Plant phenological responses to a long-term experimental extension of growing season and soil warming in the tussock tundra of Alaska
Abstract
Climate warming is strongly altering the timing of season initiation and season length in the Arctic. Phenological activities are among the most sensitive plant responses to climate change and have important effects at all levels within the ecosystem. We tested the effects of two experimental treatments, extended growing season via snow removal and extended growing season combined with soil warming, on plant phenology in tussock tundra in Alaska from 1995 through 2003. We specifically monitored the responses of eight species, representing four growth forms: (i) graminoids (Carex bigellowii and Eriophorum vaginatum); (ii) evergreen shrubs (Ledum palustre, Cassiope tetragona, and Vaccinium vitis-idaea); (iii) deciduous shrubs (Betula nana and Salix pulchra); and (iv) forbs (Polygonum bistorta). Our study answered three questions: (i) Do experimental treatments affect the timing of leaf bud break, flowering, and leaf senescence? (ii) Are responses to treatments species-specific and growth form-specific? and (iii) Which environmental factors best predict timing of phenophases? Treatment significantly affected the timing of all three phenophases, although the two experimental treatments did not differ from each other. While phenological events began earlier in the experimental plots relative to the controls, duration of phenophases did not increase. The evergreen shrub, Cassiope tetragona, did not respond to either experimental treatment. While the other species did respond to experimental treatments, the total active period for these species did not increase relative to the control. Air temperature was consistently the best predictor of phenology. Our results imply that some evergreen shrubs (i.e., C. tetragona) will not capitalize on earlier favorable growing conditions, putting them at a competitive disadvantage relative to phenotypically plastic deciduous shrubs. Our findings also suggest that an early onset of the growing season as a result of decreased snow cover will not necessarily result in greater tundra productivity.
Introduction
High-latitude regions are projected to be especially sensitive to increases in temperature as a result of climate change (Bonan et al., 1995; Intergovernmental Panel on Climate Change (IPCC), 2014). Although most changes are projected to take place during the cold season (Overpeck et al., 1997; Serreze et al., 2000), changes in surface air temperature at the beginning of the growing season have significant effects on timing of snow melt (Cleland et al., 2007; Wipf & Rixen, 2010). Earlier snow melt and reduced snow cover have been reported in North America, particularly for northern Alaska (Frei et al., 1999; Stone et al., 2002). These changes in snow properties will drive the growing season as well as affect other important abiotic factors such as thaw depth, soil temperature, water and nutrient availability, and energy fluxes (Billings & Mooney, 1968; Van Wijk et al., 2003). Environmental changes will, in turn, have cascading effects on population dynamics and community composition of these northern ecosystems (Totland & Alatalo, 2002). Thus, understanding responses of arctic plants to changes in snow melt will be critical to predict shifts in plant community structure and function.
One important consequence of changes in timing of snow melt is the onset and duration of the growing season. Evidence points to a mean of 10–20 days increase in the growing season over the past three decades (Linderholm, 2006). Tundra plants have the unique characteristic of undergoing extremely rapid leaf and flower expansion (i.e., 1–2 weeks) (Billings & Mooney, 1968). Thus, even a small increase of a few days available for plant growth has major implications for plant populations and their pollinators/seed dispersers. However, not all species, nor plant functional types, may show plastic responses to changes in the growing season (Arft et al., 1999; Starr et al., 2000). It is imperative, therefore, to consider variation in species and growth form when predicting how the vegetation landscape will be modified as the climate changes.
In general, arctic regions are experiencing earlier green-up and flowering, as well as longer periods of greening and flowering than in the past (Myneni et al., 1997; Menzel & Fabian, 1999; McCarty, 2001; Jia et al., 2003; Oberbauer et al., 2013; Wu & Liu, 2013). Earlier onset of phenophases is a response to changes in the abiotic environment and seasonality (Ernakovich et al., 2014). For example, the spring flowering index correlates with the incidence of El Niño events and rising sea surface temperatures (Beaubien & Freeland, 2000), which cause increased surface air temperatures and earlier snow melt (Billings, 1987). Earlier phenology may also be explained by shifts in vegetation cover. Growing evidence suggests that rapid climate changes during the twentieth century have caused poleward shifts in species distributions across a wide range of taxonomic groups (Hughes, 2000; Walther et al., 2002; Parmesan & Yohe, 2003). One notable example of a rapid range shift is the increase in abundance of woody shrubs (namely Betula, Salix, and Alnus) in the Arctic, presumably a response to changes in temperatures, snow melt, and the growing season length (Sturm et al., 2001a; Tape et al., 2006; Higuera et al., 2008; Myers-Smith et al., 2011). Deciduous shrubs in arctic environments have been shown to initiate phenological events before other growth forms (Arft et al., 1999; Molau et al., 2005), potentially advancing phenology of the entire community.
Given the pervasive changes in arctic plant communities and vegetation distribution, we urgently need field studies that document species-specific and growth form-specific responses to climate change. Although we know that plant functional types respond in different ways to environmental manipulation, we still do not clearly understand which functional types will acclimate best to altered environmental conditions (Dormann & Woodin, 2002). As part of the ongoing work of the International Tundra Experiment (ITEX) (Webber & Walker, 1991), our study examined the effects of lengthened growing season and soil warming on the vegetative and reproductive phenology of eight species in the tussock tundra of northern Alaska. We simulated conditions predicted for climate change, namely earlier loss of winter snowpack, later development of snowpack, and greater soil temperature during the growing season. We accomplished these simulations by manually removing snow early in the spring and maintaining plots snow free at the end of the season, as well as warming the soil (Oberbauer et al., 1998; Starr et al., 2000). While experimental warming and/or season manipulation is an increasingly common approach used to quantify phenological responses of plants to climate change (Chapin et al., 1995; de Valpine & Harte, 2001; Dunne et al., 2003; Hollister et al., 2005; Buizer et al., 2012), our research is unique both in its methodology and duration. First, as opposed to using snow fences, which can result in large nutrient inputs (Fahnestock et al., 2000), we manipulated snow cover by manually removing snow. Second, our study separates the effects of earlier snow melt and soil warming. Finally, the duration of our study is one of the longest for active season extension experiments in the low Arctic (Wipf & Rixen, 2010).
