Applications of fuzzy logic to the prediction of soil erosion in a large watershed

There is a need in many emerging nations to develop simple methods for predicting areas of extensive soil erosion using imprecise, but real-world, input data at low cost with considerable Ž . accuracy. The objectives of this study are to 1 develop fuzzy logic models that predict soil Ž . erosion in a relatively large watershed using a limited number of input variables, 2 determine the Ž . effects of scale and grid-size variations in input data on fuzzy logic model output, and 3 compare the predictions of soil erosion using fuzzy logic methodologies with those of the Universal Soil Ž . Ž . Loss Equation USLE . Two fuzzy-logic rule bases were constructed: 1 a two-variable-based Ž . model which required inputs of slope angle and landuse ratio or landuse, and 2 a three-variableŽ . Ž . based model which required inputs of slope length FLS , soil erodibility FK and vegetative Ž . cover FC . Reducing the grid size of input data resulted in a decrease in the areal extent of predicted high soil erosion. The more dispersed pattern of soil erosion, observed with a 4-mile grid at a scale of 1:250,000, became more clustered in the 2-mile grid and even more clustered in a 2-mile grid at a scale of 1:24,000. The trend of clustering towards smaller areas of high erosion potential as map scale increased was also found at the grid sizes 1 and 0.4 miles. Soil erosion, predicted with the two-variable fuzzy logic model using 30 m resolution GIS-based data sets of slope and landuse, was similarly distributed among the low, moderately low and moderate soil erosion categories. This was attributed to the lack of inherent fuzziness in the input data obtained from raster format. The areal extent and locations of soil erosion predicted by the USLE model and by the two-variable-fuzzy logic model were similar. Differences between the predictions of the USLE model and the three-variable fuzzy logic model were primarily in the moderately low and moderate soil erosion categories and were attributed to the effects of the two non-fuzzy variables used in the USLE model, annual rainfall erosivity and cropping practices, which were not used in the fuzzy logic model. Compared with the USLE model predictions, the fuzzy ) Corresponding author. Fax: q1-501-575-7465; E-mail: [email protected] 0016-7061r98r$ see front matter q 1998 Elsevier Science B.V. All rights reserved. Ž . PII: S0016-7061 98 00050-0 ( ) B. Mitra et al.rGeoderma 86 1998 183–209 184 logic-soil erosion prediction models were successful at locating and differentiating areas of soil erosion with minimum input data. q 1998 Elsevier Science B.V. All rights reserved.