Articles 514 BioScience • July 2011 / Vol. 61 No. 7 www.biosciencemag.org Articles Spring Flowering Response to Climate Change between 1936 and 2006 in Alberta, Canada ElisabEth bEaubiEn and andrEas hamann In documenting biological responses to climate change, the Intergovernmental Panel on Climate Change has used phenology studies from many parts of the world, but data from the high latitudes of North America are missing. In the present article, we evaluate climate trends and the corresponding changes in sequential bloom times for seven plant species in the central parklands of Alberta, Canada (latitude 528–578 north). For the study period of 71 years (1936–2006), we found a substantial warming signal, which ranged from an increase of 5.3 degrees Celsius (8C) in the mean monthly temperatures for February to an increase of 1.58C in those for May. The earliest-blooming species’ (Populus tremuloides and Anemone patens) bloom dates advanced by two weeks during the seven decades, whereas the later-blooming species’ bloom dates advanced between zero and six days. The early-blooming species’ bloom dates advanced faster than was predicted by thermal time models, which we attribute to decreased diurnal tem- perature fluctuations. This unexpectedly sensitive response results in an increased exposure to late-spring frosts. Keywords: climate change, global warming, phenology, flowering, Canada observations, including Thomas Jefferson, Henry David Thoreau, and Aldo Leopold (Stoller 1956, Miller-Rushing and Primack 2008), long-term records of phenology obser- vation are scarce for North America as compared with Europe (Schwartz and Beaubien 2003). A notable analysis was carried out by Aldo Leopold’s daughter N. L. Bradley and his son A. C. Leopold. They compared Aldo Leopold’s 1935–1945 Wisconsin farm records (Leopold and Jones 1947) with data on 36 plant species collected in the same area from 1976 to 1998 (Bradley et al. 1999). Another major long-term observation effort is the phenology network established by Caprio (1957), in which the phenology of lilac (Syringa vulgaris) and honeysuckle cultivars (Lonicera spp.) was recorded with the help of local garden clubs in 12 western US states until 1994 (Cayan et al. 2001). A simi- lar lilac–honeysuckle network, which still exists today, was established in 1959 in the northeastern US states and eastern Canadian provinces (Schwartz and Reiter 2000). However, there is a notable lack of phenology data for western Canada and Alaska, where the change in the spring warming signal over the last 50 years has been most pronounced globally (Rosenzweig et al. 2007). Besides being a documentation of global change, trends in plant phenology can reveal important ecological conse- quences associated with climate change (Parmesan 2006, Cleland et al. 2007). Plant populations are finely tuned to local frost-risk environments at the beginning and end of the growing season, and phenological traits are usually highly heritable and are often subject to strong selection The scientific field of phenology, the study of the seasonal timing of life-cycle events, has seen a recent revival in light of climate change’s growing prominence. Sparks and colleagues (2009) noted that the use of the term phenology in the scientific literature became seven times more common between 1990 and 2008. In documenting biological responses to global climate change, the Intergovernmental Panel on Climate Change has relied on phenology studies as compel- ling evidence that species and ecosystems respond to global climate change (Rosenzweig et al. 2007). Particularly for perennial plants in temperate zones, temperature exposure over time is the main driver for spring development, includ- ing the timing of bloom and leafout (Rathcke and Lacey 1985, Bertin 2008). This makes spring phenology one of the most sensitive, immediate, and easily observed responses to chang- ing climate in temperate regions (e.g., Schwartz et al. 2006). The use of phenology observations to document climate variability and climate change has a long history. Arakawa (1955, 1956) analyzed a long-term record of the dates of the annual cherry blossom festival in Japan that reached back to the ninth century. Remarkable phenology records covering more than two centuries also exist for European countries, starting with observations made by Linnaeus in the eighteenth century (Parmesan 2006). In a meta-analysis for Europe, Menzel and colleagues (2006) compiled an astonishing 125,000 time series recorded for more than 500 plant species in 21 countries. Although a number of famous North American his- torical figures were involved in early systematic phenology BioScience 61: 514–524. ISSN 0006-3568, electronic ISSN 1525-3244. © 2011 by American Institute of Biological Sciences. All rights reserved. Request permission to photocopy or reproduce article content at the University of California Press’s Rights and Permissions Web site at www.ucpressjournals.com/ reprintinfo.asp. doi:10.1525/bio.2011.61.7.6 Articles www.biosciencemag.org July 2011 / Vol. 61 No. 7 • BioScience 515 Articles pressures (Campbell and Sugano 1975, Vitasse et al. 2009, Li et al. 2010). The timing of spring plant development balances the need to avoid damage due to late-spring frosts with a maximization of the use of the available growing season in competition with other species (Lechowicz 1984, Leinonen and Hänninen 2002). Therefore, plants at higher northern latitudes and at high elevation break bud relatively early; that is, the need to utilize the growing season takes precedence over avoiding late-spring frost damage. This has been documented in many common garden studies for widespread plant species (reviewed in Li et al. 2010). The timing of spring development in virtually all temper- ate perennial plants is primarily controlled by temperature (Rathcke and Lacey 1985, Hunter and Lechowicz 1992). Plants require a certain amount of exposure to warm tem- peratures before leafout or flowering occurs. Exposure to warm temperatures over time can be measured in degree days, which is the sum of the average daily temperatures above