Landslide Prediction via Multispectral Satellite Imagery: A Multi-Band Convolutional Neural Network Approach

Sliding of land of a particular location due to heavy rain, vegetation change, or deforestation is a landslide. The landslides lead to loss of life, damage to infrastructure, and disruption of economic conditions. With the advancement of remote sensing technology, multispectral satellite images are available with various bands. These bands are highly useful for predicting landslides in advance. Along with these bands from satellite images, environmental features, such as rainfall, temperature, soil moisture, and topographical features, such as elevation, slope, aspect, and curvature, provide additional information for prediction. With these layers of information, satellite images of landslide and non-landslide locations in India were trained and tested using a Convolutional Neural Network (CNN) model. The CNN model was built by choosing 3 bands, 5 bands, 10 bands, and 16 bands of satellite images, resulting in prediction accuracies of 67.97%, 73.51%, 78.03%, and 79.47%, respectively. The model performance was assessed using metrics such as accuracy, precision, recall, F1-score, and confusion matrices. The results suggest that selecting a higher number of features leads to higher accuracy. The need for such models in countries like India, where landslides frequently occur, is highlighted. This study demonstrates that integrating topographical and environmental data significantly improves the accuracy of the prediction model.