Predicting Vegetation Indices in Yangtze River Delta based on climate and remote sensing data from 2001 to 2021 by Machine Learning Approach

Vegetation in the Yangtze River Delta (YRD) is significantly influenced by climatic variables such as potential evapotranspiration (PET), precipitation, and temperature. However, the prediction of vegetation dynamics in this ecologically sensitive region remains limited. To address this, the present study employed machine learning algorithms—Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Support Vector Machines (SVM)—to predict the Normalized Difference Vegetation Index (NDVI) based on climatic and remote sensing data from 2001 to 2021. Model performance was evaluated using R² and Root Mean Square Error (RMSE) metrics. Among the tested models, RF exhibited the highest accuracy and generalization across different sub-regions of the YRD, with R² values exceeding 0.85 on testing datasets and low RMSE values, indicating its superior capability in capturing non-linear environmental interactions. Temporal trend analysis revealed a steady increase in NDVI, rising from 0.475 in 2001 to 0.547 in 2021, supported by a consistent rise in precipitation and PET. Meanwhile, TCI declined from 0.534 to 0.402, suggesting increasing temperature stress on vegetation. These results underscore the complex interactions between vegetation and climate variables. Pearson correlation analysis revealed strong positive correlations between NDVI and both precipitation (r = 0.62) and PET (r = 0.80), while TCI exhibited a weaker correlation, indicating a lesser but notable impact on vegetation. SHAP (SHapley Additive exPlanations) analysis further highlighted the dominant influence of PET in predicting NDVI, followed by precipitation, with TCI having a marginal impact. These findings underscore the critical role of water availability and evapotranspiration demand in shaping vegetation dynamics. The study demonstrates the potential of machine learning in ecological forecasting and provides actionable insights for land-use planning, climate adaptation, and environmental monitoring in the YRD and similar climate-sensitive ecosystems.