a base value. A common base temperature is 5 degrees Celsius (8C), which is widely used to calculate growing degree days in agriculture. For a given species, this amount of warmth over time, referred to as the required heat sum, is nearly constant and can be used to predict bloom times from daily temperature records in what is called a thermal time model (Bertin 2008). The required heat sum for spring development is a genetically controlled adaptive trait (Lei- nonen and Hänninen 2002). Heat sum accumulation allows plants to respond to an unpredictable onset of the grow- ing season, which can easily vary by a month in northern latitudes. If spring development were exclusively driven by expo- sure to warm temperatures, climate change would not be expected to affect the match of plant development with the available growing season. However, additional factors are known to modulate the timing of spring development. Photoperiod may delay bud break if warm temperatures arrive unusually early (Menzel et al. 2005). Some plants also require a certain amount of exposure to cool temperatures following bud set in the fall before they start development in response to warm spring temperatures. This is referred to as a chilling requirement and is measured by summing the exposure to moderately cool temperatures, typically between 08C and 108C. The chilling requirement is thought to guard plants from prematurely breaking bud during midwinter thaws. In both cases, climate warming would be expected to delay the spring response. Plants may be constrained by photoperiod effects that prevent early development, or in warmer regions, they may not receive sufficient exposure to cold temperatures to release them from dormancy (Bertin 2008). Another factor that affects spring phenological response at high latitudes and high elevation is the prevalence of snow (Inouye and Wielgolaski 2003, Wielgolaski and Inouye 2003). A deep spring snowpack further shortens the growing season, and once the snow has melted, the plant response is often immediate, suggesting very low heat sum requirements, and making the release from snow a primary driver of spring phenology. This also has important implica- tions for the effects of climate change. A smaller snowpack due to either higher temperatures or less precipitation would lead to an earlier release from snow, an earlier start of plant development, and potentially higher frost exposure (Inouye 2008). In the present article, we report the results from spring flowering observations conducted over approximately seven decades (1936–2006) in Alberta, western Canada. We ana- lyzed the first-bloom dates for seven plant species that come into flower in a temporal sequence between early April and June. The first objective of this study was to attempt to pro- vide evidence of a plant response to global climate change for a higher-latitude location in western North America, a region for which long-term data are scarce. Second, we asked whether phenology trends correspond to observed tem- perature trends according to spring thermal time models or, alternatively, whether other factors influence spring devel- opment, which would potentially lead to altered sequences of bloom time. Finally, we investigated whether shifts in bloom time have led to changes in the exposure of species to late-spring frosts. Phenology observations in central Alberta We evaluated observations from a phenology network across the central parkland of Alberta (figure 1). This ecological subregion covers approximately 50,000 square kilometers and is situated between the boreal forest to the north and the warmer and drier grasslands to the south. The native vegetation consists of open forests dominated by two poplars (Populus tremuloides Michx. and Populus balsamifera L.), white spruce (Picea glauca [Moench] Voss), and birch (Betula spp.), as well as prairie vegetation found under drier microsite conditions. However, much of the native vegetation has been converted to agricultural use because the area has some of the best soils in Canada. Intensive phenology observations began in 1936 with a program by Agriculture Canada, in which the timing of wheat development and the bloom times for 50 native plant species were recorded over 26 years. The purpose of this program was to identify indicator events to guide the timing of agricultural activities (Russell 1962). This program ended in 1961, which resulted in a data gap of 11 years before botanist Charles Bird initiated a new research program, which tracked the bloom times for 12 native spe- cies between 1973 and 1986. The data were collected by a network of citizen scientists (Bird 1983), supplemented by Bird’s own observations (figure 1). This network was extended by EB in 1987, and in its current form, the vol- unteer observers record data for one or more of 25 species (www.plantwatch.fanweb.ca). Since 1987, this network has collected data from approximately 650 observers, with up to 240 observers reporting each year. The plant species for this phenology network were selected primarily on the basis of the plants’ wide distribution and short bloom period in 516 BioScience • July 2011 / Vol. 61 No. 7 www.biosciencemag.org Articles Articles spring, the ease of their identification by citizens, and the lack of similar-looking species. For additional background on these data series, see Beaubien and Johnson (1994) and Beaubien and Freeland (2000). In the present study, we evaluated the dates of the first bloom for several plant species. First bloom was defined as a plant stage at which the first flower buds had opened in an observed tree or shrub or in a patch of smaller plants. We requested that the observers report on plants that were situated in flat areas away from heat sources such as the walls of houses. Observers were asked to select plants that approximately represented the average bloom time for that species in their area (i.e., that were not the first or last of that species to bloom). Therefore, our first-bloom data do not rep- resent the earliest-blooming individuals of a popula- tion (as in Miller-Rushing et al. 