Vulnerability assessment of heat waves within a risk framework using artificial intelligence.

Area of Science:

• Environmental science
• Climate change adaptation
• Public health

Background:

• Existing climate change adaptation efforts are inadequate to mitigate projected impacts.
• Vulnerability assessments are crucial for effective resource allocation but current methods, like principal component analysis, have limitations.
• Principal component analysis-based assessments are difficult to translate into actionable strategies.

Purpose of the Study:

• To propose a novel data-driven vulnerability assessment framework within a risk context.
• To clarify definitions of sensitivity and adaptive capacity based on the IPCC's Sixth Assessment Report.
• To validate the proposed framework using heat wave data and mortality rates in Quebec, Canada.

Main Methods:

• Developed a risk framework incorporating clarified IPCC definitions.
• Utilized heat wave data (2001-2018) and daily mortality rates in Quebec.
• Employed socioeconomic and environmental variables to calculate impact and vulnerability.
• Compared AutoGluon, Gaussian process, and deep Gaussian process for modeling impact and vulnerability.
• Used SHAP values for model interpretability.

Main Results:

• Generated a vulnerability map illustrating summer mortality rate distributions across Quebec.
• Gaussian process modeling demonstrated promise for the proposed framework.
• SHAP values confirmed model results align with existing vulnerability knowledge.
• Identified specific areas with higher vulnerability to heat wave impacts.

Conclusions:

• The proposed data-driven framework offers a more actionable approach to vulnerability assessment.
• Gaussian processes provide a suitable method for modeling climate change impacts and vulnerability.
• The framework aids in identifying high-risk areas for targeted adaptation interventions.
• Recommendations are provided for both quantitative and qualitative implementation of the framework.