Developing a Community-Specific Daily Weather Health Risk Index Across Australia Using Explainable Machine Learning

Weather conditions are closely related to human health, yet effective methods for communicating the joint health risks associated with weather-related factors remain limited, especially when accounting for the complex interactions among weather exposures. To address this gap, we collected daily mortality and meteorological data (temperature, relative humidity, surface pressure, wind speed, rainfall and ultraviolet B radiation) between 2009 and 2019 for each community across Australia. We employed an advanced explainable machine learning framework integrating the eXtreme Gradient Boosting (XGBoost) model with Shapley Additive exPlanations (SHAP) algorithm to quantify the joint health risks associated with the meteorological factors and constructed a daily weather-health risk index (WHRI) for each Statistical Area Level 3 community across Australia. Among the examined weather-related factors, temperature was the dominant contributing factor for mortality risks. Communities in southern Australia generally had greater weather-related mortality risks and higher WHRI compared to those in northern Australia. Evident seasonal patterns were observed for WHRI, with peaks occurring in winter and its lowest point occurring in summer. These findings highlight the spatiotemporal heterogeneity of weather-related health risks and emphasize the need for developing a WHRI to capture and quantify the dynamic risk patterns. The integration of WHRI into public health dashboards could effectively inform the empirical community-specific health risks in real time, thereby supporting evidence-based, tiered interventions for adverse weather conditions in a changing climate.