Dominant drivers of spatiotemporal variations in carbon and water use efficiency across the Yellow River Basin revealed by interpretable machine learning

Guangchao LiLiqin Han^*Wenjie HaoMengjia FengYanjie LiZhaoqin YiYayan LuKangjia Zuo

• Henan Normal University, Xinxiang, China

Precisely quantifying the spatiotemporal variation patterns of ecosystem water use efficiency (WUE) (i.e., WUENPP and WUEGPP) and carbon use efficiency (CUE) across diverse regions, as well as identifying the spatial heterogeneity of their principal influencing factors, are crucial for elucidating the complex underlying mechanisms governing carbon and water cycles in the Yellow River Basin (YRB). In this study, we utilized multi-source remote sensing data, and employed Ensemble Empirical Mode Decomposition (EEMD) to explore the nonlinear spatiotemporal trends and patterns of WUENPP, WUEGPP, and CUE within the YRB ecosystem. Additionally, we applied the optimally parameterized XGBoost and SHAP models to discern the spatial heterogeneity of the key factors driving their spatiotemporal variations. The results showed that: (1) The WUENPP, WUEGPP, and CUE of the YRB ecosystem exhibited a spatial distribution pattern characterized by higher values in the southeast and lower values in the northwest, with these metrics were predominantly concentrated at elevations ranging from 1000 to 1500 meters. (2) The interannual change rates of the yearly average values of WUENPP, WUEGPP and CUE in the YRB ecosystem were 0.008, 0.005, and 0.001, respectively. The predominant change patterns for WUENPP and WUEGPP were monotonic increases, covering approximately 42.44% and 41.97% of the watershed area, respectively. In contrast, the change pattern for CUE was primarily a decrease followed by an increase, observed across 42.51% of the watershed area. (3) In the YRB ecosystem, the leaf area index (LAI) emerged as the primary determinant of WUENPP and WUEGPP. Specifically, WUENPP and WUEGPP both showed an upward trend in tandem with the increase in LAI. Furthermore, temperature was identified as the key driving factor for CUE within the YRB ecosystem. (4) In the YRB ecosystem, LAI exhibited the highest importance index for both WUENPP and WUEGPP. It played a dominant role in approximately 42.80% and 45.35% of the study areas for WUENPP and WUEGPP, respectively. Conversely, temperature was a crucial factor influencing the spatial variability of CUE in the YRB ecosystem, exerting a predominant influence in 38.88% of the study areas.