Interannual variations in grassland carbon fluxes and attribution of influencing factors in Qilian Mountains, China.

Abstract

Clarifying the driving factors of grassland carbon sequestration is essential for understanding its role in the regional carbon balance. However, there is a lack of studies on the upscaling of carbon flux in the Qilian Mountains (QLMs) and the driving factors of its interannual variation (IAV). Based on long-term eddy covariance observations in the QLMs, this study estimated the net ecosystem CO₂ exchange (NEE), gross primary productivity (GPP), and ecosystem respiration (ER) of the QLMs grassland using four machine learning methods (random forest regression (RF), extremely randomized tree regression (ETR), support vector regression (SVR), and extreme gradient boosting (XGBoost)) to obtain the optimal estimation model. Subsequently, the spatiotemporal variations of GPP, ER, and NEE in the QLMs grasslands were conducted in a comprehensive analysis. The factors influencing the IAV of carbon flux, the contribution of monthly NEE to NEE IAV, and the contribution of different grassland types of NEE to NEE IAV were explored. Our findings revealed that the accuracy and resolution of the grassland carbon flux estimated by the RF method in this study are higher than those of global products. The grassland exhibited a weak carbon sink from 2000 to 2022, with an average NEE of -26.46 ± 6.80 g Cm^-2 yr^-1, and it acted as a carbon sink from May to September. The spatial distribution pattern of carbon sequestration was "low in the northwest and high in the southeast". LAI was the key driving factors of IAV for GPP and ER, while NEE IAV was primarily influenced by precipitation and temperature. Climate and vegetation factors primarily regulated NEE IAV by affecting the GPP and ER of plants, and NEE IAV was primarily driven by GPP. Furthermore, NEE in alpine meadows and alpine steppes dominated the NEE IAV of the entire grassland, and summer NEE contributed the most to the NEE IAV. The results will help us to better understand the carbon cycling mechanism in grassland ecosystems and provide new data support and a theoretical foundation for regional carbon cycling research.