Improving the reliability of a combined phenological time series by analyzing observation quality

Abstract

Collecting phenological data is a slow process. Although such data have been collected by a number of organizations, the reliability of these data is not known because the data-generating process cannot be repeated. No further observations to improve the reliability can be obtained. However, the data usually consist of several overlapping observation series and this overlap can be utilized to construct a combined phenological time series and to improve its reliability. We have developed two techniques for selecting the most reliable observations or observation series and thereby improve the reliability of the combined time series. Both techniques require that the method used to combine the separate phenological time series adjusts the individual series to eliminate possible systematic differences between them. A data set of bud burst in Betula pendula Roth collected in Central Finland during 1896–1955 was adjusted and used to test both techiques. Both techniques considerably improved the reliability of the combined time series; the mean of the confidence intervals of the annual means decreased by 12%. Despite the improvement in reliability, the resulting changes in the annual values of the combined time series were small, the largest change being 2.5 days. Removing outliers was the most effective method of improving reliability, i.e., it resulted in the greatest improvement with the smallest number of discarded observations.