Estimation of Wind Erosion Threshold Velocity Based on Spectroscopy Data Using Random Forest Algorithm

Abstract

Threshold Velocity (TV) of soil is considered a great indicator in order to assess Potential wind erosion (PWE). However, TV is difficult to measure and some techniques such as wind tunnels can be quite time-consuming and hard. To deal with this challenge, spectroscopy could be considered as an advantageous method to estimate TV. In the current research, the potential of Vis-NIR spectroscopy in TV estimation with the help of machine learning algorithm namely Random Forest (RF) was assessed. for this reason, in the Fars Province, Iran, 100 in-situ wind tunnel tests were executed, and soil samples spectral reflectance were examined with the help of spectroscopy apparatus. Results Showed that outputs of TV estimation with the aid of RF model were (R2 = 0.74, RMSE = 0.65 m s-1, RPD = 1.78, and RPIQ = 2.83 m s-1). This study has shown the utilization of the reflectance spectroscopy with the assist of machine learning algorithm is a reassuring method for worldwide evaluation of wind erosion phenomena.