Spatial representativeness matters for Climate-Driven Dengue Forecasting

Area of Science:

• Epidemiology
• Biostatistics
• Environmental Health

Background:

• Accurate dengue incidence forecasting is crucial for public health interventions.
• Statistical models must account for overdispersion, temporal dependence, and delayed environmental factors.
• Dengue hemorrhagic fever (DHF) poses a significant threat in endemic regions like Thailand.

Purpose of the Study:

• To develop and assess a Bayesian negative binomial dynamic regression model for monthly DHF incidence forecasting.
• To evaluate the impact of temporal dependence and delayed climatic factors on forecast accuracy.
• To investigate the role of spatial representativeness of climate data in model performance.

Main Methods:

• Development of a Bayesian negative binomial dynamic regression model.
• Inclusion of lagged dengue incidence for transmission persistence.
• Incorporation of maximum temperature and relative humidity for delayed climate effects.
• Model validation using posterior predictive checks and leave-one-out cross-validation (LOO-CV).

Main Results:

• The negative binomial model significantly outperformed Poisson models in the presence of overdispersion.
• Forecasting performance was critically dependent on the spatial representativeness of climate inputs.
• Models using climatically representative data produced more stable and robust out-of-sample forecasts.
• Marginal climate-dengue associations alone did not determine forecasting success.

Conclusions:

• The study highlights the importance of spatial representativeness in climate-informed disease modeling.
• Distinguishes between explanatory association and predictive utility in infectious disease forecasting.
• Provides practical guidance for developing effective climate-informed dengue early warning systems.