Introduction

Coastal prairie wetlands (CPWs) form a locally abundant landscape feature that store and convey surface water drainage in the low-relief landscape around Galveston Bay, Texas, USA. Since U.S. Supreme Court rulings in the cases of SWANCC (531 U.S. 159 [2001]) and Rapanos (547 U.S. 715 [2006]), federal jurisdiction has generally not been extended to wetlands without well-established surface water connections to navigable waters (i.e., “isolated wetlands”). Based on these rulings, the U.S. Corps of Engineers Galveston District now considers most CPWs outside of the 100-year floodplain to be isolated. As a consequence, CPWs along the Texas coastal plain are being lost at an alarming rate (Jacob and Lopez 2005). There is concern from local officials and regulatory agencies (e.g., Texas Commission on Environmental Quality and Galveston Bay Estuary Program) that the cumulative loss of CPWs will increase local flooding and adversely affect water quality in the region, particularly in Galveston Bay, an important commercial and recreational resource. Unfortunately, there are few data that describe the distribution and morphology of CPWs. Moreover, a comprehensive inventory of CPWs is needed to facilitate estimates of cumulative water quality and water storage functions.

Coastal prairie wetlands along the Gulf Coast have been called geographically isolated (Comer et al. 2005) and, as such, fall into a category with vernal pools, prairie potholes, and playa lakes. The concept of isolation is not well defined, as stated by Leibowitz (2003: 518): “Isolated is generally a matter of degree and for this reason there is no accepted scientific definition of ‘isolated’ ponds or wetlands”. Rather, such wetlands may serve as headwaters within the hydrologic setting, with intermittent but regular discharge to streams, rivers, and estuaries (Wilcox et al. 2011). Nonetheless, we submit that these small wetlands are undervalued because they are viewed as individual units rather than a system. This in turn contributes to a lack of understanding of their role in regulating local, regional, and global processes. While the loss of a single small wetland may go unnoticed, the cumulative impacts that result from decisions such as those in SWANCC and Rapanos can lead to a critical loss of landscape-level ecosystem services such as flood control (Brody et al. 2008). Development of accurate, comprehensive inventories of such wetland systems is the first step in estimating their cumulative functions. For example, recent models indicate that the global extent of natural water bodies is twice as large as previously known because small (<1 km2) water bodies have been under-surveyed (Downing et al. 2006). These findings stress the importance of accounting for abundant small water bodies such as CPWs when calculating global and regional rates of nutrient cycling.

Evaluating ecosystem services provided by a system of thousands of small wetlands presents challenges for researchers. Site access and limited resources constrain field investigations; however, with the availability of Geographic Information Systems (GIS) and high resolution imagery, detailed regional inventories are now being conducted (Bowen et al. 2010; Lane and D’Amico 2010).

This project used a GIS to describe CPWs and their structural features to support estimates of their water quality and water storage functions. Catchment size (i.e. drainage area) is an important feature of CPWs, yet accurate information on catchment size for numerous small wetlands is not generally available. Our study used light detection and ranging (lidar) data to derive wetland volumes and catchment areas for CPWs. We are aware of only one GIS-based analysis, Lane and D’Amico (2010), that utilized high-resolution lidar data to estimate regional wetland water volumes. Zahina et al. (2001) developed a GIS-based approach to rapidly assess wetland functions in southern Florida with models that were based on variables derived from soils and land use/land cover geospatial datasets. Our study expanded the Zahina et al. approach by including high-resolution lidar data to estimate wetland volumes and delineate catchment areas.

The primary study objective was to develop a framework for inventorying the 10,000-plus CPWs in a study area surrounding Galveston Bay. The inventory describes important wetland hydrogeomorphic features (area, volume, catchment area, hydroperiod) and structural attributes (soil, vegetation, land use). These data were subsequently used to apply hydrogeomorphic (HGM) models to CPWs in the study area, with the development and analyses of those models being addressed in a separate manuscript. The inventory provides critical information to regional policy makers, ecologists, wetland scientists, and land managers. Additionally, the approach is applicable to other regions where the majority of wetlands are located on private lands (e.g., prairie potholes, vernal pools) and where wetland systems are comprised of small wetlands too numerous to characterize individually. A secondary objective was to assess the accuracy of GIS-derived structural features by comparing GIS values to field data from 12 CPWs. Greater analyses of the error associated between GIS-derived estimates are needed due to the increasing trend of using geospatial data to characterize wetlands.

