American Journal of Botany 101(8): 1293–1300, 2014.
DETERMINING PAST LEAF-OUT TIMES OF NEW ENGLAND’S DECIDUOUS FORESTS FROM HERBARIUM SPECIMENS1
PETER H. EVERILL2, RICHARD B. PRIMACK2,5, ELIZABETH R. ELLWOOD3, AND ELI K. MELAAS4 2Boston University Department of Biology, 5 Cummington Mall, Boston, Massachusetts 02215 USA; 3Department of Biological Science, Florida State University, Tallahassee, Florida 32306 USA; and 4Boston University Department of Earth & Environment, 685 Commonwealth Avenue, Room 130, Boston, Massachusetts 02215
• Premise of the study: There is great interest in studying leaf-out times of temperate forests because of the importance of leaf-out in controlling ecosystem processes, especially in the face of a changing climate. Remote sensing and modeling, combined with weather records and field observations, are increasing our knowledge of factors affecting variation in leaf-out times. Herbarium specimens represent a potential new source of information to determine whether the variation in leaf-out times observed in recent decades is comparable to longer time frames over past centuries. • Methods: Here we introduce the use of herbarium specimens as a method for studying long-term changes in leaf-out times of deciduous trees. We collected historical leaf-out data for the years 1834–2008 from common deciduous trees in New England using 1599 dated herbarium specimens with young leaves. • Key results: We found that leaf-out dates are strongly affected by spring temperature, with trees leafing out 2.70 d earlier for each degree C increase in mean April temperature. For each degree C increase in local temperature, trees leafed out 2.06 d earlier. Additionally, the mean response of leaf-out dates across all species and sites over time was 0.4 d earlier per decade. Our results are of comparable magnitude to results from studies using remote sensing and direct field observations. • Conclusions: Across New England, mean leaf-out dates varied geographically in close correspondence with those observed in studies using satellite data. This study demonstrates that herbarium specimens can be a valuable source of data on past leaf-out times of deciduous trees. Key words: climate change; forests; herbarium specimens; leaf-out; New England; phenology; remote sensing; trees.
Spring leaf-out of deciduous trees engages many ecosystem processes (Richardson et al., 2009; Hufkens et al., 2012) and influences plant community structure (Sherry et al., 2007; Willis et al., 2010). In temperate ecosystems, leaf development is tightly controlled by temperature (Linkosalo et al., 2006) and to some extent by day length (Körner and Basler, 2010; Laube et al., 2013, Polgar et al., 2013); as spring temperatures rise, leaf-out occurs earlier (Morin et al., 2010; Polgar and Primack, 2011). Thus, warmer temperatures could lengthen the growing season and increase carbon uptake in some forests, acting as a negative feedback on climate change (Migliavacca et al., 2012; Richardson et al., 2013). Additionally, earlier leaf-out could alter trophic interactions and the cycling of water and nutrients (Tabacchi et al., 2000; Parmesan, 2006; Morisette et al., 2009). Understanding the dynamics of these relationships is increasingly important considering recent global increases in temperature, predictions for continued warming (IPCC, 2007), and the impacts a changing climate will have on spring leaf-out and associated ecosystem processes.
1 Manuscript received 31 January 2014; revision accepted 21 July 2014. The authors thank the curators of the various herbaria for the use of their collections. Support for this project came from the iDigBio program of the National Science Foundation. Comments on the manuscript were provided by M. Boeni, A. Miller-Rushing, A. Gallinat, C. Polgar, C. MacKenzie, J. Silander, and P. Sweeney. 5 Author for correspondence (e-mail: [email protected])
doi:10.3732/ajb.1400045
Given the ecological importance of spring leaf-out, there is much interest in understanding how trees have responded to past climatic variability to predict how the timing of leaf development will change in the future. Previous studies of leaf-out phenology have generally used one of three methods (Polgar and Primack, 2011): direct field observations, near-surface remote sensing with canopy or understory cameras, and satellitebased remote sensing. The simplest of these methods is direct observation of plants, categorizing buds and young leaves by phenophase. In practice, this is time-consuming, and historical data are generally available only for specific locations, such as a garden or forest plot (e.g., Morin et al., 2009). Citizen science programs in which large numbers of volunteers monitor plant phenology offer expanded geographic coverage for some species, particularly for recent years (Menzel et al., 2001; Wolfe et al., 2005). In near-surface remote sensing, scientists have begun to use phenocams—time-lapse cameras that capture images of a landscape at scales that can range from individual trees to entire landscapes. The images can be analyzed for various measurements of leaf cover development (Richardson et al., 2007; Ahrends et al., 2008; Sonnentag et al., 2012). At larger geographical scales, multispectral images from satellite instruments (primarily the Landsat Thematic Mapper and Enhanced Thematic Mapper, the moderate resolution imaging spectroradiometer (MODIS), and the advanced very high resolution radiometer) can be used to monitor the phenology of landscape greenness (Fisher et al., 2006). Unfortunately, despite increased geographic coverage, phenocam and satellite data are available only for a short historical period—since 1972 for Landsat, 2000 for MODIS, and more recently for phenocams.
