%, 15%, 350, 150 and 6 ft. respectively. Shape
3 was selected and the shape angle was set to 150.
Surface Water Inundation Map Methodology
The surface water inundation map, SWIM, was calculated by subtracting ALSM
topographic grid values from SWEM grid values. Surface water and topographic grids
were exported as raster surfaces with a 3 meter resolution. The raster math calculator in
ArcGIS Spatial Analyst extension was used to subtract topographic grids from surface
water grids, and the resultant inundation grids also had a 3 meter spatial resolution. A
value of 0 ft. in elevation was inserted into the L31W canal to prevent aliasing caused by
3D ALSM DEMs were useful for determining elevation values that represented
vegetation. Converting the raster surface into a TIN created 3D TIN DEMs, and the TIN
was imported into ArcGIS Scene.
Table 3.1 Area 2 static GPS points used to determine the elevation filter. Zl is the
elevation of the GPS point, and Z2 is the measured ALSM elevation for that
point. AZ is the difference in elevation between Zl and Z2.
Test Id Z1 ft. X ft. Y ft. Z2 ft. aZ f t.
663 4.50 640246.31 394360.81 4.890 0.390
665 4.46 640248.90 394361.10 4.890 0.430
729 4.72 640518.40 392310.00 4.660 0.060
732 4.31 640707.30 392308.60 4.300 0.010
733 4.29 640716.80 392308.50 4.400 0.110
734 4.60 640728.50 392308.40 4.400 0.200
735 4.61 640739.40 392307.80 4.630 0.020
736 4.67 640752.00 392306.70 4.660 0.010
737 4.64 640763.10 392307.20 4.660 0.020
738 4.51 640774.10 392305.30 4.760 0.250
739 4.66 640785.90 392304.50 4.630 0.030
741 4.59 640495.50 392311.90 4.860 0.270
Table 3.1 Continued.
Table 3.2 Inverse distance weighting search parameters for topography .
SThe shape angle is in degrees. Shape type refers to the search shape used for all interpolation tests. Shape
1 is an open circle and shape 2 is a circle divided by four perpendicular lines running north, south, east and
west. Shape 3 is a circle divided by four perpendicular lines running northeast to southwest and northwest
to southeast. Shape 4 is a circle divided by eight lines that possess the same directions as Shapes 2 and 3.
Table 3.2 Continued
Test Id Power
Table 3.3 Global polynomial search parameters for topography.
Test Id Power
Table 3.4 Local polynomial search parameters for topography.
Test Id Global Influence (%) Local Influence (%) Power
1 10 90 1
2 15 85 1
3 20 80 1
4 25 75 1
5 30 70 1
6 0 100 1
7 10 90 2
8 20 80 2
9 25 75 2
10 30 70 2
Table 3.5 Radial based function search parameters for topography. SWT is spline with
tension, MQ is multi-quadratic, CRS is completely regularized spline, IM is
inverse multi-quadratic and TPS is thin plate spline.
Kernal Shape Z Prediction
Test Id Function Parameter Shape Angle Value ft. Neighbors
1 SWT 1.3715 3 15 2.9500 60
2 SWT 1.3715 1 15 2.9494 15
3 SWT 1.3715 2 15 2.9500 60
Table 3.7 Proj section Parameters for ALSM. Source 30010 Area 2 Data Quality Report.
Description NAD 27 NAD 83
Map Units Feet Feet
Distance Units Feet Feet
Standard Proj section SP 27 Florida East SP 83 Florida East
Custom Proj section Transverse Mercator UTM
Central Meridian 81 00 00 81 00 00
Latitude of Orgin 24.3333 24.3333
Scale Factor 0.999942 0.999917
False Easting 500,000 200,000
False Northing 0 0
Table 3.5 Continued
Test Id Function
1 e 20
Table 3.6 Kriging search parameters for topography.
