The year 1980 was selected as a break point for these data as it has previously been identified by the Intergovernmental Panel on Climate Change as the year after which global land temperatures rose consistently above the 1961–1990 normal . We then tested for significant differences in mean estimated frost risk between historical and recent time periods across all taxa by conducting a comparison of the pre- 1980 and the 1980–2015 frost risk estimated for each species using a Wilcoxon paired rank test. It should be noted that some late-spring and summer-flowering species were not exposed to frost within the duration of this study, or were exposed to frost only rarely. However, to avoid the inherent biases that would result from selectively including only those species that were exposed to frost , all species were included in this analysis regardless of frost exposure. The number of specimens of each species that were collected prior to 1980 often differed from the number collected post-1979, as did the mean frost date normals of the locations from which those specimens were collected. This difference in sampling had the potential to bias estimates of the change in frost risk between these two periods. We therefore conducted a multiple regression to test whether significant differences in frost risk between pre-1980 and 1980–2015 samples persisted after controlling for variation in the mean flowering dates among species, the number of specimens of each species that were collected pre-1980 and post-1979, and the mean normal frost dates of the locations from which specimens were collected pre-1980 and post-1979. In this analysis, frost risk was the dependent variable, plant pots with drainage while species-specific mean flowering dates, species-specific mean frost date normals , and the time period of collection were treated as independent variables.
Pre- 1980 values for each species were distinguished from post-1979 values using a categorical variable coded as 0 for pre-1980 values, and 1 for post-1979 values .To determine whether clades differ with respect to mean frost risk or either the magnitude or direction of the change in their frost risk in response to recent climate change, we estimated Bloomberg’s K, a metric of phylogenetic conservatism for both the pre-1980 frost risk for each species and the magnitude and direction of changes in frost risk post-1979. Phylogenetic distances used in this study were calculated using angiosperm nodes aged according to the V.Phylomaker package in R , which was also used to estimate Blomberg’s K. Note that all genera used in this study were present in this tree, and that all genera were considered to be polytomies.To evaluate the relationship between frost risk and short-term climate deviations, we first calculated the mean frost risk for all species using a 20-year moving window average calculated at 1-year increments from 1920 through 2015. To eliminate those taxa for which sample density was too low to produce robust estimates of frost risk throughout the entire time series, only those taxa that were represented by a minimum of 40 specimens within all 20-year windows were included in this analysis. A total of 268 species distributed throughout North America met this criterion. Overall frost risk within each 20-year window was then assessed by calculating the mean estimated frost risk across these 268 species within that time period. For each specimen that fell within a given 20-year window, we also calculated its annual deviation from the local normal date of last frost at the location in which it was collected . For each species, we then calculated both the frost risk and the mean annual deviation from normal dates of last frost among all the specimens represented within each 20-year window. Within each successive 20-year window, we then calculated the mean frost risk and mean annual deviation from normal dates of last frost among all species.
We predicted that, among the 20-year windows, as the mean deviation between current and historical last frost dates increases , the mean frost risk of the species in each successive 20-year window would also decline. To test for the effects of annual deviations in local dates of last frost on frost risk, we conducted a linear regression of the mean frost risk within each 20-year window on the mean annual deviation in the date of last frost within that 20-year window. To determine whether the relationship between frost risk and short-term climate deviations persisted across regions of North America characterized by both early and late dates of last frost, we separated taxa into two groups: those for which the mean last frost date normal of the collection sites was earlier than the median last frost date normals across the entire dataset and those for which the mean late frost date normal of the collection sites was later than the median last frost date normal. Having separated taxa that were predominantly sampled from locations characterized by early versus late last frost dates, we conducted the above analyses separately on each group of taxa.Across all 1,653 species that were well documented both pre-1980 and post-1979, mean frost risk has declined significantly since 1979, from 12.1% to 8.7% , representing a ~28% reduction in frost risk. Moreover, the number of species that experience minimal frost risk has increased from 234 species pre-1980 to 291 species post-1979 . Of the 66.1% of species that exhibited a reduction in frost risk post-1979, the mean frost risk declined from 16.0% to 9.7%. Conversely, of the 24.5% of species that exhibited a post-1979 increase in frost risk over this pre-1980 frost risk, mean frost risk increased from 6.3% to 9.3% , with only 77 species exhibiting a net increase in absolute frost risk greater than 5% . Median rates of directional change in dates of last frost over time were 3.3 times greater than rates of directional changes in flowering time among species . Consistently, while both frost dates and mean flowering dates among most species were found to occur earlier post-1979 than pre-1980, differences in the mean date of last frost between reference periods were of greater magnitude than those in flowering phenology .
