Effect of Spatial Data Resolution on Lumped Hydrology Models

Introduction

The growing use of GIS in hydrologic modeling and the availability of spatial digital data in recent years has allowed users to perform hydrologic simulations efficiently and in a timely manner by making spatial overlays of raster layers and performing needed computations on corresponding tabular data. The data sources commonly used in most hydrologic models are soils, land cover, and topographic data. Their digital forms are available at different scales and cell resolutions. While most research in the area has concentrated on developing digital data layers and building GIS/model interfaces, little has been done to evaluate how data layers of different scales and cell resolutions affect model input parameters and simulation outputs. This study was prima rily initiated to:

Watershed Study Areas

The study areas consist of three nested drainage basins of different sizes located within the SRB in Pennsylvania (Figure 22). The Mahantango Creek watershed covers approximately 420 km 2 and is located in the non-glaciated section of North Appalachian Ridge and Valley Region of eastern Pennsylvania. The annual precipitation ranges from 104 to 124 cm. The general geol ogy of the watershed consists of, from northwest to southeast, folded Pennsylvanian sandstone and shale, Mississippian sandstone and shale, and Devonian sandstone, siltstone, and shale. GK-27 is located in the northeastern part of the Mahantango Creek watershed.

The WE-38 watershed is the intensive hydrologic study area for the Northeast Watershed Research Center, one of the six Regional Watershed Research Centers in the USDA Agricultural Re search Service's (ARS) watershed research program. Rainfall and streamflow data have been collected in the watershed since 1967 from three rain gages and one streamflow gage.

Data

Elevation data for the WE-38 watershed were 5-m, 7.5-minute, 3-arcsecond, and 30-arcsecond digital elevation models. The 5-m DEMs were derived from scanned aerial stereophotography ac quired in April 1994. The remaining DEMs represent roughly 30 m, 90 m, and 1 km ground resolution. The 30-m and 90-m DEMs were resampled to 25 m and 100 m in order to allow proper registration of DEMs with other data layers. Soil property data for the three watersheds were derived from the SSURGO and STATSGO soil data bases. SSURGO soil maps of the Mahantango Creek watershed were digitized from orthophotographs using county soil survey reports by the SCS personnel at the Land Analysis Laboratory, Department of Agronomy, The Pennsylvania State University.

Land cover classes were derived from low-altitude infrared aerial photographs, Landsat TM and AVHRR imagery. The infrared photographs were acquired as part of the MACHYDRO-90 overflights. They were taken from a DC-8 aircraft on July 17 and 18, 1990 at an average scale of 1: 13,500. Map-oriented and system-corrected cloud-free Landsat TM imagery taken on August 12, 1990 was obtained from EOSAT (Earth Observation Satellite) Company, Lanham, Maryland. The AVHRR imagery taken on July 18, 1990 was obtained from NOAA. Both TM and AVHRR imagery were geometrically rectified and registered to the UTM (Universal Transverse Mercator) projection.

Methodology

The development of input soil data was performed using the Arc and Grid modules of the Arc/Info software. The HSGs for soil series in SSURGO were derived from records in soil survey reports from Northumberland, Dauphin, and Schuykill counties. HSGs in STATSGO were determined by assigning the HSG of the dominant soil component within a map unit to the whole map unit using the Arc Macro Language (AML) in Arc/Info. SSURGO and STATSGO HSG vector layers were gridded to 25 m and 200 m, respectively.

Land cover classes from aerial photographs were delineated, digitized using 7.5 minute-USGS topo graphic quadrangles as base maps, and then gridded to 10 m. Landsat TM and AVHRR imag ery was processed using ERDAS Imagine software, Version 8.1. Classes were determined using supervised and maximum likelihood classifiers. Training sites for this classification were defined with reference to visual features identified on infrared aerial photographs according to USGS Level I classes. The layer of land cover classes derived from Landsat TM was resampled to 25 m to allow proper overlays of other data layers.

CNs for each of the study areas were generated by overlying soil layers of HSGs to layers of land cover classes . The average weighted CN for each watershed was then computed for different data resolutions. The hydrologic modeling tools of the Arc/Info Grid module were used to generate stream network, and watershed and subarea boundaries for the WE-38 watershed, slope layers from the different DEM resolutions for upland and channel segments for each subarea. The rainfall recorded on 18 June 1990 was used in all simulations.

Results

The SCS Runoff Model is one of the most widely used watershed hydrologic models. The model uses the CN, a coefficient that reflects soil and surface cover conditions of the watershed. A sum mary of curve number results is presented in Table 3. SSURGO- and STATSGO-derived CNs were slightly different for all three land cover sources and watersheds as a result of differences in HSG distributions. Average CNs determined using SSURGO were lower by 1 to 3 units than those from STATSGO for the Mahantango Creek and GK-27 watersheds because HSG B cov ered relatively large areas in these watersheds as compared to HSG D. On the contrary, SSURGO in the WE-38 watershed provided higher CNs than STATSGO as a result of a high percent area covered by HSG D in this watershed.

The three land cover sources yielded similar CNs for the Mahantango Creek and GK-27 watersheds. CNs derived from aerial photographs were lower than Landsat TM and AVHRR by 2 and 6 units and this difference resulted from differences in percent areas covered by the major land cover categories. All three land cover sources provided higher CNs for the WE-38 watershed than for the 2 remaining watersheds because of higher percent areas covered by the agricultural land category in this watershed. HSGs affected the CN more than land cover classes for all three watersheds and at all cell sizes.

Results of model simulations for soil and land cover sources are presented in Table 4. On average, a CN decrease from SSURGO to STATSGO of 1 unit caused a difference of 0.28 cm in runoff depth (or 12% decrease) for the WE-38 watershed. Little difference in peak discharges was observed between SSURGO and STATSGO because topographic parameters were the same in both cases. Runoff depth increased by 0.30 cm (15%) and 0.84 cm (37%) for a 1 CN-unit increase from aerial photographs to Landsat TM, and for a 3 CN-unit increase from Landsat TM to AVHRR, respectively for the WE-38 watershed. Peak discharge varied in the same manner by 17% and 38%, respectively, as a result of differences in times of concentration between the three land cover sources.

Runoff depth and peak discharge summaries for different DEM cell resolutions are presented in Table 5. These data were derived using CN values determined from SSURGO and aerial photo graphs. The runoff depth varied little between 5 m and 25 m DEM cell resolutions. It increased slightly as the cell resolution was increased from 25 m to 100 m. The peak discharge tended to decrease with increasing DEM cell resolution due to the decrease in slope and channel length that increased the times of concentration. The increase in peak discharge when the resolution was increased from 100 m to 1 km could have been caused by the increase in watershed area. Therefore, peak discharge variations with cell resolution were caused by changes in channel slope and length, and probably watershed area.

The differences in runoff depth and peak discharge found between SSURGO and STATSGO, aerial photographs and Landsat TM, and between the 5-m DEM, 25-m DEM and 100-m DEM cell resolutions would not be critical in humid temperate climates. Major changes in these param eters, however, are expected if AVHRR or 1 km DEMs are used in simulations for small drain age basins similar to WE-38 watershed. Results of this investigation are specific to the water shed study areas and are subject to error associated with lumped models, particularly the SCS Runoff Model. Therefore, different results would be expected if watersheds of different distri butions in soils, land uses and elevation data were used. However, this study indicates the magnitude of changes in model input parameters and simulation outputs to be expected when different data sets and resolutions are used in hydrologic modeling (Nizeyimana, 1995).

Table of Contents Previous Section Next Section Download Report
Graphics and HTML text markup by Phil Kolb
Comments to: R. White, raw@essc.psu.edu