A forecasting method tested by York University could help Ontario public health units anticipate short-term changes in mosquito abundance – the primary vectors of West Nile virus – before they happen.
Using 22 years of mosquito surveillance and weather data from 34 Ontario public health units, the research led by PhD student Zhaowan Li tested different methods to understand what information is most useful for predicting mosquito populations.
The study, conducted with Professor Huaiping Zhu and Ziwang Deng, senior modeller and data scientist, found that relatively simple models can provide useful short-term changes in mosquito populations.
The goal, says Zhu, was to address current gaps in mosquito-borne disease prevention. While researchers have made progress with prediction models, producing credible short-term forecasts for individual health units remains a challenge.
Working through the Laboratory of Mathematical Parallel Systems and Centre for Disease Modelling, the Faculty of Science researchers designed and compared forecasts using recent mosquito counts with weather information, including weather data from different time periods.
They analyzed three modelling approaches – linear regression, neural networks and random forest regression – to determine how accurately each could predict mosquito abundance one week in advance.
In practice, a reliable one-week estimate could give public health units more time to adjust mosquito surveillance and control efforts.
Our strongest models accounted for about 69 per cent of the variation in weekly mosquito abundance and outperformed a simple persistence forecasts that assumed mosquito abundance would remain unchanged the following week," says Zhu. "These results give us a stronger foundation for anticipating where populations may be headed next.”
The researchers then tested the methods using 2022-23 data and found forecasting error was 14.4 per cent lower than predictions that assumed current mosquito levels would persist into the following week.
One of the study's key findings, Li notes, is that more complicated frameworks did not substantially improve predictions. Recent mosquito abundance was the strongest predictor of what would happen the following week, while recent weather conditions added useful information. Additionally, older weather data contributed less to the forecasts.
“Effective forecasting doesn’t necessarily require more data or more complicated models," Li says. "By using information already being collected, this kind of approach could give public health units earlier insight into changing mosquito populations and help them target surveillance and control efforts where they are needed most.”
The study, published in the Journal of Medical Entomology, suggests that a simple, scalable framework for short-term mosquito forecasting across Ontario could improve surveillance, resource allocation and targeted mosquito control, potentially helping to reduce the risks of West Nile virus.