Development and application of two hybrid metaheuristic algorithms to identify the most important parameters influencing wind erosion

Wind erosion is an important cause of land degradation and desertification in arid and semi-arid regions of the world. This phenomenon occurs more severely in dry and bare soils. During wind erosion, soil particles are transported by three mechanisms known as creeping, saltation and suspension, which result in numerous on- and off-site damages. Wind erosion as a natural phenomenon, is affected by many factors. Our understanding of wind erosion is commonly constrained by the multiplicity and complexity of factors in this process. In recent years, solving pattern recognition and optimization problems with metaheuristic algorithms has received considerable attention among researchers. Genetic algorithms is a search technique inspired by the process of natural selection, which is well applied to multimodal, non-linear, and non-derivable objective functions. In the current research, for identifying the most important parameters affecting wind erosion rate, two GA-ANN and NSGA-II hybrid algorithms were developed using genetic algorithm and artificial neural networks. In order to prepare a suitable and reliable data set; after designing a grid sampling strategy, soil samples were collected from 51 study sites in the Narmashir plain, Kerman and then some soil parameters were measured. In addition, wind erosion rate was determined at each study site using a portable wind tunnel device. Based on the GA-ANN algorithm results, gravel coverage, sand, clay, aggregate stability, surface crust, moisture, and organic matter were identified as the main determinant features affecting spatial variation of wind erosion rate. However, the selected feature subset by NSGA-II algorithm included gravel coverage, sand, aggregate stability, surface crust, and moisture. The calculated error function for the GA-ANN algorithm performance was 3.58%. It was 1.70% for the NSGA-II algorithm performance. According to the results, both algorithms had acceptable performance to achieve the purpose of the present study. Therefore, the algorithms developed in this study can be applied to identify the most important parameters affecting wind erosion rate in other areas with similar challenges.