A novel statistical framework for exploring the population dynamics and seasonality of mosquito populations

Charles Whittaker, Peter Winskill, Marianne Sinka, Samuel Pironon, Claire Massey, Daniel J. Weiss, Huong Lan Thi Nguyen, Peter W. Gething, Ashwani Kumar, Azra C. Ghani, Samir BhattProceedings of the Royal Society B Biological Sciences 2022 · DOI

TL;DR

This paper introduces a novel statistical framework to analyze and classify the complex temporal patterns of mosquito populations. By applying this framework to extensive Indian mosquito data, the researchers identified four distinct "dynamical archetypes" of mosquito seasonality, revealing that these patterns are driven by a combination of species-specific traits and a diverse set of environmental factors, not just rainfall.

What Problem Does This Paper Address?

Controlling vector-borne diseases like malaria hinges on understanding the population dynamics of disease-carrying mosquitoes. These populations fluctuate significantly over time, and their seasonality (when they peak and decline) is crucial for timing interventions like insecticide spraying or preventative medication. However, most previous studies focused on single locations or species, making it difficult to identify generalizable patterns or understand the complex interplay of factors driving these dynamics. The key challenge is to move beyond localized observations and develop a systematic way to classify and explain the diverse temporal behaviors of mosquito populations across different species and environments.

Methods

The researchers developed a novel statistical framework with several key components:

  • Systematic Review and Data Collation: They conducted a comprehensive literature search to identify and extract entomological data (mosquito catch data) from India, focusing on time series with monthly or finer temporal resolution spanning at least 12 months. This resulted in 272 time series across seven mosquito species complexes from 117 locations.
  • Time Series Smoothing and Interpolation: Raw mosquito catch data is often noisy. They used a Gaussian Process model with a Negative Binomial likelihood to smooth these time series, accounting for the overdispersion common in count data. A periodic kernel function was employed to capture the inherent seasonal or annual repeating patterns.
  • Time Series Characterization and Clustering: To classify the temporal patterns, they calculated a suite of summary statistics for each smoothed time series. These statistics captured aspects like the degree of seasonality, the timing of peaks, and the overall distribution of catches. Principal Component Analysis (PCA) was then used to reduce the dimensionality of these features, followed by k-means clustering to group time series with similar temporal properties into distinct "dynamical archetypes."
  • Statistical Modeling of Drivers: A multinomial logistic regression model was used to link the identified "dynamical archetypes" to a range of environmental variables (rainfall, temperature, land cover, water bodies, etc.) and species complex identity. This allowed them to quantify the association between specific environmental factors and each archetype.

Datasets

The primary dataset used was a compilation of 272 time series of mosquito catch data from 117 locations across mainland India. This data was extracted from a systematic review of the Indian entomological literature. Environmental data used included:

  • Rainfall: Time-period and location-specific data from the Climate Hazards Group Infrared Precipitation With Stations (CHIRPS) dataset.
  • Temperature: BioClimatic variables derived from monthly rainfall and temperature satellite data.
  • Aridity: Various measures of aridity.
  • Water Bodies: Covariates describing the seasonality and extent of water bodies.
  • Land Cover: Data from land cover datasets.
  • Other Environmental Variables: A total of 25 environmental covariates were used.

The paper also mentions integrating these results with spatial predictions of mosquito species complex presence/absence, drawing on data from the Humbug Project.