Methods

Study Area

The study area was defined by 32 USGS 1:24,000 topographical quadrangles surrounding Galveston Bay, Texas, USA (Fig. 1). Covering parts of four counties (Brazoria, Chambers, Galveston, and Harris), the 5,376 km2 land area (excludes Gulf of Mexico) comprises much of the lower watershed of Galveston Bay, an important economic resource that accounts for approximately 33% of the annual Texas commercial fishing income (Lester and Gonzalez 2002). The region has a sub-tropical climate and a year-round growing season (Miller and Bragg 2007). CPWs included in this study are located within the Beaumont geologic formation of fluvial-deltaic sediments with circular to irregular depressions (McGowen et al. 1976). The area has been described as a “clay plain” (Smeins et al. 1991) with poorly drained soils, heavy vegetation, and a general lack of incised drainage (Sipocz 2005). The landscape is extremely flat (0–1% slopes), gently sloping towards the Gulf of Mexico. Within this landscape lie pimple mounds, undrained depressions and segments of meandering and abandoned stream channels (Aronow 1986) with local microtopography of 1.5 to 3 m. The dominant soil types are Vertisols and Alfisols that developed over Pleistocene deposits. Kishné et al. (2009) reported that most CPW soils are episaturated with a zone of unsaturated soils between the surface saturated soils and the unconfined groundwater table. Thus with the exception of coastal dune wetlands, most CPWs have little groundwater exchange.

Fig. 1
figure 1

Study area defined by 32 USGS topographic quadrangles showing 12 coastal prairie wetland study sites

Full size image

Figure 2 shows NWI polygons for CPWs located in a rapidly urbanizing area of southern Harris–northern Galveston counties. Wetlands in the westernmost part of the study area (Brazoria County) are primarily relic drainage channels within the 100-year floodplain and many are located in wildlife refuges. Wetlands on the eastern side of Galveston Bay (Chambers County) include large rice-farmed or grazed wetlands and wildlife areas.

Fig. 2
figure 2

Areal imagery showing NWI boundaries (white lines) of LG and other CPWs located in southern Harris County where urban development is heavy (Aerial photo credit: ESRI, 2009). Inset is a photo taken at LG

Full size image

Structural Features

Structural features (i.e., model variables) of CPWs were identified, defined, and eventually normalized to values between 0 and 1.0 as required for eventual use in HGM models (Table 1). Geospatial datasets used to evaluate the structural features (Table 1) were defined or re-projected to the Universal Transverse Mercator Zone 15 N coordinate system (North American Datum 1983). ESRI® ArcGIS 9.3.1 and ArcHydro Tools 1.3 (Environmental Systems Research Institute, Redlands, California) were used to calculate values for the variables.

Table 1 Overview of structural features including definitions and data sources. Lidar was obtained from Harris County Flood Control District (2008) and Texas Natural Resource Information Systems (2006)
Full size table

We used the National Wetland Inventory (NWI) data as a baseline for wetlands in the study area because it is the best available dataset for such an extensive region. The NWI was also used to determine wetland surface areas and wetland classifications such as freshwater emergent, freshwater pond, scrub/shrub, aquatic bed, and freshwater forested wetlands. The majority of the NWI data for the study area were updated in 2006; however, six quadrangles were last updated in 1992. The 1992 NWI data were mapped using 1:80,000 color-infrared imagery except for one quad that was mapped using color-infrared imagery at a scale of 1:40,000. The 2006 NWI data were mapped using color-infrared imagery at a scale of 1:40,000. The minimum mapping unit for NWI data was one acre (United States Fish and Wildlife Service USFWS 2004).

Catchments for wetlands were delineated using a tiled approach. The digital elevation model (DEM) dataset was divided into tiles using natural breaklines such as roads and rivers to avoid edge effects. A smoothing process was used to minimize edge effects caused by tile boundaries. This process included buffering the tile boundaries by 0.5 km and individually defining catchments for wetlands falling inside these buffers. The catchments delineated in the buffers of the tiles were analyzed and merged with those delineated within the tiles. All catchments were combined to make a seamless dataset for the study area. Catchment boundaries were based on two lidar-based DEM datasets. Harris County lidar data had a cell size of 1 m and the remaining data had a cell size of 1.4 m. The Harris County DEM was resampled to a cell size of 1.4 m to match the resolution of the other elevation data using a nearest neighbor approach. Lidar data used in this study were post-processed by the vendors to remove vegetation, buildings, and other structures, resulting in a “bare earth” DEM.