Our study aims to answer three primary questions: (i) Do early snow removal and soil warming affect the timing of leaf bud break, flowering, and leaf senescence, as well as the proportion of individuals in each phenophase? (ii) Are responses to treatments species-specific and/or growth form-specific? (iii) Which environmental factors best predict timing of leaf bud break, flowering, and senescence? First, we hypothesized that leaf bud break, flowering, and senescence would begin earliest in the soil warming + snow removal plots and latest in the control plots, with an intermediate effect in the snow removal only plots. Second, we hypothesized that treatment responses would vary among species and growth forms. Deciduous shrubs and graminoids would be the most responsive to treatment effects, demonstrated by earlier greening, flowering, and senescence than other growth forms. This hypothesis is based on convincing evidence of an increase in deciduous shrubs and graminoids in the Arctic in response to experimental warming (Chapin et al., 1995; Chapin & Shaver, 1996). Finally, we hypothesized that no single environmental variable would be the most important for the duration of the study. Rather, we expected specific environmental variables to have shifting relative importance during different periods of the growing season.
Materials and methods
Experimental design
Data presented in this study cover the growing seasons from May through August over the period of 1995 through 2003. Fieldwork was conducted in the northern foothills of the Brooks Range, Alaska, in the low Arctic. The site is approximately 1.5 km southwest of Toolik Field Station (68°38′ N, 149°36′ W, elevation 730 m). The landscape referred to as tussock dwarf shrub tundra is defined by the sedge, cotton grass (Eriophorum vaginatum). Intertussock spaces are occupied by deciduous and evergreen shrubs, herbaceous forbs, and peat-forming mosses (Bliss & Matveyeva, 1992).
Our experiment consisted of three treatments: control (C), extended Season (ES), and extended season + warming (ESW). In this study, experimental treatment refers to the two latter treatments. We extended the growing season by carefully removing snow and preventing new snow accumulation at the beginning of the growing season as well as preventing snow accumulation and light frosts at the end of the growing season (Oberbauer et al., 1998). Here, we define growing season as the period of favorable conditions under which growth can occur. The active period is defined as the period of activity lasting from bud break to leaf senescence or browning. Snow was removed over the days May 1–2, which is 2–5 weeks before normal snow melt. Once snow was removed, open-ended, polyethylene-covered A-frames were maintained on the ES and ESW plots to keep plots snow free and protect against light frosts until the early-season danger of frosts had passed (usually June 1). The tents were open-ended to minimize changes in air temperature. Our intent was to lengthen the season by accelerating the loss of snow, not to provide a season-long warming of air temperature. In the event of an impending snow or windstorms, the ends were closed temporarily with transparent fiberglass panels. The tents were placed back on experimental plots during the second week of August and maintained until the end of August, to protect against early winter snow accumulation or light frosts. Because early snow removal did not ensure that the soil would thaw sufficiently for belowground plant metabolism, we included a soil warming treatment. In the warming plots, we installed greenhouse-heating wires uniformly throughout the plots approximately 10 cm below the ground surface in 1994, the year before data collection began. A 1400-watt generator powered the heating wires for two hours per day, releasing 0.4 MJ m−2 day−1 of energy to the soil at midday. This energy input is equivalent to 20–30% of the daily soil heat flux for this ecosystem during summer months (Eugster et al., 2005), which has been compared to the magnitude of increase in energy predicted with global warming (Maxwell, 1992; Starr et al., 2008).
We used a randomized block design, consisting of 18 plots, six replicates per treatment. Each plot was measured 1.5 by 1.5 m, and each plot was separated by 3 m. Our sample size was limited by the intense labor required to (i) manually remove the snow over a short period to maintain a uniform growing season length in all snow removal plots and (ii) maintain the experimental treatment plots snow free. We accessed plots using an elevated boardwalk so as to minimize effects on the plant community. In each plot, six individual shoots belonging to eight species were tagged and monitored throughout the season. We studied eight species belonging to the four broad functional types of the Alaskan tundra, as outlined by Molau (1997): sedges (Eriophorum vaginatum L.) and (Carex bigellowii Torr.), evergreen shrubs (Ledum palustre L., Cassiope tetragona (L.) D. Don, and Vaccinium Vitis-idaea L.), deciduous shrubs (Betula nana L. and Salix pulchra Cham.), and forbs (Polygonum bistorta L.). We chose these species because excluding Cassiope tetragona, these seven species occupy greater than 90% of the ground cover of the tussock tundra (McKane et al., 1997). We included C. tetragona because it is one of the focal species in the ITEX study.