American Journal of Botany 101(8): 1293–1300, 2014; http://www.amjbot.org/ © 2014 Botanical Society of America
1293
1294
[Vol. 101
AMERICAN JOURNAL OF BOTANY
At present, researchers are actively working to relate greenness measures made by phenocams and satellites to species-specific stages of leaf development and key physiological stages such as the onset of photosynthesis (White et al., 2009; Sonnentag et al., 2012). Herbarium specimens provide a largely untapped source of historical data on tree leaf-out dates. An estimated 350 million plant specimens reside in over 3000 herbaria worldwide, and in some cases, the specimens were collected several hundred years ago (New York Botanical Garden’s Virtual Herbarium: http:// sweetgum.nybg.org/ih/). Herbarium specimens have in the past been used to study flowering phenology of various species in New England (Primack et al., 2004, Miller-Rushing et al., 2006, Primack and Miller-Rushing, 2009), Europe (Diskin et al., 2012), and elsewhere (e.g., Gaira et al., 2011). Additionally, many herbarium specimens contain important information regarding leaf-out times. For example, research based on herbarium specimens in Munich, Germany demonstrated climate tracking in four woody species (Zohner and Renner, 2014). Many of the common deciduous trees of the temperate forest, such as species of Quercus and Acer, produce their first flush of young leaves at approximately the same time as their flowers. Collectors interested in flowers often unwittingly collect herbarium specimens that contain young leaves, which could provide a measure of leaf-out phenology. Leaf-out of most New England tree species, including those researched here, occurs over a relatively short period of time for each species and does not recur later in the season. In this study, we investigated whether herbarium specimens of common New England trees can be used as a source of information about historical leaf-out times and how leaf-out times are affected by climatic variation. We also investigated whether recent regional patterns of leaf-out indicated by satellite-based remote sensing match long-term historical leaf-out data collected from herbarium specimens. MATERIALS AND METHODS Data collection—We examined herbarium specimens of 27 of the most common canopy tree species (Table 1) of New England’s deciduous forests and identified specimens with young leaves. These species were ones that leafed out around the time of flowering, so that specimens often had both flowers and young leaves. We only examined specimens that had been collected from March through June, the time of spring leaf-out. We examined herbarium specimens at seven of the largest herbaria in the region: Brown University’s Stephen T. Olney Herbarium, Harvard University Herbaria, George Safford Torrey Herbarium at the University of Connecticut, University of Maine Herbaria, Hodgdon Herbarium at the University of New Hampshire, University of Vermont Herbarium, and the Yale University Herbarium. We considered leaves as young if they were emerging from the breaking bud, i.e., they were unfolded and had the recognizable shape of a mature leaf, but were less than half the length of a typical mature leaf of that species (Fig. 1). When possible we also used other visible characteristics of young leaves—e.g., pubescence, color variation, and transparency—to indicate whether a leaf was young. Specimens that contained only breaking buds, with the young leaves not yet emerged, and specimens with leaves that were larger than half of full size were not used. We recorded species, date of collection, location (town, county, state), and collector for each specimen with young leaves. To avoid over-representation of well-collected localities in the data set, for any given combination of town and year, we considered only the earliest specimen collected of each species that fit the above criteria. The specimens that were removed by this procedure comprised a small percentage of the overall data set (103 specimens, ~6%), and many were collected on the same day, or within a day or 2 d of the specimen that was included in the data set. These “duplicates” were often the result of multiday field trips made by numerous collectors and therefore reflect repeated effort as opposed to phenological
changes. We used the town (or city) of collection (and the latitude and longitude of the current town centroid) to define the location for all specimens, because that was the most specific location information available in most cases. A complete list of specimens used in this project is given as Appendix S1 (see Supplemental Data with the online version of this article). To determine historical interannual variation in temperature, we obtained mean monthly temperature records from 12 weather stations dispersed across New England from NOAA’s Global Historical Climatology Network (GHCN) (National Oceanic and Atmospheric Agency, 2013; http://www.ncdc.noaa.gov/ oa/climate/ghcn-daily/) (Menne et al., 2012). Temperature data from the 19th century are limited, but this method allowed us to broadly characterize temperature patterns in a given year. Ten of the 12 stations entered operation during or before 1893, beginning with Middletown, Connecticut, in 1875. We calculated geographical variation in climate by estimating the mean monthly temperature from 1982 to 2011 for specific areas of 1 km × 1 km using the Daymet database (daymet.ornl.gov/gridded, Thornton et al., 1997). We used these data to characterize the local spring climate around each town of sample collection during the months prior to leaf-out. Analysis and modeling—For each of the 12 GHCN stations, we calculated a temperature anomaly value for every month in the record. The 12 weather stations’ anomalies were then averaged for each month and year, creating one value representing the degree to which a given month was relatively warm or cold across New England. We then tested the strength of the correlation between leaf-out dates in a year and temperature means from all possible combinations of consecutive months from January to May, to determine which metric of temperature was most relevant to leafing phenology. After this initial analysis, we used mean April temperatures for both annual and geographic variation metrics, because they were found to be most tightly correlated with leaf-out times. Including other months in addition to April did not improve the models. We chose to use mean monthly temperatures rather than a growing degree model in our analysis for simplicity of both analysis and interpretation. Growing degree models are considered to better represent the environment experienced by plants, but their results are often more difficult to interpret and, in any case, are generally strongly correlated with mean temperature models (Archetti et al., 2013). Due to the fact that the data set did not include leaf-out date for every species in every year, we investigated normalizing the data set (giving all of the species the same mean leaf out dates) by adding or subtracting correction factors. This procedure would standardize the mean leaf-out dates of all species, but we found that using correction factors made the models more complicated and did not improve them. We therefore combined data from all species in our analyses. Since both herbarium and historical temperature data were sparse in the state of Maine, we excluded these data points from regression and mapping analyses. Additionally, we excluded all herbarium specimens from before 1875, as we had no weather data from that period. Using the statistical package R (R Development Core Team, 2008), we calculated mean leaf-out dates across species and years and tested multivariate models derived from this most inclusive model: DOY ∼ B0 x1 YEAR x2 ANNCLIM
