The ATLSS High Resolution Multi-Data Source Topography (HMDT).

Scott M. Duke-Sylvester
sylv@tiem.utk.edu
Introduction
The ATLSS High Resolution Multi-Data Source Topography (HMDT) is a replacement for the ATLSS High Resolution Topography (HRT). This new topography is created by merging a number of different topographic data sources into a single topographic map. This map provides topographic information at a 500x500 meter resolution for most of the natural areas of South Florida (SF). These data are used to estimate water depths, based on the 2x2 mile South Florida Water Management Model (SFWMM) stage heights, at the 500x500 meter resolution. Spatial heterogeneity at this scale strongly influences the individual and population level dynamics of both plants and animals. It is important to have estimates of water depths at this resolution to understand better how Everglades restoration will affect the fauna and flora of SF.

Creation of HMDT
The current version of the HMDT is 1.0 (2003091200). This version is based on four sources of topographic information: 1) Light Detection and Ranging (LIDAR), 2) USGS High Accuracy Elevation Data (HAED) (Desmond, 2003), 3) South Florida Water Management Topography (SFWMMT) (Hinton 2001), and 4) ATLSS High Resolution Topography (HRT) (Duke-Sylvester 2000). The LIDAR and HAED data sets are products of US Geological Survey (USGS) programs that measure ground surface elevation in SF. HAED data represent ground surface elevation measurements taken on an irregular lattice where sample points are approximately 400 meters apart. The LIDAR data used here are estimated ground surface elevations on a lattice where sample points are 100 meters apart. ( LIDAR estimates are based on data sampled on a much finer spatial lattice (4 meters) (Hinton 2001). ) For both of these data sets the geographic location and ground surface elevation is reported for each point in their respective lattices. The higher density of sampling points is an important difference between these and other topographic data sets. The sampling density also makes LIDAR and HAED particularly relavent to the ATLSS project.

The SFWMMT data are organized into a raster grid. Each cell of the grid represents a plot 2 miles (3218 meters) on a side, and the grid uses 41 rows and 65 columns to cover SF. For each grid cell the SFWMMT provides an elevation in feet, NGVD 1929. This single elevation value is interpreted as the ground surface elevation for the entire 2x2 mile plot. Hinton (2001) provides details on the creation of the SFWMMT.

The HRT data are also organized into a raster grid. Each cell of the grid represents a plot 30 meters on a side and the grid uses 6973 rows and 4399 columns to cover SF. For each grid cell the HRT provides an estimate of elevation. Unlike the LIDAR and HAED elevations, HRT elevations are estimates made using the ATLSS HRT model. The model estimates elevations based on a hypothesized relationship between the kind of vegetation present at a location and the elevation of that location. The elevation of a location should result in local hydrologic conditions that are appropriate for the local vegetation. Duke-Sylvester (2000) provides details on the creation of the HRT.

For the purposes of this project, all geographic locations are converted, where necessary, to UTM coordinates, in meters, NAD 1983, using ArcGIS 8 and elevations are converted to meters, NGVD 1929. This makes the input data and, more importantly, the resulting HMDT data compatible with the SFWMM output and earlier ATLSS results.

The four data sets are used in a hierarchical manner when computing elevations for the HMDT. In areas where LIDAR or HAED samples are available these data sets are use to the exclusion of the other two. Figure 1 shows the areas for which LIDAR and HAED are available. The areas covered by these two data sets do not overlap, so there is no conflict about where to use each data set. LIDAR coverage is currently limited to the portion of Water Conservation Area 3A north of I-75 (this area is referred to as WCA-3A north). LIDAR coverage is shown in purple in figure 1. HMDT elevations in this area are based on the mean of the LIDAR data points that fall with the boundaries of a HMDT 500x500 meter cell. However, if the number of LIDAR measurements in a HMDT 500x500 meter cell is less than 16, LIDAR is not used to estimate the elevation for that cell. The result of this restriction is that there are two areas within WCA-3A north where elevations are not based on LIDAR. One of these areas follows the northern section of the Miami canal. The other is a small region in the center of WCA-3A north (see figure 1).

