Climate predicts geographic and temporal variation in mosquito-borne disease dynamics on two continents

The study addresses how climate drives mosquito-borne disease dynamics and whether laboratory-measured, climate-sensitive mosquito traits can predict field dynamics across diverse contexts. Although climate is a major ecological driver, its effects are often nonlinear, interactive, and seemingly context-dependent, making field prediction difficult. Arboviral transmission by Aedes aegypti spans endemic to epidemic regimes, potentially due to climatic variation. The authors hypothesize that a climate-driven mechanistic model with limited calibration can capture key epidemic characteristics (number of outbreaks, peak timing, duration) across regions with different ecology and socio-demographics (Ecuador and Kenya), because Aedes aegypti life-history and vector-virus traits (development, survival, biting, extrinsic incubation) depend strongly on temperature and humidity, and rainfall influences larval habitat through multiple pathways.

Prior work shows Aedes aegypti and dengue often increase with warm, wet conditions in South America and sub-Saharan Africa, though African arboviruses (e.g., chikungunya) may peak in warm, dry conditions. Laboratory and mechanistic studies demonstrate temperature-dependent mosquito and viral traits with intermediate optima, and humidity positively affects survival due to desiccation risk. Rainfall relationships are complex: it can create larval habitat, flush containers at high intensity, or water-storage during drought can increase habitat and human–mosquito contact. A previous simulation study suggested temperature shapes timing and duration of outbreaks but not peak magnitude in climates suitable year-round. Epidemiological patterns also differ regionally: South America often has seasonal epidemics; many African settings have low-level year-round transmission. Existing mechanistic models incorporate mosquito life cycles and climate but have limited validation with vector and disease time series, prompting evaluation of which epidemic characteristics can be reliably captured across contexts.

Study design: Sites in Ecuador and Kenya were selected to span gradients of temperature, humidity, and rainfall, reflecting diverse ecological and socio-demographic conditions. Climate data: Daily mean temperature, relative humidity, and rainfall were recorded in situ (HOBO loggers, rain gauges); missing values were interpolated via regressions with nearby stations and historical databases. Saturation vapor pressure and saturation vapor pressure deficit (SVPD) were derived from temperature and relative humidity. Vector data: Aedes aegypti were sampled using Prokopack aspirators. Ecuador: adult mosquitoes collected every 1–2 weeks during multiple 4-month periods (2016–2018) and aggregated monthly. Kenya: monthly collections (2014–2018) across life stages (eggs, larvae, pupae reared to adults) with aggregation to monthly means per house. Human cases: Laboratory-confirmed dengue, chikungunya, and Zika cases were compiled monthly. Ecuador MoH data (2011–2018 for dengue; chikungunya and Zika subsets) and passive pediatric febrile surveillance in Kenya (2014–2018) with RT-PCR and serology. Model: A climate-driven SEI-SEIR compartmental model with mosquito compartments (susceptible, exposed, infectious) and human compartments (susceptible, exposed, infectious, recovered). Mosquito life-history and vector-virus traits (eggs/female/day, egg-to-adult survival, development rate, adult lifespan, biting rate, probability of infection, parasite/extrinsic incubation rate, infectiousness) were parameterized with thermal response functions (Brière or quadratic) from laboratory experiments at constant temperatures. Mosquito mortality was modeled as a function of temperature and humidity (via SVPD). Mosquito carrying capacity depended on temperature, humidity, and rainfall. Because the rainfall–habitat relationship is uncertain, three alternative rainfall functions for carrying capacity were tested per site: (1) inverse (abundance peaks at low/no rainfall, consistent with water-storage), (2) left-skewed Brière (increasing with rainfall to a threshold then decreasing due to flushing), and (3) symmetric quadratic (peak at intermediate rainfall). For each site, time lags between climate and outcomes (0–5 months) were explored. Outbreak definition: continuous periods with peak cases exceeding site mean plus one standard deviation. Validation: Pairwise correlations between model outputs and observed vector abundance and arboviral cases were computed for each site, selecting the rainfall function and lag maximizing correlation, with adjusted p values accounting for temporal autocorrelation (Modified Choi method). Sensitivity analyses assessed dependence on initial conditions and stability of predicted dynamics. Simulations were implemented in R (deSolve).

