This study applies the Multilayer Perceptron (MLP) and Random Forest (RF) models, utilizing remote sensing and ground-based net primary productivity (NPP) data from 1992 to 2020, along with meteorological data and soil properties, to model the NPP in the alpine grassland and alpine meadow ecosystems of the Qinghai-Tibetan Plateau (TP) and assess their sensitivity to climate change. As a vital ecological barrier, the TP’s grassland ecosystems are critical for understanding the impacts of climate change. However, sensitivity analysis of the NPP in the TP grasslands has been limited, which this study aims to address by focusing on the effects of maximum temperature, solar radiation, and wind speed on the NPP. The results show that the MLP model outperforms the RF model in prediction accuracy (R2 = 0.98, RMSE = 16.24 g C·m−2·a−1, MAE = 9.04 g C·m−2·a−1). NPP responses to climate factors are diverse: linear with temperature and nonlinear with solar radiation and wind speed. Under multi-factor scenarios, the NPP in both alpine meadow and alpine grassland exhibit nonlinear trends, with a higher sensitivity to changes in all three factors than to single- or two-factor changes. Spatial distribution analysis revealed that the NPP in alpine meadows was more sensitive to climate change in the southern regions, while alpine grassland showed greater sensitivity in the central regions. This study, using machine learning models and sensitivity analysis, sheds light on the complex response of the NPP in the TP grasslands to climate change, offering valuable insights for carbon cycle research in cold ecosystems and regional climate adaptation management.
Read full abstract