Abstract
Dengue fever is a mosquito-borne viral disease rapidly creating a significant global public health burden, particularly in urban areas of tropical and sub-tropical countries. Hydroclimatic variables, particularly local temperature, precipitation, relative humidity, and large-scale climate teleconnections, can influence the prevalence of dengue by impacting vector population development, viral replication, and human-mosquito interactions. Leveraging predictions of these variables at lead times of weeks to months can facilitate early warning system preparatory actions such as allocating funding, acquisition and preparation of medical supplies, or implementation of vector control strategies. We develop hydroclimate-based statistical forecast models for dengue virus (DENV) at 1-, 3-, and 6- month lead times for four cities across Colombia (Cali, Cúcuta, Medellín, and Leticia) and compare with standard autoregressive models conditioned on dengue case counts. Our results indicate that (a) hydroclimate-based models are particularly skillful at 3- and 6- month lead times when autoregressive models often fail, (b) sea surface temperatures are the most skillful predictor at 3- and 6- month leads and (c) application of hydroclimate models are most beneficial when average DENV incidence is low, autoregressive relationships are weak, but outbreaks may still occur.