Zhixiao Xie (Florida Atlantic University, xie@fau.edu), Zhongwei Liu (University of Florida), John W. Jones (U.S. Geological Survey)

Hydrology regime is a critical limiting factor in the delicate ecosystem in the Greater Everglades area in Southeastern Florida, and "making the water right" is regarded as the key to the successful restoration of this unique wetland ecosystem. One essential component to represent and model the hydrological regime is a reliable and accurate ground Digital Elevation Model (DEM). The Everglades Depth Estimation Network (EDEN) products (including ground DEM) developed by the USGS have been a great success and well received by scientists and resource mangers involved in Everglades restoration. This study intends to improve the EDEN DEM in the north Everglades, by adopting a landscape unit (LSU) based interpolation approach. The study first filters the input elevation data based on newly available vegetation dataset and then creates a separate geostatistical model (universal kriging) for each LSU. The resultant DEMs have encouraging cross-validation and validation results, especially since the validation is based on an independent secondary elevation dataset (derived by subtracting water depth measurements from EDEN water surface elevations). The LSU based DEMs is finally mosaic-ed and compared with the current release DEM at 400 m EDEN cell and at selected sample points. It appears that the revised DEM leads to a better representation of the terrain variation in the study area.

The USGS Everglades Depth Estimation Network (EDEN) provides critical datasets that support analysis and models of the Everglades ecosystem for the Comprehensive Everglades Restoration Plan (CERP) (Telis, 2006), including water depth, hydro-periods, ground DEM, daily water surface level, and etc. Reported at a uniform 400-m EDEN cell grid system, EDEN data are being actively utilized by various research groups and for restoration decisions. Yet, the development of EDEN is an iterative evolving process as additional high accuracy elevation data are collected, water surfacing algorithms improve, and additional ground-based ancillary data become available (Jones and Price, 2007). This study represents a recent effort for DEM improvement in A.R.M. Loxahatchee National Wildlife Refuge (LNWR), in the northern Everglades.

1. DATA
USGS High Accuracy Elevation Data (HAED, Figure 1): collected at a 400-m grid, through Airborne Height Finder (AHF), (Desmond, 2003), to suit the unique terrain surface in the Everglades, typically under water, obscured by vegetation, and extremely flat.

EDEN daily water surface: created with the radial basis function (RBF) interpolation of real-time gage data (Pearlstine et al., 2007), from January 1, 2000 to present.

PI dataset (Figure 2): water depth measurements by different scientists over 8 years (2000-2007). Ground elevation at these sites was derived by subtracting from the daily EDEN water surface the depth measurement.

Vegetation map: result of a stereoscopic analysis of 1:24,000 scale color-infrared positive transparencies flown in December 2003, with a minimal mapping unit of 50 m x 50 m, by South Florida Water Management District (Rutchey, 2008, email communication).

Figure 1 [larger image] Figure 2 [larger image]

2. Filter out unrepresentative HAED points
The study area has numerous small-size elevated "spikes" (pop-up peats colonized by vegetation, degraded and dissected tree islands, etc.). The survey HAED elevation at alike locations may not be representative of elevation at the target scale, 400 m x 400 m EDEN cell. Based on the 2008 vegetation map (SFWMD), the unrepresentative HAED points are filtered with the following protocol: (1) HAED point falling on upland + others ; and (2) areal coverage of upland + others in the corresponding EDEN cell is less than 33%

3. Landscape Unit-based DEM development
The input HAED/PI points are divided into 3 groups based on 3 landscape units (LSU) (Figure 3). A landscape units (landscape subregions) represents distinct landscape patterns and/or management in the Greater Everglades (RECOVER, 2004). After examining the data, a geostatistical (kriging) model is developed separately for each landscape unit. Each model is assessed by cross-validation and validation with PI data. The three models are used to create the DEM at the 400 m EDEN cell grid and mosaic into the whole EDEN DEM. The revised DEM is compared with released DEM (Figure 4).

Figure 3 [larger image] Figure 4 [larger image]

Results

For each landscape unit, the HAED data are examined with spatial trend determined and outliers removed (1 outlier for center and south LSU respectively).

Different kriging methods are experimented. In the end, universal kriging (1^st order trend, anisotropy) is used to create DEM for each LSU at a 400-m resolution (Figure 5). The cross-validation and validation results are shown in Tables 1 and 2. Compared with the Root-Mean-Square Error (RMSE) of 16.16 cm in the current release DEM, the LSU based DEMs has lower RMSE in north and center LSUs.

The 3 DEMs are then fitted into EDEN 400-m grid and mosaic-ed into one DEM (Figure 6). Different blend methods are tested.

Figure 5 [larger image] Figure 6 [larger image] Figure 7 [larger image]

The revised DEM is compared with the current release DEM (2007) at EDEN cell grid (Figure 7), as well as at selected HAED points and summarized by land cover types (Table 3). The average differences over independently described land cover classes suggest that the revised DEM decreases the overall terrain variation (e.g., lower tree islands, higher open water areas).

Anisotropic universal kriging by landscape units, with input HAED data filtered based on newly available vegetation map, leads to better cross-validation and validation results, although the southern part warrants further examination. In particular, the validation results with independent PI dataset are encouraging. Compared with the current release DEM, the decreased terrain variation in the revised DEM is a better representation of the actual terrain trends.

The authors greatly appreciate Pamela Telis, USGS, for her leadership and support. Aaron Higer is always a source of vision and inspiration. Thanks are also extended to the scientists who shared their field measurements in the Everglades, including Todd Z. Osborne (University of Florida), Dale E. Gawlik (Florida Atlantic University) Michael S. Ross (Florida International University), and Joel C. Trexler (Florida International University). The project is funded by the USGS through CESU agreements.

Desmond, G.D. 2003. Measuring and mapping the topography of the Florida Everglades for Ecosystem Restoration. U.S. Geological Survey Fact Sheet 021-03.

Jones, John. W. and Susan D. Price. 2007. Everglades Depth Estimation Network (EDEN) digital elevation model research and development. U.S. Geological Survey Open-File Report 2007-1034.

Pearlstine, L., A. Higer, M. Palaseanu, I. Fujisaki, and F. Mazzotti. 2007. Spatially Continuous Interpolation of Water Stage and Water Depths Using the Everglades Depth Estimation Network (EDEN): CIR1521. Institute of Food and Agricultural Sciences, University of Florida.

RECOVER. 2004. CERP Monitoring and Assessment Plan: Part 1 Monitoring and Supporting Research. Restoration Coordination and Verification Program, c/o United States Army Corps of Engineers, Jacksonville District, Jacksonville, Florida, and South Florida Water Management District, West Palm Beach, Florida.

Telis, P.A. 2006. The Everglades Depth Estimation Network (EDEN) for Support of Ecological and Biological Assessments. U.S. Geological Survey Fact Sheet 2006-3087.