Maj or Minor Partial
Type Range Range Direction Sill Nugget
OK 6306.9 5546.8 274.4 0.015308 0.0406
Lag Z Prediction
Number Value ft.
Table 3.8 Conyinued
Test Id Maj or Minor Angle Shape Z Prediction
Range Range Direction Partial Sill Shape Angle Value ft.
Table 3.8 Universal kriging search parameters for SWEM.
Table 3.9 Simple kriging search parameters for SWEM.
Angle Angle Shape
Test Id Direction Tolerance Shape Angle
1 0 35 1 30
2 0 35 2 30
3 0 35 3 30
4 0 35 4 30
5 15 35 1 30
6 15 35 2 30
7 15 35 3 30
Table 3.10 Ordinary Kriging search parameters for SWEM.
Table 3.10 Continued
Test Id Direction
Table 3.11 Disjunctive kriging search parameters for SWEM. D in the column heading is
the distribution, PD is probability distribution and CD is cumulative
di stributi on.
Angle Angle Shape Z Prediction
Test Id D Direction Tolerance Shape Angle Neighbors Value ft.
1 PD 15 45 3 20 5 3.1977
2 PD 30 45 2 15 5 3.1977
3 PD 30 45 1 15 5 3.2058
4 PD 20 35 4 25 5 3.1977
5 PD 0 30 3 10 5 3.1977
6 PD 10 30 3 10 5 3.1977
7 PD 20 30 3 30 5 3.2058
8 PD 20 30 3 15 5 3.1977
9 PD 10 30 3 15 5 3.1977
10 PD 10 30 3 30 5 3.2058
11 PD 25 30 4 30 5 3.1977
12 PD 15 35 3 15 5 3.1977
13 CD 15 45 3 15 5 3.1029
14 CD 15 45 4 20 5 3.1977
15 CD 20 40 1 10 5 3.2058
16 CD 20 40 4 10 5 3.1977
17 CD 20 40 1 20 5 3.2058
18 CD 20 40 1 5 5 3.2058
19 CD 20 40 1 30 5 3.2058
20 CD 20 40 2 35 6 3.1977
21 CD 20 40 3 35 5 3.2058
22 CD 20 40 1 35 6 3.2058
23 CD 20 40 4 35 6 3.1977
24 CD 20 40 1 0 5 3.2009
25 CD 15 40 1 20 5 3.2058
26 CD 15 40 2 20 6 3.1977
27 CD 15 40 3 20 6 3.1977
28 CD 15 40 4 20 6 3.1977
29 CD 15 45 1 15 5 3.2058
30 CD 15 45 2 15 6 3.1977
31 CD 15 45 4 15 6 3.1977
32 CD 15 45 1 30 5 3.2058
33 CD 15 45 2 30 6 3.1977
34 CD 15 45 3 30 5 3.2058
35 CD 15 45 1 10 5 3.2058
36 CD 15 45 2 10 6 3.1977
37 CD 15 45 3 10 6 3.1977
38 CD 15 45 4 10 6 3.1977
39 CD 15 45 1 20 5 3.2058
40 CD 15 45 2 20 6 3.1977
41 CD 15 45 3 20 6 3.1977
Z Prediction Value ft.
Table 3.11 Continued
Angle Angle Shape Z Prediction
Test Id D Direction Tolerance Shape Angle Neighbors Value ft.
42 CD 15 45 4 20 6 3.1977
43 CD 15 45 1 10 5 3.2058
44 CD 15 45 2 10 6 3.1977
45 CD 15 45 3 10 6 3.1977
46 CD 15 45 4 10 6 3.1977
47 CD 15 45 4 20 6 3.1977
48 CD 15 45 3 20 6 3.1977
49 CD 15 45 2 20 6 3.1977
50 CD 15 45 1 20 5 3.2058
Table 3.12 Universal Kriging for SWEM, October 12-22, 1999. All values represent
surface water interpolation for NAD 83 and NGVD 88.
