Herbarium records indicate variation in bloom-time sensitivity to temperature across a geographically diverse region.

Anthropogenic warming's effects on phenology across environmental and temporal gradients are well recognized. Long-term phenological monitoring data are often limited in duration and geographic scope, but recent efforts to digitize herbaria collections make it possible to reliably reconstruct historic flowering phenology across broad geographic scales and multiple species, lending to an increased understanding of community response to climate change. In this study, we examined collection dates (1901 to 2015) of 8540 flowering specimens from 39 native species in the Pacific Northwest (PNW) region of North America. We hypothesized that flowering phenology would be sensitive to temperature but that sensitivity would vary depending on blooming season and geographic range position. As expected, we found that early-season bloomers are more sensitive to temperature than later-season bloomers. Sensitivity to temperature was significantly greater at low elevations and in the maritime (western) portion of the PNW than at higher elevations and in the eastern interior, respectively. The elevational and longitudinal effects on flowering sensitivity reflect spring "arriving" earlier at low elevations and in the maritime portion of the PNW. These results demonstrate that phenological responses to warming vary substantially across climatically diverse regions, warranting careful and nuanced consideration of climate warming's effects on plant phenology.

Keywords: Phenology; Pacific Northwest; Herbarium specimens; Climate change; Phenological sensitivity; Geographic gradients

Electronic supplementary material The online version of this article (10.1007/s00484-020-01877-1) contains supplementary material, which is available to authorized users.

Variable responses of phenology (the timing of life history events) (Walther et al. [29]; Parmesan [17]; Cleland et al. [2]; Rosenzweig et al. [26]) to anthropogenic warming across environmental (Parmesan [18]; Inouye [6]) and temporal gradients (Menzel and Fabian [11]; Wolkovich et al. [34]) are well recognized. For example, across gradients of latitude, plant species closer to the poles, where anthropogenic warming is most pronounced (Socker et al. [27]), are experiencing the most significant shifts in phenology (Parmesan [18]; Prevéy et al. [21]). At the upper reaches of elevational gradients, progressively earlier snowmelt dates (Mote et al. [13]) correspond to greater shifts in flowering phenology (Inouye [6]) and leaf-out (Vitasse et al. [28]). Across seasonal gradients, early-season flowering plant species (spring bloomers) are more sensitive to warming temperatures than later-season bloomers (Menzel and Fabian [11]; Wolkovich et al. [34]) (but see Prevéy et al. [21]). Gradients of longitude, however, are often deemphasized when studying climate change's effects on biotic communities (but see Myking and Skrøppa [14]; Hamilton and Aitken [4]; Wang et al. [30]).

The primary factor controlling spring phenology of temperate and boreal vegetation across the Northern Hemisphere is temperature (Zhang et al. [35]). While it is widely recognized that plant phenology is shifting earlier more rapidly at high elevations and latitudes (Parmesan [18]; Inouye [6]; Vitasse et al. [28]), it is less clear if species and populations at these locations have a greater sensitivity to warming temperatures than those at lower elevations and latitudes, or if the rapid shift in phenology is primarily a response to the degree of warming that has taken place at these locations. One way to answer this question is through the calculation of temperature sensitivity, the change in phenology per unit change in temperature (Matsumoto et al. [9], Menzel et al. [12], Wolkovich et al. [34], Lapenis et al. [8], Kopp and Cleland [7], Wang et al. [30], [31], Park et al. [16], Prevéy et al. [21]) of multiple species distributed over a wide region. For example, in the Northern Hemisphere, Wang et al. ([30]) used satellite-generated time-series normalized difference vegetation index (NDVI) data and geographically corresponding climate data to calculate temperature sensitivity of vegetation green-up. However, for flowering phenology, there do not exist long-term monitoring stations that have the wide spatial and temporal coverage needed to calculate phenological sensitivity to temperature that would reflect fine-scale differences across gradients of elevation, latitude, longitude, and seasonality. Herbarium collections can help fill this need.

Large-scale digitization of herbarium collections has opened the door to broader analyses of species responses to climate change across expansive geographic and temporal scales (Hart et al. [5]; Matthews and Mazer [10]; Willis et al. [33]; Park et al. [16]; Park and Mazer [15]). For example, herbaria specimens have been used to examine long-term trends in a single species' flowering phenology across a region from northern California to the USA-Canada border (Matthews and Mazer [10]). Recently, phenological sensitivity of a mix of native and introduced species' was examined using herbarium specimens matched with coarsely calculated mean climate data in the eastern USA (Park et al. [16]). The ability to simultaneously examine multiple species over expansive, topographically diverse regions will provide greater clarity of fine-scale variation in phenological response patterns to past and future anthropogenic warming.