x3 LOCCLIM x4 LAT x5 LONG
x6 ANNCLIM ¸ LAT x7 LOCCLIM ¸ LAT ,
where DOY = numerical day of year of leaf-out; B0 = intercept; each x (1–7) represents the strength of the proceeding variable; ANNCLIM = New Englandwide temperature anomaly in April of the year the sample was collected, capturing annual variations in temperature; LOCCLIM = mean average temperature in April of the town where the sample was collected, capturing geographic variation in temperature; LAT = latitude of geographic center of town; LONG = longitude of geographic center of town. ANNCLIM and LOCCLIM provide distinct variables by which to target different aspects of New England-wide temperature over time (ANNCLIM) and geographically across the region (LOCCLIM). This model was reduced, based on AIC, until the most robust model was created. Once a model with only significant factors was determined, the relative strength of variables within the model was calculated using the function calc.relimp (lmg) within the R package relaimpo (Grömping, 2006). Mapping—Using geographic information system (GIS) data on the area and location of each political unit downloaded from each state’s respective government
Acer rubrum Acer saccharinum Acer saccharum Betula alleghaniensis Betula lenta Betula papyrifera Betula populifolia Carpinus caroliniana Carya glabra Carya ovata Fagus grandifolia Fraxinus americana Fraxinus nigra Fraxinus pennsylvanica Juglans cinerea Ostrya virginiana Populus grandidentata Populus tremuloides Quercus alba Quercus bicolor Quercus coccinea Quercus palustris Quercus prinus Quercus rubra Quercus velutina Sassafras albidum Ulmus americana Grand total
Species
SD 10.9 11.8 7.4 6.2 7.2 8.7 7.5 7.9 8.2 10.6 7.8 11.8 N/A 5.5 8.7 10.4 9.8 5.9 9.0 8.8 8.8 7.3 7.1 7.5 8.0 10.4 5.6 9.8
Mean
132.9 (20) 122.8 (6) 125.2 (17) 131.2 (15) 128.1 (30) 124.4 (5) 128.0 (38) 126.5 (26) 142.3 (11) 143.6 (7) 134.5 (10) 128.0 (8) 134* (1) 137.7 (3) 140.8 (18) 127.3 (25) 135.0 (9) 127.8 (5) 139.5 (17) 141.9 (7) 132.5 (13) 135.3 (10) 133.6 (10) 130.3 (9) 133.6 (19) 134.3 (37) 133.1 (9) 132.0 (385)
CT
133.8 (51) 131.4 (15) 127.0 (19) 128.8 (31) 130.7 (30) 130.0 (6) 131.1 (48) 131.4 (25) 141.9 (22) 137.9 (27) 135.7 (20) 134.7 (24) 141.0 (7) 137.0 (6) 140.7 (32) 128.0 (28) 135.3 (10) 127.1 (13) 142.2 (38) 143.4 (18) 137.8 (16) 137.0 (3) 140.6 (5) 139.1 (23) 139.9 (35) 140.8 (39) 138.9 (23) 135.6 (614)
Mean
MA
12.3 11.9 7.7 6.9 7.4 7.1 7.3 5.5 7.8 9.3 7.8 8.3 5.6 7.4 7.2 7.8 10.4 8.5 6.9 6.1 7.1 5.6 5.3 8.2 7.5 12.0 8.5 9.7
SD
ME
139.8 (13) 125.5 (2) 135.3 (8) 138.4 (12) 131.0 (2) 133.7 (20) 139.0 (8) 138.4 (7) N/A (0) 157.3 (4) 138.0 (6) 144.6 (5) 136* (1) 136.3 (4) 151.5 (4) 140.3 (13) N/A (0) 134.9 (8) 136.5 (2) 144* (1) N/A (0) N/A (0) N/A (0) 144.4 (8) N/A (0) 137* (1) 146.3 (4) 139.0 (133)
Mean 9.0 17.7 6.4 8.4 15.6 7.8 6.6 7.2 N/A 8.2 7.1 10.1 N/A 3.3 4.7 6.2 N/A 5.8 20.5 N/A N/A N/A N/A 9.8 N/A N/A 13.9 9.4
SD
NH
138.1 (19) 134.6 (12) 134.4 (7) 134.1 (13) 138.7 (7) 136.4 (13) 140.6 (24) 139.0 (5) N/A (0) 145.8 (4) 143.1 (7) 141.3 (6) N/A (0) N/A (0) 146.5 (17) 136.1 (9) 140.5 (2) 135.5 (8) 150.6 (5) 146.7 (3) N/A (0) N/A (0) N/A (0) 144.2 (18) 145.0 (4) 144.9 (12) 142.6 (12) 140.4 (207)
Mean 13.3 11.1 6.6 8.7 5.9 11.7 7.9 5.6 N/A 4.7 7.1 11.1 N/A N/A 6.8 8.7 4.9 3.3 3.6 13.1 N/A N/A N/A 6.5 9.2 9.0 17.8 10.2
SD
RI
128.3 (7) 126.3 (3) N/A (0) 130* (1) 132.3 (9) 122* (1) 130.6 (7) N/A (0) 143.0 (3) 141* (1) 130.0 (2) 134.0 (5) N/A (0) 147.5 (2) 139.5 (2) 133.8 (4) 132.5 (2) N/A (0) 142.5 (4) 144.3 (4) 136.0 (2) N/A (0) 138.7 (3) 137.0 (3) 137.3 (3) 133.9 (13) 135.0 (2) 134.7 (83)
Mean 14.9 7.6 N/A N/A 5.7 N/A 11.1 N/A 5.2 N/A 2.8 6.4 N/A 2.1 2.1 2.6 0.7 N/A 3.0 7.5 5.7 N/A 2.1 6.1 6.7 6.4 2.8 8.4
SD
139.3 (4) 139.5 (4) 131.5 (12) 136.1 (15) 135.5 (8) 144.3 (3) 134.6 (7) 139.0 (3) N/A (0) 147.6 (7) 135.9 (8) 138.9 (10) 132.8 (4) 141.3 (3) 145.1 (19) 135.0 (10) 143.4 (8) 132.9 (10) 145.0 (8) 138.0 (3) 146* (1) N/A (0) 144.4 (5) 141.5 (12) 144.0 (5) N/A (0) 128.9 (8) 138.7 (177)
Mean
VT
12.5 9.3 5.9 4.0 9.3 17.0 5.1 3.5 N/A 3.2 7.5 10.1 8.7 6.7 6.2 8.1 7.7 5.4 3.8 5.6 N/A N/A 4.0 4.7 4.5 N/A 13.7 8.6
SD
134.9 (114) 131.2 (42) 129.3 (63) 132.6 (87) 131.1 (86) 133.4 (48) 132.6 (132) 131.2 (66) 142.1 (36) 142.3 (50) 136.5 (53) 136.0 (58) 137.5 (13) 138.8 (18) 143.1 (92) 131.5 (89) 137.5 (31) 131.4 (44) 142.3 (74) 143.0 (36) 135.8 (32) 135.7 (13) 138.1 (23) 140.2 (73) 138.6 (66) 138.0 (102) 137.8 (58) 135.9 (1599)
Mean
TOTAL
12.2 11.8 7.8 7.6 7.9 10.3 8.8 8.0 7.6 10.0 7.8 10.1 7.1 6.3 7.6 9.6 9.4 7.1 7.6 7.2 8.1 6.8 7.0 8.4 8.2 11.1 12.3 10.0
SD
Mean and standard deviation of leaf-out dates with sample sizes below (in parentheses) for each combination of species and state. Species are arranged alphabetically by genus. Values with an asterisk (*) are single values of leaf-out and do not represent a true mean. The right hand column represents all specimens for New England. The bottom row represents all specimens for a state. These data are not all from the same time period and are presented here to show the range of values and sample sizes. Species were combined in all of the analyses presented in the present study; as a result, the bottom row gives a better representation of the sample size for the analyses than particular combinations of species and states. The original data on which the table is based are provided in the Supplemental Files.
TABLE 1.
August 2014] EVERILL ET AL—LEAF-OUT FROM HERBARIUM SPECIMENS 1295
1296
AMERICAN JOURNAL OF BOTANY
[Vol. 101
Fig. 1. Two specimens of Populus grandidentata accessed online from the G. S. Torrey Herbarium at the University of Connecticut (EEB, 2004). Young leaves and flowers are visible on the left specimen (collected 10 May 1936), but only mature leaves on the right specimen (collected 8 June 1932). Scale at bottom, in cm. website (Connecticut Department of Environmental Protection Natural Resources Center, 2013; Maine Office of GIS, 2013; Commonwealth of Massachusetts, 2013; NH GRANIT, 2013; Rhode Island GIS, 2013; Vermont Office of GIS, 2013), we then mapped the long-term, all-species mean leaf-out date by state, county, and town. These results were then compared visually and using regression analysis with maps and data from Landsat images of landscape greenness from 1982 to 2012. Specifically, we estimated remotely sensed longterm mean spring onset using a new algorithm that exploits dense time series of Landsat Thematic Mapper and Enhanced Thematic Mapper data at 30 m spatial resolution (Melaas et al., 2013). Using this methodology and systematic analysis, spring onset dates roughly correspond to the timing when leaf lengths reach 25% of their seasonal maximum (Melaas et al., 2013). Pixels with low seasonal amplitude in greenness were classified as nondeciduous forest (i.e., fields, evergreen forest, bodies of water. built-up areas, etc.) and removed from the analysis. Since the exact geographic coordinates of the origin of the herbarium specimens were unknown, we computed the mean leaf-out value for each town using all of the Landsat values for that town and compared these values with the respective mean herbarium leaf-out dates for the town.
