Studying airborne pollen concentrations is an essential part of aerobiology owing to its important applications in allergology. A time-varying first order autoregressive (AR(1)) model able to describe the annual cycle of both the expectation and variance as well as the highly skewed probability distribution of daily ragweed pollen concentrations conditioned on previous-day pollen concentration values is developed. Confidence bands for forecasts obtained with these conditional lognormal distributions are analysed. The probability of exceeding specific pollen concentration thresholds is also addressed with the model based on a refinement of the AutoRegressive To Anything process. In order to have more accurate forecasts for the next-day pollen concentration level, eight meteorological variables influencing pollen concentrations are considered. Based on a procedure similar to the stepwise regression method, only one predictor has been retained, namely the daily mean temperature. Using root mean square error, the percentage variance of the ragweed pollen concentration level accounted for by this extended AR(1) model is 53. 5%, while the mean absolute error produced by the model is 32.2 pollen grains m-3. The probability of exceeding pollen concentration thresholds obtained from the conditional lognormal distributions under the extended AR(1) model fits well the observed exceedance events.
ASJC Scopus subject areas
- Atmospheric Science