2008). Rather, they represent a developmental stage sampled to represent a local population. Gener- ally, the first-bloom stage is the simplest to observe and yields more temporally pre- cise data than later bloom stages, which can be harder to estimate. Because many of the data (e.g., those col- lected between 1987 and 2006) were compiled from multiple individual plant observations, we used the mean annual bloom date from all available points in the central parkland. The annual first-bloom dates were categorized by species and year for all three data sets and were used for sta- tistical analysis and graphi- cal presentation. Except for the first data set (collected between 1936 and 1961; Russell 1962), we excluded phenology data from the greater Edmonton area. Edmon- ton’s human population has grown at an exponential rate to over one million from 85,000 at the beginning of this research (Statistics Canada 2010). It is therefore possible Figure 1. Central parkland study area in Alberta, western Canada. The figure indicates the location of long-term weather stations and the locations of phenology observations. The white symbols indicate long-term observations. For the Alberta PlantWatch network, the size of the circles indicates the length of data collection by a volunteer. Articles www.biosciencemag.org July 2011 / Vol. 61 No. 7 • BioScience 517 Articles that urban heat-island effects on temperature may confound data on climate-change trends (e.g., Roetzer et al. 2000). The three observation programs (those of Russell [1962], Bird [1983], and Beaubien [Beaubien and Johnson 1994, Beaubien and Freeland 2000]) included the same four woody and three herbaceous (nonwoody) plant species (figure 2). The first species to bloom is the prairie crocus (Anemone patens L.), which is found in grasslands through- out the Northern Hemisphere and blooms soon after a snowmelt. Usually blooming within two days of the prairie crocus is the trembling aspen (P. tremuloides), one of the most common and widely distributed tree species in North America. It is the first tree in Alberta to shed pollen and to produce leaves in spring. About 25 days later, saskatoon, or serviceberry, (Amelanchier alnifolia Nutt.) blooms. Saska- toon is a widespread tall woody shrub with edible berries that were the most important plant food for the prairie Blackfoot First Nations. The remaining four species follow in approximately eight-day intervals, starting with the choke cherry (Prunus virginiana L.), a tall woody shrub that is also widespread throughout North America. The wolf willow (Elaeagnus commutata Bernh. ex Rydb.), or silverberry, is a nitrogen-fixing, medium-sized shrub with a short, well- defined bloom period and an overpowering smell that aids in correct identification. The northern bedstraw (Galium boreale L.) is another widely distributed and easily identi- fied herbaceous species. The last species in this sequence is the yarrow (Achillea millefolium L.), perhaps one of the best known and most widely distributed herbaceous species in the world. In this section, we followed the scientific nomen- clature of Moss (1983). Climate and phenology trends We used daily minimum, maximum, and mean tempera- ture data obtained from the Adjusted Historical Cana- dian Climate Database (AHCCD 2009) to analyze climate trends. This database includes four weather stations with long-term records for the study area (figure 1): Edmonton Figure 2. The species included in the study were prairie crocus (Anemone patens L.); trembling aspen (Populus tremuloides Michx.); choke cherry (Prunus virginiana L.); wolf willow (Elaeagnus commutata Bernh. ex Rydb.), or silverberry; saskatoon (Amelanchier alnifolia Nutt.), or serviceberry; yarrow (Achillea millefolium L.); and northern bedstraw (Galium boreale L.). Photos by Linda Kershaw. 518 BioScience • July 2011 / Vol. 61 No. 7 www.biosciencemag.org Articles Articles International Airport (ID #3012205), which is well outside the city of Edmonton; Lacombe (ID #3023722); Calmar (ID #3011120); and Coronation (ID #3011887). To visualize temperature trends and to compare station records, we also calculated the mean monthly minimum, the mean monthly maximum, and the mean temperature values for February, March, April, and May from the daily data. In addition, we generated interpolated monthly data following the method of Mbogga and colleagues (2009) for the central parkland ecoregion. The interpolated climate data and station data suggest that the central parkland ecoregion is climatically very homogenous. The mean monthly temperatures for Feb- ruary, March, and April for the 1961–1990 normal period differ by less than 18C between any two of the four weather stations and among the grid cells of the interpolated surface. The average correlation coefficient between pairs of stations is .97 for the mean monthly temperatures from February through May. Because of the climatic homogeneity of the study area, we used the mean climate values from the four weather stations for subsequent analysis, which matches the data preparation of phenology observations as regional averages for the central parkland. We observed a substantial warming trend between 1936 and 2006 that was most pronounced in late winter and early spring (figure 3). For the 70-year period of this research, the slope of a linear regression equates to a 5.38C increase in the mean February temperature, a 2.78C increase in the mean March temperature, and a 1.88C increase in the mean April temperature. These trends were even more pronounced in the mean monthly minimum temperatures (6.08C, 3.98C, and 2.28C for February, March, and April, respectively), whereas the mean maximum temperature changes over the study period were 4.58C, 1.58C, and 1.58C. A Mann–Kendall test for identifying trends in time-series data following the method of Hipel and McLeod (1994) revealed that the warming trends for the monthly temperatures from Febru- ary to April were statistically significant at a = .05 (table 1). The annual sequence of the species’ first-bloom dates was fairly consistent between years (figure 4a). The plants responded