RESULTS Herbarium data set—Specimens with young leaves represented approximately 15% of all specimens we examined. We identified 1599 total herbarium specimens with young leaves, representing 27 species in six states (Table 1). The geographic distribution of specimens was heterogeneous (Fig. 2), with the greatest numbers drawn from academic and urban centers such as Burlington, Vermont (University of Vermont); Boston, Massachusetts (Harvard University, Boston University, and many others); Amherst, Massachusetts (University of Massachusetts); Storrs, Connecticut (University of Connecticut); New Haven, Connecticut (Yale University); and Providence, Rhode Island (Brown University), as well along the Connecticut River Valley in Connecticut and Massachusetts. Large areas of New Hampshire, Vermont, and
Maine had only limited numbers of specimens with young leaves. After omitting data from Maine and data prior to 1875, we had a sample size of 1431 for multivariate analyses. The specimens were collected between 1834 and 2008, with about half of the specimens dating to the period 1880–1920 (51%) (Fig. 3). With the full data set containing all species, we were able to obtain at least one leaf-out value for 127 of the 140 yr (91%) over the full length of the study period, and for 100% of the years during the period 1880–1920. Species were not all equally represented, with sample sizes ranging from 13 (Fraxinus nigra and Quercus palustris) to 130 (Betula populifolia). Species and regional differences— The overall mean leafout date across all species and all years was day 136, or 16 May (SD = 10 d). Individual observations ranged from day 96 to 166, corresponding to 6 April–15 June. Mean leaf-out date differed significantly among species (F27, 1430 = 12.71, P < 0.001); the range of species’ means was 14 d, small in comparison to the 70-d range of the entire data set. In general, species of Betula and Acer leafed out early (e.g., B. lenta, mean = day 131, n = 86; A. saccharum, mean = day 129, n = 63). Species of Quercus and Juglans cinera leaf out relatively late (e.g., Q. alba, mean = day 142, n = 74; J. cinera, mean = day 143, n = 92). Species’ leaf-out dates were highly correlated among states (i.e., CT:MA, slope = 0.85, P < 0.001, R2 = 0.57), indicating that the order of leafing out is relatively constant across different geographical areas. In addition to variation among species, mean leaf-out dates varied significantly among states (F4,1430 = 30.59, P < 0.001). Leaf-out occurred earliest in Connecticut (mean = 132, n = 376), and latest in the more northern, mountainous state of New Hampshire (mean = 140, n = 196).
August 2014]
EVERILL ET AL—LEAF-OUT FROM HERBARIUM SPECIMENS
1297
DOY 301.70 − 0.02 YEAR − 2.70 ANNCLIM
− 2.06 LOCCLIM − 1.56 LONG .
Warmer years (ANNCLIM), warmer locations (LOCCLIM), more western longitude (LONG), and changes over time (YEAR) were all significantly associated with earlier leaf-out (R2 = 0.17, P < 0.001, Table 2). The strongest effect was caused by annual variation in April temperature, with trees on average leafing out 2.70 d earlier for every degree C increase in April temperature (Fig. 4). Annual temperature explains the majority, 63.6%, of the total variance of the model. Leaf-out also occurred 2.06 d earlier for every degree C increase in local temperature, which corresponds to 30.6% of the overall variance. The weakest effects were longitude, with leaf-out occurring 1.56 d earlier for every additional degree of longitude, i.e., travel westward (3.9% of the variance), and −0.02 d early for each successive year (YEAR, 1.8% of the variance).
Fig. 2. The total number of herbarium specimens with young leaves for each town and city in New England. Towns in white were not represented in our data set. Locations of major university herbaria are indicated; note that they tend to be densely sampled.
Factors correlated with leaf-out— Using the full data set of 1599 specimens, we found that leaf-out date changed significantly over time, advancing roughly 5 d over the period of study (slope = −0.04, R2 = 0.02, P < 0.001) or 0.4 d earlier per decade (Appendix S2, see online Supplemental Data). Even though trees are leafing out earlier over time, the amount of variation explained is relatively small. The advance in leaf-out times was likely driven by annual temperatures that warmed during this same period (slope = 0.01°C/yr, R2 = 0.02, P < 0.001). Among all possible configurations of factors, the strongest multivariate linear model, i.e., that with the lowest AIC, is the following:
Fig. 3. Number of herbarium specimens with young leaves that were collected per decade; labels show the sample size in each decade.
Spatial patterns of leaf-out— Spatial patterns in leaf-out times were distinctive and easily noticeable (Fig. 5). The Connecticut River Valley and major cities, including Boston, Worcester, and Providence, have relatively early leaf-out times. Areas with late leaf-out times are concentrated in cool, mountainous towns in northern New Hampshire, Vermont and Western Massachusetts, and on the relatively cool Cape Cod and islands of Massachusetts. The spatial patterns observed using both herbarium specimens and Landsat data (Melaas et al., 2013; E. K. Melaas, unpublished data) are similar. Using the towns for which there are both herbarium data (with at least 5 specimens) and remote sensing data (n of towns = 86), regression analysis shows that observed leaf-out dates determined by the two methods are significantly related (slope = 0.34, R2= 0.32, P < 0.001; online Appendix S3). However, the low slope indicates that the herbarium specimen method is recording more variation and later leaf-out dates than is the remote sensing method. DISCUSSION Validity and quality of methodology— The main purpose of this study was to determine whether scientifically useful data on leaf-out times could be obtained from herbarium specimens. We found that herbarium specimens contain large quantities of historical information on leaf-out dates, at least in well-studied areas of New England. Focusing on 27 tree species, we found that approximately 15% of the herbarium specimens examined in this study had young leaves, resulting in 1599 leaf-out dates over a 175 yr period. We also found that the process of classifying specimens for their phenophase was repeatable. Though the period of time during which trees have visually distinctive young leaves might vary by species, it does not appear that there is undue uncertainty in identifying young leaves. A number of lines of evidence suggest that leaf-out data from herbarium specimens is an accurate reflection of true patterns in leaf-out and can contribute to the study of climate change and leaf-out phenology. First, in our analysis we found a consistent order of leaf-out times among New England states, with some species leafing early in each state (such as species of Acer and Betula) and some late in each state (such as species of Quercus and Juglans cinerea).
1298
AMERICAN JOURNAL OF BOTANY
TABLE 2.
Results of the best-fit multivariate model that describes variation in leaf-out dates in New England. Model: DOY ~ B0 + LONG + ANNCLIM + LOCCLIM.
