Prediction of Gross Primary Productivity Change in Central Asia Under Climate Change Using Deep Learning

Accurate prediction of future gross primary productivity (GPP) change in Central Asia is crucial for assessing the health of terrestrial ecosystems in the region and adapting to impending climate change. This study investigates the feasibility of utilizing deep learning (DL) models for forecasting future annual GPP based on historical data. Specifically, we introduce a novel model, LSTM-Transformer-Group model (LTG), tailored to the unique characteristics of GPP prediction in Central Asia. Based on MODIS GPP data, we evaluate the future annual GPP prediction performance of the LTG model and classical DL models, and we further examine the impact of different input time steps on prediction accuracy and carefully investigate the influence of climatic factors on the accuracy of annual GPP predictions. Furthermore, we evaluate the LTG model's long-term forecasting capability. Our experiments reveal several key findings: 1) DL models adeptly capture the temporal dynamics within GPP time series, facilitating direct prediction of future annual GPP based on past data. The proposed LTG model, optimized with an appropriate input time step, achieves superior prediction accuracy, yielding an <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">${{\mathbf{R}}^2}$</tex-math></inline-formula> of 0.945; 2) GPP prediction performance varies across different climatic zones, but integrating climatic variables as direct inputs to the DL model does not enhance accuracy and may even diminish performance; 3) Over the long term (2019–2023), the LTG model consistently outperforms, exhibiting an average <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">${{\mathbf{R}}^2}$</tex-math></inline-formula> exceeding 0.95. Our findings offer methodological insights for future GPP prediction endeavors in diverse geographical contexts.