Application of neural network model-based framework approach to identify gully erosion potential hotspot zones in sub-tropical environment
Asish Saha, Subodh Chandra Pal, Indrajit Chowdhuri, Abu Reza Md. Towfiqul Islam, Rabin Chakrabortty, Paramita Roy
Abstract
Gully erosion is a common type of soil erosion form that results in significant soil loss in a variety of climatic environment. The amount of sediment produced by gully erosion is several times higher in comparison to other forms of erosion. Therefore, an effort has been made to improve gully susceptibility assessment in sub-tropical environment of India, using neural network algorithms. Based on satellite image data, a gully inventory map was created and twenty gully conditioning variables were considered for modelling perspective. Study revealed causative factors like slope, land use and drainage density are most significant for gully occurrences. The result of this study revealed the ‘multi-layer perceptron (MLP)’ algorithm is the most robustness model with the AUC (area under curve) is 0.95 compared to the remaining applied models. The findings may provide an outline for more biophysical planning of gullies and associated planning strategies throughout the study area.