Spatial-Prediction-of-Landslide-Susceptibility-in-Kodagu-Using-Random-Forest-Models

Spatial Prediction of Landslide Susceptibility in Kodagu Using Random Forest Models: Insights from the 2018 Landslide Inventory Landslides are natural disasters distributed worldwide, often resulting in socio-economic losses and loss of life, especially in Kodagu, the South Western Ghats of India. Identifying the zones susceptible to landslides in Kodagu essential for safe living in the region and reducing the losses due to landslides. The six influencing factors namely elevation, slope, aspect, stream power index (SPI), curvature and topographic wetness index (TWI) were considered in the study. The article presents a machine learning model for accurately mapping landslide susceptibility by utilizing a newly generated landslide inventory map of Kodagu District, based on the 2018 landslide event, to assess the effectiveness of Random Forest (RF) techniques in evaluating landslide vulnerability. Optimization technique namely grid and random search was employed to enhance the effectiveness of RF models in predicting landslide-prone areas, ensuring more accurate and reliable landslide susceptibility map. The result analysis shows that the model optimized using grid search demonstrate nearly the same predictive performance but require substantially greater computational time than models optimized using random search. The RF classifier identified 2,502 km² of low and 1,579 km² of high susceptibility areas. The RF regressor identified 876 km² of very high, 595 km² of high, and 574 km² of moderate susceptibility areas. The slope, elevation, curvature, and SPI were major factors influencing landslide occurrences. The RF classifier focusing on high-likelihood landslide areas, while the RF regressor provides a continuous likelihood of susceptibility, capturing a broader range of landslide-prone areas. The regressor's flexibility allows it to detect gradual environmental changes and include borderline regions often excluded by the classifier. It also revealed that the accuracy obtained for the random forest regressor (0.95) outperformed the random forest classifier (0.85). Therefore, RF regression model for LSM is more reliable as the model has more discrimination power compare to RF classifier model.