Predicting Net Ecosystem Carbon Exchange of Typical Forest Ecosystems in China Based on ChinaFLUX

Abstract. Accurate prediction of forest carbon sinks is crucial for achieving carbon neutrality, peak carbon emissions goals, and advancing the Sustainable Development Goals (SDGs). Due to the complexity of forest ecosystems and the limited application and accessibility of ChinaFLUX observation data, previous studies generating Net Ecosystem Exchange (NEE) products largely relied on global flux observation data. The relatively sparse observations in China introduce significant uncertainties in regional carbon sink estimations. While Long Short-Term Memory (LSTM) models have been widely applied to remote sensing image time-series analysis and vegetation index prediction, their use in carbon sink prediction remains limited. This study assesses the ability of the LSTM model to predict NEE dynamics in typical Chinese forest ecosystems using ChinaFLUX data and multi-source remote sensing data. Using long-term Eddy Covariance (EC) observation data from 11 forest sites, alongside meteorological information and multi-source remote sensing data, we analyzed the carbon sink characteristics of five typical forest types: Deciduous Broadleaf Forest (DBF), Deciduous Needleleaf Forest (DNF), Evergreen Needleleaf Forest (ENF), Evergreen Broadleaf Forest (EBF), and Mixed Forest (MF). The results indicate that the LSTM model effectively captures the main trends of NEE, though some fluctuations persist in predictions for certain data points. During training and testing, the average R2 values between model-predicted NEE and EC-derived NEE were 0.83 and 0.73, respectively, with RMSE values ranging from 9.75 to 31.04 g C m−2 mon−1.Furthermore, this study identifies key driving factors behind NEE variations across forest types. Environmental factors and vegetation physiological conditions exhibit significantly differing impacts on NEE. This study offers theoretical foundations and technical support for improving forest carbon sink assessments in China and informing climate change responses. It also presents a novel approach for accurately predicting and evaluating forest carbon sink dynamics.