North America's Pacific Northwest (PNW) region (Fig. 1) contains dense lowland coastal forests in the west and more open montane forests in the interior east. Mean annual temperatures and seasonal temperature variability in the region vary across elevation, latitude, and longitude, producing a diverse canvas to investigate the effects of geographic range position on the sensitivity of flowering phenology to temperature (i.e., change in collection date per °C). Aggregation of digitized herbaria collections across the PNW has produced a robust online database of publicly accessible plant specimens via the Consortium of Pacific Northwest Herbaria (www.pnwherbaria.org). When data gleaned from these specimens (i.e., specimen phenology, specimen collection date, and location information) are paired with extrapolated localized historic climate data (Wang et al. [32]), it is possible to efficiently and reliably (Robbirt et al. [23]) examine the phenological sensitivity of multiple species to spring temperatures across wide temporal and geographic scales.

Graph: Fig. 1 Delineation of the study region. Herbarium collection records were gathered from the Consortium of Pacific Northwest Herbaria online database (http://www.pnwherbaria.org) in a region bounded to the west by the Pacific Ocean and to the east by roughly the eastern edge of the Rocky Mountains. The southern boundary was 40° N and the northern boundary was 57° N

In this study, we investigated if collection dates (a proxy for flowering time) between 1901 and 2015 of plant specimens in the PNW were sensitive to spring temperatures. Consistent with global responses to recent climate warming (Walther et al. [29]; Parmesan and Yohe [19]; Root et al. [24]; Parmesan [17], [18]; Cleland et al. [2]; Inouye [6]; Wolkovich et al. [34]; Vitasse et al. [28]; Prevéy et al. [21]), we expected that day of year (DOY) of collection of flowering specimens would be sensitive to spring temperatures (i.e., for every 1 °C increase in temperature, date of collection would occur earlier). Further, we expected that spring-blooming species would be more sensitive to climate warming than species blooming later in the season. Finally, we expected the level of sensitivity would vary across the region, but increase with increasing latitude and elevation.

To address our hypotheses, we selected native perennial species (Table 1 and Appendix S1: Fig. S1) that produce conspicuous flowers, such that phenology at the time of collection could be reliably determined from digitized herbarium sheets. Records were gathered from the Consortium of Pacific Northwest Herbaria (CPNWH) online database (www.pnwherbaria.org) in a region bounded to the west by the Pacific Ocean and to the east by roughly the eastern edge of the Rocky Mountains. We set these bounds in order to have a robust study region that conforms to the traditional definition of the PNW while also having adequate coverage by the CPNWH database. The southern boundary was 40° N and the northern boundary was 57° N (Fig. 1). We excluded specimens that were not in flower (i.e., vegetative or in fruit) or contained both fruits and flowers at the time of collection. This was done by utilizing determinations assigned by source herbaria or by visually assessing specimens that had not previously been classified. Further, we eliminated any duplicates, here defined as specimens from the same species, collected on the same date and from the same location, and then distributed across multiple herbaria. We manually verified specimen collection locations for specimens lacking geographic coordinates (Global Positioning System (GPS) or Universal Transverse Mercator (UTM)) and/or elevation using specimen collection notes and Google Earth. Specimens with questionable collection location information (e.g., GPS coordinates rounded to the nearest degree, broad or ambiguous location descriptions, and no elevation information) were eliminated from our analysis. Species with 100 or more records remaining after the above filters were applied were retained for analysis and those with less were discarded.

Sensitivity and mean location data for the 39 species investigated across the study region

We next gathered localized historic climate data (4 km × 4 km resolution) extrapolated by ClimateNA (Wang et al. [32]) for each specimen in the year of its collection at its geographic point of collection. Regional temperature shift between 1901 and 2015 was determined by subtracting the average annual temperature for each collection location (for the specific year of collection) from 1961 to 1990 mean for that location to produce location and year-specific temperature anomalies. The 1961–1990 baseline is a useful reference ecologically since it precedes significant anthropogenic warming (Socker et al. [27]). These anomalies were then regressed across the period of the study (1901–2015) to determine regional temperature shift.