Factor YEAR LONG ANNCLIM LOCCLIM
Estimate
SE
t value
P
−0.02 −1.56 −2.70 −2.06
0.01 0.38 0.23 0.24
−2.05 −4.14 −11.55 −8.61
0.04
DETERMINING PAST LEAF-OUT TIMES OF NEW ENGLAND’S DECIDUOUS FORESTS FROM HERBARIUM SPECIMENS1
PETER H. EVERILL2, RICHARD B. PRIMACK2,5, ELIZABETH R. ELLWOOD3, AND ELI K. MELAAS4 2Boston University Department of Biology, 5 Cummington Mall, Boston, Massachusetts 02215 USA; 3Department of Biological Science, Florida State University, Tallahassee, Florida 32306 USA; and 4Boston University Department of Earth & Environment, 685 Commonwealth Avenue, Room 130, Boston, Massachusetts 02215
• Premise of the study: There is great interest in studying leaf-out times of temperate forests because of the importance of leaf-out in controlling ecosystem processes, especially in the face of a changing climate. Remote sensing and modeling, combined with weather records and field observations, are increasing our knowledge of factors affecting variation in leaf-out times. Herbarium specimens represent a potential new source of information to determine whether the variation in leaf-out times observed in recent decades is comparable to longer time frames over past centuries. • Methods: Here we introduce the use of herbarium specimens as a method for studying long-term changes in leaf-out times of deciduous trees. We collected historical leaf-out data for the years 1834–2008 from common deciduous trees in New England using 1599 dated herbarium specimens with young leaves. • Key results: We found that leaf-out dates are strongly affected by spring temperature, with trees leafing out 2.70 d earlier for each degree C increase in mean April temperature. For each degree C increase in local temperature, trees leafed out 2.06 d earlier. Additionally, the mean response of leaf-out dates across all species and sites over time was 0.4 d earlier per decade. Our results are of comparable magnitude to results from studies using remote sensing and direct field observations. • Conclusions: Across New England, mean leaf-out dates varied geographically in close correspondence with those observed in studies using satellite data. This study demonstrates that herbarium specimens can be a valuable source of data on past leaf-out times of deciduous trees. Key words: climate change; forests; herbarium specimens; leaf-out; New England; phenology; remote sensing; trees.
Spring leaf-out of deciduous trees engages many ecosystem processes (Richardson et al., 2009; Hufkens et al., 2012) and influences plant community structure (Sherry et al., 2007; Willis et al., 2010). In temperate ecosystems, leaf development is tightly controlled by temperature (Linkosalo et al., 2006) and to some extent by day length (Körner and Basler, 2010; Laube et al., 2013, Polgar et al., 2013); as spring temperatures rise, leaf-out occurs earlier (Morin et al., 2010; Polgar and Primack, 2011). Thus, warmer temperatures could lengthen the growing season and increase carbon uptake in some forests, acting as a negative feedback on climate change (Migliavacca et al., 2012; Richardson et al., 2013). Additionally, earlier leaf-out could alter trophic interactions and the cycling of water and nutrients (Tabacchi et al., 2000; Parmesan, 2006; Morisette et al., 2009). Understanding the dynamics of these relationships is increasingly important considering recent global increases in temperature, predictions for continued warming (IPCC, 2007), and the impacts a changing climate will have on spring leaf-out and associated ecosystem processes.
1 Manuscript received 31 January 2014; revision accepted 21 July 2014. The authors thank the curators of the various herbaria for the use of their collections. Support for this project came from the iDigBio program of the National Science Foundation. Comments on the manuscript were provided by M. Boeni, A. Miller-Rushing, A. Gallinat, C. Polgar, C. MacKenzie, J. Silander, and P. Sweeney. 5 Author for correspondence (e-mail: [email protected])
doi:10.3732/ajb.1400045
Given the ecological importance of spring leaf-out, there is much interest in understanding how trees have responded to past climatic variability to predict how the timing of leaf development will change in the future. Previous studies of leaf-out phenology have generally used one of three methods (Polgar and Primack, 2011): direct field observations, near-surface remote sensing with canopy or understory cameras, and satellitebased remote sensing. The simplest of these methods is direct observation of plants, categorizing buds and young leaves by phenophase. In practice, this is time-consuming, and historical data are generally available only for specific locations, such as a garden or forest plot (e.g., Morin et al., 2009). Citizen science programs in which large numbers of volunteers monitor plant phenology offer expanded geographic coverage for some species, particularly for recent years (Menzel et al., 2001; Wolfe et al., 2005). In near-surface remote sensing, scientists have begun to use phenocams—time-lapse cameras that capture images of a landscape at scales that can range from individual trees to entire landscapes. The images can be analyzed for various measurements of leaf cover development (Richardson et al., 2007; Ahrends et al., 2008; Sonnentag et al., 2012). At larger geographical scales, multispectral images from satellite instruments (primarily the Landsat Thematic Mapper and Enhanced Thematic Mapper, the moderate resolution imaging spectroradiometer (MODIS), and the advanced very high resolution radiometer) can be used to monitor the phenology of landscape greenness (Fisher et al., 2006). Unfortunately, despite increased geographic coverage, phenocam and satellite data are available only for a short historical period—since 1972 for Landsat, 2000 for MODIS, and more recently for phenocams.