Key Insights and Ideas

  • Beyond Rainfall: The study strongly suggests that rainfall alone is insufficient to explain mosquito population dynamics. While rainfall is a significant driver, its influence varies greatly, and other factors play crucial roles.
  • Four "Dynamical Archetypes": Mosquito seasonality isn't a single phenomenon. The framework successfully clustered diverse temporal patterns into four distinct archetypes: monsoon-peaking, bimodal, dry-season peaking, and perennial. This provides a powerful way to categorize and compare mosquito behaviors.
  • Species-Specific and Environment-Specific Interactions: The research highlights that both the mosquito species itself and its local environment are critical. Some species (like Anopheles fluviatilis s.l.) show consistent dry-season peaks regardless of location, while others (like Anopheles culicifacies s.l.) exhibit much greater plasticity, adapting their dynamics to local conditions.
  • Urbanicity as a Driver: Urban environments were found to be positively associated with monsoon-peaking dynamics, likely due to the abundance of artificial water-holding structures that can serve as breeding sites.
  • Hydrology Matters for Perennial Dynamics: The presence of permanent water bodies and flow accumulation were strongly linked to perennial mosquito abundance, suggesting that local hydrological features can decouple mosquito populations from strict rainfall seasonality.

Key Results

  • The framework successfully identified and classified mosquito population dynamics into four distinct "dynamical archetypes."
  • Significant variation in seasonality was observed both between different mosquito species complexes and across different locations within India.
  • Environmental factors like temperature seasonality, rainfall seasonality, proximity to water bodies, and urbanicity were found to be strongly associated with specific dynamical archetypes. For instance, perennial dynamics (Cluster 4) were strongly associated with water bodies and negatively associated with temperature and rain seasonality.
  • Predictive maps generated using the framework show distinct spatial distributions for each dynamical archetype across India, indicating where each type of mosquito seasonality is most likely to occur.

Limitations and Future Work

  • Sibling Species Disaggregation: The study acknowledges a limitation in its inability to disaggregate time series by sibling species within complexes (e.g., Anopheles culicifacies s.l.). These sibling species can have different habitat preferences and thus different temporal dynamics, which could explain some of the observed variation.
  • Methodological Heterogeneity: The collated data came from studies using various mosquito sampling methods (e.g., landing catch, resting collections, light traps). While efforts were made to normalize catch data, residual biases due to these methodological differences might exist.
  • Spatio-Temporal Averaging: Environmental covariates were often averaged over study areas and sometimes across multiple years, potentially obscuring fine-scale inter-annual variations that could influence population dynamics.
  • Limited Epidemiological Data: The study focused on entomological data and could not fully integrate epidemiological metrics like sporozoite positivity or malaria incidence, which would provide a more complete picture of disease transmission risk.

Future work could involve integrating these findings with malaria case data to better understand the entomological inoculation rate (EIR) and exploring the dynamics of other mosquito species relevant to public health. Further research into the specific contributions of sibling species within complexes would also be valuable.

Summary updated

Abstract

Understanding the temporal dynamics of mosquito populations underlying vector-borne disease transmission is key to optimizing control strategies. Many questions remain surrounding the drivers of these dynamics and how they vary between species-questions rarely answerable from individual entomological studies (that typically focus on a single location or species). We develop a novel statistical framework enabling identification and classification of time series with similar temporal properties, and use this framework to systematically explore variation in population dynamics and seasonality in anopheline mosquito time series catch data spanning seven species, 40 years and 117 locations across mainland India. Our analyses reveal pronounced variation in dynamics across locations and between species in the extent of seasonality and timing of seasonal peaks. However, we show that these diverse dynamics can be clustered into four 'dynamical archetypes', each characterized by distinct temporal properties and associated with a largely unique set of environmental factors. Our results highlight that a range of environmental factors including rainfall, temperature, proximity to static water bodies and patterns of land use (particularly urbanicity) shape the dynamics and seasonality of mosquito populations, and provide a generically applicable framework to better identify and understand patterns of seasonal variation in vectors relevant to public health.
BibTeX
@inproceedings{whittaker2022novel,
  author = {Charles Whittaker and Peter Winskill and Marianne Sinka and Samuel Pironon and Claire Massey and Daniel J. Weiss and Huong Lan Thi Nguyen and Peter W. Gething and Ashwani Kumar and Azra C. Ghani and Samir Bhatt},
  title = {{A novel statistical framework for exploring the population dynamics and seasonality of mosquito populations}},
  booktitle = {Proceedings of the Royal Society B Biological Sciences},
  year = {2022},
  doi = {10.1098/rspb.2022.0089},
}
Figure