Partial Sill ft.
RESULTS AND DISCUSSION
The initial step for flood analysis began with detecting clouds in the October 16,
1999, Landsat 7 ETM+ scene. Clouds can obstruct the spectral signature of water bodies,
therefore clouds were mapped in the study area. Clouds found within Hurricane Irene
were used as a reference feature to identify cloud classes in all maps of vegetative
indices. Hurricane Irene is located in the north east quadrant of the October 16, 1999,
Landsat 7 ETM+ scene. Vegetative index two, vegetative index three and Band 8 were
used to map clouds in the study area. For an initial analysis, Band 8 was selected to
identify clouds in the study area, because of its 15 meter spatial resolution. Clouds are
clearly displayed as irregular shapes comprised of white pixels, and each cloud possesses
a shadow located northwest of the cloud shape, Figure 4. 1(A).
Three clouds are visible in the study area, however only two full cloud shadows are
visible, Figure 4.1(B). Clouds in the study area were found to completely obstruct the
ground signature and their shadows produced darker pixels on the land surface.
Figure~~~~~~~~~i: 4. A a fteFo odwt ad .Cod pera ht reua
Fig ~ hpe faures (B) Zoo in)M of the study area with Band 8. Cod pera ht reua
Vegetation Index Two and Vegetation Index Three
Unsupervised classification of vegetation index two and vegetation index three
were also used to detect dense clouds and verify clouds identified with Band 8. As with
Band 8, the cloud formation of Hurricane Irene was used as the reference feature to
identify potential cloud classes. The clouds from Hurricane Irene were located in the
northeast quadrant of the vegetative index two and vegetative index three maps.
Classes 1-2 in vegetation index two were determined to be cloud classes and
classes 3-8 were determined to be partial classes. Classes 3-7 were determined to be
cloud classes in vegetation index three. Clouds that were identified with Band 8 in the
study area produced similar but not exact shapes with vegetation index two and
I rl~i I II1 12
vegetation index three. Vegetation index two uses more classes to map these clouds than
vegetation index three, Figure 4.2(A) and Figure 4.2(B).
.1~ Class 28
Figure 4.2(A) Vegetation index two map of south Florida, October 16, 1999. Clouds from
Hurricane Irene are most visible in the northeast section with class 1 and class
2. (B) Vegetation index three map of south Florida, October 16, 1999. Clouds
from Hurricane Irene are most visible in the northeast section with classes 3-7.
| Class ?
50 0 50
1 0 1 2 Miles
Figure 4.3(A) Vegetation index two map of the study area, October 16, 1999. All three
clouds are visible in the study area outlined in red. The legend for Figure
4.2(A) applies to Figure 4.3(A). (B) Vegetation index three map of the study
area, October 16, 1999. Clouds in the study area, outlined in yellow, are most
visible with classes 15-30. The legend for Figure 4.2(B) applies to Figure
I 0 1 2 Miles
The analysis of vegetative index two and vegetative index three showed that the
combination of high and low gain thermal bands was superior for mapping clouds. It is
important to note that the lack of cloud cover displayed with vegetative index three could
possibly lead to the incorrect assessment that clouds do not exist in the study area. To
conclude, three clouds were located in the study area, and their signature completely
dominated the signature of the ground, however it is inconclusive whether or not cloud
shadows prevented water detection.
NDVI was useful for mapping water under both dry and severe flood conditions,
and the Atlantic Ocean was the primary feature used to identify open water classes in
both NDVI maps. Open water represents a severely inundated condition where only the
spectral signature of water is visible; however this may also represent a condition where
emergent canopy does not exceed inundation depth. Atlantic Ocean open water classes
were found to conform to the east coast of south Florida' s peninsular land boundary.
Open water classes were found to be clearly distinguishable and separated from land
classes along the east coast boundary, Figure 4.4(A) and Figure 4.4(B).