- The model reproduced key epidemic characteristics across sites: number of outbreaks (R² = 0.79, P < 0.01), peak timing (R² = 0.11, P < 0.01), and outbreak duration (R² = 0.51, P < 0.01). It did not reliably predict final outbreak size or maximum number of infections within outbreaks. - Vector dynamics: Model-predicted Aedes aegypti abundances significantly correlated with field-collected abundances in 8 sites, explaining 28–85% of site-level temporal variation (depending on site and lag). Performance tended to be better in Ecuador than Kenya. - Human disease dynamics: Model-predicted arboviral cases significantly correlated with laboratory-confirmed incidence in 7 of 8 sites, explaining up to 88% and on average about 48% of temporal variation among sites with significant correlations. - Rainfall mechanisms and lags: No single rainfall function fit all contexts. The inverse function best described dynamics most often in Kenya; Brière and quadratic relationships were often selected in Ecuador. Optimal lags for mosquito abundance were 0–1 months; for human cases, lags of 3–4 months were most common (at least 2 months in most sites). - Predictability mediators for vectors: Higher predictability was associated with a smaller proportion of the population under 5 years (R² = 0.89, P < 0.01), higher prevalence of piped water (R² = 0.76, P < 0.01), lower mean temperature within the studied range (22–28 °C; R² = 0.63, P < 0.05), and more cement housing (R² = 0.69, P < 0.05). Predictability declined as temperatures approached the thermal optimum for transmission (~29 °C). Socioeconomic factors examined did not explain variability in human case correlations. - Regime differences: Coastal sites with stronger seasonal climate cycles exhibited more regular seasonal epidemics; inland sites with greater day-to-day variability showed more irregular transmission patterns.

Findings support the central hypothesis that climate-driven, trait-based mechanistic models can capture core epidemic characteristics across ecologically and socio-demographically distinct settings without extensive calibration. By grounding transmission in temperature-, humidity-, and rainfall-dependent mosquito and viral traits, the model reconciles context-dependent observations across continents: it reproduces when outbreaks occur and how long they last, even if magnitude depends more on host susceptibility than on climate when conditions are suitable year-round. The study highlights multiple rainfall-driven pathways and time scales influencing transmission, necessitating flexible rainfall functions and lags. Differences between coastal (seasonal epidemics) and inland (variable endemic) regimes underscore how climate seasonality versus variability modulates outbreak regularity. Model predictability for vectors is enhanced in contexts with infrastructural features (piped water, cement housing) that likely reduce human–mosquito contact or modify microclimate, and diminishes near thermal optima where biological and behavioral nonlinearities may be strongest. These insights can guide targeted interventions (e.g., prioritization based on likely outbreak frequency and timing) and improve climate-informed public health planning.

A climate-driven SEI-SEIR model, parameterized with laboratory-derived mosquito and viral trait responses to temperature, humidity, and rainfall, successfully reproduces the number, timing, and duration of arboviral outbreaks and captures a substantial fraction of observed spatiotemporal variation in vectors and cases across Ecuador and Kenya. The work demonstrates that limited-calibration mechanistic models can generalize across diverse settings and be useful for prioritizing surveillance and interventions. Future research should incorporate: (i) vector control activities and infrastructure, (ii) immunological dynamics including dengue serotype interactions and cross-reactivity among arboviruses, (iii) age-structured susceptibility and contact heterogeneity, (iv) human behavior and mobility, and (v) pathogen-specific parameters to model dengue, chikungunya, and Zika separately. Improved characterization of rainfall–habitat functions across time scales and microclimatic effects of housing will further enhance predictive performance and climate change projections.

- Magnitude of outbreaks and maximum infections were not well predicted, likely due to host susceptibility dynamics not explicitly modeled. - Limited data for chikungunya and Zika precluded pathogen-specific validation; most analyses were driven by dengue. - Kenyan human case data derived from pediatric outpatient passive surveillance may not represent all age groups or community transmission. - Rainfall–carrying capacity relationships were assumed via three generic functional forms; true relationships may be site-specific and time-scale dependent. - Laboratory-derived trait data (constant temperatures) may not fully capture effects of fluctuating microclimates or behavioral responses. - Socio-environmental factors such as vector control, water-storage practices, and infrastructure were not explicitly modeled, potentially contributing to residual variability.