ArcHydro Tools 1.3 was used to delineate catchments for wetlands. Using the DEM, water flow direction was determined and catchments were delineated by identifying pixels in the DEM that flow toward the wetland. A major issue with wetland catchment delineation is how “sinks” are managed. A “sink” is defined as a location where surface water flow in interrupted. Sinks are often errors created during the interpolation process when creating a DEM; however, sinks can also represent natural depressions. The ArcHydro Toolset allows natural sinks, in this case wetlands, to be preserved while other sinks, such as errors in the DEM, to be filled. The NWI dataset was used to confirm which sinks were natural depressions (i.e., wetlands). The NWI inventory includes many wetland systems that are divided into smaller, conjoined wetlands, based on characteristics such as water regime and vegetation. However, for catchment delineations and wetland volume calculations, conjoined wetlands were treated as one wetland system (n = 7,372). All catchment areas include the area of the receiving wetland.

Using high resolution lidar data to define catchments for thousands of individual wetlands was challenging. Lidar data were used to delineate catchments only for wetlands outside of the mapped 100-year floodplain (n = 3,337). Floodplain boundaries were based on Federal Emergency Management Agency maps for Brazoria, Chambers, and Galveston Counties and Harris County Flood Control District maps for Harris County. Catchments could not be delineated for wetland systems within the 100-year floodplain due to their potential connectivity to adjacent rivers, computer processing limitations, and data limitations. Some catchments for wetlands that were associated with an adjacent river system may have extended several kilometers upstream and potentially outside the study area. Attempts to automate the delineation of those catchments were unsuccessful due to computer processing limitations. For these reasons, use of lidar to delineate catchments was abandoned for wetlands within the 100-year floodplain. While each wetland’s boundary could have been manually determined with digital topographic maps, that was not considered feasible given the large number of wetland systems involved (n = 4035). Therefore, based on professional judgment, 100 m wide buffer areas were substituted for wetland catchment areas within the 100-year floodplain. Overlap in buffers occurred when wetlands were within 100 m of each other. Where overlap occurred, the overlapped area was included in the catchment area measurement for each wetland. Hawth’s Tools, an add-on for ArcMap, was used to calculate features within the overlapping buffers because it processes each buffer separately.

Wetland volumes were obtained using a modified version of the Antonić et al. (2001) methodology for estimating sink depth. The “fill sinks” function in ArcGIS, which allows proper identification of basins and streams, was used to fill wetland depressions. Wetland volumes were based on the differences between elevations of the newly filled DEM and the original DEM. Specifically, the elevation of each pixel in the original DEM was subtracted from the elevation value in the filled DEM. If there was no difference in elevation, the resulting output pixel value was 0. If the pixel was filled, the difference resulted in a positive elevation value representing potential depth of water at that point. Next, the depth per pixel was multiplied by the area of the pixel (1.96 m2) to obtain the volume per pixel. Zonal statistics in ArcGIS is a function that summarizes pixels in a raster dataset within “zones”. NWI boundaries provided the “zones”, resulting in the total volume for each wetland. Assumptions with this approach include: 1) the bottom of the wetland is represented by the original DEM; and 2) the resulting volume represents the maximum or potential wetland volume at full pool. The method is best suited for natural systems, as hydrologically modified, deep wetlands with steep contour typically yield exaggerated depths and volumes.

National Land Cover Data (NLCD) from 2001 were used to map land use/land cover in the study area. These classifications were compared to the observed land uses for the 12 study sites.

One-meter resolution color infrared imagery from the National Agriculture Imagery Program (NAIP) collected in 2004 was used to calculate a Normalized Difference Vegetation Index (NDVI). NDVI is a standard vegetation index based on a plant’s absorption of chlorophyll in red visible light and reflection of near-infrared light. The index is obtained using the following equation: NDVI=(NIR – RED)/(NIR+RED) where NIR is the brightness value recorded in the near infrared band and RED is the brightness value recorded in the visible red band in the imagery. Application of the equation produces a raster dataset with pixel values ranging from −1.0 to 1.0. Negative values and near zero values represent open water features and bare soil; generally, values of 0.1 to 1.0 represent vegetated areas. Using true-color NAIP imagery collected in 2005, a threshold value (0.1) for the presence of vegetation was determined. Values below this threshold were considered unvegetated pixels, allowing calculation of percent vegetated cover of a wetland and its immediate surroundings (30 m buffers). Hawth’s Tools were used to process data in overlapping 30 m buffers.

