The Urban Dry-Wet Island (UDWI) effect reflects urbanization-induced alterations in regional hydrothermal cycles, yet its spatial patterns and nonlinear drivers remain poorly quantified. This study develops an integrated framework coupling Geographically Weighted Regression (GWR), XGBoost, and SHapley Additive exPlanations (SHAP) to investigate summer diurnal and nocturnal UDWI dynamics across 329 Chinese cities. Spatial patterns are characterized across five representative years (2015, 2018, 2020, 2022, 2024) and their 2015–2024 mean, with 2020 selected as a representative baseline for detailed mechanistic attribution. Results reveal a persistent ‘drier northwest-wetter southeast’ spatial pattern across the multi-year period. Focusing on the representative year 2020, dry island effects intensify markedly at night (frequency increasing from 65.1% to 67.2%, and intensity rising from 1.569 to 1.716). UDWI generally exhibits negative coupling with the Urban Heat Island (UHI) effect, strengthened under nocturnal conditions, with the R2 increasing from 0.127 ~ 0.129 during the day to 0.354 ~ 0.428 at night. Driving mechanisms shift diurnally: daytime UDWI is primarily governed by background humidity and evapotranspiration, whereas nighttime dynamics are increasingly controlled by land surface temperature and anthropogenic activities. Notably, urban impervious surface fraction exhibits a nonlinear threshold (≈40%), beyond which drying effects accelerate. These findings underscore that UDWI emerges from coupled hydrological-thermal processes characterized by strong nonlinearity and spatial heterogeneity. The proposed framework provides an interpretable pathway for disentangling urban climate mechanisms and supports evidence-based strategies for urban climate regulation and ecological optimization.