Application of a Random Forest Method to Estimate the Water Use Efficiency on the Qinghai Tibetan Plateau During the 1982–2018 Growing Season

Water use efficiency (WUE) reflects the quantitative relationship between vegetation gross primary productivity (GPP) and surface evapotranspiration (ET), serving as a crucial indicator for assessing the coupling of carbon and water cycles in ecosystems. As a sensitive region to climate change, the Qinghai Tibetan Plateau’s WUE dynamics are of significant scientific interest for understanding carbon water interactions and forecasting future climate trends. However, due to the scarcity of observational data and the unique environmental conditions of the plateau, existing studies show substantial errors in GPP simulation accuracy and considerable discrepancies in ET outputs from different models, leading to uncertainties in current WUE estimates. This study addresses these gaps by first employing a machine learning approach (random forest) to integrate observed GPP flux data with multi-source environmental information, developing a predictive model capable of accurately simulating GPP in the Qinghai Tibetan Plateau (QTP). The accuracy of the random forest simulation results, RF_GPP (R2 = 0.611, RMSE = 69.162 gC·m−2·month−1), is higher than that of the multiple linear regression model, regGPP (R2 = 0.429, RMSE = 86.578 gC·m−2·month−1), and significantly better than the accuracy of the GLASS product, GLASS_GPP (R2 = 0.360, RMSE = 91.764 gC·m−2·month−1). Subsequently, based on observed ET flux data, we quantitatively evaluate ET products from various models and construct a multiple regression model that integrates these products. The accuracy of REG_ET, obtained by integrating five ET products using a multiple linear regression model (R2 = 0.601, RMSE = 21.04 mm·month−1), is higher than that of the product derived through mean processing, MEAN_ET (R2 = 0.591, RMSE = 25.641 mm·month−1). Finally, using the optimized GPP and ET data, we calculate the WUE during the growing season from 1982 to 2018 and analyze its spatiotemporal evolution. In this study, GPP and ET were optimized based on flux observation data, thereby enhancing the estimation accuracy of WUE. On this basis, the interannual variation of WUE was analyzed, providing a data foundation for studying carbon water coupling in QTP ecosystems and supporting the formulation of policies for ecological construction and water resource management in the future.