Temporal changes in frost risk during this period are clearly associated with annual deviations from normal dates of last frost . From 1970 to 2015, as the annual dates of last frost have generally become progressively earlier than the 1901–1990 mean last frost date, the mean frost risk has declined. Moreover, among 20-year moving windows from 1920 to 2015, advances in the frost-free date explain 65% of the observed variation in mean frost risk . When examined separately, this pattern remained consistent among taxa occupying regions with late frosts and those occupying regions with early frosts, although late-frost regions experienced slightly greater reductions in frost risk during years experiencing earlier frosts . Analysis of 801 weather stations across North America revealed that rates of change over time for the date of last occurrence of 0°C temperatures were highly correlated to changes in last occurrences for −1°C, −2°C, −3°C, and −4°C temperatures . The magnitude of change over time in last 0°C dates closely matched that of changes for last −1°C and −2°C dates of occurrence . In turn, directional changes in dates of last frost defined using −3°C, and −4°C were of moderately lesser magnitude than changes in 0°C dates of last frost . Furthermore, advances in the date of last frost from 1920 to 2015 have typically outpaced advances in flowering phenology, despite advances in both frost dates and mean flowering dates among most species. We recognize that, for any given species, the frost risk during any particular time period could have been overestimated if the species was disproportionately sampled from sites where phenological advancement exceeded the advancement of the date of last frost, or underestimated if the species was disproportionately sampled from sites where the advancement of the date of last frost exceeded phenological advancement. However, the fact that these results remained consistent across so many taxa, and remained similar even when taxa occupying relatively warm regions and those occupying relatively cold regions were examined separately, indicates that such biases did not contribute significantly to the observed temporal changes in frost risk. Moreover, by constructing models in which we controlled statistically for spatial variation in the climate normals of each species’ specimens sampled before 1980 versus after 1979, plastic plants pots we further minimized this possibility. Thus, although it is well documented that early-flowering species, which flower in closest proximity to the last spring frost, are typically more phenologically sensitive to local temperature and show more rapid advances under warming conditions than late-flowering species , this study indicates that early-flowering species do not typically advance more rapidly than the date of last frost throughout North America. This pattern is relatively consistent among taxa, with less than 25% of sampled taxa exhibiting increases in frost risk post-1980. However, these results also indicate that, while significant phylogenetic signal for exposure to frost exists among the taxa sampled here, the degree to which frost risk is conserved among closely related taxa is minimal. Instead, similarities in frost risk among related taxa are more likely to be the result of convergent evolution in traits that are correlated with attributes that lead to exposure to frost rather than to strong phylogenetic conservatism . Thus, it does not appear that there are systematic differences among angiosperm clades with respect to the effect of recent warming on frost risk. Similarly, while a higher proportion of invasive species was observed in relatively warm regions , no systematic differences in frost risk were apparent between native and invasive taxa after controlling for differences in long-term frost dates across their ranges . Thus, although previous studies have indicated that invasive species often exhibit more responsive phenologies and weaker chilling requirements than their native counterparts, this study indicates that more successful avoidance of frost risk is unlikely to be a systematic factor that determines the invasive ability of most exotics. Dates of last frost have typically advanced more rapidly than dates of flowering for most of the 1,653 North American species sampled in this study , and temporal advances in the mean date of last frost are closely associated to reductions in frost risk . These results are supported by previous examinations of exposure to frost by growing leaves among North American tree species by Zohner et al. who also found reduced frost exposure in association with recent warming. Collectively, these results demonstrate that the risk of frost exposure and damage to both reproductive and vegetative tissues are systematically decreasing across North America. The ecological implications of these reductions in frost risk may be complicated by variation among species in their resilience to frost-related damage and in the degree to which frost damage to floral tissues may reduce their reproductive success, both of which remain largely unknown for most species. We were not able to directly quantify changes in frost risk defined using temperature thresholds other than 0°C because, to our knowledge, no gridded products provide dates of last frost defined using a lower temperature threshold at high spatial resolutions and spanning our study period . However, weather station data revealed that rates of temporal change in dates of last frost defined using different temperature thresholds were highly correlated and of similar magnitude. Temporal changes in dates of last frost defined using thresholds of −3°C and −4°C were moderately lower in magnitude than rates of change for 0°C dates of last frost . However, given that rates of change over time in dates of last frost at 0°C were more than three times greater than those in flowering time , the moderately slower rate of change of −3°C and −4°C dates of last frost likely outpaced changes in flowering time as well. Therefore, although the identity and exact proportion of taxa experiencing changes in frost risk over time might vary when using freezing temperatures more severe than 0°C to define frost, the general trend toward lower frost risk over time is likely to be robust to different definitions of the temperatures that is likely to induce tissue damage. In addition to direct effects of frost damage on vulnerable vegetative or reproductive tissues, the effects of these changes in frost risk on the survival and local abundances of many species will likely be mediated by a variety of factors.