Contribution and behavioral assessment of physical and anthropogenic factors for soil erosion using integrated deep learning and game theory

Ensuring sustainable management of soil erosion is of utmost importance to prevent its adverse effects. Unfortunately, this issue has received limited attention in the past, therefore, a comprehensive study was undertaken to identify key indicators for soil erosion and develop an Ensemble Deep Neural Network (EDNN) model for predicting soil erosion probability (SEP) zones in the Guwahati urban watershed. To conduct this study, Revised Universal Soil Loss Equation (RUSLE) was employed to estimate potential soil loss (PSL). The study identified 17 physical and landscape parameters as priority factors for soil erosion initiation based on the PSL. These priority variables were analyzed using three hyper-tuned models i.e., Gradient Boosting Machine (GBM), Random Forest (RF), and Deep Neural Network (DNN) within the H2O framework. Furthermore, the study developed EDNN and hyper-tuned DNN models to predict SEP and analyzed the behavior of priority variables in high and low soil erosion probability zones using DNN-based game theory. The findings indicated that the study area experiences soil loss ranging from 140 to 181 tonnes/ha/year. Rainfall, Ff, Dd, slope, and elevation identified as the most critical factors influencing soil erosion. The EDNN model successfully predicted very high and high SEP zones, covering 172.13 km2 and 22.60 km2 respectively. Further, the result shows that high values of drainage density, cohesion, TWI, LULC, and rainfall contribute to significant soil erosion, while low values of slope, SHDI, ED, and Ff act as preventive factors. The study offers valuable insights into the construction of an EDNN model and the integration of deep learning with game theory for robust and scientific soil erosion modeling. The analysis of behavior of the priority parameters provides a novel approach for understanding the impact of individual parameter samples on soil erosion, facilitating the development of effective prevention strategies.