to warming temperatures by blooming earlier in spring, with the most pronounced changes in the earliest- blooming species (A. patens and P. tremuloides). These spe- cies’ flowering dates advanced by approximately two weeks, whereas the later-blooming species’ flowering dates advanced Figure 3. Temperature trends for the central parkland study area for the change in the mean monthly minimum temperature (in degrees Celsius) and the change in the mean monthly maximum temperature. Articles www.biosciencemag.org July 2011 / Vol. 61 No. 7 • BioScience 519 Articles This result corresponds to the observed temperature changes, with considerable warming in late winter but minimal warming in late spring. Figure 4. (a) Trends in observations of first bloom for seven species. The species names are abbreviated to the first four letters of the genus and the first three letters of the species name provided in figure 2. (b) The predicted day of first bloom from a thermal time model (the best model is highlighted in bold in table 3). between zero and six days over the study period. A Mann– Kendall test also confirmed the advanced blooming in the earliest-blooming species as statistically significant (table 2). Table 1. Mann–Kendall test statistics for time-series trends in minimum, maximum, and mean monthly temperature (shown in figure 3). Month Minimum temperature Maximum temperature Mean temperature Temperature change (8C per decade) Mann–Kendall statistic () p(>) Temperature change (8C per decade) Mann–Kendall statistic () p(>) Temperature change (8C per decade) Mann–Kendall statistic () p(>) February 0.86 0.31 .0001 0.64 0.28 .0004 0.75 0.28 .0003 March 0.56 0.21 .0052 0.21 0.07 .1854 0.39 0.15 .0365 April 0.31 0.18 .0157 0.21 0.07 .1928 0.26 0.12 .0756 May 0.05 0.10 .1275 0.00 0.04 .5874 0.02 0.04 .3010 8C, degrees Celsius 520 BioScience • July 2011 / Vol. 61 No. 7 www.biosciencemag.org Articles Articles Trends toward an earlier onset of spring phenology in the Northern Hemisphere are well documented in the literature. In a meta-analysis for the Northern Hemisphere, Root and colleagues (2003) revealed an average three-day advance per decade in tree phenology, with somewhat more pronounced trends at higher latitudes. For Western Europe, Menzel and colleagues (2006) and Schleip and colleagues (2009) analyzed phenology time series of at least 30 years between 1955 and 2000. They found that changes in the spring phenology of plants were most pronounced in central and western maritime Europe, advancing around 3.5 days per decade. These changes appear to be larger than our observed changes for our earliest-blooming species (which advanced approximately two days per decade). However, the difference arises mainly from the observation period. For example, in a long-term study of UK plant communities, Amano and col- leagues (2010) found approximately the same 3.5 days per decade rate of change as did Menzel and colleagues (2006) over the last 30 years. Conversely, the rate of change over the 70 years corresponding to our study showed an advance of only approximately one day per decade for the data from Amano and colleagues (2010), because most of the observed warming at their study site occurred over the last 30 years. Our observation of a total advance of 14 days for A. patens and P. tremuloides over 70 years appears to be on the high end of changes observed in the Northern Hemisphere. Thermal time models of spring development Thermal time models use daily temperature data to predict the timing of bud break or flowering. Daily temperatures are, however, not directly used as predictor variables. Instead, daily temperature values are integrated over time by adding daily temperature measurements. The derived predictor variable for bud break or flowering is the date at which the tempera- ture sum reaches a certain value (the required heat sum). De Réaumur (1735) was the first to establish the principle of thermal time and the concept of degree days as predictors for plant development. Degree days are calculated as the sum of the daily average temperature values from a chosen start date (often arbitrarily set as 1 January) and a threshold value (often 08C to 58C for early spring–blooming species). This summa- tion continues up to the day of a phenology event, yielding a required heat sum for the observed event. This classical thermal time model has been modified in various ways to account for the nonlinearity of the physio- logical response to temperature (for a review, see Bonhomme 2000). Other modifications include accounting for the chill- ing requirements of plants before temperature accumulation begins or for additional environmental factors (for a review, see Chuine et al. 2003). Nevertheless, the simple linear model has proven to be surprisingly accurate, often having just one variable parameter: minimum temperature threshold. This parameter bounds the lower end of the temperature range that is assumed to be approximately linearly correlated with a spring physiological response (Bonhomme 2000). Sometimes, start dates of heat sum accumulation other than 1 January are tested in order to approximately account for dormancy release or photoperiod effects (e.g., Wielgolaski 1999). Complex mechanistic or statistical models often yield only minor improvements—if any—over the classical ther- mal time model, particularly for studies that are not carried out in controlled environments (e.g., Hannerz 1999, Schaber and Badeck 2003, Linkosalo et al. 2006). In figure 4b and table 3, we show the results from a classical thermal time model applied to our data. The development of a thermal time model involves the selection of a base tempera- ture for degree-day calculations—for example, 08C. The next step is to calculate the required heat sum for an observed phe- nology event to occur. This required heat sum is a mean value based on the phenology events of a species observed over multiple years that can be estimated with a standard error (SE; see table 3). With a species-specific required heat sum value, we can now use daily temperature data to predict a bloom time for each year (figure 4b). The correlation between the observed bloom dates in each year (figure 4a) and the bloom date predicted by the thermal time model (figure 4b) serves as a measure of model fit. The model fit may be improved by modifications of base temperatures or start dates. We tested a wide range of base temperatures for degree- day calculations from –108C to 108C, in 18C intervals. Fur- thermore, we tested multiple start dates for temperature accumulation (1 January, 1 March, and 31 March) in order to account for possible unmet chilling requirements. The best thermal time model—the one with the highest cor- relation between the observed and the predicted flowering dates—was obtained with threshold values between –38C and 38C (bold correlation coefficients in table 3). This is a fairly typical result for northern temperate and boreal plant species, which usually have optimal threshold parameters between 08C and 58C (e.g., White 1995, Hannerz 1999). As was expected for a northern environment, later start dates did not improve the correlations, suggesting that species’ chilling requirements were met before winter. We therefore report only the statistics for a start date of 1 January in table 3. Table 2. Mann–Kendall test statistics for time-series trends in first-bloom dates for seven plant species, expressed in number of days per decade shift to earlier bloom time. Species Change in bloom time (days per decade) Mann–Kendall statistic () p(<) Anemone patens –2.1 –0.26 .0039 Populus tremuloides –2.0 –0.29 .0008 Amelanchier alnifolia 0.0 0.03 .6181 Prunus virginiana –0.6 –0.09 .1759 Elaeagnus commutata –0.7 –0.16 .0735 Galium boreale –0.4 –0.03 .3735 Achillea millefolium –0.9 –0.09 .1673 Note: See figure 4a. Articles www.biosciencemag.org July 2011 / Vol. 61 No. 7 • BioScience 521 Articles also checked for the effects of chilling requirements in the previous fall, with chilling degree days calculated between the upper and lower thresholds of 08C and 58C and 28C and 88C, following the method of Linkosalo and colleagues (2006). None of these additional factors could account for a significant portion of the variance that was not already explained by the thermal time model (data not shown). A nonlinear, Q 10 -based thermal time model following the method of Bonhomme (2000) yielded model accuracies for all species (measured as r2 between the observed and predicted events and as the SE of heat sum) that were similar to previous results (figure 4b). However, they actu- ally increased the discrepancy between the observed and predicted temporal trends for the early-blooming species by a small amount (data not shown). A possible remaining explanation is that spring phenology is not only a function of mean daily temperatures; it is also influenced by the amplitude of diurnal temperature variations. Karl and colleagues (1993) were the first to demonstrate that global minimum temperature increased faster than maximum temperature, resulting in a significant decrease of diurnal temperature variation, which was subsequently confirmed by Easterling and colleagues (1997). This differential warm- ing pattern in minimum and maximum temperatures clearly applies to our study area as well, where the minimum night temperatures in March increased more than twice as fast as the daily maximum temperatures (figure 3, table 1). We therefore hypothesize that the increase of the minimum night tempera- tures relative to the mean daily temperatures used in the heat sum model results in a more rapid heat sum accumulation. Correlation coefficients are a good measure to use in assess- ing statistical error, but they do not detect statistical bias (sys- tematic over- or underprediction). We therefore validated the thermal time model using a second statistical measure: mean absolute error (MAE). MAE is calculated as the absolute dif- ference between the observed and predicted bloom dates. We carried out an independent cross-validation using a temporal split of the temperature and phenology data. The first two- thirds (1936–1986, with approximately 40 years of data) were used for the development of the thermal time model, and the last third (1987–2006) was used for model validation. Generally, the classical thermal time model appears to be very accurate in predicting the mean bloom dates of the species in the central parkland (table 3). The MAE values in predicting bloom time in the independent cross-validation ranged from 1.6 to 4.5 days. The species with the largest MAE values in table 3 were the earliest-blooming species: A. patens and P. tremuloides. For these species, the predictions were biased, underpredicting the rate of change in bloom time (compare figure 4a and 4b). Observed versus predicted phenology trends In an attempt to explain the discrepancy between the observed and predicted trends in A. patens and P. tremuloides, we used a multiple regression approach in order to incorporate other climatic and environmental factors (see equation 1 in Chuine et al. 2003). The environ- mental factors that we tested include the amount of winter precipitation that fell as snow, the depth of snowpack at the end of February and March, and several dryness indices. We Table 3. Correlations between flowering date and thermal time for different base temperature values for heat sum accumulation. Species Heat sum Base temperature (°C) Years of recorded data HS0 (8C) HS (8C) SE –3 –2 –1 0 1 2 3 MAE (days) Anemone patens 50 94 187 9.6 .81 .80 .79 .78 .75 .72 .68 3.9 Populus tremuloides 60 103 202 6.4 .90 .89 .88 .88 .86 .84 .81 4.5 Amelanchier alnifolia 60 303 172 3.1 .84 .86 .87 .89 .90 .91 .92 1.6 Prunus virginiana 57 419 258 3.8 .84 .86 .87 .89 .90 .90 .91 2.4 Elaeagnus commutata 44 511 385 7.0 .75 .76 .78 .79 .80 .81 .81 3.5 Galium boreale 58 690 467 6.3 .70 .72 .74 .75 .75 .76 .77 2.1 Achillea millefolium 52 782 696 8.9 .69 .71 .72 .73 .74 .73 .72 2.2 Note: The threshold value for the best model (highest correlation) is shown in bold. °C, degrees Celsius; HS, heat sum for the best model; HS 0 , heat sum for a threshold of 0°C, provided for comparison of thermal time requirements across