Phenological variables
Weekly or more frequent phenological observations were performed for each of the eight species in each plot. During these observations, we recorded the status of three response variables: (i) leaf bud break; (ii) flowering; and (iii) leaf senescence. Leaf bud break was defined as any point where bud scales parted and leaf emergence was visible. We equate leaf bud break to greening and use the terms interchangeably. Direct bud break observations cannot be made for the graminoids, C. bigellowii and E. vaginatum, and the forb, P. bistorta, because the buds are located beneath the soil surface. For C. bigellowii and P. bistorta, leaf greening equates to leaf appearance above the surface. Eriophorum vaginatum is a wintergreen species and some leaves remain green over the winter.
We considered both the timing of flowering as well as the intensity of flowering (Larigauderie & Kummerow, 1991). Plants of all species were assessed for flowering status and assigned one of three categories during each phenological observation: (i) floral bud, (ii) anthesis, and (iii) postanthesis. We define anthesis as it relates to functional pollination: The flower is open and reproductive structures (styles and anthers) are intact and not wilted. The total proportion of individuals in bloom was also calculated for each species and plot. For end of season leaf changes, leaves of the current year (new leaves) were assigned one of the three categories: (i) 1–30% brown, (ii) 31–60% brown, and (iii) >60% brown. Leaf senescence was defined as when the third category occurred. In this study, leaf senescence is synonymous with leaf browning. Leaf senescence in evergreens typically does not occur at the end of the growing season, but rather as new leaves expand. However, leaves of evergreens that are retained undergo browning at the end of the season as plant photo-protective pigments, specifically anthocyanins, accumulate in preparation for winter (Oberbauer & Starr, 2002). We used the timing of this leaf browning to indicate the end of the active period for a given year. We specifically analyzed current-year leaves to be able to compare evergreen and deciduous species.
This study was only concerned with aboveground changes in phenology. Although snow removal and soil warming may have caused significant changes belowground, we restricted our observations to aboveground because these long-term study plots were nondestructive. Leaf data (greening and senescence) were recorded for a total of 9 years (1995 through 2003). Funding constraints shortened the field season in 2003 resulting in no data in August of that year. Flowering data were recorded for 6 years (1995, 1996, 1997, 1998, 2000, and 2003).
Abiotic factors
Mean daily air temperature data were obtained from the meteorological station adjacent to Toolik Field Station. Total day length or photoperiod was calculated for each day of each study year. In each plot, the active layer was measured on a weekly basis using standard soil probing techniques (Oberbauer et al., 1991). Two readings were taken at randomly selected intertussock locations in each plot to compute a mean measurement. Soil moisture (volumetric water content) was also measured in each plot using Campbell Scientific 615 Water Content Reflectometers (Campbell Scientific Inc., Logan, UT, USA), following methods of O'Brien & Oberbauer (2001). Steel probes of the soil moisture sensors were inserted at a 45 degree angle relative to the soil surface, integrating soil moisture over the top 20 cm of soil. Sensors were left in the soil throughout the study. Weekly measurements were used to calculate monthly means of depth of thaw and volumetric water content. Soil temperature was not included in our analyses. Prior work on these plots shows that although soil temperatures were highest in the warming plots, these differences were not statistically significant (Starr et al., 2008).
Statistical analyses
One-way anovas were used to compare the duration of each phenophase (greening, flowering, and senescence) among treatments, as well as the total active growing period among treatments. Chi-square tests were conducted to quantify the relationship between treatment and frequency, or proportion of individuals demonstrating leaf bud break, flowering, and leaf senescence. To determine whether the timing of each of the phenological response variables differed among treatments, a 2-way anova was performed for each variable, analyzing main effects (species and treatment) and the interaction between the two terms. We also conducted a 2-way anova testing the effects of growth form, treatment, and the interaction on timing of each of the response variables. A one-way anova was used to determine whether the mean day of year (DOY) of flowering differed significantly among years. Least significant differences, Tukey, and Dunnett C adjustments were applied to pairwise comparisons. We conducted a simple linear regression between DOY and year for each phenological response variable to determine whether the timing of phenophases was progressively earlier as the study progressed. A simple linear regression was also performed to determine the relationship between date of snow melt and bud break in each plot. The relationship between each of the three response variables and air temperature was determined using a Pearson's correlation. Thawing degree days (Molau & Mølgaard, 1996) were calculated for each plot, species, and year by summing the temperatures between the date each plot was recorded as snow free and the first record of flowering. Flowering frequency, defined as the proportion of population in bloom, was calculated for each DOY postsnow melt. The relationship between flowering frequency and thawing degree days was determined using a Spearman's rank correlation. Data were tested for normality before proceeding with parametric tests. Data that were unsuccessfully transformed were analyzed using nonparametric procedures (Zar, 1984).
In addition to evaluating the effect of treatment and species on the timing of response variables, we also asked how abiotic factors influenced phenology and which factors most strongly predicted changes in phenological outcomes. We conducted a binomial logistic regression for greening and flowering using the Enter Method. An ordinal logistic regression was used for leaf senescence. Species, year, treatment, and snow-free status of plot comprised categorical variables while air temperature, depth of thaw, photoperiod, and soil moisture comprised continuous variables. The CS615 soil moisture sensors were not installed until 1997, so soil moisture data were not available for 1995 and 1996. Consequently, these years were excluded from the bud break and flowering analyses in which all years were analyzed together. Soil moisture was not significant in the leaf senescence model (all years), so 1995 and 1996 were included in the ordinal regression. We defined snow melt for each particular plot as when 50% or more of the plot was snow free (Rumpf et al., 2014). The regression coefficient (β), a measurement of how strongly each predictor variable influences the dependent variable, and the odds ratio (Exp(β)), the likelihood that each phenophase will occur, were calculated for each independent variable. Correlations between independent variables were verified before proceeding with the regression. Regressions were conducted for all years together to determine general patterns, as well as separately, to explain sources of variation among years and identify the most consistent independent variables across years. The goodness-of-fit for each logistic model was assessed by evaluating the −2 log likelihood, model X2, and pseudo R2 values before reporting results.
