Development of Different Machine Learning Ensemble Classifier for Gully Erosion Susceptibility in Gandheswari Watershed of West Bengal, India

Abstract

In various types of geo-environmental problems in the fringing area of Chhotanagpur plateau in India, gully erosion is one of the vulnerable issue. In our current research, using Multi-layer perception approach (MLPC) and its ensembles (MLPC-Bagging, MLPC-Dagging and MLPC-Decorate) models, we have identified potentiality zone of gully erosion in Gandheswari watershed. Considering 20 geo-environmental factors, namely; rainfall, slope, slope aspect, elevation, drainage density, Land use and land cover (LULC), Normalized difference vegetation index (NDVI), geology, geomorphology, soil texture, soil moisture, distance from road, distance from river, plan curvature, profile curvature, topographical wetness index, stream power index, terrain ruggedness index, soil erodibility and distance from lineament, the susceptible areas are indentified. The five susceptible zones are identified with the help of the MLPC computational approaches and different ensemble classifier. All the models are predicted of fitted with good manner but the MLPC-Decorate is comparatively better than other models. The Area under curve (AUC) values of Receiver operating characteristic (ROC) curve for the MLPC-Decorate model in the training and validation database are 0.924 and 0.906 respectively. This model can be use in any type of environmental modelling in sub-tropical region.