Soil clay content was taken from the Soil Survey Geographic Database (SSURGO), which is organized by soil map units. In cases where more than one soil type was mapped within a wetland boundary, a mean weighted clay percentage was calculated.

Water regimes were assigned according to NWI classifications. The NWI does not assign water regime classifications to farmed wetlands, thus the water regime of an adjacent wetland was applied to farmed wetlands. Wetlands classified as grazed pasture were assigned the water regime of the adjacent wetlands and those farmed in rice were assigned an intermittently exposed hydroperiod.

Error Verification

Error associated with water regime, soil, vegetation, and land use/land cover were based on a direct comparison of GIS values to those observed at 12 CPW sites (Fig. 1). The selected CPWs were located in a variety of environmental settings (e.g., urban, natural areas, agricultural). Six of the sites (CR, WD, KS, TH, LC, and SW) were located in wildlife refuges or preserves and were selected with input from regional wetland managers as being representative of CPWs in the region. To ensure adequate coverage for sites outside the 100-year floodplain, another six sites were selected from CPWs outside the 100-year floodplain (Table 2). Four of those (SE, DW, LG, UH) were randomly selected; however, as most of the CPWs outside the floodplain were on private land, it was not feasible to gain access to six randomly selected sites, thus the two final sites (HA and KI) were selected based on access. One of these (HA) was slightly outside of the study area (Fig. 1).

Table 2 Summary characteristics of 12 wetland study sites. Catchment size includes wetland area. Land use categories: WR=wildlife refuge, NA=natural area, GP=grazed pasture. ND=not determined. a. HA catchment size estimated from distribution of other wetlands
Full size table

To evaluate the error of lidar-derived elevations, three study sites with good visibility were selected (KS, SW, LC; Fig. 1) and topographical surveys were performed using a Trimble GeoXT GPS unit, transit, and stadia rod. For each survey, elevation differences between a control point and randomly-selected, individual survey points located within the wetland were compared to the lidar elevation for the same locations. Thus, differences in elevation between lidar and manually surveyed points were partially attributable to differences in horizontal position. Although error associated with manual topographical surveys is unknown, lidar horizontal error was ±0.73 m (Table 1) and the GPS horizontal error was approximately 1 m. To minimize horizontal error, lidar elevations were compared to GPS elevations using both a surface spot check approach and a neighborhood approach. The surface spot check approach involves directly comparing the elevation from the topographic survey to the elevation for the exact DEM location. The neighborhood approach calculates the mean elevation for a nine-pixel (pixel size 1.4 m2) neighborhood and assigns that elevation to the pixel in the center of the neighborhood. A root mean square error was calculated from the differences between manual and lidar-derived elevations.

Results and Discussion

Wetland Class Distribution and Area Coverage

Based on NWI data, there are 10,349 CPWs that cover 51,125 ha within the 560,000 ha study area. CPWs cover 17% of the land area. By comparison, the original 221 million acres of wetlands in the conterminous USA covered 11.6% of the land area, with that coverage reduced to approximately 5% by the 1980s (Wilen and Bates 1995).

Nearly half (46%) of CPWs were classified as emergent, accounting for 83% of the total wetland area (Table 3). Approximately 2,480 wetland polygons (3.7% by area) were classified as permanently or artificially flooded (PFAF, NWI classes H and K) and an additional 820 had hydrologic modifications such as excavation, dikes, or levees. Many PFAF wetlands are used as holding ponds and provide low flood water storage. Removing PFAF wetlands decreased CPW coverage to 45,820 ha, about 15.3% of the land area. The median wetland size of the 7,372 wetland systems was 0.37 ha and 72% of CPW systems were less than 1 ha. Wetlands outside the 100-year floodplain were smaller (median 0.3 ha) than those inside the floodplain (median 0.4 ha).

