QLS Seminar Series - Sara Zapata-Marin
Understanding the spatio-temporal distribution of pollen in Toronto
Sara Zapata-Marin, McGill University
Tuesday November 15, 12-1pm
Zoom Link:聽
Abstract:聽Many people around the world suffer from respiratory allergies during the pollen season. Nowadays, some apps can tell us the pollen levels in a city based on the measurement of a single monitoring site. However, one monitoring site is not enough to describe what happens in the entire city.
My Ph.D. research focused on developing spatio-temporal models to investigate the distribution of different types of pollen across Toronto. Available data consist of daily and weekly measurements of pollen concentration collected between March and October 2018 across 18 fixed monitoring sites. Measurements of tree, weed, grass, and total pollen concentration were obtained daily for eleven sites and weekly for the other seven sites.
In the first part of the project, we modeled the weekly concentration for all pollen types. Instead of considering the temporal window with only positive values, a hurdle model was proposed to account for the high number of measurements equal to zero. This allowed us to estimate the probability of the presence of different pollen types at any given week. Additionally, a dynamic linear model was used to capture the city's weekly trend of pollen concentration.
In the second part of the project, rather than aggregating the data to the weekly scale, a temporal misalignment model was proposed to account for the difference in temporal scale across sites, borrowing strength from sites with available daily data.
Higher levels of tree pollen were found at the beginning of the pollen season, followed by an increase in grass and weed pollen as the season progressed. The predicted surfaces obtained in this project will help inform future health-related studies.