We present a linked model of plant productivity, plant phenology, snowmelt and soil thaw in order to estimate interannual variability of arctic plant phenology and its effects on plant productivity. The model is tested using 8 years of soil temperature data, and three years of bud break data of Betula nana. Because the factors that trigger the end of the growing season of arctic vegetation are less well known than those of the start of the growing season, three hypotheses were formulated and tested for their effects on productivity and its sensitivity to climate change; the hypothesised factors determining the end of the growing season were frost, photoperiod and periodic constraints.
The performance of the soil thermal model was good; both the onset of soil thaw in spring and the initiation of freezing in autumn were predicted correctly in most cases. The phenology model predicted the bud break date of Betula nana closely for the three different years. The soil thaw model predicted similar growing season start dates under current climate as the models based on sum of temperatures, but it made significantly different predictions under climate change scenarios, probably because of the non-linear interactions between snowmelt and soil thaw. The uncertainty about the driving factors for the end of the growing season, in turn, resulted in uncertainty in the interannual variability of the simulated annual gross primary productivity (GPP). The interannual variability ranged from − 25 to + 26% of the mean annual GPP for the frost hypothesis, from − 20 to + 20% for the photoperiod hypothesis and only from − 7 to + 7% for the periodic hypothesis. The different hypotheses also resulted in different sensitivity to climate change, with the frost hypothesis resulting in 30% higher annual GPP values than the periodic hypothesis when air temperatures were increased by 3 °C.