Finally, given results from the logistic regressions, we asked how air temperature of previous years affected bud break and flowering. Simple linear regressions were carried out to determine the relationship between the monthly proportions of individuals in bud break (and anthesis) and mean monthly air temperature 1, 2, and 3 years prior to the study year. Regressions were conducted for all species together, as well as each species separately. All analyses were performed in IBM SPSS statistics version 21 (SPSS, Chicago, IL, USA).
Results
Effects of treatment, growth form, and species on phenology
Treatment plots became snow-free a mean of 24 days before control plots, although we observed large interannual variation among treatments, ranging from 1 to 35 days. The earliest snow-free status was recorded in 1998 (DOY 119, April 28), and the latest in 2000 (DOY 160, June 8). We used our plant phenophase data to define total period of activity, that is, the total number of days from onset of leaf bud break to the onset of leaf senescence. Overall, total active period lasted 114 days, independent of treatment. Even when we conducted separate analyses by year, total duration of the active period for all species combined did not differ among the three treatments (F2,24 = 0.003, P > 0.05). However, treatment did significantly affect the timing of each of the three phenophases and the proportion of the population in each phenophase.
Bud break
The percent of bud break over the course of the entire study period was lowest in the control plots and highest in the experimental treatment plots (X2 = 62.73, df = 2, P < 0.001). Bud break occurred 2 days later in control plots compared to experimental treatment plots (F2,36102 = 17.72, P < 0.001), although no difference was found in timing of bud break between snow removal (ES) and snow removal + warming (ESW) treatments (Fig. 1a). We also found no significant difference in duration of bud break among the experimental treatments and control (F2,24 = 0.07, P > 0.05). A significant linear relationship was found between timing of snow melt and timing of bud break (F1,36118 = 481.31, P < 0.001) using the overall data set. Bud break occurred progressively earlier as the study progressed (Fig. 2a) (F1,36118 = 152.82, P < 0.001). The timing of bud break was also significantly affected by growth form (F2,36111 = 397.83, P < 0.001) and the interaction between growth form and treatment (F4,36111 = 2.41, P < 0.05). Deciduous shrubs broke their buds a mean of 7 days earlier than evergreen shrubs and 4 days earlier than the forb. The mean DOY of bud break was up to 3 days earlier in the experimental treatment plots than the control plots for all growth forms.
Flowering
The vast majority of flowering took place each year between June 1 (DOY 151) and July 9 (DOY 191). However, Eriophorum vaginatum began flowering earlier than all other species, as early as May 11 (DOY 131). Salix pulchra also tended to flower early, while Polygonum bistorta and Vaccinium vitis-idaea tended to flower later than other species. The proportion of individuals in bloom was significantly highest in the treatment plots and lowest in the control plots (X2 = 49.05, df = 6, P < 0.001). Flowering in the experimental treatment plots occurred significantly earlier than in the control plots, by a mean of 5 days (F2,2388 = 23.20, P < 0.001), although no significant difference was found between experimental treatments (Fig. 1b). The duration of flowering did not differ among treatments (F2,21 = 0.11, P > 0.05). The mean day of flowering differed among years (F5,2586 = 45.06, P < 0.001). Flowering tended to occur earlier as the study progressed (Fig. 2b). Year was a good predictor of DOY of flowering for both experimental treatments (ES: F1, 751 = 14.41, P < 0.001; ESW: F1,846 = 15.59, P < 0.001), but not for the control (F1,788 = 0.74, P > 0.05). Growth form (F3,2379 = 386.54, P < 0.001) and the interaction between growth form and treatment (F6,2379 = 3.56, P = 0.002) significantly affected the timing of flowering. Graminoids flowered a mean of 9 days earlier than deciduous shrubs, and 20 and 25 days before evergreen shrubs and the forb, respectively. Although flowering began earlier in treatment plots compared to the control plots, it did not last longer (F2,21 = 0.11, P > 0.05).
Leaf senescence
Treatment significantly affected the timing of leaf senescence (F2,11707 = 46.45, P < 0.001) (Fig. 1c). Leaf senescence occurred a mean of 8 days earlier in the experimental treatment plots than in the control plots, but timing did not differ between the treatment plots. A downward trend was detected between year and DOY, that is, senescence occurred earlier in the later years of the project. Negative slopes for the control and treatments suggest a 4-day advance in onset of leaf senescence for the control and 5-day advance for the experimental treatments with each additional study year (Fig. 2c). Year accurately predicted DOY of senescence for the control and experimental treatments (C: F1,3229 = 219.31, P < 0.001; ES: F1,4156 = 366.21, P < 0.001; ESW: F1,4340 = 404.25, P < 0.001). Growth form (F3,9314 = 427.53, P < 0.001) and the interaction between growth form and treatment (F6,9314 = 10.26, P < 0.001) also affected the timing of leaf senescence. In other words, leaf senescence occurred significantly later in the control plots compared to the experimental plots in every single growth form. Each pairwise comparison among the four growth forms was also significantly different, with evergreen shrubs and graminoids browning up to 3 weeks before deciduous shrubs and the forb.