Table 3 Wetland class distribution and area coverage
Full size table

Wetlands converted to upland since the 2006 NWI update would contribute to overestimation of CPW abundance and coverage reported here. On the other hand, Jacob and Lopez (2005) found NWI boundaries from 1988 NWI data in this region to be conservative with few false positives resulting in underestimation of wetland coverage. One of the sites (SW) was not mapped by NWI because it was an agricultural field recently restored to an emergent wetland. While we did not formally delineate the wetland areas of the 12 sites, the NWI boundaries of those CPWs were reasonably accurate.

Catchment Delineation

The median catchment size of CPW wetland systems was 3.0 ha for lidar-derived catchments and 6.0 ha for wetlands within the 100-year floodplain (Fig. 3). Although data for both groups were log normally distributed (Fig. 3), lidar-based catchments were more evenly distributed around the median. Examples of catchments delineated for CPWs located in developed and undeveloped areas are shown in Fig. 4. The total CPW catchment plus buffer area represents 122,152 ha, or 40.8% of the land portion of the study area. Eliminating PFAF wetlands increased median catchment size from 3.0 ha to 4.0 ha, and median buffer size from 6.0 ha to 6.4 ha, and reduced areal coverage from 40.8% to 36.6% of the study area. Thus, excluding PFAF wetlands, over one-third of the precipitation that falls on land within the study area is captured within CPW drainage basins.

Fig. 3
figure 3

a Histogram of CPW catchment sizes derived from lidar DEMs (wetlands outside the 100-yr floodplain). b Histogram of catchment size derived from 100-m wide buffer strips (wetlands inside the 100-yr floodplain)

Full size image
Fig. 4
figure 4

Lidar-derived catchment boundaries (solid lines) for NWI wetlands (shaded area) in (a). developed area, and (b). undeveloped area. Note that SE was one of the 12 study sites

Full size image

The root mean square error (RMSE) analysis for relative elevation averaged 0.14 m for the surface spot check approach and 0.13 m using the neighborhood approach (Table 4). This is similar to the 0.15 m lidar error reported by Reutebuch et al. (2003) for open canopies, although the error was higher (0.3 m) in forested canopies. The present study did not evaluate error in forested sites, which comprise 10% of the total CPW area.

Table 4 Summary of lidar elevation error analyses. KS vertical accuracy was not calculated because the error was not normally distributed
Full size table

Because the CPW region is relatively flat, the lidar-derived catchments delineated for this study may be considerably larger during periods of moderate to heavy precipitation. During some rainfall events, individual wetlands can discharge into the drainage area of the adjacent, down-gradient wetland. This is consistent with observations of Sipocz (2002) and Wilcox et al. (2011) that CPWs function as headwater systems that fill and drain in a “stair step” fashion. This type of intermittent hydrologic connectivity is likely to be widespread in landscapes with flat topography and emphasizes both the connectivity of CPWs and an additional source of uncertainty in using lidar-derived catchment areas to predict flood storage.

Wetland Volumes

Wetland volumes ranged from zero to 10,354,000 m3. The largest wetlands were in the agricultural and wildlife refuge lands located east of Galveston Bay. However, median volumes were similar in CPWs inside (240 m3) and outside (230 m3) of the 100-yr floodplain, there were more large wetlands (>100 ha) in the floodplain. The median CPW volume was 235 m3 and the total full-pool volume of all CPWs was approximately 47 million m3. Excluding PFAF wetlands, that total reduces to 35.5 million m3.

Volumes could not be obtained for 0.1% of the wetland systems due to voids in lidar data that occurred when a pulse is not returned from the surface. These voids did not appear to be associated with a particular land cover type. Additionally, a volume of zero was calculated for 0.3% of CPWs, and approximately 140 CPWs (1.9%) had volumes less than 1 m3, accounting for 0.06% of the total CPW area. These CPWs were particularly small (median area 620 m2). Zero volumes could be due to a variety of factors, including: 1) wetlands that had been filled since NWI data were created; 2) surface features that were not accurately reflected by lidar; or 3) depressions with maximum depths within the error range (approximately 15 cm) of the lidar data.

