Dryland ecosystems are highly sensitive to global change, with their carbon-water cycling strongly regulated by soil moisture (SM). The critical soil moisture content at which the ecosystem shifts from energy limitation to water limitation during drought can be regarded as a soil moisture threshold that characterizes this regulation. Accurately quantifying critical soil moisture thresholds (Bt) remains a key scientific challenge in elucidating the mechanisms of carbon-water interactions. This study focuses on the Horqin Sandy Land, one of the four major sandy regions in northern China, using long-term eddy covariance and environmental data collected from dune and meadow ecosystems during the growing seasons (May-September) from 2013 to 2024. Three methods-evaporative fraction (EF), covariance (Cov), and correlation difference (corr)-were used to identify to Bt, marking the transition from energy-limited to water-limited states during drought. Furthermore, interpretable machine learning (XGBoost-SHAP) and dominance analysis (DA) were employed to elucidate the driving mechanisms of Bt. Results showed that meadow ecosystems had stronger carbon uptake capacity, water regulatory potential, and environmental stability than dune ecosystems. The three methods yielded highly consistent Bt for both dune (BEFt : 0.0676, BCov t : 0.0564, Bcorr 0.3934 m3/m3) ecosystems, with standard deviations of 0.005 and 0.023 m3 /m3, respectively. In dunes, Bt was primarily driven by rainfall (Rain) and evapotranspiration (ET), with gross primary productivity (GPP) and ET jointly contributing over 50 % to its formation. In meadows, Bt was regulated by Rain and canopy conductance (gc), with GPP and gc contributing 28.98 % and 19.57 %, respectively. As shown for the two ecosystems examined, the dune and meadow ecosystems exhibit pronounced differences in the critical soil moisture thresholds at which the system shifts between water and energy limitation. Given that drylands encompass a wide range of ecosystems-including grasslands, forests, and croplands. This highlights the necessity of establishing coordinated, multi-ecosystem observational networks across drylands to compare responses systematically and clarify vegetation-water interactions. This is essential for effective water resource management and ecosystem restoration in arid regions.