Neuro-fuzzy classification of the landscape for soil mapping in the central plains of Venezuela

Abstract

The application of geomorphology to soil survey has encouraged the study of genetic relationships between soil and geoforms. However, the qualitative classifi cation of the landscape can be slow and expensive, and the outcome often depends on the perception of the classifi er. This work applied a quantitative method based on artifi cial neural network and fuzzy logic to classify the landscape into land-surface units from a digital elevation model (DEM) of 5 × 5 m cells. The method helped explore the data to determine the optimal combination of number and fuzziness of classes. The classifi cation output included the values of the geomorphometric parameters at the centre of each class, the memberships of the model cells to each class, and a map showing the spatial distribution of the land-surface classes. This output was transformed into a map of geoforms that was used as a framework for soil sampling and mapping. The resulting map disclosed the landscape structure consisting of a plateau dissected into mesas, hilltops, slopes, and valleys, with predominance of well-drained Alfi sols in steep lands and imperfectly drained Vertisols in valleys. The method proved to be effective for establishing soillandscape relationships in the study area.