Compound climate extremes sharpen dengue forecasting across Brazil
Dengue risk in Brazil does not move with rainfall or temperature alone. It rises and falls with combinations of climate extremes that unfold over different time windows, and this preprint argues that those interacting signals can improve public health forecasting across a country with sharp regional climatic contrasts.
Using a Bayesian spatiotemporal model, investigators analyzed monthly dengue cases from 2010 to 2025 across 435 health regions. The framework linked relative risk to interacting long-lag and short-lag drought indicators, temperature anomalies, absolute temperature, and El Niño-Southern Oscillation, then examined the pattern across nine Köppen-Geiger climate zones. The design matters because dengue transmission is rarely driven by a single environmental pulse; mosquito breeding, adult survival, human exposure, and viral amplification each respond to climate on different clocks.
One striking result emerged across multiple zones: long-lag dry conditions, mid-lag warm conditions, and short-lag wet conditions were tied to the highest relative risk in five climatic zones. In other words, the danger did not come from drought or rain alone, but from sequencing — a dry stretch that precedes warming, then wet conditions that can rapidly expand mosquito habitat and accelerate transmission. For vector-borne disease, this is the kind of compound pattern that can quietly build pressure long before case counts surge.
Temperature behaved differently depending on geography. Temperature anomalies modified hydrometeorological dynamics more strongly in temperate areas than in tropical ones, suggesting that the same climate shock can have a different epidemiologic meaning depending on local background conditions. In tropical settings, where mosquito populations and transmission potential are already close to perennial viability, incremental shifts may matter less than in temperate areas where thresholds are more fragile. The biology is straightforward: Aedes mosquitoes respond to moisture, heat, and survival windows in ways that are highly nonlinear, so climate interactions often create more useful signals than any single variable alone.
Forecast gains appeared at operational lead times
The forecasting component gives the study its public health relevance. Climate covariates improved prediction of case counts and outbreaks compared with baseline models at lead times of 1, 3, 6, and 12 months, and for the upcoming season five months ahead. Gains were largest in the Northeast, Southeast, and Midwest, regions where climate variability may offer stronger predictive structure for dengue dynamics.
Recent dengue incidence carried the greatest value at the 1-month lead. When the horizon extended to 3 and 6 months, combining recent incidence with climate information performed best. That pattern fits routine surveillance logic: recent cases capture what is already happening, while climate variables add an early signal about where transmission is likely to intensify next. At shorter horizons, the epidemic trajectory is already visible; at intermediate horizons, climate helps fill the gap between known transmission and future risk.
A separate weekly prediction framework extended the approach to state-level season-ahead forecasting, aligning the model with national public health planning. This is not merely a technical detail. State and regional health authorities need a forecast that can inform larval control, diagnostic readiness, staffing, and outbreak communication before the season unfolds. A model that works only in hindsight has little operational value; a model that can anticipate a surge five months out can shape decisions when they still matter.
Why compound climate signals matter more than single-variable alerts
Dengue ecology in Brazil is shaped by the interaction of human settlement, vector abundance, climate heterogeneity, and imported or residual viral circulation. A single hot month may not do much on its own. A dry period that alters water storage behavior, followed by warming and then rain, can create an environment that favors container breeding and rapid vector proliferation. Those staggered exposures can also influence mosquito development, adult biting frequency, and the extrinsic incubation period of the virus.
The preprint’s multi-zone approach is important because Brazil is not climatically uniform. A model calibrated only to national averages can miss local thresholds, especially when the same rainfall pattern has different consequences in dry, wet, temperate, and tropical settings. By separating the analysis across climate zones and testing interaction terms, the investigators treated dengue transmission as a spatially adaptive process rather than a single national curve.
The work also places absolute temperature and El Niño-Southern Oscillation in the model, which helps separate local weather from broader ocean-atmosphere variability. That distinction matters for forecasting because ENSO can shape rainfall and heat patterns months in advance, giving public health teams a longer planning window than local weather alone. When those signals are folded into surveillance, the result is a more layered warning system rather than a one-dimensional alert.
What the model adds to surveillance practice
For health services, the practical implication is not just better statistical fit. It is a more usable early warning architecture. Baseline surveillance can identify current burden, but climate-informed prediction can indicate where pressure is likely to rise next, especially in regions that routinely experience recurrent dengue seasons. The strongest improvements in this study appeared when climate information was added to recent incidence over 3- to 6-month horizons, exactly the period when response planning still has room to bend the curve.
The study also suggests that forecasting should not rely on a single model form. A multi-staged approach, with recent incidence dominating short-term predictions and climate variables contributing more at intermediate and seasonal horizons, may better reflect the real structure of dengue transmission. That tiered strategy mirrors the disease itself: immediate case trends, delayed environmental forcing, and season-scale climatic context all contribute different pieces of the risk signal.
This remains a preprint and has not undergone peer review. Even so, the analysis offers a clear operational message for Brazil and other dengue-endemic settings: compound climate extremes carry useful predictive information, and that information can be translated into forecasts that are meaningful at public health lead times. The main unresolved question is how well these gains will hold when the framework is tested prospectively during future dengue seasons, especially as climate volatility continues to reshape transmission conditions.