Climate data sharpens disease forecasts: heat islands, compound extremes, wetland risk maps

City blocks, not grid cells

Dengue forecasting is going to the city block. In an ESS Open Archive study, a 30-meter machine-learning framework trained on location-specific satellite imagery and local weather-station data was used to project dengue transmission suitability across contrasting US cities, and urban heat island effects were associated with extended projected dengue seasons for both Aedes aegypti and Ae. albopictus. m:…8072

The methodological point is worth stating plainly: global gridded reanalysis like ERA5, at 0.25 degrees, lacks the spatial detail needed to model urban air-temperature variation. The framework is automated and retrainable on local data, which makes it a forecast a city can actually run, not just read. m:…8072

The policy read is straightforward. If urban heat extends transmission season, then heat mitigation is disease mitigation, and the cities that cool their neighborhoods get an epidemiological dividend. Forecasts at 30 meters make that argument block by block, which is the resolution at which planting trees and painting roofs actually happens. m:…8072

Brazil's compound extremes

In a preprint, Fletcher and colleagues built a Bayesian spatiotemporal model for dengue across Brazil and found that compound climate extremes were the strongest signal: long-lag dry spells, mid-lag warmth, and short-lag wet conditions, in combination, were associated with the highest relative dengue risk in five climatic zones. m:…4603

Climate covariates improved both case-count and outbreak prediction at 1, 3, 6, and 12-month lead times over baseline models. Months-ahead operational forecasts are the kind of output a health ministry can staff and stock around. m:…4603

The compound nature of the extremes is the operationally useful part. Single-variable thresholds fail because mosquitoes do not respond to temperature alone; they respond to sequences, dry then warm then wet. A forecast that encodes the sequence gives vector-control teams a decision rule they can actually schedule spraying and larviciding around. m:…4603

One question the papers leave for the next round is ground truth: projected suitability and lead-time skill are only as good as the case reporting beneath them. The forecasts will sharpen as surveillance densifies, but under-reported regions will stay under-forecast, which is where equity and accuracy collide. m:…4603

Where encephalitis circulates

The third forecast is a risk map for Japanese encephalitis in Australia. In a preprint, Harrison's geospatial assessment identified 45 human infections and more than 85 piggery detections over 2021 to 2023, with floodplain wetlands in the north and the inland east showing both the highest estimated environmental suitability for transmission and the strongest evidence of virus circulation. m:…5134

Suitability plus circulation evidence is the pair that turns a map into a plan: it says where the virus can spread and where it already is. For planners weighing surveillance spend, that is the actionable slice of the three-paper arc: city heat, compound extremes, and wetland hotspots, all pointing at forecasts that operate at the scale decisions actually get made. m:…5134

Pair this with the Brazil result and the week's thesis is clear: vector-borne disease is becoming forecastable at the scales where public-health money gets spent. The remaining gap is not the modeling. It is whether agencies fund the surveillance that feeds it: wetlands in Australia, health-region case counts in Brazil, weather stations in US cities. m:…5134