HAED data cover most of the Everglades National Park (ENP), Big Cypress National Preserve (BCNP) and portions of WCA-3A south of I-75 (Desmond 2003). The HAED used in the SFWMMT is a subset of the total HAED collected by USGS. To make the initial version of the HMDT consistent with the SFWMMT, the use of HAED has been limited to the area used in the SFWMMT. HMDT elevations in this area are based on the mean of the HAED data points falling within the boundaries of a HMDT 500x500 meter cell. As a result of the interaction between the irregular HAED lattice spacing and the HMDT raster grid, there can be between 0 and 4 HAED data points in each HMDT 500x500 meter cell. Mean HAED elevation are used when there are between one to four samples in a HMDT 500x500 meter cell. When there are no HAED samples, HAED is not used to estimate elevations. This results in a number of scattered points within the HAED region where HAED is not used to estimate elevations (see figure 1). Locations where HMDT elevations are based on HAED are show in red in figure 1.

Where LIDAR and HAED elevations are unavailable, HMDT elevations are based on either the SFWMMT or the HRT data. The choice of SFWMMT or HRT as the source of data for HMDT is based on the presence of absence of urban or agricultural areas. Urban and agricultural areas are identified by the presence of urban and/or agricultural habitat types from version 6.6 of the Florida GAP (FGAP) map. The FGAP is a raster grid with a 30x30 meter cell size. Each cell is classified as being one of 71 habitat types. These 71 habitat types can be readily grouped into natural, agricultural and urban categories. Area classified as either agricultural or urban include most of the eastern coast and the Everglades Agricultural Area.

An HMDT 500x500m elevation is based on SFWMMT if it intersects any FGAP 30x30 m cell that is classified as either urban or agricultural. In these cases the estimated elevation for an HMDT cell is based on the weighted mean of elevations from the intersecting 2x2 mile SFWMMT cells. Most HMDT 500x500 meter cells are completely contained in a 2x2 SFWMMT cell, in which case the HMDT elevation is simply the elevation of the containing 2x2 mile SFWMMT cell. In some cases the HMDT 500x500 meter cell intersects with two to four SFWMMT 2x2 mile cells. In these cases the HMDT elevation is the weighted mean of the SFWMMT elevations where the weights are based on the relative area of intersection between each 2x2 mile cell and the 500x500 meter cell. The areas with elevations based on SFWMMT are shown in gray in figure 1. An HMDT 500x500m elevation is based on HRT if the entire cell is composed of natural vegetation. In these cases HMDT, elevations are based on the weighted mean of the HRT elevations from the 30x30 meter HRT cells that intersect the HMDT 500x500 meter cell. Weights are based on the area of intersection between each 30x30 meter cell and the 500x500 meter cell. Locations where HMDT elevations are based on HRT are shown in yellow in figure 1.

Currently the software developed to create the HMDT data has been written for PV-Wave (Visual Numerics, Inc., San Ramon, CA, USA). The decision to use PV-Wave was influenced by the limitations on time available to complete this project. The high level language features available in PV-Wave simplifies the manipulation of complex data objects and permits rapid code development. The time and effort savings achieved by using PV-Wave are critical for ATLSS development where both commodities are at a premium.

Results
Figure 2 shows the HMDT version 1.0. In this figure locations are colored according to their HMDT elevation.

Arc data files are available for the maps shown in figures 1 and 2.
Figure 1 : Data Source Map
  ASCII grid format (.asc)
  Arc exchange format (.e00)

Figure 2 : HMDT Map
  ASCII grid format (.asc)
  Arc exchange format (.e00)


Figure 3 shows a plot of SFWMMT v5.0 elevations against mean HMDT 500x500m elevation. For each of the 1745 2x2 mile SFWMMT cells I have computed the mean 500x500m HMDT elevation and plotted the SFWMMT elevation against the mean HMDT elevation. A line is fitted to the data using least squares and is shown in black in figure 3. The correlation coefficient for this regression is 0.97, indicating a strong linear relationship between the SFWMMT elevations and the mean HMDT elevations. The slope and intercept are 1.08 and -0.137, respectively. While these values deviate from the slope and intercept representing a perfect, one-to-one, relationship between the data (slope = 1.0, intercept = 0), the differences are small. The relevant statistics for the regression are shown in table 1.

