Application of machine learning algorithms for estimation of aboveground biomass of <italic>Tamarix </italic>nebkha

Tamarix chinensis, a dominant shrub species in arid ecosystems, is characterized by high stress resistance and adaptive growth strategies. It is crucial for stabilizing sand dunes, reducing wind erosion, and facilitating ecological restoration. Consequently, the aboveground biomass (AGB) of Tamarix nebkha is an essential indicator of vegetation status and desertification control in arid regions. The research object of this study was Tamarix nebkha in the lower reaches of the Tarim River. In total, 92 spectral, vegetation, and texture indices were extracted as feature variables from Landsat 8 images. For subsequent variable selection, stepwise regression, least absolute shrinkage and selection operator (LASSO), and extreme gradient boosting (XGBoost) algorithms were employed. Random forest (RF), support vector regression (SVR), and backpropagation neural network (BPNN) models were then constructed to estimate the AGB of Tamarix nebkha. The adaptability and estimation accuracy of different feature selection methods and models were compared, and the applicability of multivariate algorithms for the estimation of Tamarix nebkha AGB were explored. The results indicated that: (1) The variable sets selected by the stepwise regression, LASSO, and XGBoost algorithms were significantly correlated with the AGB of Tamarix nebkha, and the low multicollinearity among variables (VIFR2=0.73, RMSE=447.63 g·m-2). The incorporation of multivariate combinations substantially enhanced the predictive capability of the models. (3) The LASSO-based RF model estimated the mean AGB of Tamarix nebkha in the lower Tarim River region to be 1733.63 g·m-2 and the total biomass to be approximately 1.71×108 kg. The developed model exhibited high reliability and strong applicability across the regional scale. The findings of this study provide a scientific basis for optimizing the selection of remote sensing inversion methods and enhancing the accuracy of AGB estimation in desert shrub ecosystems.