Assessing net primary productivity variation and potential future impacts based on machine learning and contribution analysis in the Yangtze River Delta, China.

Net primary productivity (NPP) is a critical indicator of ecosystem carbon sequestration and vegetation growth, directly influencing global carbon cycles and climate change mitigation. However, understanding the spatiotemporal patterns of NPP in rapidly developing regions like the Yangtze River Delta (YRD) remains challenging due to complex interactions between climatic factors and human activities. This study addresses this gap by analyzing the spatiotemporal variations of NPP in the YRD from 2001 to 2020 using multi-source data and an advanced convolutional neural network (CNN) optimized by the dung beetle optimization (DBO) algorithm. The dynamic search strategy of the DBO algorithm enables it to to avoid local optima effectively and is particularly suitable for complex nonlinear problems such as NPP modelling. The proposed model demonstrated strong performance, with an R of 0.914 and an RMSE of 21.1 gC/m/yr, indicating reliable NPP estimation. The average NPP was 0.795 KgC/yr, with a significant annual increase of 3.72 gC/m/yr. Spatially, NPP exhibited a northwest-to-southeast gradient, reflecting regional climate and land cover variations. Shapley additive explanations (SHAP) analysis highlighted sunshine hours, elevation, precipitation, and temperature as dominant factors influencing NPP. Partial correlation analysis further confirmed that elevation and precipitation were positively correlated with NPP, while sunshine hours and temperature showed negative correlations. Future predictions based on shared socioeconomic pathways (SSP) scenarios suggest a gradual decline in NPP from 2030 to 2050, with SSP245 maintaining higher NPP values than SSP585. The findings indicate that climate factors are the primary drivers of NPP increases, while human activities are primarily responsible for NPP declines. This study offers new insights into the complex dynamics of NPP in the YRD, providing a more accurate model for predicting future NPP trends and their implications for carbon management strategies.