From Grid Cells to Policy in Heat-Stress Analysis

Abstract

Heat stress poses a growing risk to human health across India, yet many climate-health analyses introduce systematic biases by applying nonlinear metrics to aggregated data. This thesis develops and applies an aggregation-bias aware methodological framework for analysing spatio-temporal patterns of heat stress using high-resolution climate reanalysis data (ERA5 and ERA5-Land datasets for the years 1980–2024). We compute heat stress indicators such as the Universal Thermal Climate Index (UTCI) and Wet Bulb Globe Tem perature (WBGT) from the reanalysis datasets. This thesis makes three primary contributions. First, it identifies and characterises the Jensen gap. The gap arises when nonlinear functions such as UTCI are applied to spatiotem porally averaged meteorological inputs rather than averaging the function’s outputs. We show that these biases are not random: they are systematic, spatially clustered, threshold contingent, and non-additive when they propagate through exceedance calculations. This has direct consequences for district-level heat action planning in India, where aggregation to administrative boundaries is routine. Furthermore, the framework of Jensen Gaps is broadly generalisable to all kinds of environmental exposure analyses that use geo-spatial and/or temporal data. The second contribution looks at how the temporal structure of heat events across India has changed from 1980 to 2024. Rather than asking simply whether heat is increasing, we ask: Are heat seasons starting earlier and ending later? Are individual heat events getting longer? Are the breaks between events — which matter for physiological recovery — getting shorter? We find that the answer to most of these questions is yes for large parts of India, but the specifics vary considerably across space. North-East India, the eastern coast, and the Indo-Gangetic plains show up repeatedly across multiple metrics and thresholds. After this, we ask whether the meteorological character of heat events is itself changing. We fit logistic regression models at each gridpoint to track the temporal sensitivity of heat stress exceedance to temperature, humidity, radiation, and wind speed over the 45-year record. The results point towards at least two broad regimes emerging across India. Along the coasts and much of peninsular India, humidity and radiation are playing an increasingly prominent role. This could point to increasing humid heat that limits the body’s ability to cool itself through sweating, and makes evaporative measures like desert coolers progressively less effective. In parts of northern India, wind speed appears to be gaining importance, consistent with the well-known role of hot dry winds like the loo in driving dangerous daytime heat in that region. Throughout, the analysis is conducted at the finest available spatio-temporal resolution, motivated directly by the Jensen gap findings. The framework developed here is not only generalisable to other composite heat-stress metrics, such as WBGT, but also to alternative data sources, such as station-level data. Therefore, the thesis highlights the bias of the Jensen Gap in existing methodologies and also develops a generalisable Jensen-aware framework that drives insights for heat adaptation and heat-related climate change research.