American Journal of Botany 101(8): 1293–1300, 2014; http://www.amjbot.org/ © 2014 Botanical Society of America
1293
1294
[Vol. 101
AMERICAN JOURNAL OF BOTANY
At present, researchers are actively working to relate greenness measures made by phenocams and satellites to species-specific stages of leaf development and key physiological stages such as the onset of photosynthesis (White et al., 2009; Sonnentag et al., 2012). Herbarium specimens provide a largely untapped source of historical data on tree leaf-out dates. An estimated 350 million plant specimens reside in over 3000 herbaria worldwide, and in some cases, the specimens were collected several hundred years ago (New York Botanical Garden’s Virtual Herbarium: http:// sweetgum.nybg.org/ih/). Herbarium specimens have in the past been used to study flowering phenology of various species in New England (Primack et al., 2004, Miller-Rushing et al., 2006, Primack and Miller-Rushing, 2009), Europe (Diskin et al., 2012), and elsewhere (e.g., Gaira et al., 2011). Additionally, many herbarium specimens contain important information regarding leaf-out times. For example, research based on herbarium specimens in Munich, Germany demonstrated climate tracking in four woody species (Zohner and Renner, 2014). Many of the common deciduous trees of the temperate forest, such as species of Quercus and Acer, produce their first flush of young leaves at approximately the same time as their flowers. Collectors interested in flowers often unwittingly collect herbarium specimens that contain young leaves, which could provide a measure of leaf-out phenology. Leaf-out of most New England tree species, including those researched here, occurs over a relatively short period of time for each species and does not recur later in the season. In this study, we investigated whether herbarium specimens of common New England trees can be used as a source of information about historical leaf-out times and how leaf-out times are affected by climatic variation. We also investigated whether recent regional patterns of leaf-out indicated by satellite-based remote sensing match long-term historical leaf-out data collected from herbarium specimens. MATERIALS AND METHODS Data collection—We examined herbarium specimens of 27 of the most common canopy tree species (Table 1) of New England’s deciduous forests and identified specimens with young leaves. These species were ones that leafed out around the time of flowering, so that specimens often had both flowers and young leaves. We only examined specimens that had been collected from March through June, the time of spring leaf-out. We examined herbarium specimens at seven of the largest herbaria in the region: Brown University’s Stephen T. Olney Herbarium, Harvard University Herbaria, George Safford Torrey Herbarium at the University of Connecticut, University of Maine Herbaria, Hodgdon Herbarium at the University of New Hampshire, University of Vermont Herbarium, and the Yale University Herbarium. We considered leaves as young if they were emerging from the breaking bud, i.e., they were unfolded and had the recognizable shape of a mature leaf, but were less than half the length of a typical mature leaf of that species (Fig. 1). When possible we also used other visible characteristics of young leaves—e.g., pubescence, color variation, and transparency—to indicate whether a leaf was young. Specimens that contained only breaking buds, with the young leaves not yet emerged, and specimens with leaves that were larger than half of full size were not used. We recorded species, date of collection, location (town, county, state), and collector for each specimen with young leaves. To avoid over-representation of well-collected localities in the data set, for any given combination of town and year, we considered only the earliest specimen collected of each species that fit the above criteria. The specimens that were removed by this procedure comprised a small percentage of the overall data set (103 specimens, ~6%), and many were collected on the same day, or within a day or 2 d of the specimen that was included in the data set. These “duplicates” were often the result of multiday field trips made by numerous collectors and therefore reflect repeated effort as opposed to phenological
changes. We used the town (or city) of collection (and the latitude and longitude of the current town centroid) to define the location for all specimens, because that was the most specific location information available in most cases. A complete list of specimens used in this project is given as Appendix S1 (see Supplemental Data with the online version of this article). To determine historical interannual variation in temperature, we obtained mean monthly temperature records from 12 weather stations dispersed across New England from NOAA’s Global Historical Climatology Network (GHCN) (National Oceanic and Atmospheric Agency, 2013; http://www.ncdc.noaa.gov/ oa/climate/ghcn-daily/) (Menne et al., 2012). Temperature data from the 19th century are limited, but this method allowed us to broadly characterize temperature patterns in a given year. Ten of the 12 stations entered operation during or before 1893, beginning with Middletown, Connecticut, in 1875. We calculated geographical variation in climate by estimating the mean monthly temperature from 1982 to 2011 for specific areas of 1 km × 1 km using the Daymet database (daymet.ornl.gov/gridded, Thornton et al., 1997). We used these data to characterize the local spring climate around each town of sample collection during the months prior to leaf-out. Analysis and modeling—For each of the 12 GHCN stations, we calculated a temperature anomaly value for every month in the record. The 12 weather stations’ anomalies were then averaged for each month and year, creating one value representing the degree to which a given month was relatively warm or cold across New England. We then tested the strength of the correlation between leaf-out dates in a year and temperature means from all possible combinations of consecutive months from January to May, to determine which metric of temperature was most relevant to leafing phenology. After this initial analysis, we used mean April temperatures for both annual and geographic variation metrics, because they were found to be most tightly correlated with leaf-out times. Including other months in addition to April did not improve the models. We chose to use mean monthly temperatures rather than a growing degree model in our analysis for simplicity of both analysis and interpretation. Growing degree models are considered to better represent the environment experienced by plants, but their results are often more difficult to interpret and, in any case, are generally strongly correlated with mean temperature models (Archetti et al., 2013). Due to the fact that the data set did not include leaf-out date for every species in every year, we investigated normalizing the data set (giving all of the species the same mean leaf out dates) by adding or subtracting correction factors. This procedure would standardize the mean leaf-out dates of all species, but we found that using correction factors made the models more complicated and did not improve them. We therefore combined data from all species in our analyses. Since both herbarium and historical temperature data were sparse in the state of Maine, we excluded these data points from regression and mapping analyses. Additionally, we excluded all herbarium specimens from before 1875, as we had no weather data from that period. Using the statistical package R (R Development Core Team, 2008), we calculated mean leaf-out dates across species and years and tested multivariate models derived from this most inclusive model: DOY ∼ B0 x1 YEAR x2 ANNCLIM
