Assessment of digital land cover maps for hydrological modeling of the Yampa River Basin, Colorado, USA
Date
2005
Authors
Repass, Julie Mae, author
Fassnacht, Steven R., advisor
Journal Title
Journal ISSN
Volume Title
Abstract
In order to produce satisfactory results from hydrologic models, it is imperative to use good input data. Today there is a multitude of different land cover maps available, and determining which input data map for the model can be unclear. The goal of this study was to quantify the differences between several readily available land cover maps to determine their relative suitability for hydrological modeling of the Yampa River Basin, Colorado. The land cover maps compared in this study are derived from Advanced Very High Resolution Radiometer (AVHRR), Landsat Thematic Mapper (TM), and Moderate Resolution Imaging Spectroradiometer (MODIS) imagery. These maps were compared to a 30-m land cover map modeled from ground data, Landsat imagery, and MODIS imagery, all collected in 2004. This map was regarded as "truth" in this study due to its fine resolution and use of recent ground data and imagery, and was used to rank the public domain land cover data sets. In order to compare the different land cover data sets, all data were first degraded to a common spatial resolution (~30-m) and a common species resolution. Once this was accomplished, the maps were assessed on four levels. The four assessments were based on: (i) the relative agreement of the total aggregated land class percentages after the data had been cross-walked with respect to the reference map; (ii) pixel accuracy; (iii) scene accuracy; and (iv) cumulative streamflow model output from the United States Geological Survey (USGS) Precipitation-Runoff Modeling System (PRMS) in relation to observed cumulative streamflow. The results showed that the pixel and scene accuracies did not correlate with model performance within the Yampa River Basin using the PRMS model. The qualitative comparison of the total aggregated land class percentages helped explain the general trends in the simulation results. It was found that maps with the correct proportion of forested and non-forested areas generally had simulated cumulative streamflow that matched closest to observed cumulative streamflow. Overall, the MODIS-derived land cover maps performed the best in terms of hydrological modeling using PRMS in the Yampa River Basin. However, the model was not found to be particularly sensitive to accurate land cover conditions. As a result, the scene and pixel accuracy results would not necessarily correlate with the model results.