Deep learning-based dynamic monitoring technology for soil moisture in upland turf stripping

Abstract

In this paper, the inversion of soil moisture in highland turf stripping was carried out by constructing an artificial neural network model, improving the vegetation-adjusted vertical drought index (VAPDI), and processing CYGNSS data. The accuracy of the inversion model was verified by linearly fitting the VAPDI calculated from remote sensing images in the observation area and the measured soil moisture data in the same period in SPSS. More accurate reflectance was obtained by surface roughness attenuation correction. After integrating the topographic feature quantity, the accuracy of the model inversion situation was analyzed by comparing the index situation of the artificial neural network model with other models. The linear regression model for a total of three periods of data from 2020-2022 in the observation area passed the significance test of 0.05. The interpolated Bias, RMSE, and R values of the artificial neural network model were better than those of the other comparative models, and the training time was 30% and 15% less than that of the other models. The model inversion results were consistent with the actual soil moisture situation with high accuracy.