x3 LOCCLIM x4 LAT x5 LONG
x6 ANNCLIM ¸ LAT x7 LOCCLIM ¸ LAT ,
where DOY = numerical day of year of leaf-out; B0 = intercept; each x (1–7) represents the strength of the proceeding variable; ANNCLIM = New Englandwide temperature anomaly in April of the year the sample was collected, capturing annual variations in temperature; LOCCLIM = mean average temperature in April of the town where the sample was collected, capturing geographic variation in temperature; LAT = latitude of geographic center of town; LONG = longitude of geographic center of town. ANNCLIM and LOCCLIM provide distinct variables by which to target different aspects of New England-wide temperature over time (ANNCLIM) and geographically across the region (LOCCLIM). This model was reduced, based on AIC, until the most robust model was created. Once a model with only significant factors was determined, the relative strength of variables within the model was calculated using the function calc.relimp (lmg) within the R package relaimpo (Grömping, 2006). Mapping—Using geographic information system (GIS) data on the area and location of each political unit downloaded from each state’s respective government
Acer rubrum Acer saccharinum Acer saccharum Betula alleghaniensis Betula lenta Betula papyrifera Betula populifolia Carpinus caroliniana Carya glabra Carya ovata Fagus grandifolia Fraxinus americana Fraxinus nigra Fraxinus pennsylvanica Juglans cinerea Ostrya virginiana Populus grandidentata Populus tremuloides Quercus alba Quercus bicolor Quercus coccinea Quercus palustris Quercus prinus Quercus rubra Quercus velutina Sassafras albidum Ulmus americana Grand total
Species
SD 10.9 11.8 7.4 6.2 7.2 8.7 7.5 7.9 8.2 10.6 7.8 11.8 N/A 5.5 8.7 10.4 9.8 5.9 9.0 8.8 8.8 7.3 7.1 7.5 8.0 10.4 5.6 9.8
Mean
132.9 (20) 122.8 (6) 125.2 (17) 131.2 (15) 128.1 (30) 124.4 (5) 128.0 (38) 126.5 (26) 142.3 (11) 143.6 (7) 134.5 (10) 128.0 (8) 134* (1) 137.7 (3) 140.8 (18) 127.3 (25) 135.0 (9) 127.8 (5) 139.5 (17) 141.9 (7) 132.5 (13) 135.3 (10) 133.6 (10) 130.3 (9) 133.6 (19) 134.3 (37) 133.1 (9) 132.0 (385)
CT
133.8 (51) 131.4 (15) 127.0 (19) 128.8 (31) 130.7 (30) 130.0 (6) 131.1 (48) 131.4 (25) 141.9 (22) 137.9 (27) 135.7 (20) 134.7 (24) 141.0 (7) 137.0 (6) 140.7 (32) 128.0 (28) 135.3 (10) 127.1 (13) 142.2 (38) 143.4 (18) 137.8 (16) 137.0 (3) 140.6 (5) 139.1 (23) 139.9 (35) 140.8 (39) 138.9 (23) 135.6 (614)
Mean
MA
12.3 11.9 7.7 6.9 7.4 7.1 7.3 5.5 7.8 9.3 7.8 8.3 5.6 7.4 7.2 7.8 10.4 8.5 6.9 6.1 7.1 5.6 5.3 8.2 7.5 12.0 8.5 9.7
SD
ME
139.8 (13) 125.5 (2) 135.3 (8) 138.4 (12) 131.0 (2) 133.7 (20) 139.0 (8) 138.4 (7) N/A (0) 157.3 (4) 138.0 (6) 144.6 (5) 136* (1) 136.3 (4) 151.5 (4) 140.3 (13) N/A (0) 134.9 (8) 136.5 (2) 144* (1) N/A (0) N/A (0) N/A (0) 144.4 (8) N/A (0) 137* (1) 146.3 (4) 139.0 (133)
Mean 9.0 17.7 6.4 8.4 15.6 7.8 6.6 7.2 N/A 8.2 7.1 10.1 N/A 3.3 4.7 6.2 N/A 5.8 20.5 N/A N/A N/A N/A 9.8 N/A N/A 13.9 9.4
SD
NH
138.1 (19) 134.6 (12) 134.4 (7) 134.1 (13) 138.7 (7) 136.4 (13) 140.6 (24) 139.0 (5) N/A (0) 145.8 (4) 143.1 (7) 141.3 (6) N/A (0) N/A (0) 146.5 (17) 136.1 (9) 140.5 (2) 135.5 (8) 150.6 (5) 146.7 (3) N/A (0) N/A (0) N/A (0) 144.2 (18) 145.0 (4) 144.9 (12) 142.6 (12) 140.4 (207)
Mean 13.3 11.1 6.6 8.7 5.9 11.7 7.9 5.6 N/A 4.7 7.1 11.1 N/A N/A 6.8 8.7 4.9 3.3 3.6 13.1 N/A N/A N/A 6.5 9.2 9.0 17.8 10.2
SD
RI
128.3 (7) 126.3 (3) N/A (0) 130* (1) 132.3 (9) 122* (1) 130.6 (7) N/A (0) 143.0 (3) 141* (1) 130.0 (2) 134.0 (5) N/A (0) 147.5 (2) 139.5 (2) 133.8 (4) 132.5 (2) N/A (0) 142.5 (4) 144.3 (4) 136.0 (2) N/A (0) 138.7 (3) 137.0 (3) 137.3 (3) 133.9 (13) 135.0 (2) 134.7 (83)
Mean 14.9 7.6 N/A N/A 5.7 N/A 11.1 N/A 5.2 N/A 2.8 6.4 N/A 2.1 2.1 2.6 0.7 N/A 3.0 7.5 5.7 N/A 2.1 6.1 6.7 6.4 2.8 8.4
SD
139.3 (4) 139.5 (4) 131.5 (12) 136.1 (15) 135.5 (8) 144.3 (3) 134.6 (7) 139.0 (3) N/A (0) 147.6 (7) 135.9 (8) 138.9 (10) 132.8 (4) 141.3 (3) 145.1 (19) 135.0 (10) 143.4 (8) 132.9 (10) 145.0 (8) 138.0 (3) 146* (1) N/A (0) 144.4 (5) 141.5 (12) 144.0 (5) N/A (0) 128.9 (8) 138.7 (177)
Mean
VT
12.5 9.3 5.9 4.0 9.3 17.0 5.1 3.5 N/A 3.2 7.5 10.1 8.7 6.7 6.2 8.1 7.7 5.4 3.8 5.6 N/A N/A 4.0 4.7 4.5 N/A 13.7 8.6
SD
134.9 (114) 131.2 (42) 129.3 (63) 132.6 (87) 131.1 (86) 133.4 (48) 132.6 (132) 131.2 (66) 142.1 (36) 142.3 (50) 136.5 (53) 136.0 (58) 137.5 (13) 138.8 (18) 143.1 (92) 131.5 (89) 137.5 (31) 131.4 (44) 142.3 (74) 143.0 (36) 135.8 (32) 135.7 (13) 138.1 (23) 140.2 (73) 138.6 (66) 138.0 (102) 137.8 (58) 135.9 (1599)
Mean
TOTAL
12.2 11.8 7.8 7.6 7.9 10.3 8.8 8.0 7.6 10.0 7.8 10.1 7.1 6.3 7.6 9.6 9.4 7.1 7.6 7.2 8.1 6.8 7.0 8.4 8.2 11.1 12.3 10.0
SD
Mean and standard deviation of leaf-out dates with sample sizes below (in parentheses) for each combination of species and state. Species are arranged alphabetically by genus. Values with an asterisk (*) are single values of leaf-out and do not represent a true mean. The right hand column represents all specimens for New England. The bottom row represents all specimens for a state. These data are not all from the same time period and are presented here to show the range of values and sample sizes. Species were combined in all of the analyses presented in the present study; as a result, the bottom row gives a better representation of the sample size for the analyses than particular combinations of species and states. The original data on which the table is based are provided in the Supplemental Files.
TABLE 1.
August 2014] EVERILL ET AL—LEAF-OUT FROM HERBARIUM SPECIMENS 1295
1296
AMERICAN JOURNAL OF BOTANY
[Vol. 101
Fig. 1. Two specimens of Populus grandidentata accessed online from the G. S. Torrey Herbarium at the University of Connecticut (EEB, 2004). Young leaves and flowers are visible on the left specimen (collected 10 May 1936), but only mature leaves on the right specimen (collected 8 June 1932). Scale at bottom, in cm. website (Connecticut Department of Environmental Protection Natural Resources Center, 2013; Maine Office of GIS, 2013; Commonwealth of Massachusetts, 2013; NH GRANIT, 2013; Rhode Island GIS, 2013; Vermont Office of GIS, 2013), we then mapped the long-term, all-species mean leaf-out date by state, county, and town. These results were then compared visually and using regression analysis with maps and data from Landsat images of landscape greenness from 1982 to 2012. Specifically, we estimated remotely sensed longterm mean spring onset using a new algorithm that exploits dense time series of Landsat Thematic Mapper and Enhanced Thematic Mapper data at 30 m spatial resolution (Melaas et al., 2013). Using this methodology and systematic analysis, spring onset dates roughly correspond to the timing when leaf lengths reach 25% of their seasonal maximum (Melaas et al., 2013). Pixels with low seasonal amplitude in greenness were classified as nondeciduous forest (i.e., fields, evergreen forest, bodies of water. built-up areas, etc.) and removed from the analysis. Since the exact geographic coordinates of the origin of the herbarium specimens were unknown, we computed the mean leaf-out value for each town using all of the Landsat values for that town and compared these values with the respective mean herbarium leaf-out dates for the town.
