[Current Status and Prospects of Machine Learning Applications in Terrestrial Ecosystem Carbon Sink Modeling]

Under the "Dual-Carbon" strategy， assessing the spatial distribution of carbon sinks in typical regions represented by terrestrial ecosystems is essential for accurately characterizing regional carbon sink functions. Net ecosystem exchange （NEE） serves as a key indicator of ecosystem carbon fluxes. To accurately simulate the spatial patterns of regional NEE， commonly used methods include the inventory method， process-based models， and atmospheric inversion techniques. In recent years， machine learning has emerged as a powerful data-driven approach for spatial NEE modeling， offering distinct advantages in capturing complex nonlinear relationships. This study reviews the main methods applied in NEE spatial distribution modeling， elaborates on the principles and current applications of machine learning-based approaches， analyzes key feature selection and algorithm choices across different ecosystem types， and discusses the potential of applying machine learning to investigate NEE spatial characteristics in complex environments such as urban ecosystems.