1. Introduction
Late spring frost poses a major threat for vegetation’s development and may result in considerable environmental and economic losses [
1,
2]. Recently, a warm start of spring 2017 was followed by an incursion of cold Arctic air into Western and Central Europe, which seriously harmed prematurely grown vegetation [
3]. In general, the severity of such damage is linked to the lag between an onset of spring plant growth and a subsequent frost event [
4].
Due to the ongoing climatic change and consequent rising temperatures, both start of growing season and frost-free period are being shifted towards the beginning of the calendar year. According to Jeong et al. [
5], a growing season for temperate vegetation over the Northern Hemisphere has been starting earlier, the mean trend being about 2 days per decade in the 1982–2008 period. The prolonged growing season is linked to changes in the timing of plants’ phenophases. For example, Menzel et al. [
6] concluded that the average trend of spring/summer phenophase dates was −2.5 days per decade in Europe (phenophases tend to occur sooner). It should be noted, however, that these changes are species- and region-dependent [
7,
8]. According to Kolářová et al. [
9], who analysed 18 common tree species in Central Europe, the largest advancements in spring phenophases were observed for shorter-lived, early-successional species (e.g., Prunus spinosa or Robinia pseudoacacia). The largest shifts of phenophases towards the start of the year were observed in the northern parts of Central Europe, while advancements on the Mediterranean coastline were smallest (regarding apple trees) [
10].
A key question in spring frost risk assessment is whether the advancement of phenophases is analogous to the shift of last spring frost. Wypych et al. [
11] analysed trends of last spring frost dates over Central and Eastern Europe in the 1951–2010 period and showed that their magnitude is relatively variable over the domain. The western part of the area (roughly between 46–54 °N and 5–20 °E) had negative trends (last spring frost tends to occur sooner) of a magnitude ranging from –1 to −4 days per decade. This is in accordance with Bigler and Bugmann [
12], who showed a negative trend of last spring frosts since the 1980s in Switzerland. By contrast, the trends in Eastern Europe (Ukraine, Belarus, western parts of Russia) were indistinct or even positive.
Many authors concluded that spring frost damage on vegetation has been increasing over middle latitudes in Northern hemisphere. Liu et al. [
13] reported this phenomenon roughly across 43% of the hemisphere, especially in Europe. Kim et al. [
14], using satellite data for the USA, showed that spring frost damage is linked to lower vegetation growth in spring and consequent lower vegetation greenness in summer. In addition, Augspurger [
4] reported increased risk of spring frost damage, using 124 years of temperature records for the State of Illinois (USA). By contrast, Vitasse and Rebetez [
3] concluded that the risk of damaging frost events to vegetation has remained unchanged in the 1864–2017 period, over the lowlands of Switzerland and Germany, due to comparable shifts in an onset of spring plant growth and late spring frosts, implying that changes in spring frost risk vary among regions and species analysed. This was shown by Vitasse et al. [
15], who demonstrated that spring frost risk in Switzerland increased predominantly at stations located at elevations higher than 800 m a.s.l., while it remained mostly unchanged in lower altitudes.
Projections of changes in spring frost risk in a possible future climate are even more challenging. A probable decrease of spring frost risk was reported by Bennie et al. [
16], who focused on deciduous trees (Betula pubescens) in Finland. Using climate change projections combined with phenological modelling, Molitor et al. [
17] concluded that Luxembourg’s winegrowing region will be less exposed to dangerous spring frost. By contrast, Leolini et al. [
18] found an increased frequency of frost events at bud break in Central Europe for future scenarios. It should be noted, however, that these projections contain substantial uncertainties, most of which are related to the choice of climate model chains (greenhouse gas concentration scenario/global climate model/regional climate model/phenological model/estimated vegetation parameters [
2]). This is in accordance with Mosadale et al. [
19], who obtained opposite results when using different phenological models.
In this study, we endeavour to overcome uncertainties originating from a selection of climate model chains by developing a more robust approach. An onset of spring plant growth is estimated using temperature series only, which allows us to apply the definition within various data sets. The computed onsets of spring plant growth from temperature data only are evaluated against the Swiss spring index [
20], calculated from actual phenological observations. The main aim of the study is to analyse changes in spring frost risk in a possible future climate of Switzerland. We focus on the Aare river catchment, which is probably the most vulnerable Swiss area in terms of late spring frost due to its agricultural importance. Moreover, it was one of many regions struck by a severe spring frost in April 2017 [
3].