To allow for the comparison of phenological response to climate warming across a broad region and multiple species, we chose to assess phenological change as a measure of species' sensitivity to climate warming. To test if day of year (DOY) was sensitive to temperatures experienced during the 3 months preceding a species mean collection DOY, we calculated the average temperature for that 3-month period in the year of collection and then subtracted the 1961–1990 mean value for that 3-month period at that specimen's specific collection location. For example, if a species had a study-wide mean collection date of June 5, we used temperature data for March, April, and May. We then conducted a linear regression between DOY and temperature anomaly to produce a sensitivity value for each individual species, as measured by the slope of the regression line. Negative slope values indicate advancement of flowering date per 1 °C increase in temperature while positive slopes signify delays. Once sensitivity values were calculated for each species (n = 39), we performed species-specific linear regressions between a species' mean DOY of collection, mean latitude, mean longitude, and mean elevation (independent variables) and sensitivity values (dependent variable) to determine if sensitivity is dependent on species flowering time. Next, to determine if there were interactions between temporal and geographic positions, we performed multiple linear regression with species sensitivity as the dependent variable and mean elevation, latitude, longitude, and DOY as the independent variables. We then performed an ANOVA on the resulting model to test the level of significance for each factor. All statistics were performed using R vers. 3.4.4 (R Core Development Team [22]).

Our conservative filtering protocol resulted in a dataset of 8540 specimens from 39 species. Each species had at least 100 specimens (Table 1). Based on climate estimates (Wang et al. [32]) for each collection location and year, mean annual study region temperature increased 0.12 °C per decade between 1901 and 2015 (F1,8538 = 1752, p < 0.001, r2 = 0.170; Fig. 2), a rate that is similar to the global rate (Socker et al. [27]). Mean sensitivity across the 39 species was − 4.4 (± 1.5) days per 1 °C increase in temperature in the 3 months preceding mean species collection date, ranging from − 9.3 days (Rubus ursinus) to + 0.3 days (Gentianella amarella) (Table 1). Earlier flowering species were more sensitive to warming temperatures than late bloomers, with sensitivity decreasing 1.4 days for every 30-day delay in date of flowering (F1,37 = 8.782, p = 0.005, r2 = 0.192; Fig. 3a). Sensitivity decreased 2.2 days for every 1000-m increase in elevation (F1,37 = 9.250, p = 0.004, r2 = 0.200; Fig. 3b). Contrary to our predictions, there was no effect of latitude on sensitivity (F1,37 = 0.135, p = 0.715, r2 = 0.004). Westwardly distributed species closer to the Pacific coast were more sensitive to warming than eastern interior species, with sensitivity decreasing 0.9 days for every 1° eastward shift in mean longitude (F1,37 = 14.94, p < 0.001, r2 = 0.288; Fig. 3c). Our ANOVA test of multivariate linear regressions found no significant interactions of independent variables with phenological sensitivity (Appendix 1: Table S1).

Graph: Fig. 2 Regional temperature trend (1901–2015). Between 1901 and 2015, mean annual temperatures, as measured by temperature anomaly from 1961 to 1990 means, in the study region increased 0.12 °C per decade (F1,8541 = 1752, p < 0.001, r2 = 0.170). The shaded area around the trendline represents the 95% confidence interval

Graph: Fig. 3 Species blooming sensitivity. Species blooming sensitivity decreased a 1.4 days for every 30-day delay in flowering (F1,37 = 8.782, p = 0.005, r2 = 0.192), b 2.2 days for every 1000-m increase in elevation (F1,37 = 9.250, p = 0.004, r2 = 0.200), and c 0.9 days for every 1° eastward shift in mean longitude (F1,37 = 14.94, p < 0.001, r2 = 0.288). The shaded area around the trendline represents the 95% confidence interval

The study-wide level of sensitivity of bloom time to temperature (− 4.4 days shift per 1 °C) is consistent with global findings of a 2.5–5 day shift per 1 °C increase in temperature (Menzel et al. [12]; Amano et al. [1]), and reinforces the validity of using herbarium specimens to reconstruct historical phenology (Robbirt et al. [23]). However, joining the region-wide rate of warming (0.12 °C per decade) with region-wide temperature sensitivity results in a mean advance in bloom date of 0.5 days per decade. This is below the reported global mean advances of 2.3 days (Parmesan and Yohe [19]) and 5.1 days (Root et al. [24]) per decade and reveals an important disparity between methods using sensitivity vs. overall temporal shift to measure phenological responses to climate warming. However, the mean sensitivity calculated in our study includes 11 species that are not significantly sensitive to temperature (Table 1). Across researchers, many non-significant results often go unpublished (i.e., the file drawer problem (Rosenthal [25])), potentially skewing previously reported global decadal rates of shift (Parmesan and Yohe [19]; Root et al. [24]). The use of temperature sensitivity to measure phenological response to climate is now widely reported (Matsumoto et al. [9]; Menzel et al. [12]; Wolkovich et al. [34]; Lapenis et al. [8]; Wang et al. [31]; Wang et al. [30]; Kopp and Cleland [7]; Park et al. [16]; Prevéy et al. [21]) and should be adopted as the standard method for describing phenological responses to climate change both within the scientific and lay communities.

