Abstract
<jats:p>Urban heat is a significant cause of mortality. In London, during the summer of 2022, the United Kingdom experienced its first recorded 40°C day, resulting in excess mortality levels not observed since pre-antibiotic influenza seasons. On 19 July 2022, London reached a temperature of 40.3°C, the highest recorded in the United Kingdom, which corresponded to 18,435 excess deaths in England and Wales. This study develops the first Lower Layer Super Output Area (LSOA)-scale heat vulnerability index (VI) for all 33 London Boroughs (n = 4,994) by integrating eight distinct datasets using a reproducible six-script Python pipeline. A critical methodological step involved identifying and excluding seven variables that leaked data, thereby reducing the maximum feature-VI correlation from 1.000 to 0.772. Five machine learning models were trained and validated spatially using borough-based data splitting. The four best-performing models achieved Test R² scores between 0.864 and 0.873, with overlapping 95% bootstrap confidence intervals, indicating that model learnability is not dependent on specific architectural choices. XGBoost is the main model (Test R² = 0.869, MAE = 0.161); the SHAP attribution shows HVI_mean (|SHAP| mean = 0.272) and imd_employment_score (0.216) as co-dominant drivers in all five architectures. Temporal trend analysis from 2015 to 2024 identified a pronounced asymmetry: the number of deaths increased substantially during the summer period (τ = 0.644, p = 0.012, with 1,544 additional deaths annually), whereas temperature trends were not statistically significant (p = 0.371). However, seasonal-trend decomposition using LOESS (STL) indicated an underlying warming trend (p = 0.032). The environmental justice analysis demonstrated a vulnerability ratio of 1.30 between the most and least deprived decile groups, with 60.3% of the vulnerability index attributable to physical thermal exposure. UKCP18 projections indicate that Barnet, currently ranked 32nd among 33 boroughs, is projected to become the primary location for emerging hot spots.</jats:p>