Remote-sensing data as an alternative input for the ‘STREAM’ runoff model

Abstract

Water erosion of cropland constitutes an issue for natural environments along runoff flowpaths due to property damage by soil-laden water and the associated transfer of nutrients and pesticides. In the Pays de Caux region of northwestern France, the silty soils with crusting properties induce a high risk of runoff and erosion. Changes in agricultural practices, land use and landscape patterns appear to have increased the occurrence of erosion and mud flows over the past few decades. A runoff and erosion model called STREAM, applicable to single rainfall events at catchment scale, has been developed to simulate the impacts of land-use modifications. The model takes into account processes that degrade surface states when calculating infiltration rates, as well as agricultural aspects when computing the runoff circulation network. STREAM is based on an expert-system approach that focuses on the dominant processes whilst having only a few input parameters: three of these are used to determine the runoff circulation network, and the other four to calculate infiltration rates. Input nevertheless requires field observations, which restricts application of the model to small catchments. Satellite data covering large areas is considered as an alternative input for such a model, the main objectives being to adapt STREAM accordingly, and to compare the obtained results with field data. In view of previous work involving the extraction and validation of roughness indices using RADARSAT data, this study is based on RADARSAT and LANDSAT TM data collected during the winter of 1998. After adaptation to receive remote-sensing data, the resulting STREAM-TED model requires less input, namely (1) slope and orientation, (2) land-use classification from optical remote-sensing data, (3) roughness indices from radar remote-sensing data, and (4) previous rainfall. Runoff volumes at a gauged catchment outlet (Bourville in Upper Normandy, France) are simulated by four successive versions of the model ranging from the original STREAM to the adapted STREAM-TED. Predictions of the four versions are compared, and performance of the successive simulations is assessed in relation to measured values and according to five statistical indices. Predictions of runoff volume at the catchment outlet using STREAM-TED are similar to those using the original STREAM model, but with a tendency towards overestimation. The final STREAM-TED version is capable of identifying areas sensitive to runoff within a group of catchments and could be used as a planning decision tool in the implementation of conservation practices.