Further refinements
There are a number of future refinements that can be made to the HMDT. Listed in no particular order they are:
1) Update the version of the HRT model output being used the create HMDT. HMDT version 1.0 (2003091200) is based on HRT version 2.0. HRT version 2.0 is based on FGAP version 6.6 and SFWMM Cal/Ver run 3.7. Once SFWMM Cal/Ver version 5.0 becomes available I can update the HRT to version 3.0 and make a new version of the HMDT.
2) Refine the LIDAR and HAED data points being used. I have made every effort to use the same LIDAR and HAED data points used by the SFWMD personnel to create the SFWMMT. However, when I compute the mean LIDAR and HAED elevations for the SFWMMT 2x2 mile cells I get elevations that are close to, but not identical to, the elevations reported in Hinton (2001). I have identified two possible sources of error:
  A) I have followed the methods described in Hinton (2001) but, I may not be excluding the same set of outliers as those excluded by Hinton (2001).
  B) The registration of the 2x2 mile cells used to compute mean elevations may be different than the one used in Hinton (2001).
I'm currently investigating both of these options.
3) Based on the comments from Matthew Hinton from the SFWMD, I will create a TIN data layer at the 500x500 meter resolution for the HAED data and use elevations from the center of each 500x500 meter cell as the estimated elevation for the HMDT.
4) The SFWMD personnel also have new topographic information for the Rotenberger region. This information will be included in the SFWMMT used for the new SFWMM version 5.0. (Winnie Said, SFWMMD, personal communication). I will be able to use these data to obtain HMDT for this region when it becomes available.
5) There are a number of areas in SF for which HAED data are available but omitted from the SFWMMT and the HMDT. These data could be used where available to estimate elevation instead of the SFWMMT or the HRT model output.

Conclusions
The analysis comparing the HMDT to the SFWMMT indicates that use of HMDT as a basis for estimating water depths at the 500x500 meter resolution should be appropriate, at least in the sense of producing elevations consistent with the 2 mile SFWMMT when the HMDT elevations are averaged to that level of resolution.

If you have comments or suggestions about the HMDT please contact me at sylv@tiem.utk.edu

Acknowledgments
The work presented here has been supported by the ATLSS project, with funding provided in significant part by DOI's Critical Ecosystems Studies Initiative, through the U.S. Geological Survey, Florida Caribbean Science Center, under Cooperative Agreement #1445-CA09-95-0094, Subagreement #1 with the University of Tennessee. The statements, findings, conclusions, recommendations, and other data in this report are solely those of the authors and do not necessarily reflect the views of the U.S. Geological Survey. I am grateful to Winnie Park Said and Matthew Hinton as well as other personnel at the South Florida Water Management District who have been very helpful in obtaining and understanding the LIDAR and HAED data, as well as the SFWMM topography.

Citations

Desmond, G. et. al. 2003. High Accuracy Elevation Data Collection. www.sofia.usgs.gov/projects/elev_data/. Visited September 16, 2003.

Hinton, M. 2001. Topo200 Elevation Update (SFWMM200). South Florida Water Management District internal memorandum. www.sfwmd.gov/org/pid/hsm/models/sfwmm/v5.0/sfwmm_v50_data.zip. Visited September 16, 2003.

Duke-Sylvester, S. M. 2000. ATLSS High Resolution Topography/High Resolution Hydrology. Visited September 16, 2003.

Tables and Figures

Figure 1 : Data source for each HMDT 500x500 meter cell.

Figure 2 : Map of the HMDT elevations.

Figure 3 : Comparison of the SFWMMT elevations and the mean of the of the 500x500 meter HMDT elevation for each 2x2 mile cell.

Correlation Coef. 0.973506 R Squared 0.947713 Observations 1745 Intercept -0.137 Slope 1.08

Table 1 : Summary of Fit for the regression of SFWMMT 2x2 mile elevations on mean HMDT 500x500 meter elevations.

Change Log
In this section I will notate changes and corrections to this document. Only changes made since this document became public (Oct. 21, 2003) are included.
Oct. 21, 2003 : First version of this document made public.
Oct. 27, 2003 : In the Further refinements section I stated that HMDT 1.0 was based on HRT version 1.0. This was incorrect. HMDT 1.0 is based on HRT version 2.0. I have made this changes as well as the sentence that describes the bases for HRT 2.0.

---

Copyright © 2003 Scott M. Duke-Sylvester.
Last Updated : 2003/10/27 20:18:40
Version : 1.8