| Popis: |
We developed a temperature sum model to predict the daily pollen release of alder, based on pollen data collected with pollen traps at seven locations in Finland over the years 2000–2014. We estimated the model parameters by minimizing the sum of squared errors (SSE) of the model, with weights that put more weight on binary recognition of daily presence or absence of pollen. The model results suggest that alder pollen ripens after a couple of warm days in February, while the whole pollen release period typically takes up to 4 weeks. We tested the model residuals against air humidity, precipitation and wind speed, but adding these meteorological features did not improve the model prediction capacity. Our model was able to predict the onset of pollen season with similar accuracy as models describing only the start of the pollen release period (average prediction error 8.3, median 5.0 days), while for the end of the pollen release period the accuracy of our predictions was not as good. We split the pollen data into odd and even years, and fitted our model separately to each half. Difference in the parameter values suggests a biennial behavior in the onset of pollen ripening, with almost two weeks of difference in the modeled starting date of the pollen development. Monte Carlo resampling of the observation data confirmed that the difference is not just a random anomaly in the data. peerReviewed |