species; MAE, mean absolute error of the observed versus the predicted bloom time; SE, standard error 522 BioScience • July 2011 / Vol. 61 No. 7 www.biosciencemag.org Articles Articles to –208C (represented by the gradient). To give an example, in 1994 and 1995, there were frost events of –208C as late as 28 and 30 April (days 118 and 120, respectively). That means that virtually all of the reported flowering individuals were exposed to these extreme frost events. As a contrasting exam- ple, in 2000, we had a –128C event that occurred on 14 April (day 104). This affected only the early-blooming portion of the population. Most individuals bloomed after that late frost event and were exposed only to a –48C frost that occurred as late as day 133 (13 May). In figure 5, trends toward an earlier bloom would be represented by the violin plot points being located higher on the left side than on the right side. Higher frost exposure experienced by blooming populations would be visible as darker colors toward the right side. Note that we have population-level information from many observers of the PlantWatch Alberta network only since 1987. Before that date, we assume a normal distribution around the known annual average reported by Russell (1962) and Bird (1983). In the case of A. patens, we can see a slightly increased exposure of blooming populations to frost events over time, with overall darker shades toward the right side of figure 5. To test whether this trend is statistically significant, we can- not directly use the distributions shown in figure 5, since we lack population-level data from before 1987. Instead, we analyzed trends in the value of the coldest frost event follow- ing the average bloom times shown in figure 4a for each year (table 4; later-blooming species that were not exposed to frost were excluded). For example, A. patens individuals with an average bloom time were exposed to colder spring frost events, at a rate of –0.578C per decade. This means that frost events to which blooming plants were exposed were on aver- age 48C colder at the end than at the beginning of the study period. This trend was not significant for any other species at an a level of .05. However, two other early-blooming species showed similar trends toward increased exposure to Although heat sum accumulation calculated from min- imum night temperatures is not biologically reasonable because it does not incorporate daytime temperature expo- sure, we explored this option as well. The result is a relatively poor model fit (r = .79 for P. tremuloides), but the flowering advance over time was predicted more accurately (14 days observed versus 13 days predicted over the study period). It makes adaptational sense that the minimum temperature values (which could represent damaging frost events) modu- late daytime thermal time accumulation to control spring development. This would allow plants to fine-tune spring development for microsites with different diurnal temperature variation but may also increase the exposure of P. tremuloides and A. patens to late-spring frosts under climate change. Exposure to late-spring frost Late-spring frosts of –108C occurred earlier, at a rate of 0.7 days per decade, and very severe spring frosts of –208C occurred earlier, at a rate of 1.1 days per decade over the study period (data not shown). This is a considerably slower rate than the advance of bloom times for the early-blooming species, which occurred at a rate of approximately 2 days per decade (figure 4a). This discrepancy raises the question of whether early-blooming species might be exposed to increased risks of late-spring frosts because of climate change. To answer this question, we compared the incidence of late-spring frost events with the timing of first bloom. In figure 5, we show the variance of bloom times observed across a population sample of A. patens for different years by means of a special form of boxplot, the so-called violin plot, which reveals the frequency of bloom time for different dates. This plot quantifies the bloom dates of the sampled popula- tion (the width of the violin plot points indicates frequency, the number of reported observations with that date). It also shows the latest dates of spring frost events ranging from 08C Figure 5. The distribution of the day of the year when flowers appear in Anemone patens individuals. The width of the points in this form of boxplot indicates the frequency of reported observations. The gradient indicates the severity of frost events to which blooming individuals are exposed, with the lighter part of the gradient representing less severe frost events. We have population-level data available only since 1987. Before that date, we assume a normal distribution (which is used only for visualization in this figure). Articles www.biosciencemag.org July 2011 / Vol. 61 No. 7 • BioScience 523 Articles tures in this month). The phenology response of two early- blooming species—A. patens and P. tremuloides—appears to be unexpectedly sensitive to these temperature changes. Their bloom times changed twice as fast as did the frost events, thus shifting their bloom period closer to the receding winter and increasing the danger of damage from late-spring frost. The database that we analyzed was assembled as a col- laborative effort among university biologists, government researchers, and over 650 members of the general public. This effort has both harnessed the energy of concerned citizens and provided them with biological insights and a raised awareness of climate-change issues in Alberta. Besides documenting a biological response to global climate change, citizen contributions are invaluable for the validation of remote-sensing data and the calibration of carbon uptake models in terrestrial ecosystems (Badeck et al. 2004). In con- clusion, we would like to encourage interested readers to join local phenological networks that make this research possible. Links to local networks can be found at www.plantwatch.ca for Canada and www.usanpn.org for the United States. Acknowledgments Funding to carry out the analysis presented in this paper was provided by NSERC Discovery Grant RGPIN-330527-07 and Alberta Ingenuity Grant #200500661. We thank all citizen sci- entists who contributed to the data collection and appreciate their enthusiasm and continued support of this program. Botanist Linda Kershaw contributed the photographs, and we thank Mryka Hall-Beyer for her helpful comments. References cited Amano T, Smithers RJ, Sparks TH, Sutherland WJ. 2010. A 250-year index of first flowering dates and its response to temperature changes. Pro- ceedings of the Royal Society B 277: 2451–2457. AHCCD. 