Our finding of declining sensitivity as mean species bloom date becomes later (Fig. 3a) consistent with other studies that have found flowering phenologies of later-blooming species to be less responsive to warming (Menzel and Fabian [11]; Wolkovich et al. [34]), with tundra species being an exception (Prevéy et al. [21]). This is a pattern that is reflected across elevation, with species at lower elevations (where the growing season begins earlier) more sensitive than those found higher in the mountains (Fig. 3b). Our finding of a maritime (i.e., longitudinal) gradient of phenological sensitivity (Fig. 3c) was unexpected and, to our knowledge, has not been previously observed in a specimen-based study. This novel phenomenon corresponds to a west to east decrease in mean spring (March–May) temperatures (F1,8542 = 863, p < 0.001, r2 = 0.168; Appendix 1: Fig. S2) east of ~ 123° W, as the moderating effect of the Pacific Ocean dissipates. Similar to the pattern of sensitivity observed across elevation (Fig. 3b), early-flowering species in the PNW are more sensitive to warming than later-flowering species (Fig. 3a). Therefore, the maritime gradient reflects the seasonal gradient of temperature sensitivity. A similar longitudinal pattern in the Northern Hemisphere has been observed using satellite NDVI data (Wang et al. [30]). The likely driver of this phenomenon is that species with ranges near the climatically less variable coast (i.e., lower within spring warming speed (Wang et al. [30])) have lower winter chilling requirements than those in the continental interior. The result is that coastal species more effectively track climate warming (Polgar et al. [20]; Wang et al. [30]). With regard to conservation planning, species that are more sensitive to temperature have increased performance (Cleland et al. [3]), suggesting that coastal and low elevation species in the PNW may be better able to track ongoing anthropogenic warming than interior and high elevation species.

Contrary to reported findings of latitudinal variation in phenological response to warming (Parmesan [18]; Park et al. [16]), we did not observe latitudinal variation in temperature sensitivity across the study region, possibly an artifact of the narrowness of our study region. In the eastern USA, Park et al. ([16]) reported that populations at southern latitudes have greater sensitivity to warming than those further north. However, this study included many non-native species and used climate data averaged across non-uniform political boundaries in a topographically complex region. Examining phenological sensitivity using satellite NDVI for terrestrial regions north of 30° N, Wang et al. ([30]) found phenological sensitivity was greater at southern latitudes, although the latitudinal pattern was not as strong as the longitudinal pattern. Further, it has been reported that species at higher elevations are shifting their phenologies faster than those at lower elevations (Inouye [6]; Vitasse et al. [28]), while we found that species at lower elevation are actually more sensitive to warming. These findings regarding latitude and elevation should not necessarily be seen as contradictory to previous work reporting overall rates of shift, however. In response to climate warming, species at higher latitudes and elevations have likely shifted their phenologies more rapidly over the past several decades compared with lower latitude and elevation species (Parmesan [18]; Inouye [6]; Vitasse et al. [28]). This is likely a direct result of these high-latitude and elevation populations experiencing faster rates of warming than populations and species at lower latitudes and elevations (Socker et al. [27]), thus resulting in greater overall shifts in phenology despite lower temperature sensitivity. This supports the shift in focus on temperature sensitivity from the overall rate of shift in flowering dates, as mentioned above. Finally, it is anticipated that future winter warming will be more pronounced than spring warming (Socker et al. [27]), resulting in more seasonally equitable climate, potentially increasing phenological sensitivity to temperature in populations that currently have low sensitivity.

The methods in this study allowed us to draw out subtle effects of geographic position that are not always apparent in studies focused on a single species or a geographically confined study area, thus providing an unbiased measure of the effect of climate warming on flowering phenology. Further, the use of only native species ensured that all species in the study were highly adapted to the environment they were collected in. Our results demonstrate the power of natural history collections as tools for answering questions with large geographic, temporal, and biotic breadth. Continued digitization of, and open-access to, such collections can only broaden the avenues of inquiry and discovery and should be embraced across all fields of biology. Finally, these results demonstrate that more careful consideration of variation across environmental gradients, particularly in climatically complex regions, is needed when examining biotic responses to climate change.

We thank the botanists that collected the specimens examined in this study. Jonathan Davies and Elizabeth Wolkovich provided feedback and advice on an early draft of this manuscript. Louisa Hsu assisted with data collection. We also thank the anonymous reviewers of this paper for their feedback and insight.

Graph: (PDF 1945 kb)

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.