RESULTS Herbarium data set—Specimens with young leaves represented approximately 15% of all specimens we examined. We identified 1599 total herbarium specimens with young leaves, representing 27 species in six states (Table 1). The geographic distribution of specimens was heterogeneous (Fig. 2), with the greatest numbers drawn from academic and urban centers such as Burlington, Vermont (University of Vermont); Boston, Massachusetts (Harvard University, Boston University, and many others); Amherst, Massachusetts (University of Massachusetts); Storrs, Connecticut (University of Connecticut); New Haven, Connecticut (Yale University); and Providence, Rhode Island (Brown University), as well along the Connecticut River Valley in Connecticut and Massachusetts. Large areas of New Hampshire, Vermont, and
Maine had only limited numbers of specimens with young leaves. After omitting data from Maine and data prior to 1875, we had a sample size of 1431 for multivariate analyses. The specimens were collected between 1834 and 2008, with about half of the specimens dating to the period 1880–1920 (51%) (Fig. 3). With the full data set containing all species, we were able to obtain at least one leaf-out value for 127 of the 140 yr (91%) over the full length of the study period, and for 100% of the years during the period 1880–1920. Species were not all equally represented, with sample sizes ranging from 13 (Fraxinus nigra and Quercus palustris) to 130 (Betula populifolia). Species and regional differences— The overall mean leafout date across all species and all years was day 136, or 16 May (SD = 10 d). Individual observations ranged from day 96 to 166, corresponding to 6 April–15 June. Mean leaf-out date differed significantly among species (F27, 1430 = 12.71, P < 0.001); the range of species’ means was 14 d, small in comparison to the 70-d range of the entire data set. In general, species of Betula and Acer leafed out early (e.g., B. lenta, mean = day 131, n = 86; A. saccharum, mean = day 129, n = 63). Species of Quercus and Juglans cinera leaf out relatively late (e.g., Q. alba, mean = day 142, n = 74; J. cinera, mean = day 143, n = 92). Species’ leaf-out dates were highly correlated among states (i.e., CT:MA, slope = 0.85, P < 0.001, R2 = 0.57), indicating that the order of leafing out is relatively constant across different geographical areas. In addition to variation among species, mean leaf-out dates varied significantly among states (F4,1430 = 30.59, P < 0.001). Leaf-out occurred earliest in Connecticut (mean = 132, n = 376), and latest in the more northern, mountainous state of New Hampshire (mean = 140, n = 196).
August 2014]
EVERILL ET AL—LEAF-OUT FROM HERBARIUM SPECIMENS
1297
DOY 301.70 − 0.02 YEAR − 2.70 ANNCLIM
− 2.06 LOCCLIM − 1.56 LONG .
Warmer years (ANNCLIM), warmer locations (LOCCLIM), more western longitude (LONG), and changes over time (YEAR) were all significantly associated with earlier leaf-out (R2 = 0.17, P < 0.001, Table 2). The strongest effect was caused by annual variation in April temperature, with trees on average leafing out 2.70 d earlier for every degree C increase in April temperature (Fig. 4). Annual temperature explains the majority, 63.6%, of the total variance of the model. Leaf-out also occurred 2.06 d earlier for every degree C increase in local temperature, which corresponds to 30.6% of the overall variance. The weakest effects were longitude, with leaf-out occurring 1.56 d earlier for every additional degree of longitude, i.e., travel westward (3.9% of the variance), and −0.02 d early for each successive year (YEAR, 1.8% of the variance).
Fig. 2. The total number of herbarium specimens with young leaves for each town and city in New England. Towns in white were not represented in our data set. Locations of major university herbaria are indicated; note that they tend to be densely sampled.
Factors correlated with leaf-out— Using the full data set of 1599 specimens, we found that leaf-out date changed significantly over time, advancing roughly 5 d over the period of study (slope = −0.04, R2 = 0.02, P < 0.001) or 0.4 d earlier per decade (Appendix S2, see online Supplemental Data). Even though trees are leafing out earlier over time, the amount of variation explained is relatively small. The advance in leaf-out times was likely driven by annual temperatures that warmed during this same period (slope = 0.01°C/yr, R2 = 0.02, P < 0.001). Among all possible configurations of factors, the strongest multivariate linear model, i.e., that with the lowest AIC, is the following:
Fig. 3. Number of herbarium specimens with young leaves that were collected per decade; labels show the sample size in each decade.
Spatial patterns of leaf-out— Spatial patterns in leaf-out times were distinctive and easily noticeable (Fig. 5). The Connecticut River Valley and major cities, including Boston, Worcester, and Providence, have relatively early leaf-out times. Areas with late leaf-out times are concentrated in cool, mountainous towns in northern New Hampshire, Vermont and Western Massachusetts, and on the relatively cool Cape Cod and islands of Massachusetts. The spatial patterns observed using both herbarium specimens and Landsat data (Melaas et al., 2013; E. K. Melaas, unpublished data) are similar. Using the towns for which there are both herbarium data (with at least 5 specimens) and remote sensing data (n of towns = 86), regression analysis shows that observed leaf-out dates determined by the two methods are significantly related (slope = 0.34, R2= 0.32, P < 0.001; online Appendix S3). However, the low slope indicates that the herbarium specimen method is recording more variation and later leaf-out dates than is the remote sensing method. DISCUSSION Validity and quality of methodology— The main purpose of this study was to determine whether scientifically useful data on leaf-out times could be obtained from herbarium specimens. We found that herbarium specimens contain large quantities of historical information on leaf-out dates, at least in well-studied areas of New England. Focusing on 27 tree species, we found that approximately 15% of the herbarium specimens examined in this study had young leaves, resulting in 1599 leaf-out dates over a 175 yr period. We also found that the process of classifying specimens for their phenophase was repeatable. Though the period of time during which trees have visually distinctive young leaves might vary by species, it does not appear that there is undue uncertainty in identifying young leaves. A number of lines of evidence suggest that leaf-out data from herbarium specimens is an accurate reflection of true patterns in leaf-out and can contribute to the study of climate change and leaf-out phenology. First, in our analysis we found a consistent order of leaf-out times among New England states, with some species leafing early in each state (such as species of Acer and Betula) and some late in each state (such as species of Quercus and Juglans cinerea).
1298
AMERICAN JOURNAL OF BOTANY
TABLE 2.
Results of the best-fit multivariate model that describes variation in leaf-out dates in New England. Model: DOY ~ B0 + LONG + ANNCLIM + LOCCLIM.
Factor YEAR LONG ANNCLIM LOCCLIM
Estimate
SE
t value
P
−0.02 −1.56 −2.70 −2.06
0.01 0.38 0.23 0.24
−2.05 −4.14 −11.55 −8.61
0.04