2009. Adjusted Historical Canadian Climate Data. (26 April 2011; http://ec.gc.ca/dccha-ahccd ) Arakawa H. 1955. Twelve centuries of blooming dates of the cherry blossoms in the city of Kyoto and its own vicinity. Geophysica Pura e Applicata 30: 147–150. Arakawa H. 1956. Climatic change as revealed by the blooming dates of the cherry blossoms at Kyoto. Journal of Meteorology 13: 599–600. Badeck F-W, Bondeau A, Böttcher K, Doktor D, Lucht W, Schaber J, Sitch S. 2004. Responses of spring phenology to climate change. New Phytolo- gist 162: 295–309. Beaubien EG, Freeland HJ. 2000. Spring phenology trends in Alberta, Canada: Links to ocean temperature. International Journal of Biome- teorology 44: 53–59. Beaubien EG, Johnson DL. 1994. Flowering plant phenology and weather in Alberta, Canada. International Journal of Biometeorology 38: 23–27. Bertin RI. 2008. Plant phenology and distribution in relation to recent cli- mate change. Journal of the Torrey Botanical Society 135: 126–146. Bird CD. 1983. Alberta flowering dates. Alberta Naturalist 13 (suppl. 1): 1–4. Bonhomme R. 2000. Bases and limits to using ‘degree day’ units. European Journal of Agronomy 13: 1–10. Bradley NL, Leopold AC, Ross J, Huffaker W. 1999. Phenological changes reflect climate change in Wisconsin. Proceedings of the National Acad- emy of Sciences 96: 9701–9704. Campbell RK, Sugano AI. 1975. Phenology of bud burst in Douglas-fir related to provenance, photoperiod, chilling, and flushing temperature. Botanical Gazette 136: 290–298. frost, and trends of this magnitude or larger would arise by random chance only once in 19 times for P. tremuloides or once in 16 times for A. alnifolia. This paradoxical result of increased frost risk with climate warming differs from the results of comparable studies at lower latitudes. Scheifinger and colleagues (2003) observed that frost risks generally decreased, because the retreat of late frosts outpaces the advance of spring development. However, the low heat sum requirements of species from environments with short growing seasons results in a finely tuned adaptive balance between avoiding spring frost and using the avail- able growing season (Li et al. 2010). Earlier snowmelt due to higher temperatures or lower winter precipitation may affect this balance in high-elevation environments (Inouye 2008). In our study, at relatively high latitude, we excluded snow as a factor that can explain the unexpectedly sensitive pheno- logical response of A. patens and P. tremuloides to warming trends. Instead, changes in diurnal temperature fluctuations may be responsible for a faster advance in bloom dates than can be explained by standard thermal time models. Although this explanation is speculative, it could guide future experimental research toward the development of improved thermal time models that take diurnal temperature fluctuations into account. Regardless of whether diurnal tem- perature variations are the ultimate cause of the discrepancy between the observed and predicted trends, our results sug- gest that projections of phenology response into the future, although they are important (e.g., Leinonen and Kramer 2002), should be made with caution. Models that explain interannual variation of plant response very well over a limited observation period may not always provide reliable long-term projections. In the case of P. tremuloides and A. patens, it appears that we would underpredict climate change response by 23% and 44%, respectively, with a standard thermal time model. Conclusions In this study, we documented considerable advances in phe- nology over time that were driven by what we perceive as astonishing warming trends in spring temperature. Particu- larly in March, we also found large changes in diurnal temper- ature fluctuations (the average daily minimum temperatures increased 2.7 times faster than the daily maximum tempera- Table 4. Mann–Kendall test statistics for time-series trends in the value of the coldest frost event following average bloom time, expressed in degrees Celsius change per decade. Species Temperature change (°C per decade) Mann–Kendall statistic () p(<) Anemone patens 20.57 20.20 .0231 Populus tremuloides 20.44 20.15 .0521 Amelanchier alnifolia 20.13 20.06 .0671 Prunus virginiana 20.01 0.01 .4909 524 BioScience • July 2011 / Vol. 61 No. 7 www.biosciencemag.org Articles Parmesan C. 2006. Ecological and evolutionary responses to recent climate change. Annual Review of Ecology, Evolution, and Systematics 37: 637–669. Rathcke B, Lacey EP. 1985. Phenological patterns of terrestrial plants. Annual Review of Ecology and Systematics 16: 179–214. Roetzer T, Wittenzeller M, Haeckel H, Nekovar J. 2000. Phenology in Cen- tral Europe—Differences and trends of spring phenophases in urban and rural areas. International Journal of Biometeorology 44: 60–66. Root TL, Price JT, Hall KR, Schneider SH, Rosenzweig C, Pounds JA. 2003. Fingerprints of global warming on wild animals and plants. Nature 421: 57–60. Rosenzweig C, Casassa G, Karoly DJ, Imeson A, Liu C, Menzel A, Rawlins S, Root TL, Seguin B, Tryjanowski P. 2007. Assessment of observed changes and responses in natural and managed systems. Pages 79–131 in Parry ML, Canziani OF, Palutikof JP, van der Linden PJ, Hanson CE, eds. Cli- mate Change 2007: Impacts, Adaptation, and Vulnerability. Contribution of Working Group II to the Fourth Assessment Report of the Intergovern- mental Panel on Climate Change. Cambridge University Press. Russell RC. 1962. Phenological records of the prairie flora. Canadian Plant Disease Survey 42: 162–166. Schaber