Tensor based spatiotemporal attribution of compound dry heat exposure in the Lanzhou–Xining urban agglomeration from 2000 to 2025

Introduction Compound dry-heat exposure is an intensifying climate risk in arid and high-altitude cities, yet its spatiotemporal evolution and drivers are not fully resolved. The Lanzhou–Xining urban agglomeration is a representative hotspot where population exposure and environmental constraints intersect. Methods A three-dimensional temperature cube with 305 monthly dry-bulb layers (2000–2025) encodes each pixel as a high-dimensional temporal vector. From this cube, three compound dry-heat indices (CDH1–CDH3) are derived to represent surface thermal intensity, radiative saturation, and hydrothermal imbalance. Six exposure metrics (EXP1–EXP6) quantify demographic, ecological, and infrastructural vulnerability dimensions. Annual XGBoost models with SHAP interpretation are used to attribute exposure dynamics to evolving CDH drivers and to map spatial heterogeneity. Results Four coupling regimes are identified: an early temperature-dominated phase (2000–2004), a radiative intensification period (2005–2011), a compound maturation stage (2012–2018), and a post-2019 fragmentation era marked by sharp attribution volatility and strong spatial heterogeneity. CDH3 acts as a latent but volatile disruptor, with pronounced influence on soil-moisture-related exposure (EXP6), whereas CDH2 shows a dominant, structurally embedded role in vegetation-constrained zones. The exposure response exhibits asymmetry and threshold sensitivity across the urban-rural gradient. Discussion The findings indicate that dry-heat exposure in Lanzhou–Xining is co-controlled by radiative saturation and hydrothermal imbalance rather than temperature alone. Adaptation should prioritize radiative redistribution, evaporative buffering, and localized risk governance, with policies targeted to regime-specific drivers and spatial heterogeneity.