Integrating Forecast Uncertainty into Robust Predictions of Heat-Stress Risk Using Full Thermoregulatory Models

Presented by Christine Killip at the 24th International Congress of Biometeorology
Written by Christine Killip, Matt Young and Marko Uksanovic
Presented by Christine Killip at the 24th International Congress of Biometeorology
Biometeorology: Addressing 21st century global climate issues and perspectives of living beings and places
https://www.dgt.uns.ac.rs/icb2026/
Accurate prediction of human heat-stress risk remains challenging due to the extremely narrow margin between environmental conditions under which the body can maintain thermal equilibrium and those that precipitate uncompensable heat strain. Even small forecast errors in temperature, humidity, wind speed, or radiant load can materially change predicted outcomes. This study evaluates how uncertainty in meteorological forecasts propagates through three advanced heat-stress indices grounded in full thermoregulation modelling: Predicted Heat Strain (PHS), the Thermal Work Limit (TWL), and the Universal Thermal Climate Index (UTCI).
We first assessed the performance of operational ensemble forecasts to determine whether ensemble spread meaningfully improves the reliability of predicted heat-stress categories. Ensemble approaches, however, showed limited benefit. The ensemble spread was typically narrower than the true forecast error distribution, leading to overconfident predictions and offering little improvement in capturing the threshold-sensitive transition into dangerous heat-stress states.
We then applied a Monte Carlo perturbation framework using empirically derived forecast-error distributions for each meteorological driver. Thousands of realisations were generated for each forecast time step and propagated through the PHS, TWL, and UTCI models. This method substantially improved the robustness of risk prediction, producing probabilistic heat-stress profiles that more accurately represented the likelihood of exceeding critical physiological thresholds. The Monte Carlo approach captured nonlinear sensitivities inherent in thermoregulation modelling and provided a more realistic operational decision-support tool.
Our findings highlight that explicit representation of forecast uncertainty—rather than reliance on ensemble systems alone—is essential for reliable heat-stress prediction in occupational contexts.
