Inversion of herbage aboveground biomass in a multi-feature group mining area based on UAV remote sensing

The aboveground biomass of vegetation can be used as an important indicator to evaluate the ecological function of mined land restoration. To achieve a rapid and accurate prediction of the aboveground vegetation biomass of restored land, the aboveground biomass of herbaceous plants in Tongguanshan mining area, Tongling City, Anhui Province, China, was selected. Unmanned aerial vehicle (UAV) high-resolution multispectral images were used to extract single band spectral reflectance, vegetation index, and texture feature variables of each band. High precision digital elevation model (DEM) was used to generate terrain features, and then grey correlation and entropy weight methods were used to screen the spectral and texture feature variables, respectively. The filtered and terrain feature variables were divided into spectral and multiple feature groups as the input variables of the model. Finally, back propagation neural network (BPNN), convolutional neural network (CNN), and Elman neural network were used to build biomass prediction models based on spectral and multiple feature groups, as well as compare and select herbaceous aboveground biomass inversion models with higher accuracy in mining areas. The results show that the three inversion models had improved accuracy to a certain extent after introducing texture and terrain features based on spectral features. Among them, the BPNN model based on multiple feature groups has the highest accuracy, with a determination coefficient (R²) and root mean square error (RMSE) of 0.841 and 11.813g·m⁻², respectively. Furthermore, cross-verification of the three models showed that the BPNN model based on multiple feature groups is more stable and has optimal inversion accuracy. Then, the optimal inversion model was used to evaluate the vegetation biomass in the study area, the classification results of vegetation biomass show that the biomass of the study area is generally low, concentrated at 20～40 g·m⁻². These results can provide theoretical support for the inversion of herbaceous biomass in mining areas.