Species-specific responses
Phenological responses to treatment were species-specific. The highest proportion of bud break was recorded in Polygonum bistorta while the highest proportion of flowering occurred in Eriophorum vaginatum. Species significantly affected timing of greening (F5,36102 = 288.14, P < 0.001), flowering (F7,2367 = 284.91, P < 0.001), and leaf senescence (F7,11707 = 53.44, P < 0.001). The mean day of bud break, flowering, and leaf senescence differed significantly among experimental treatments and controls for all species except Cassiope tetragona. In general, C. tetragona showed no response to either experimental treatment. In contrast, Betula nana, Salix pulchra, and Vaccinium vitis-idaea were highly responsive to both treatments (Fig. 1a–c). The duration of flowering differed significantly between Eriophorum vaginatum and each of the other species (F7,16 = 13.01, P < 0.001). Duration of flowering did not differ significantly between any other species. Not only did treatment and species affect the timing of each phenological variable, the interaction between species and treatment was also significant (greening: F10,36102 = 2.42, P = 0.007; anthesis: F14,2367 = 2.46, P = 0.002; leaf senescence: F14,11707 = 4.65, P < 0.001).
Abiotic variables as determinants of phenology
Bud break
When all years were pooled together, each of the independent variables (year, species, treatment, air temperature, photoperiod, soil moisture, depth of thaw, and snow-free status of plot) significantly predicted the probability of bud break (Table S1). Within these explanatory variables, the highest probability of bud break (indicated by a high odds ratio) was associated with Polygonum bistorta, the snow removal and warming treatments, soil moisture and timing of snow melt. However, when the analysis was separated by year, species (specifically Polygonum bistorta) and air temperature were consistently the best predictors across all years (Table 1). Photoperiod, depth of thaw, and treatment were also good predictors, although less so than species and air temperature.
| Year | Species – PB | Air temp. | Photoperiod | Depth of thaw | ES | ESW |
|---|---|---|---|---|---|---|
| 1997 | 28518.13, <0.001 | 1.72, <0.001 | 2.89, <0.001 | 1.05, 0.004 | 1.43, NS | 0.76, NS |
| 1998 | 154.67, <0.001 | 1.43, <0.001 | 0.75, 0.05 | 1.12, <0.001 | 1.13, NS | 0.66, 0.003 |
| 1999 | 63.98, <0.001 | 1.06, <0.001 | 1.06, <0.001 | 1.06, <0.001 | 0.79, 0.025 | 0.98, NS |
| 2000 | 2.86, <0.001 | 1.14, <0.001 | 1.14, <0.001 | 1.11, <0.001 | 0.62, <0.001 | 0.96, NS |
| 2001 | 4.91, <0.001 | 1.06, <0.001 | 1.06, <0.001 | 1.07, <0.001 | 2.16, <0.001 | 0.71, <0.001 |
| 2002 | 39.23, <0.001 | 1.06, <0.001 | 1.06, <0.001 | 1.07, <0.001 | 0.93, NS | 0.98, NS |
| 2003 | 1.57, <0.001 | 1.08, 0.004 | 1.08, <0.001 | 1.02, NS | 0.75, NS | 0.99, NS |
Flowering
When all years were analyzed together, the following independent variables were associated with a high probability of flowering: year, species, treatment, air temperature, photoperiod, and depth of thaw. Soil moisture and date of snow melt were not significant predictors in the model. The highest probability of anthesis was associated with Eriophorum vaginatum, air temperature, and photoperiod (Table S2). On a year-to-year basis, Eriophorum vaginatum, air temperature, and depth of thaw were consistently associated with flowering, making them the best predictors of flowering (Table 2). We also found a significant relationship between flowering frequency and thawing degree days for all species (Spearman's rank correlation = 0.72, P < 0.001).
| Year | Species – EV | Air temp. | Depth of thaw |
|---|---|---|---|
| 1995 | 31.33, <0.001 | 1.07, 0.035 | 1.07, <0.001 |
| 1996 | 0, <0.001 | 0.91, <0.001 | 1.15, <0.001 |
| 1997 | 292.57, <0.001 | 0.94, 0.021 | 1.07, <0.001 |
| 1998 | 118.52, <0.001 | 1.52, <0.001 | 1.08, <0.001 |
| 2000 | 29.04E8, <0.001 | 2.99, <0.001 | 1.17, <0.001 |
| 2003 | 0.87, NS | 1.67, <0.001 | 1.04, NS |
Leaf senescence
When all years were analyzed together, year, species, treatment, air temperature, photoperiod, depth of thaw, and snow-free status of plot significantly predicted timing of leaf senescence (Table S3). Of the significant explanatory variables, air temperature, depth of thaw, and year best predicted leaf browning, (indicated by a high odds ratio). Soil moisture was not significant. On a year-to-year basis, air temperature, photoperiod, and depth of thaw most consistently predicted senescence; these three variables were significant in 100% of the models (Table 3). Soil moisture, treatment, and snow-free status of plot were significant predictors of leaf senescence for fewer than half of the study years.
