An interpretable LAI time series prediction model of subtropic forests using a ConvLSTM coupling spatiotemporal attention mechanism model

• An interpretable LAI time series prediction model is proposed based on ConvLSTM-STAM and SHAP model. • The model captures key LAI-predicted variables during the growth process of forest. • The contribution of variables to the predicted LAI is opposite in spring-winter and summer-autumn seasons. Leaf Area Index (LAI) is a critical biophysical parameter for understanding vegetation dynamics and carbon cycling, yet achieving high-precision LAI prediction in structurally complex forest environments remains a significant scientific challenge. This study proposes an interpretable ConvLSTM model coupled with a Spatio-Temporal Attention Mechanism (ConvLSTM-STAM) to predict forest LAI in Zhejiang Province (2013–2018) by integrating multi-source remote sensing, meteorological, and topographic data. Unlike conventional deep learning approaches, the model combines predictive power with interpretability through spatiotemporal attention weights and SHAP (SHapley Additive exPlanations) analysis. The ConvLSTM-STAM achieved state-of-the-art accuracy (R 2 = 0.887, RMSE = 0.349 m 2 /m 2 ), significantly outperforming MODIS products and benchmark models. More importantly, it provides spatiotemporal interpretability by quantifying feature contributions and revealing seasonal mechanisms of LAI variability, with Tmin and Tavg dominating in spring and winter, and LAI itself leading in summer and autumn. This dual capability of accurate prediction and mechanistic insight bridges remote sensing observations and ecological process modeling, advancing both theoretical understanding and practical applications in ecological research.