Predicting climate impacts on Quercus mongolica carbon sequestration: A case study from temperate forests of northeast China

Abstract

Global ecosystems are being significantly impacted by climate change and rising carbon dioxide (CO₂) levels. Trees can adjust the photosynthetic process in reaction to climate change, playing an important role in moderating global warming. However, there has been little systematic research into how climate conditions interact to influence tree carbon sequestration ability. This study focused on Quercus mongolica in the Maoershan forest of northeastern China, with the objectives of (1) Through wood anatomy, investigate the response mechanisms of xylem anatomical characteristics to climatic factors; (2) Integrating wood anatomical data with local climate data, constructing a fine-scale carbon sequestration prediction model PINN-LSTM4CSP based on physical knowledge and historical data. Wood anatomy revealed that elevated temperatures typically cause a reduction in vessel diameter and thickening of the vessel cell wall. Furthermore, increased precipitation was found to substantially alleviate the adverse impacts of high temperatures on vessel diameter. Changes in these anatomical characteristics directly correlate with and influence tree carbon sequestration capacity. The results of the prediction model validation demonstrate that PINN-LSTM4CSP outperforms existing models (Data-driven GRU, Data-driven LSTM), with much improved prediction accuracy ($R^2=0.9335$) and over 40 % reduction in key error metrics (MSE, RMSE). These findings not only support the significant potential of combining physical mechanisms and deep learning to improve carbon sequestration prediction accuracy, but more importantly, they provide key technical support for achieving more reliable and high-resolution dynamic assessments of regional forest carbon sequestration, and lay the groundwork for future research on sustainable forest management, carbon sink assessment, and climate reconstruction in the future.