| Year | Air temp. | Photoperiod | Depth of thaw |
|---|---|---|---|
| 1995 | 1.05, <0.001 | 0.85, <0.001 | 1.03, <0.001 |
| 1996 | 1.08, <0.001 | 0.68, <0.001 | 1.09, <0.001 |
| 1997 | 0.83, <0.001 | 0.60, <0.001 | 0.98, 0.034 |
| 1998 | 0.90, <0.001 | 0.65, <0.001 | 0.97, <0.001 |
| 1999 | 1.06, <0.001 | 0.51, <0.001 | 0.93, <0.001 |
| 2000 | 0.96, <0.001 | 0.60, <0.001 | 0.95, <0.001 |
| 2001 | 1.02, <0.001 | 0.84, <0.001 | 0.99, <0.001 |
| 2002 | 1.13, <0.001 | 15.53, <0.001 | 1.08, <0.001 |
| 2003 | 1.14, <0.001 | 0.39, 0.002 | 1.15, <0.001 |
Delayed effects of air temperature on phenology
We found a significant relationship between air temperature of previous years and the frequency of bud break of all species combined; 1-year time lag (F1,610 = 137.39, P < 0.001), 2-year time lag (F1,610 = 175.96, P < 0.001), and 3-year time lag (F1,610 = 156.30, P < 0.001). Each of the six species was responsive to historic air temperature. Previous years’ temperature also significantly predicted frequency of flowering across all species; 1-year time lag (F1,524 = 17.31, P < 0.001), 2-year time lag (F1,524 = 18.44, P < 0.001), and 3-year time lag (F1,526 = 15.54, P < 0.001), were all significant. However, when analyzed alone, not each species showed a significant response to air temperature carryover effects. This pattern was consistent for 1-, 2-, and 3-year time lags (Table 4). Eriophorum vaginatum, Ledum palustre, Polygonum bistorta, and Vaccinium vitis-idaea consistently responded to air temperature up to 3 years prior to actual flowering, while Betula nana, Carex bigellowii, Cassiope tetragona, and Salix pulchra showed no response.
| Species | 1-year time-lag | 2-year time-lag | 3-year time-lag |
|---|---|---|---|
| BN | F = 1.61NS | F = 1.72NS | F = 2.46NS |
| CB | F = 0.98NS | F = 1.42NS | F = 1.84NS |
| CT | F = 1.93NS | F = 1.00NS | F = 0.99NS |
| EV | F = 16.75** | F = 4.79* | F = 9.77** |
| LP | F = 8.96** | F = 5.64* | F = 4.79* |
| PB | F = 7.98** | F = 7.12* | F = 7.28** |
| SP | F = 0.46NS | F = 0.004NS | F = 0.00NS |
| VVI | F = 35.56** | F = 27.34** | F = 23.38** |
- Df values for all categories is 64, 1. P-values <0.05 are indicated by *, P-values <0.01 are indicated by **.
Discussion
Our results clearly show that early snow removal induces earlier leaf bud break, senescence, and floral anthesis relative to the controls. Previous ITEX results also reported earlier leaf burst and flowering in season-long-warmed plots (Arft et al., 1999). However, our findings show that although earlier loss of snow cover caused earlier onset of phenophases, the total period of activity did not increase because senescence was also accelerated. These results contrast Arft et al. (1999), which reported delayed senescence in response to warming of air, resulting in longer overall periods of plant activity. Thus, our study suggests that species which do not respond immediately to favorable environmental conditions will be at a disadvantage relative to phenologically plastic species. Our findings corroborate a growing number of studies showing that plant growth at high latitudes may begin earlier but may not necessarily last longer. In fact, warming temperatures may cause the growing season to shorten (Post et al., 2001; Linderholm, 2006; Shutova et al., 2006). Previous work in our plots found that season duration did not increase physiological activity in response to the manipulation (Oberbauer et al., 1998; Starr et al., 2000). Høye et al. (2013) found an inverse correlation between duration of flowering and temperature, suggesting that as the summers get warmer in the Arctic, the flowering season will shorten. A shorter flowering season in the Arctic has been linked to decreases in floral resources and visitation rates (Potts et al., 2010).
Species variation is an important factor to consider when evaluating the phenological and physiological responses of tundra communities to climate change (Starr et al., 2008; Cooper et al., 2011). Our study shows that species do not respond uniformly to an extended snow-free period and soil warming. In contrast to the other seven species, Cassiope tetragona exhibited no response to either treatment. Molau (1997) also found little response of C. tetragona to warming treatments. Therefore, we could classify C. tetragona, and possibly Polygonum bistorta (Starr et al., 2000), as periodic species, or those which have a genetically fixed growth strategy. Highly responsive species such as Salix pulchra and Vaccinium vitis-idaea fit the classification of aperiodic, or plastic, species (Sørensen, 1941). Such species can adapt their growth strategy to changing environmental conditions and consequently may out-compete species with nonplastic growth strategies (Lechowicz, 1995).
Our data corroborate the idea that present plant growth directly reflects past environmental conditions. Many species in the Arctic produce bud primordia underground, sometimes up to several years prior to flowering, as an evolutionary adaptation to a short growing season (Sørensen, 1941; Billings & Mooney, 1968). Diggle (1997) showed that in the alpine species, Polygonum viviparum, the maximum reproductive output of each inflorescence is already determined 1-year prior to actual maturation. Moreover, the development of each leaf and inflorescence, from initiation to functional and structural maturity, requires 4 years. Not only may underground buds take longer to develop, but they also may be susceptible to abiotic conditions prior to the season of observation. In other words, the effect of air temperature on plant growth may not manifest itself until several years later (Arft et al., 1999). Our results imply that time lags in air temperature do not affect flowering in all species in the same way. The nonplastic response in C. tetragona to air temperature is what we would expect from a periodic species which cannot respond to current, nor past, abiotic conditions. In contrast, we would expect species such as Eriophorum vaginatum, which develop preformed buds, to be highly sensitive to past environmental conditions. Although it is implied in the literature that bud formation is susceptible to past abiotic conditions, few studies actually address how delayed responses in flowering and greening vary across a range of species.
