Comparative research on multi-algorithm of soil salinity monitoring based on Gaofen-5, Sentinel-1, and Sentinel-2

Abstract

ABSTRACT Soil ecosystem provides a guarantee for the ecological environment. Soil salinization is a typical land degradation problem. In order to explore the applicability of GF-5, Sentinel-1, and Sentinel-2 remote sensing data in soil salinity research, this study selected Golmud of Qaidam Basin in China as the study area, using particle swarm optimization-artificial neural network (PSO-ANN), partial least squares regression (PLSR), and whale optimization algorithm-extreme learning machine (WOA-ELM) to predict soil salinity. The research results indicated that the correlation between Sentinel-2 and electric conductivity (EC) was higher than GF-5 and Sentinel-1. The simulation accuracy of models based on Sentinel-2 was slightly higher than that based on GF-5 and Sentinel-1. The three models’ mean values of root mean square error (RMSE), R2, and mean absolute error (MAE) of Sentinel-2 were slightly better than those of GF-5 and Sentinel-1, and the fitting accuracy of GF-5 was similar to that of Sentinel-1 on the whole. The result proved that Sentinel-1, GF-5, and Sentinel-2 were suitable for salinity monitoring because the strong penetrability and sensitivity of Sentinel-1, the high spectral resolution of GF-5, and the high spatial resolution of Sentinel-2. The performance of PSO-ANN was the best, and that of PLSR was the worst. The salinity mapping result of Sentinel-2 coupled with PSO-ANN proved that machine learning methods and spaceborne data could be used to study large-scale soil salinization.