[Paper Review] Dengue Seasonality and Non-Monotonic Response to Moisture: A Model-Data Analysis of Sri Lanka Incidence from 2011 to 2016

Milad Hooshyar, Caroline E. Wagner+6 more|arXiv (Cornell University)|Sep 6, 2020

Mosquito-borne diseases and control4 citations

TL;DR

This study proposes a hydrologically driven epidemiological model (HYSIR) that integrates rainfall seasonality and moisture memory to explain the two-month lag between rainfall peaks and dengue incidence in Sri Lanka. It reveals a non-monotonic relationship where dengue transmission initially increases with moisture but declines under very high water availability, improving seasonal peak prediction without ad hoc lag parameters.

ABSTRACT

Dengue fever impacts populations across the tropics. Dengue is caused by a mosquito transmitted flavivirus and its burden is projected to increase under future climate and development scenarios. The transmission process of dengue virus is strongly moderated by hydro-climatic conditions that impact the vector's life cycle and behavior. Here, we study the impact of rainfall seasonality and moisture availability on the monthly distribution of reported dengue cases in Sri Lanka. Through cluster analysis, we find an association between seasonal peaks of rainfall and dengue incidence with a two-month lag. We show that a hydrologically driven epidemiological model (HYSIR), which takes into account hydrologic memory in addition to the nonlinear dynamics of the transmission process, captures the two-month lag between rainfall and dengue cases seasonal peaks. Our analysis reveals a non-monotonic dependence of dengue cases on moisture, whereby an increase of cases with increasing moisture is followed by a reduction for very high levels of water availability. Improvement in prediction of the seasonal peaks in dengue incidence results from a seasonally varying dependence of transmission rate on water availability.

Motivation & Objective

• To understand the mechanistic link between rainfall seasonality and dengue incidence patterns in Sri Lanka from 2011 to 2016.
• To investigate whether hydrologic memory and nonlinear dynamics can explain the observed two-month lag between rainfall and dengue peaks.
• To assess the role of moisture availability in shaping dengue transmission dynamics beyond simple monotonic relationships.
• To improve seasonal dengue incidence prediction by incorporating seasonally varying transmission responses to water availability.

Proposed method

• A minimalist bucket model simulates water availability (w) with memory through a decay parameter ρ, capturing hydrologic persistence.
• The HYSIR model couples hydrology with an SIR-type disease transmission model, where transmission rate depends nonlinearly on water availability w.
• The transmission rate is modeled as a piecewise function: linear at low w with a seasonally varying slope ψ, and decreasing at high w to reflect habitat disruption.
• Model parameters are calibrated using observed monthly dengue case data and CHIRPS precipitation data at 0.05° spatial resolution.
• Cluster analysis identifies spatial and temporal patterns linking rainfall and dengue incidence, revealing consistent two-month lags.
• The model’s performance is evaluated by comparing simulated dengue incidence seasonality with observed data across multiple districts in Sri Lanka.

Experimental results

Research questions

• RQ1 How does rainfall seasonality in Sri Lanka correlate with the timing of dengue incidence peaks, and what accounts for the observed two-month lag?
• RQ2 Can a hydrologically informed model explain the seasonal dynamics of dengue incidence without relying on arbitrary lag parameters?
• RQ3 What is the nature of the relationship between water availability and dengue transmission rate, and does it exhibit non-monotonic behavior?
• RQ4 How does hydrologic memory (via parameter ρ) influence the emergence of seasonal dengue patterns in the model?
• RQ5 To what extent does a seasonally varying transmission rate response to moisture improve prediction accuracy compared to constant-response models?

Key findings

• A two-month lag between rainfall peaks and dengue incidence peaks is consistently observed across multiple districts in Sri Lanka, with cluster analysis confirming this temporal association.
• The HYSIR model successfully reproduces the two-month lag by integrating hydrologic memory and nonlinear transmission dynamics, eliminating the need for ad hoc lag parameters.
• A non-monotonic relationship between water availability and dengue transmission is identified: transmission increases with moisture up to a point, then declines under very high water availability.
• The decline in transmission at high moisture levels is attributed to habitat disruption from intense rainfall and flooding, which can destroy breeding sites and reduce mosquito survival.
• Seasonally varying transmission rate parameters (ψ) significantly improve model fit to observed dengue incidence, particularly in capturing the amplitude and timing of seasonal peaks.
• The model’s performance is robust across districts, with simulated incidence patterns closely matching observed data, especially when the transmission rate is allowed to vary seasonally.