The timing of phenophases with respect to climate will play a key role in species resiliency and interspecific competition. In tussock tundra, earlier snow melt may affect early greening and flowering species such as E. vaginatum and S. pulchra more than late greening and flowering species such as V. vitis-idaea and B. nana. Earlier flowering species are more dependent on timing of snow melt than late-flowering species (Dunne et al., 2003; Kudo et al., 2008; Wipf, 2010). Abundance of species with early leaf expansion was higher in early snow melt plots than late snow melt plots in alpine areas of Colorado (Galen & Stanton, 1995). Molau (1993) hypothesized that late-bloomers could be short-term ‘winners’ under climate change. The probability of successful pollen transfer and seed germination would be higher for these individuals because they could take advantage of the end of the growing season. While earlier flowering species have a longer timespan to set fruit and disperse their seeds, they can potentially lose all their reproductive output because of early-season frost, a typical outcome of warming (Inouye, 2008). Although it still remains unclear if plastic growth strategies in response to early snow melt will ultimately be advantageous for a species or not, it underlines the possibility that phenotypic acclimation to warming may be more important for specific individuals and populations than general macro-evolutionary adaptations (Totland and Alatalo, 2002). Furthermore, phenotypic adaptation will have direct effects on other biotic interactions. Variation among early-flowering individuals and late-flowering individuals can alter the length of the flowering season, imposing immediate effects on population dynamics and plant–pollinator interactions (Høye et al., 2013).
In addition to the varying responses of species to treatment, we also found the contribution of each species to differ in each of the regression models. Although species, in general, significantly predicted bud break and flowering, the likelihood of bud break and flowering was an order of magnitude higher in P. bistorta and E. vaginatum, respectively. One interpretation is that such high odds ratios simply reflect the high frequency of these two species and their life histories. Rates of growth (both vegetative and reproductive) and regeneration tend to be higher in herbaceous than woody plants (Bazzaz, 1979). Thus, we must consider relative abundance of these species when interpreting the results. Alternatively, E. vaginatum and P. bistorta may be more predictable than other species and do well in a more tightly controlled environment. In contrast to bud break and flowering, we did not detect a species-specific or growth form-specific pattern with respect to leaf senescence, suggesting that new leaves of all species are equally likely to senesce at the end of the season. Oberbauer et al. (2013) found senescence of all growth forms to occur at similar thaw degree day values. It is also important to point out that the most abundant species will not necessarily be more resilient to changes in the growing season. Our results suggest that E. vaginatum will be at a competitive disadvantage to deciduous shrubs.
While plant phenology is dependent upon many abiotic factors, our study suggests that air temperature is the key determinant of the onset of vegetative and reproductive phenology in the tundra. Temperature significantly predicted each of the three phenophases, both when years were pooled and analyzed separately. The strong correlation between thaw degree days and flowering further highlights the critical role air temperature plays in phenology. Surface air temperature is one of the most useful climate change variables as it accounts for changes in the surface energy budget and atmospheric circulation (Serreze et al., 2000). Air temperature is disproportionately important as it dictates, in part, depth of thaw, with indirect effects on other abiotic variables including water and nutrient availability. There is ample evidence demonstrating that temperature determines many phenophases, in general (Badeck et al., 2004) and particularly in arctic and alpine species (Dunne et al., 2003; Huelber et al., 2006; Euskirchen et al., 2014). Timing of bud break in Betula nana and Salix pulchra was found to be a function of air temperature (Pop et al., 2000). Air temperature was the strongest predictor of the commencement of photosynthesis in an evergreen boreal forest (Tanja et al., 2003).
With the exception of air temperature, environmental variables such as snow melt, thaw, and photoperiod exert different degrees of importance at different periods in the growing season. Date of snow melt was the most relevant for bud break, which suggests that plants are limited by snow cover early in the season, when the majority of growth occurs. These results corroborate the strong, positive linear relationship we found between timing of snow melt and bud break. Other studies have also found bud break of evergreen and deciduous species to be dependent on timing of snow melt (Shaver & Kummerow, 1992; Shevtsova et al., 1997). Earlier snow melt since the 1950s is well documented in the Arctic and is predicted to be one of the primary consequences of climate change (Foster, 1989; Aurela et al., 2004; Post et al., 2009). Thus, we predict earlier bud break in the tundra as the timing of snow melt advances.
Soil thaw also explained leaf bud break in our models, consistent with other studies. Soil thaw accurately predicted bud break of B. nana 3 years in a row (Van Wijk et al., 2003). In our plots, photoperiod significantly predicted vegetative phenology for the majority of study years, but not for flowering. These results make intuitive sense; an increase in hours of light is a key trigger for plants to begin vegetative processes, while a decrease in hours of light is a key trigger for plants to prepare for winter (Håbjørg, 1972). Photoperiod is at a maximal constant during peak flowering (June/July), however. Photoperiod may play an important role in the timing of early- or late-bloomers, but has a negligible effect for the majority of flowering individuals. Our results agree with other studies suggesting that snow cover and depth of thaw may be especially important in the early season (when greening occurs), while photoperiod and genetic constraints become increasingly important in the late-season (associated with senescence) (Shaver & Kummerow, 1992; Oberbauer et al., 1998; Van Wijk et al., 2003; Estiarte & Peñuelas, 2015).
