Machine Learning-based Prediction of Carbon Fluxes in Terrestrial Ecosystems in China and Its Response to Land Use Change.

Current carbon flux data for China's terrestrial ecosystems have limitations, such as short time series and low resolution. Generating longer time series and higher-resolution carbon flux data for these ecosystems is crucial for studying their temporal and spatial variation characteristics and for analyzing the impact of land use changes on carbon fluxes. This study constructed four machine learning models using flux data from Chinese sites and relevant predictor variables, identifying the random forest model as the best predictive model. The R2 values for the training and test sets were 0.73 and 0.77, respectively, with RMSE （in C） values of 35.09 g·（m2·month）-1 and 32.85 g·（m2·month）-1. The model was validated through ten-fold cross-validation and leave-one-site-out cross-validation, and it was subsequently used to generate a monthly carbon flux prediction dataset for China's terrestrial ecosystems at 1 km resolution from 2000 to 2020. The results showed a decreasing trend in terrestrial carbon sinks from southeast to northwest China between 2000 and 2020, with the largest carbon sinks located in southern monsoon regions, North China, and the northeastern mountains. Inner Mongolia and the interior Tibetan Plateau are weak carbon sources, whereas Xinjiang and the northeastern Tibetan Plateau are stronger carbon sources. Terrestrial carbon sinks are influenced by climate factors, showing distinct seasonal variations. Over the past 20 years, China's terrestrial carbon sinks （in C） ranged from 0.38 to 0.77 Pg·a-1, with an overall increasing trend. Analysis of land use data revealed that forests have the strongest carbon sinks, followed by croplands and built-up areas. Grasslands are weak carbon sources, and unused lands are strong carbon sources. Changes in land use types significantly impact terrestrial carbon sinks. Rational land use policies help increase carbon sinks, contributing to climate change mitigation and ecological environment protection.