Earthquake Potential Zones Identification using MCDA, Machine Learning and Geospatial Techniques in Sikkim

Sikkim, nestled in the North-Eastern Himalayas, is a region frequently disturbed by earthquakes and are triggered by a myriad of factors, including tectonic plate movements, volcanic activities, subterranean explosions, human-induced quakes, and environmental conditions. This leads to loss of human lives, destruction of human properties and deformation of earth crust plates. The study aims to refine earthquake vulnerability assessment by incorporating Machine Learning Weighted Overlay methods over traditional Analytical Hierarchy Process (AHP) techniques. Unlike AHP, where users assign weightage based on their knowledge, ML models use feature importance for objective weightage assignment. A comprehensive set of parameters—fault characteristics, peak ground acceleration, earthquake magnitude, proximity, slope, and elevation which previous studies were somewhere lacking in their study. These parameters are weighted and superimposed using the AHP method. We used Google Earth Engine (GEE) to facilitates the extraction of weighted stacked images, which are then analyzed to pinpoint earthquake-prone locations via machine learning models, namely Random Forest and SmileCart classification. The earthquake vulnerability zones are stratified into five distinct categories: Very High (4.54%), High (28.37%), Medium (27.84%), Low (23.60%), and Very Low (15.65%). The Random Forest and SmileCart models outperformed the AHP method, yielding accuracies of 0.89 and 0.78, respectively, compared to 0.57 for the AHP approach by the validation points and ROC Curve/AUC values corroborate these findings, with respective scores of 0.71, 0.75, and 0.60. The integration of advanced machine learning algorithms over conventional AHP methods significantly enhances the precision of seismic susceptibility estimations. This synergy illustrates the potential of modern analytical techniques in the realm of natural disaster risk evaluation.