While early snow melt allows plants to get a ‘jump-start’ on growth, it also exposes them to early-season low temperatures, damaging frost, and drier conditions (Wookey & Robinson, 1997; Sturm et al., 2005). Soil warming likely aggravated drying effects in our study, probably explaining why we did not observe a difference in phenology between the two experimental treatments. Drier tundra organic soils are highly insulative, reducing thermal transfer from the soil surface to deeper layers (Hinkel et al., 2001); as a result, the warmed plots had slightly shallower depth of thaw than the extended season plots. Although a lengthened snow-free period offers plants the opportunity for a longer period of potential growth, it can also lead to water stress. A lack of water obstructs effective nitrogen assimilation (Kramer & Boyer, 1995) and prevents plants from taking advantage of a longer growing season. Other studies have attributed delayed reproductive phenology (Dorji et al., 2013) and declines in flower production (de Valpine and Harte, 2001) to soil drying caused by soil warming.
Growth form affects snow accumulation and has a consequential ripple effect through the landscape (Sturm et al., 2001b, 2005). Furthermore, different plant functional types with different life histories will respond heterogeneously to climate change, potentially introducing asynchronies into plant–plant and plant–animal interactions (Dunne et al., 2003; Roy et al., 2004; Post & Forchhammer, 2008; Høye et al., 2013). In our study, evergreen shrubs were the least responsive to treatment, while deciduous shrubs and graminoids were the most responsive, corroborating previous studies (Molau, 1997; Dormann & Woodin, 2002). Woody, deciduous shrubs have a competitive advantage over other growth forms as a result of their physical architecture, ability to uptake nutrients, and production of rapidly decomposing litter (Shaver et al., 1996). Evergreen shrubs tend to be slow growing and rely heavily on internal cycling of nitrogen, while deciduous shrubs are characterized by rapid growth and can access soil-available nitrogen (Aerts, 1995). Starr et al. (2008) demonstrated that photosynthetic rates (directly affecting and affected by phenology) in our plots were highest for deciduous shrubs and lowest for evergreen shrubs. Also, deciduous shrubs must ‘take a chance’ and break buds before other plant functional types (Billings & Mooney, 1968), monopolizing resources and increasing recruitment probability. In our study, deciduous shrubs senesced later than evergreens and graminoids, suggesting that deciduous shrubs are best suited to take advantage of an extended growing season. These results support growing evidence of deciduous shrub expansion in the tundra (Sturm et al., 2001b; Stow et al., 2004; Tape et al., 2006; Walker et al., 2006). We are not the first to predict a decline in evergreen shrubs including C. tetragona (Grime, 1979; Billings & Peterson, 1992; Molau, 1997; Buizer et al., 2012) and a consequential reduction in species diversity (Walker et al., 2006).
As expected, temporal variation was significant among years. Anomalous weather events explain, to an extent, seasonal phenology. In alpine communities, 15–40% of among-year variation in phytomass was attributed to interannual climate variation (Walker et al., 1994). The dynamic interaction between abiotic factors and interannual variation also may explain nonlinear phenological responses to climate (Iler et al., 2013). In our study, 1999 was marked by unusually early snow melt, warm spring conditions, and an extended dry period. Other seemingly random events such as the complete absence of flowering in S. pulchra in 1996 make generalizing temporal patterns difficult. Although our results suggest a unidirectional, downward trend in onset of phenology, we interpret these results with caution. Further work is crucial so we can describe with more certainty the relationship between timing of phenophases and year. Nevertheless, our results add to a growing body of evidence demonstrating earlier onset of phenology at northern latitudes (Zhou et al., 2001; Badeck et al., 2004; Post et al., 2009; Bock et al., 2014).
Finally, our work highlights the possibility that advanced snow melt and phenology will not necessarily be accompanied by a net increase in productivity. While greening began significantly earlier in the treatment plots, the total duration of activity did not increase, suggesting that an earlier start to the growing season, in and of itself, is not enough to increase productivity. Our phenological observations are consistent with finding of La Puma et al. (2007), who reported no increase in net ecosystem productivity at Toolik despite increased season length. Perhaps plants need additional nutrients to be able to take advantage of a longer growing season. Natali et al. (2012) point to N limitation as the primary explanation why ecosystem-level net primary productivity did not increase despite experimental lengthening of the growing season. Fertilization effects have been shown to have a greater influence on productivity in the Arctic than warming effects (Chapin et al., 1995; Dormann & Woodin, 2002; Van Wijk et al., 2004; but see Larigauderie & Kummerow, 1991). Eriophorum vaginatum appears to be more limited by nutrients than photosynthesis in the Alaskan tussock tundra (Tissue & Oechel, 1987). Growth in Ledum and Eriophorum was highly constrained by nutrients in the late-season (Chapin & Shaver, 1996). Earlier phenology may cause a feedback mechanism by which nutrient limitation is further exacerbated, thereby limiting productivity (Kremers et al., 2015). Our results provide evidence that an earlier onset of the growing season alone will not result in increased growth. These findings contrast with satellite data that indicate an increase in tundra plant growth (Myneni et al., 1997; Tucker et al., 2001), implying the role of other factors beyond changes in timing of snow melt. Late-season nutrient supply and plants’ ability to access nutrients may be key factors determining ecosystem productivity. Limitations in growth will, in turn, cause limitations in carbon sequestration, making the tundra potentially a larger source than sink of carbon (Starr & Oberbauer, 2003).
Acknowledgements
This work is based, in part, on funding from National Science Foundation grants OPP-9321626, 9615845, 9907185, and 0856710. Logistical support by the staff of the Institute of Arctic Biology Toolik Field Station is greatly appreciated. Kevin R. T. Whelan, Carlo Calandriello, Esperanza Rodriguez, Brook Shamblin, Carrie Beeler, Michael Rasser, and Tara Madsen provided much appreciated field assistance. The work benefited greatly from statistical advice by Dr. Jianbin Zhu.