Errors in wetland volume can be due to errors in NWI areas as well as elevation; however, since using lidar to estimate wetland volumes is relatively novel, we focused on the latter type of error. Additional error results from excluding interstitial spaces in soil, which could be significant and leads to an underestimation of wetland volume. Error may also derive from our use of terrestrial lidar data that has a spectral resolution of near-infrared (0.75–1.4 μm), which is partially absorbed by water. In contrast, lidar systems for mapping bathymetry include a spectral reference in the green band which indicates the bottom of the water body (Guenther 2007). It is not known exactly how deep terrestrial lidar penetrates water. A recent study by Athern et al. (2010) analyzed terrestrial lidar elevations in two tidal wetland restoration sites with those measured using a singlebeam echosounder survey. They found the mean (± SD) elevations to be lower than the ecosounder results by 5 (±0.1) cm at one site and 2 (± 0.2) cm at the other site. The findings from Athern et al. are encouraging; however, their study sites were predominantly dry when the lidar data were collected. Most natural CPWs are temporarily or seasonally flooded; therefore this type of error has a higher probability of affecting PFAF wetlands. The lidar used for Brazoria, Galveston, and Chambers counties was collected early to mid-April and mid-May 2006 when preceding rainfall was below normal. Harris County lidar data was collected on February 2, 2008. Rainfall in January and February 2008 were approximately 2.5 cm above normal, thus volumes of Harris County CPWs may have been underestimated. Despite the fact that the extent of CPW inundation during the lidar flights is not known, lidar data should be reasonably accurate for shallow, intermittently flooded wetland systems such as CPWs. Volumes of wetlands that were inundated during the lidar collection, particularly those with deep water, would likely be underestimated. Lane and D’Amico (2010) found that mean differences for volumes calculated using lidar and volumes calculated using literature based equations (Brooks and Hayashi 2002; Gamble et al. 2007) ranged between −14% and 31% for around 1,600 karst wetlands in north central Florida.

Water Regimes

According to NWI data, approximately one-third of the total CPW area (n = 4,233) is temporarily flooded and another 23% is seasonally flooded (n = 2,215). Semi-permanently flooded wetlands (18–40 weeks flooded) were also abundant (n = 2,296) but accounted for only 4.7% of total CPW area. Intermittent draw down due to evapotranspiration, a natural feature of CPWs, increases flood storage capacity, and facilitates maintenance of native flora (Jin 2008) and fauna (Boix et al. 2001). In contrast, PFAF wetlands generally have lower floral and faunal diversity (Pechmann et al. 1989).

The study period was characterized by hydrologic variation (drought, wetter than normal months, and inundation by a hurricane storm surge), making a comparison between NWI water regimes and field data more difficult. We found that, of the 11 CPWs with NWI water regimes (SW not included in NWI), three classifications were incorrect. One site classified as farmed was assigned the water regime of an adjacent rice field. In fact, the wetland was flooded less than 18 weeks annually and should have been assigned the other adjacent land use, pasture. Although one would expect this type of error to be unbiased, assigning hydroperiods to farmed wetlands where mixed crop types occur remains problematic. Two additional CPWs did not have NWI hydrologic modifiers, yet one was excavated and the other’s outlet was blocked by a large levee; these errors were a result of NWI misclassifications.

Vegetation Density

According to NDVI estimates derived from NAIP imagery, 73% (by area) of CPWs were 80–100% vegetated. Field vegetation surveys conducted at the 12 sites, when compared to NDVI percent cover (Table 5), yielded an RMSE of 0.46. Most error was associated with disturbance, seasonal variation in cover, and translation of NAIP imagery to NDVI imagery. Disturbances contributed to error at three sites, KI, SE, and SW. Vegetation surveys were conducted at KI and SE (Fig. 1) after Hurricane Ike storm surge decimated the vegetation, while NAIP imagery was obtained prior to Ike. Conversely, SW was fully vegetated during field surveys but NAIP imagery was obtained when the site was fallow. Eliminating these three sites yielded an adjusted RMSE of 0.32. Error associated with seasonal variation was evident at DW, where the field survey was conducted when the wetland was flooded (63.3% open water) and emergent/submerged vegetation was observed. An earlier visit, however, revealed dense cover by Sesbania drummondia and no standing water. It has been recommended that error would be minimized by collecting data only during the growing season (Whigham et al. 1999). However, this strategy may be ineffective in subtropical regions where the growing season is year-round but interannual variability in precipitation is high and disturbances such as drought, flooding, and hurricanes occur within the time scale of NAIP updates.