4. Discussion
Projecting changes in vegetation responses to a possible future climate is a challenging task that is accompanied by many uncertainties. The proposed method aims to provide a robust approach for assessing future spring frost risk for the complex Swiss terrain that contains several different vegetation zones, based on a widely available daily mean temperature only. The method was calibrated in 1951–2014 using the Swiss spring index constructed from phenological in-situ observation [
34]; however, it is possible that the relationship between temperature characteristics in spring and timing of vegetation’s development would be changed in the future climate [
36] because this link is probably not linear [
37]. Many authors reported that climate change is responsible for decreased temperature sensibility of bud break. For example, Wang et al. [
38] suggested that plants might be less likely to track climatic warming at locations with larger local spring temperature variance in order to avoid a frost risk and rely more on other cues, such as high chill requirements or photoperiods.
The 3.9 °C threshold value used in our method for identifying starts of spring corresponds to the growth-onset temperature identified by Breitenmoser et al. [
39] for mid-latitudinal trees, however, it has to be considered that the threshold value represents an average over the whole analysed domain. Therefore, the threshold value has to be regarded with caution, because it may vary between vegetation zones [
15]. Due to the size of the domain and its terrain heterogeneity, we employed daily mean temperature instead of commonly used daily minimum temperature when calculating the frost index. Low night-time temperature minima are often linked to geomorphologic features, such as altitude, shape of valleys, or exposition of slopes and may differ substantially between nearby locations [
40,
41]. The use of spatially more coherent daily mean temperature allows distinguishing of larger-scale frost events that are mostly related to an interruption of the prevailing westerly flow [
42] and an incursion of a cold air mass from northern or eastern directions [
43]. Cold northerly/easterly advection into Central and Western Europe is often linked to blocking anticyclones [
44].
Improper simulation of frequency and persistence of blocking anticyclones is one of the largest drawbacks of current climate models [
45,
46], and it influences projections of circulation-induced temperature and precipitation extremes in a future climate [
47]. Furthermore, Lhotka and Kyselý [
48] showed that climate models tend to have too-cold northerly advection in winter, and this deficiency may propagate also into a spring season. Although the RCMs used simulated the spring frost index relatively well in the historical climate, an analysis of frost events’ driving mechanisms is beyond the scope of this study. The 20CRv2 reanalysis did not capture the observed advancement of the spring start towards the beginning of the year, and the average date of spring start differed by 5 days compared to the observed data (considering the 1971–2000 period). Lorenz and Jacob [
49] found weaker temperature trends in the NCEP/NCAR reanalysis over Europe compared to E-OBS and CRU observed data sets. Inasmuch as the 20CRv2 reanalysis shares many features with the NCEP/NCAR reanalysis [
26,
50], the erroneous, weaker temperature trend is probably present also in 20CRv2, and it is most likely related to this discrepancy.
Although substantial progress has been made in understanding relationships between increasing temperatures and shifts of phenophases, exact physiological mechanisms are still a subject of broad discussions. Cong et al. [
51] reported that interannual variations in an onset of spring plant growth are extensively related to the number of chilling days over the Tibetan Plateau and suggested that continued future warming may lead to a deficiency in chilling and thus in changes in phenophase timing. An analogous phenomenon was reported in Switzerland by Asse et al. [
52], who concluded that warmer winters significantly delayed bud burst and flowering along the elevation gradient. The role of a chill requirement, however, was questioned by Güsewell et al. [
53], who found that reduced temperature sensitivity can result directly from spring warming alone. In addition, the importance of photoperiods for spring phenophases is not clear, and their roles are probably species- and region-dependent [
54,
55]. Another important factor is a timing of snowmelt, especially in regions with higher elevation [
56]. Finally, besides spring frost risk, other hazards such as heat waves and droughts or spread of pests [
57] should be taken into account when preparing complex climate change adaptation and mitigation strategies.