Estimating regional terrestrial ecosystem carbon sinks on multi-model coupling approach

• A multi-model coupling approach improves regional carbon storage estimation. • The enhanced InVEST model refines carbon stock assessments across four pools. • A geographical detector model reveals human and natural impacts on carbon sinks. • Findings support low-carbon policy formulation for sustainable development. The regional terrestrial ecosystems serve as primary carbon sinks, characterized by strong spatial heterogeneity and significant interannual fluctuations. In Xinjiang, one of China’s five autonomous regions, carbon storage increased from 12967.897 Tg C (1Tg = 10 12 g) to 14262.31 Tg C. Traditional carbon sink assessment methods struggle to fully account for the combined impacts of human activities and environmental factors, impeding accurate depiction of the spatial distribution and evolution of regional carbon stocks. This study proposes a regional terrestrial ecosystem carbon density estimation method based on an ARIMA-CatBoost-RNN coupled model. Firstly, the ARIMA model forecasts carbon density time series, the CatBoost model reduces the impacts of spatial heterogeneity, and the RNN model estimates ecosystem carbon density values. Secondly, terrestrial carbon storage is estimated using an parameter-updated InVEST model, with an accuracy of up to 78.4 %. Finally, the Geodetector model quantifies the influence of nine driving factors on carbon sink capacity. The results reveal that soil carbon stocks comprise 55 %–61 % of total carbon storage, making them the main component of Xinjiang’s terrestrial ecosystems. Annual average carbon sequestration is 39.02 T C/km 2 , with forests showing the highest capacity at 103.33 T C/km 2 . NDVI (Normalized Difference Vegetation Index) has the most significant impact on Xinjiang’s carbon sink capacity, contributing up to 0.615.