C...._ lass 24
(1 0 10 20 Miles
Figure 4.4(A). October 16, 1999, NDVI map of the south Florida. Open water classes are
represented with blue. (B). April 9, 2000, NDVI map of south Florida. Classes
1-7 are open water and represented with blue.
200 0 2001 400 Miles~i
To identify and map water classes in the study area for October 16, 1999, and April
9, 2000, only the NDVI classes found in the Atlantic Ocean were used. The land
boundary was clearly visible from open water in both images; however the increase of
open water classes in the October 16, 1999, NDVI map showed the flood impact of
Classes 1 -7 were determined to be open water in the April 9, 2000, NDVI map,
and classes 1-17 were determined to be open water in the October 16, 1999, NDVI map.
The October 16, 1999, NDVI map was expected to have more water classes due to the
flood condition produced by Hurricane Irene. Figures 4.5(A) and Figure 4.5(B) display
the coverage of water in the study area. Clouds from Hurricane are visible with classes
18-20; however clouds in the study area are not distinguishable with classes 18-20.
Because clouds are not be mapped in October 16, 1999, NDVI map, it is difficult to
exactly determine the separation between open water and cloud pixels in the study area.
The method used to determine an open water class was successful for separating clouds
from water in the April 9, 2000, NDVI map; however this method is not adequate for
separating clouds from water in the October 16, 1999, NDVI map. Despite this
constraint, the October 16, 1999, NDVI map does display a large increase in the coverage
of water classes that is not found in the April 9, 2000, NDVI map.
~ 1 A
Figure 4.5(A). October 16, 1999, NDVI map of the study area outlined in yellow. Open
water classes are represented with blue. (B). April 9, 2000, NDVI map of the
study area outlined in blue. Open water classes 1-7 are represented with blue.
5~~~ V Study Area
s 1 20-21
V I 26-27
2 0 2 4 Miles i
High NDVI classes in the October 16, 1999, NDVI map are found where the
spectral signature of water is obstructed by the signature of canopy. These pixels were
mostly found in areas where high and dense canopy exists. This is most visible in the
wetland shrub/scrub areas and in the row crop areas where high NDVI class pixels are
located adj acent to NDVI water pixels.
Although NDVI was determined to be useful for verifying the flood extent, several
constraints became obvious during the analysis. First, the 30 meter spatial resolution of
NDVI maps failed to distinguish vegetation from water where vegetation canopy
exceeded ponded water depth. Second, clouds that were mapped with Band 8, vegetation
index two and vegetation index three were not mapped with NDVI. Finally, NDVI maps
could not display the duration, change in magnitude and extent of flooding for Hurricane
Irene, due to the low frequency of available Landsat 7 ETM+ images.
The procedure used to create bare earth topographic grids involved bare earth
modeling and spatial modeling. Bare earth modeling initially began by identifying ALSM
vegetation and artifact points with color infrared imagery and DEMs, and then the points
were removed. Point removal was based under the assumption that the topography in the
C-111 basin is extremely flat and that a low variability exists between neighborhood
elevation points. Spatial modeling was employed to predict elevation values where large
gaps were left from ALSM point removal. The spatial modeling procedure involved the
use of multiple interpolators and assessment of the generated statistics. The optimum
interpolation method was used to create both NAD 27 and NAD 83 ALSM DEMs.
The four deterministic interpolators that were used to create ALSM elevation grid
surfaces are inverse distance weighting, global polynomial, local polynomial and radial
based functions. The lowest root mean square value was used as the decision statistic for
selecting the optimum test method, however several tests were found to possess the
lowest value. To resolve this problem, the test that possessed a mean absolute error
closest to zero was selected as the optimum method.
