Standardising pollen monitoring: quantifying confidence intervals for measurements of airborne pollen concentration

摘要

Measurement of atmospheric pollen concentrations is inexact, yet pollen concentrations are universally reported without estimate of accuracy. These imprecise values are nevertheless used for modelling, forecasting and public health decision-making. Estimation of the variability in reported pollen concentrations would help resolve associations between weather and pollen aerobiology as well as associations between pollen exposure and health. For any given daily atmospheric pollen level, a statistical variability would be expected in the set of possible measures. This variability is introduced and compounded by many factors including human error, classification error or instrument variability, as well as variability derived from strategies used to count and scale the airborne pollen sample. Here, we performed numeric simulations of pollen deposition and modelled the variability in contemporary pollen density estimates. Statistical distribution of the mean and variance of these simulated counts was compared with an existing pollen count dataset. Both simulations and actual pollen data showed that a significant range of atmospheric pollen concentrations could be inferred from the same daily pollen collection. The range of possible concentrations varied both with the atmospheric pollen density and with the portion of the daily pollen sample that is counted. Furthermore, pollen concentration data were shown to be non-normal and heteroscedastic, which has implications for a variety of tests (e.g. ANOVA), for regression analysis, and for pollen forecasting and forecast verification. These results reinforce the importance of counting as much of the collected pollen impaction surface as feasible to minimise the uncertainty in reported pollen levels. The outcomes of this study suggest that confidence intervals for daily pollen concentrations should be reported.