Large‐scale assessment of soil erosion using a neuro‐fuzzy model combined with GIS: A case study of Hubei Province, China

Abstract

AbstractReliable and cost‐effective soil erosion assessment is an important precondition for soil conservation measures, which remains a major challenge at large scale. Considering that the neuro‐fuzzy model has the special advantage in multi‐index comprehensive assessment and GIS technology is adept at geo‐spatial information processing, through the combination of them, it is possible to provide an effective approach for this difficult problem. Taking Hubei Province as a case study area, five evaluating indicators were selected for the large‐scale assessment, in which the GIS technology was used to construct the classification maps of evaluating indicators and to divide basic assessment units, and the neuro‐fuzzy model was adopted to extract fuzzy rules for individual units assessment from available ground truth data. According to the optimized assessment criteria generated by the neuro‐fuzzy model, the soil erosion state of the entire study area was then assessed. To represent the spatial distribution of soil erosion, a detailed map was produced by statistical mapping, which was represented with six erosion levels (from slight to severe) at a map scale of 1:250 000. The resulting map showed that about 30.1% of the total land area in Hubei was affected by different levels of soil erosion problem. Western high mountains and eastern low mountains suffered from the most serious erosion damage, a strong level of soil erosion was widely observed in these mountains. Large areas of moderate level erosion occurred in the northern hills. In contrast, most of the central plains were characterized as slight level erosion effect. The validation indicated that an overall accuracy of 88% and a κ of 0.89 were achieved, proving that the resulting map was in conformity with actual conditions, which indicates this assessment approach was reasonable and applicable. Copyright © 2009 John Wiley & Sons, Ltd.