The radial based function interpolator produced the overall lowest root mean square
error values, and was selected as the best interpolation method for ALSM DEMs, Table
4.1. Tests 3, 12, 13 and 16 all produced the lowest root mean square value, 0.1351 ft.,
therefore the lowest mean error statistic among these tests was used to select the optimum
search parameters. Because of its low mean error value, test 13 was selected as the
optimum search parameter method for NAD 27 and NAD 83 ALSM data. Details for the
other interpolators may be found in Tables 4.2, 4.3 and 4.4.
The search parameters for test 13 were applied to ordinary kriging, universal
kriging, disjunctive kriging and simple kriging interpolators. Simple kriging produced the
lowest root mean square value, 0.1433 ft., Table 4.5. Although geo-statistical
interpolators are more rigorous than deterministic interpolators, they are not ideal for
predicting topographic grids that possess a significant variability in density with ALSM
points (ESRI 2001). Furthermore, the high variation in point density within the DEM
made analysis and interpretation of semi-variograms inconclusive.
Classified ALSM DEM
Classified ALSM DEMs were manually created by assigning elevation values into
a specified interval. An elevation interval of 0.2 ft. was used to separate vegetation from
the bare ground between the elevations of 4-6 ft. for NAD 27. The legend for elevation in
Figure 4.6(A) describes elevation intervals in feet. Elevation intervals that were above the
maximum elevation threshold of 4.8 ft. were represented with green, to represent
vegetation. Three dimensional images of classified NAD 27 and NAD 83 DEMs were
used to analyze the effect of point removal, Figure 4.6(A)-(H). The classified TINT DEM
clearly displayed field vegetation, fiducial features and the S175 culvert. Except for part
of the L3 1W canal, the NAD 83 classified TINT DEM did not map these features. This is
attributed to large gaps produced by point removal. The three dimensional views of the
NAD 83 DEM in Figures 4.6 (E-H) show the effect of point removal.
It is interesting to note that both NAD 27 DEMs show extremely false low
elevation values east of the L3 1 W canal. This may be caused by scattering of the
infrared laser beam, or a problem with post processing. Both NAD 27 DEMs also display
extremely high elevation values that are not characteristic of the topography in the study
area, and this was likely caused by the laser beam striking an obj ect in the atmosphere.
0 0.05 0 t 02 0.3 0.4
Figure 4.6(A) Planar view of the NAD 27 study area DEM. The legend applies to all
three dimensional (3D) DEMs. (B) 3D southeasterly view of the study area
using NAD 27 ALSM data. (C) 3D southerly view of the study area using
NAD 27 ALSM data. (D) 3D westerly view of the study area using NAD 27
ALSM data. (E) 3D southerly view of the study area using NAD 83 ALSM
data. (F) 3D easterly view of the study area using NAD 83 ALSM data. (G)
3D easterly view of the study area using NAD 83 ALSM data. (H 3D
westerly view of the study area using ALSM data.
~t' '~+:Ci .i'
Surface Water Elevation Map Analysis
The four geo-statistical interpolators that were used to predict elevation values for
surface water elevation grids are universal kriging, disjunctive kriging, simple kriging
and ordinary kriging. The root mean square error served as the decision statistic for
selection of the optimum interpolation method. If two tests shared an equal value, then
the test with the mean error value closest to zero was selected as the optimum method.
For universal kriging, test 15 produced the overall lowest root mean square error
value of 0.4701 ft. (see Table 4.6), and the search parameters for test 15 were applied to
create all surface water elevation maps, Table 4.7. These parameters were also used for
October 16, 1999, NAD 27 surface water elevation data; however the root mean square
error value was 0.02 ft. greater than that of NAD 83. Furthermore, NAD 27 possessed
greater mean error and mean error values; however the root mean square standardized
error and average standard error values for NAD 27 were less than NAD 83. Details for
the other interpolators are shown in Tables 4.8, 4.9 and 4.10.