Table 5 Net error, RMSE, percent of correct observations, and bias of GIS derived variables compared to field measurements at 12 sites
Full size table

The remaining error derived from the conversion of NAIP imagery to NDVI. This appeared to be a greater problem with grass dominated sites such as LG and WD, which were poorly characterized by NDVI analyses. Further NAIP conversion error results from the classification of open water as unvegetated, despite the presence of submersed vegetation. NDVI is a measure of green vegetation, thus vegetation that is brown from die off, prescribed burns, or submerged by water may not be identified as vegetated area. Forested wetlands introduce another complication in estimating cover, as tree canopies were identified in the NDVI, while the wetland ground under the canopy may be bare. The relative cover of overstory, understory, and ground level vegetation provide unique wetland structure; however, these differences were not accurately reflected by the NDVI datasets. Subdividing the study area by vegetation type and evaluating a threshold value for each area may lead to greater accuracy than using just one threshold value for the entire study area.

Soil Clay Content

According to the SSURGO database, clay content of soils in the study area ranged from zero to approximately 70%. The mean SSURGO percentage clay content of CPWs (± SD) was 29 ± 16% (n = 10,349), which corresponds to texture category ranging from loams to clay loams. Soil clay content (texture by feel) at the 12 field sites were nearly always higher than those reported in SSURGO (Table 5). SSURGO data are based on county soil surveys that analyze a small number of soil samples to derive information about a soil type. Furthermore, soil map units are often large and do not take into account local variation in depositional patterns and small wetland features that would be expected to occur in micro-topographical landscapes and drainageways. SSURGO parameters such as organic matter content and soil pH were also examined during this study but had very poor agreement with field data and the direction of the differences was not predictable. The use of SSURGO data to estimate wetland soil characteristics should be done with caution, particularly when the wetlands of interest are small relative to soil map units.

Land Use/Land Cover Data

Land use/land cover data was compared to land uses observed at study sites. Over 32% of CPWs are located in areas classified as natural areas, while the remainder were nearly evenly distributed among the other land use classes. This variable had moderately low, unbiased error when compared to field sites (Table 5).

Conclusions

Our study developed a framework for inventorying several features for an extensive number of small wetland systems. Important features of the 10,000-plus CPWs such as wetland volume and catchment area revealed that CPWs and their catchments cover 35–40% of the land area, confirming the significance of these wetlands to regional ecological processes. The calculation of wetland volumes and catchment areas with high resolution lidar datasets is a novel tool that could be used to increase our knowledge of wetlands in other regions.

A secondary objective was to assess the accuracy of GIS-derived structural features (i.e., model variables) by comparing GIS values to field data from 12 CPWs. The lidar-derived catchment areas and volumes were among the most accurate estimates developed for this project. The SSURGO soils data were biased whereas the use of NAIP imagery to predict percent cover was unreliable in this setting due to frequent disturbance, seasonal dynamics, and a wide range of vegetation type. Despite these problems, useful, screening-level information about often inaccessible, small wetlands can be gained from applying available GIS datasets.

Lidar-derived variables such as catchment size and wetland volume appear to be more reliable than SSURGO or NAIP-derived NDVI vegetation cover. One weakness of this assessment was that 12 sites (three sites for lidar elevation analysis) may be too few to adequately quantify error between GIS datasets and field measured variables. Further study is needed to better evaluate uncertainty in using GIS datasets to describe wetland structure. Lidar is becoming more readily available and its use in ecological studies is increasing. Of particular interest would be studies that deploy an approach similar to the one used by Athern et al. (2010) to examine the penetration depth of terrestrial lidar into inundated wetlands.

In regards to wetland conservation policy, results from this study provide data that, together with measures of wetland function, will facilitate the quantification of ecosystem services provided by CPWs in the study area. Ecological function assessments often require the type of data compiled in this inventory, particularly water regime, vegetation cover, land use, geomorphology, and catchment information. Wetland volumes and catchments calculated in this inventory provide critical information for planners and regulatory agencies to address cumulative wetland loss. Furthermore, CPWs’ extensive coverage and distribution demonstrates that they form an integral component of the regional landscape. Together with their catchments, approximately 40% of regional precipitation is captured and processed by CPWs, providing them a unique opportunity to control the ecological, water quality, and hydrologic processes within the lower Galveston Bay watershed.