1. Leaf senescence is a major event in a plant’s life history as autumn marks the end of the growing season. The optimal timing of leaf senescence is crucial to both, minimize risks of low temperature events and maximize carbon gain during the growing season. As abiotic conditions are currently changing at unprecedented rates, it is important to study how leaf senescence of different species is responding to these changes in order to forecast future growing season length and carbon sequestration potentials. In contrast to flowering phenology, data on autumn events is scarce and even more so for herbaceous than for woody plants, thus more information on this phenological stage is urgently needed. 2. We studied leaf senescence of 632 populations from 17 herbaceous species located along elevational gradients. We focussed on the beginning (5% of the population senesce, LS5) and peak (50% senesce, LS50) of leaf senescence. To see whether we can predict species-specific changes, we studied the link between LS5 and LS50 and flowering phenology as well as leaf functional traits related to plant performance. We looked at first and last flowering day and flowering duration as well as the traits specific leaf area, leaf dry matter content, area based leaf nitrogen and carbon content, carbon isotope discrimination (Δ13C), and the stomatal pore area index. 3. We found species-specific slopes of the beginning of leaf senescence along the elevational gradient. The peak of leaf senescence was uniformly delayed with increasing elevation across all species. Flowering phenology as well as leaf functional traits had a close relationship with leaf senescence and thus can be used to forecast species-specific responses to changes in abiotic conditions. High SLA and high leaf nitrogen were related to earlier senescence while high LDMC, high Δ13C and high SPI to later senescence. 4. Synthesis: The link between senescence, flowering phenology and plant functional traits will help to fine-tune predictions of future growing season length and ecosystem function. To date, most analyses are based on spring phenology and traits, for which data is more abundant than data on autumn senescence.

This is the phenological data used in the manuscript. Phenology was monitored weekly on the two mountains at each elevational band, i.e. every 100 m increase in elevation during the growth phase in two consecutive years (2012 and 2013). Up to three subpopulations per species, which were located on separate grasslands at least 50 m apart were monitored on each elevational band (1024 subpopulations in total). We monitored phenology on the population scale, as repeated measurements of the same, marked individuals of herbaceous plants in meadows are not feasible. Subpopulations were formed by at least 20 individuals; monitored population sizes as well as the area covered by the populations differed between species. At each census the percentage of each subpopulation showing leaf senescence displayed either as dead leaves or visual colouring was recorded. We selected two different levels of leaf senescence: the beginning of leaf senescence defined as the day when 5% of the leaves within a subpopulation showed visually observable leaf senescence (LS_5), and the day, where 50% of the leaves within a subpopulation showed visually observable leaf senescence (LS_50). For both, the day of the year is given. Equally, we here report day of first flowering (FFD), day of last flowering (LFD) and flowering duration (FD) observed on the same populations.

Not all subpopulations or species reached LS_50 during the time of observation, so they were only included in the LS_5 analysis. Additionally, not all species could be observed in both years and both elevational gradients.