Table 4. 11 lists the surface water elevation values for the study period, and Figure
4.7 shows a graph of surface water elevation data for the study period. An analysis of
surface water data showed a sharp increase in elevation that was coincident with the
impact of Hurricane Irene, and a gradual decrease associated with drainage. SWEM
contours in Figures (A-H) appear to show a directional flow towards the S332 and S178
pumping stations. The SFWMD (2000) reported that the S332 was operating at maximum
capacity on October 14, 1999, however no specific information is provided for the other
water control structures in the study area. Surface water elevation maps displayed a
smooth transition between contour intervals; however discontinuities in the elevation
intervals were more noticeable as the distance between stations increased, Figure 4.8 (A-
K). Elevation values are in feet NGVD 88.
Prediction error maps for SWEM were produced, because universal kriging was
selected as the interpolation method. Figure 4.9 (A-K) show universal kriging prediction
error maps made from surface water elevation maps. Several trends were noticed during
the production of SWEM prediction error maps. October 16, 1999 displayed the lowest
prediction error, and an increase in prediction error existed for the remainder of the study
period. Furthermore, SWEM prediction error for October 12, 1999, was observed to be
the highest for the entire study period. Prediction error values are in feet.
The low prediction error for October 16, 1999, is attributed to both high surface
water elevation values and the low variability in values for stations throughout the study
area. The increase in prediction error coincides with an increase in the variability in
elevation values between neighboring stations. This increase in variability is most likely
due to the effect of water management and variable drainage rates for water control
12 13 14 15 16 17
-*-S178 -. NP 158
-a- FP2 -* S332 -
18 19 20 21 22
- NP112 -- S177
Figure 4.7 Graph of surface water elevation values. Values are in feet NGVD 88.
SURFACE WATER STfES
0 0.25 0.5 1 1.5 2
Figure 4.8 (A) SWEM October 12, 1999. (B) SWEM October 13, 1999. (C) SWEM
October 14, 1999. (D) SWEM October 15, 1999. (E) SWEM October 16,
1999. (F) SWEM October 17, 1999. (G) SWEM October 18, 1999. (H)
SWEM October 19, 1999. (I) SWEM October 20, 1999. (J) SWEM October
21, 1999. (K) SWEM October 22, 1999. Elevation values are in feet NGVD
SSURFACE WATER STfES
0 0.25 0.5 1 1.5 2
0 11 W2
0 0.25 0.5 1 1.5 2
0 0.25 0.5 1 1.5 2
0 0.25 0.5 1 1.5 2 Z2
0 0.25 0.5 1 1.5 2 ~-2
3 5- 3.75
0 0.25 0.5 1 1.5 2
0 0.25 0.5 1 1.5 2
SU RFACE WATER SITES
0 0.25 0.5 1 1.5 2 [ ]2-2.25
E ~~2 5-275 32 .
S 3 5 3.5
13 75 4
0 0.25 0.5 1 1.5 2 O r-
0 0.25 0.5 1 1.5 2
W E us.25
*SURFACE WATER SITES
- 31 W
Figure 4.9 (A) Prediction error for SWEM October 12, 1999. (B) Prediction error for
SWEM October 13, 1999. (C) Prediction error for SWEM October 14, 1999.
(D) Prediction error for SWEM October 15, 1999. (E) Prediction error for
SWEM October 16, 1999. (F) Prediction error for SWEM October 17, 1999.
(G) Prediction error for SWEM October 18, 1999. (H) Prediction error for
SWEM October 19, 1999. (I) Prediction error for SWEM October 20, 1999.
(J) Prediction error for SWEM October 21, 1999. (K) Prediction error for
SWEM October 22, 1999. Legend error values are in feet.
0 0 25 0 5 1 1 5 2
0 0.25 0.5 1 1.5 2 SUFC WAE SIS
W E 0.5 -0.6
SSURFACE WATER STiES
lo 0 .1
S0.1 0 2
10.3- 0 4
0.5 -0 6
0.7 -0 8
0 0.25 0.5 1 1.5 2
SSURFACE WATER SITES
0 0.25 0.5 1 1.5 2