J, Badeck FW. 2003. Physiology-based phenology models for forest tree species in Germany. International Journal of Biometeorology 47: 193–201. Scheifinger H, Menzel A, Koch E, Peter C. 2003. Trends of spring time frost events and phenological dates in Central Europe. Theoretical and Applied Climatology 74: 41–51. Schleip C, Sparks TH, Estrella N, Menzel A. 2009. Spatial variation in onset dates and trends in phenology across Europe. Climate Research 39: 249–260. Schwartz MD, Beaubien EG. 2003. North America. Pages 57–73 in Schwartz MD, ed. Phenology: An Integrative Environmental Science. Kluwer Academic Publishers. Schwartz MD, Reiter BE. 2000. Changes in North American spring. Interna- tional Journal of Climatology 20: 929–932. Schwartz MD, Ahas R, Aasa A. 2006. Onset of spring starting earlier across the Northern Hemisphere. Global Change Biology 12: 343–351. Sparks TH, Menzel A, Stenseth NC. 2009. European cooperation in plant phenology: Introduction. Climate Research 39: 175–177. Statistics Canada. 2010. Statistics Canada historical census records for Edmonton. (14 April 2011; www.statcan.gc.ca) Stoller L. 1956. A note on Thoreau’s place in the history of phenology. Isis 47: 172–181. Vitasse Y, Delzon S, Bresson CC, Michalet R, Kremer A. 2009. Altitudi- nal differentiation in growth and phenology among populations of temperate-zone tree species growing in a common garden. Canadian Journal of Forest Research 39: 1259–1269. White LM. 1995. Predicting flowering of 130 plants at 8 locations with temperature and daylength. Journal of Range Management 48: 108–114. Wielgolaski FE. 1999. Starting dates and basic temperatures in phenologi- cal observations of plants. International Journal of Biometeorology 42: 158–168. Wielgolaski FE, Inouye DW. 2003. High latitude climates. Pages 175–194 in Schwartz MD, ed. Phenology: An Integrative Environmental Science. Kluwer Academic Publishers. Elisabeth Beaubien (e.beaubien@ualberta.ca) is an ecologist coordinating the Alberta PlantWatch network and also serves as science adviser for Canada PlantWatch. She is currently pursuing a doctoral degree with Andreas Hamann (andreas.hamann@ualberta.ca), who is assistant professor in the field of ecological genetics and climate change research at the University of Alberta, Department of Renewable Resources, Edmonton, Alberta, Canada. Caprio JM. 1957. Phenology of lilac bloom in Montana. Science 126: 1344–1345. Cayan DR, Kammerdiener SA, Dettinger MD, Caprio JM, Peterson DH. 2001. Changes in the onset of spring in the western United States. Bul- letin of the American Meteorological Society 82: 399–415. Chuine I, Kramer K, Hänninen H. 2003. Plant development models. Pages 217–235 in Schwartz MD, ed. Phenology: An Integrative Environmental Science. Kluwer Academic Publishers. Cleland EE, Chuine I, Menzel A, Mooney HA, Schwartz MD. 2007. Shifting plant phenology in response to global change. Trends in Ecology and Evolution 22: 357–365. De Réaumur RAF. 1735. Observations du thermometer, faites à Paris pendant l’année 1735, comparées avec celles qui ont été faites sous la ligne, à l’Isle de France, à Alger et en quelques-unes de nos isles de l’Amérique. Mémoires de l’Académie des Sciences. Easterling DR, et al. 1997. Maximum and minimum temperature trends for the globe. Science 277: 364–367. Hannerz M. 1999. Evaluation of temperature models for predicting bud burst in Norway spruce. Canadian Journal of Forest Research 29: 9–19. Hipel KW, McLeod AI. 1994. Time Series Modelling of Water Resources and Environmental Systems. Elsevier. Hunter AF, Lechowicz MJ. 1992. Predicting the timing of budburst in temperate trees. Journal of Applied Ecology 29: 597–604. Inouye DW. 2008. Effects of climate change on phenology, frost damage, and floral abundance of montane wildflowers. Ecology 89: 353–362. Inouye DW, Wielgolaski FE. 2003. High altitude climates. Pages 195–214 in Schwartz MD, ed. Phenology: An Integrative Environmental Science. Kluwer Academic Publishers. Karl TR, Jones PD, Knight RW, Kukla G, Plummer N, Razuvayev V, Gallo KP, Lindseay J, Charlson RJ, Peterson TC. 1993. A new perspective on recent global warming: Asymmetric trends of daily maximum and minimum temperature. Bulletin of the American Meteorological Soci- ety 74: 1007–1023. Lechowicz MJ. 1984. Why do temperate deciduous trees leaf out at differ- ent times? Adaptation and ecology of forest communities. American Naturalist 124: 821–842. Leinonen I, Hänninen H. 2002. Adaptation of the timing of bud burst of Nor- way spruce to temperate and boreal climates. Silva Fennica 36: 695–701. Leinonen I, Kramer K. 2002. Applications of phenological models to predict the future carbon sequestration potential of boreal forests. Climatic Change 55: 99–113. Leopold A, Jones SE. 1947. A phenological record for Sauk and Dane coun- ties, Wisconsin, 1935–1945. Ecological Monographs 17: 81–122. Li H, Wang X, Hamann A. 2010. Genetic adaptation of aspen (Populus tremuloides) populations to spring risk environments: A novel remote sensing approach. Canadian Journal of Forest Research 40: 2082–2090. Linkosalo T, Häkkinen R, Hänninen H. 2006. Models of the spring phenol- ogy of boreal and temperate trees: Is there something missing? Tree Physiology 26: 1165–1172. Mbogga MS, Hamann A, Wang T. 2009. Historical and projected climate data for natural resource management in western Canada. Agricultural and Forest Meteorology 149: 881–890. Menzel A, Estrella N, Testka A. 2005. Temperature response rates from long- term phenological records. Climate Research 30: 21–28. Menzel A, et al. 2006. European phenological response to climate change matches the warming pattern. Global Change Biology 12: 1969–1976. Miller-Rushing AJ, Primack RB. 2008. Global warming and flowering times in Thoreau’s Concord: A community perspective. Ecology 89: 332–341. Miller-Rushing AJ, Inouye DW, Primack RB. 2008. How well do first flowering dates measure plant responses to climate change? The effects of population size and sampling frequency. Journal of Ecology 96: 1289–1296. Moss EH. 1983. Flora of Alberta. 2nd ed. Revised by Packer JG. University of Toronto Press.