Challenges of rainfall erosivity prediction: A Novel GIS-Based Optimization algorithm to reduce uncertainty in large country modeling

The rainfall erosivity factor (R-factor) is the critical factor for calculating the rates of soil erosion implemented by many soil erosion modeling methods, such as the Revised Universal Soil Loss Equation (RUSLE). Therefore, choosing an appropriate technique to interpolate the amount of rainfall erosivity is critical to reducing the uncertainty of soil erosion rate. This article aims to introduce a new method to model the R-factor using artificial intelligence (AI). The proposed model consists of three algorithms, including Particle Swarm Optimization (PSO), Polynomial, and Multiquadric algorithms (PPSOMQ). Eventually, the amount of R-factor calculated by the models above was compared with the results of widely used interpolation techniques, including Simple Kriging (SK), Ordinary Kriging (OK), Universal Kriging (UK), Global Polynomial (GP), Local Polynomial (LP), Radial Basis Function (RBF), and Inverse Distance Weighting (IDW), using error evaluation indexes (RMSE، MAD، MAPE، NSE, and R). The results indicated that the uncertainty in the R-factor estimation was significantly less than in other conventional models, considering a maximum error of 15% in the proposed model in contrast with errors exceeding even more than 100% in some other models. Also, the results showed that the values of MAD, NSE, RMSE, MAPE, and R indices for the test data in the proposed model are 2.36, 0.99, 3.3, 1.7, and 1, respectively, and the relative error in estimating the rainfall erosivity factor in the proposed model did not exceed 2% in any of the watershed subareas. Moreover, the proposed model showed higher flexibility than the other common models since it presented the change process more smoothly and yielded more accurate estimations. Overall, it can be suggested that using AI algorithms, such as PSO, can increase the accuracy of environmental modeling by reducing uncertainty.