Variables controlling growing season carbon dynamics in desert shrub ecosystem

Abstract

Arid and semi-arid lands (drylands) are under the influence of extreme environmental conditions and cover large areas on Earth’s land surface. Determining temporal and spatial variations of the greenhouse gas and energy exchange in these ecosystems will provide a better understanding of both, the dynamics between the drylands’s surface and the atmosphere and the importance in climate change. In this study, CO2 exchanges over a shrub steppe ecosystem were determined using the Eddy Covariance method during both dry and wet growing seasons in the Gobi Desert, Mongolia. The net ecosystem exchange (NEE) was modeled using the extreme gradient boosting method (XGBOOST). Variables influencing NEE were estimated employing the SHapley Additive exPlanations algorithm (SHAP). The XGBOOST predictions demonstrated impressive results across all growing seasons, with high performance metrics (R2 = 0.821; NSE = 0.817), particularly notable during the wet season (R2 = 0.932; NSE = 0.929). The total NEE changed between 79.7 and -110.4 gC m−2 throughout the growing seasons. Especially the precipitation before the growing period has a positive impact on carbon sequestration in the shrub ecosystem in the following dry season. A high similarity was found in the patterns of daily NEE of all growing seasons and normalized difference vegetation index (NDVI), albedo, soil water content, temperature, sensible heat flux, the difference between air and soil temperature in desert shrub ecosystem. Although NDVI was the dominant factor during the wet season, there was a weaker relationship between these factors and NEE in the dry season than in the wet season.