Vapor pressure deficit (VPD) arises from multi-factor, nonlinear, scale dependent processes, yet many studies still rely on linear or single causal models, overlooking spatial non stationarity, synergistic/redundant interactions and thresholds. We develop an integrated China wide framework: (i) linear regression and STL for spatiotemporal heterogeneity; (ii) Synergy Uniqueness Redundancy Decomposition (SURD) to partition redundancy, synergy and information leakage; (iii) Geographical Convergent Cross Mapping (GCCM) for scale dependent causal intensities; and (iv) XGBoost-SHAP for nonlinear thresholds and contribution rates. (1) 2000-2020 VPD shows strong seasonality and heterogeneity: significant increases in 17.9 % of area (North China Plain, middlelower Yangtze, coastal South China) and decreases in 19.6 % (western Tibetan Plateau, southern Xinjiang Basin, parts of Inner Mongolia); anomalies occurred in May 2011 and October 2019. (2) Redundancy dominates temperature and precipitation effects (indices 0.624, 0.689), while humidity exhibits stronger multi factor coupling via synergy (0.192) and information leakage (21.24 %). (3) Driving intensity is scale dependent: nationally temperature (0.941) > precipitation (0.799); regionally temperature peaks in the Plateau Temperate Semiarid zone (L = 200, 0.941) whereas precipitation dominates the Marginal Tropical Humid zone (L = 120, 0.442). (4) Dominant factors display clear thresholds: maximum temperature >12.72 degrees C (contribution 0.241) and precipitation <403.70 mm (400 mm isohyet). Meteorological factors jointly contribute 0.418 versus weak topographic influence (0.031). The framework links system mechanism, spatial causal validation and threshold characterization, advancing VPD ecohydrological risk assessment.