Quantifying full phenological event distributions reveals simultaneous advances, temporal stability and delays in spring and autumn migration timing in long-distance migratory birds: Divergent changes in migration timing

Phenological changes in key seasonally expressed life-history traits occurring across periods of climatic and environmental change can cause temporal mismatches between interacting species, and thereby impact population and community dynamics. However, studies quantifying long-term phenological changes have commonly only measured variation occurring in spring, measured as the first or mean dates on which focal traits or events were observed. Few studies have considered seasonally paired events spanning spring and autumn or tested the key assumption that single convenient metrics accurately capture entire event distributions. We used 60 years (1955–2014) of daily bird migration census data from Fair Isle, Scotland, to comprehensively quantify the degree to which the full distributions of spring and autumn migration timing of 13 species of long-distance migratory bird changed across a period of substantial climatic and environmental change. In most species, mean spring and autumn migration dates changed little. However, the early migration phase (≤10th percentile date) commonly got earlier, while the late migration phase (≥90th percentile date) commonly got later. Consequently, species' total migration durations typically lengthened across years. Spring and autumn migration phenologies were not consistently correlated within or between years within species and hence were not tightly coupled. Furthermore, different metrics quantifying different aspects of migration phenology within seasons were not strongly cross-correlated, meaning that no single metric adequately described the full pattern of phenological change. These analyses therefore reveal complex patterns of simultaneous advancement, temporal stability and delay in spring and autumn migration phenologies, altering species' life-history structures. Additionally, they demonstrate that this complexity is only revealed if multiple metrics encompassing entire seasonal event distributions, rather than single metrics, are used to quantify phenological change. Existing evidence of long-term phenological changes detected using only one or two metrics should consequently be interpreted cautiously because divergent changes occurring simultaneously could potentially have remained undetected.