Systematic Review of Heat Stress in Latin America
Taking stock of what we know—and what we don’t.
Latin America faces growing heat risks, yet the research landscape is fragmented. We are conducting the first region-wide systematic review of heat stress studies across the region.
Our review catalogs the metrics used (WBGT, heat index, extended indices, etc.), the populations studied (urban residents, outdoor workers, rural communities), and the gaps in coverage—both geographically and socially. By mapping out what’s missing, we can show where new research is urgently needed.
Why it matters: Without a clear baseline, climate adaptation risks repeating old blind spots—ignoring the very communities most exposed to extreme heat. This review becomes a foundation for smarter, more inclusive climate planning across the region.
Machine Learning for Extreme Heat Forecasts
Because not all heat indices tell the same story.
The most common metric, the “heat index,” often fails in real-world conditions—especially for outdoor workers under the sun. We are comparing multiple heat stress metrics, including the Extended Heat Index (EHI/EHI-350), WBGT, and new formulations that account for solar radiation and metabolic heat load.
By applying these metrics to ERA5 and ERA5-Land data across Latin America, we can identify which ones best capture the lived risks people face. This includes testing how indices perform in humid tropics, high-altitude cities, and rural agricultural zones.
Why it matters: Choosing the right metric isn’t academic—it shapes whether a heat warning is accurate, trusted, and actionable. Better measurement means better protection for the people most at risk.
Measuring Heat Stress with Better Metrics
Pushing climate models further to see the dangerous extremes.
We are building a stacked ensemble machine learning model to forecast extreme heat stress events across Latin America. Instead of focusing on averages, this model captures the upper-tail percentiles—the hottest, most humid days that communities will actually struggle to survive.
Our approach blends ERA5 reanalysis data with bias-corrected CMIP6 projections, applying advanced methods like quantile regression and ensemble learning. The output is probabilistic maps of future risk, showing not only when and where heat stress will intensify, but also the uncertainty around those estimates.
Why it matters: Traditional climate models often smooth over the most dangerous events. By highlighting the extremes, we give cities, health agencies, and local planners the evidence they need to prepare for power outages, crop failures, and heat-related illness before the crisis hits.