Differential effects of temperature and precipitation on early- vs. late-flowering species

Transcription

1 Differential effects of temperature and precipitation on early- vs. late-flowering species LYNN M. MOORE AND WILLIAM K. LAUENROTH Department of Botany, University of Wyoming, 1000 E. University Avenue, Laramie, Wyoming USA Citation: Moore, L. M., and W. K. Lauenroth Differential effects of temperature and precipitation on early- vs. late-flowering species. Ecosphere 8(5):e /ecs Abstract. Shifts in flowering dates have been associated with climate change and warming temperatures. However, the influence of temperature and precipitation on annual phenology patterns in semiarid ecosystems is not well understood. We observed the flowering stages of 21 shortgrass steppe species from 1995 to We compared first and last flowering dates and used climatic data to interpret relationships between the timing of flowering and temperature and precipitation. On average, the first flowering dates of 21 species advanced 0.53 d over the 20-yr period. This advance was significantly related to an increase in annual March September mean temperature in that first flowering date advanced at the rate of 7.5 d C 1. The advance of the first flowering date was significantly related to increased spring temperatures in earlyseason species. Late-blooming species flowered longer; this delay in end of flowering was significantly related to wetter conditions. Significant advances in first flowering date were related to increasing temperatures over time, suggesting a response to climate change. In the water-limited shortgrass steppe, the effect of the environment on flowering phenology is controlled by warmer temperatures early in the growing season and precipitation later in the growing season. Key words: climate change; first flowering; last flowering; phenology; precipitation; shortgrass steppe; temperature. Received 11 November 2016; revised 7 April 2017; accepted 10 April Corresponding Editor: Debra P. C. Peters. Copyright: 2017 Moore and Lauenroth. This is an open access article under the terms of the Creative Commons Attrib ution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. lmoore7@uwyo.edu INTRODUCTION Analyzing changes in phenology of plants is one of the most elegant and simplest methods available to examine how climate change influences ecosystems. Reproductive processes, including flowering time, can determine the relative abundance and the presence or absence of individual species (Rathcke and Lacey 1985, Craine et al. 2011, Crimmins et al. 2011) and can influence annual aboveground net primary production. For example, flower stalk production of grasses had a positive influence on aboveground net primary production in a mesic grassland (La Pierre et al. 2011). Many studies have reported the advance of first flowering dates across ecosystems in response to climate change (Sparks and Carey 1995, Ahas et al. 2002, Scheifinger et al. 2003, Menzel et al. 2006, Zheng et al. 2006, Primack et al. 2009, Dunnell and Travers 2011). Increasing spring temperature is the most common climatic variable driving these shifts in first flowering dates (Sparks and Carey 1995, Menzel et al. 2006, Zheng et al. 2006, Miller-Rushing et al. 2007). In addition to shifts in first flowering dates, secondary effects of these shifts include disrupting flowering phenology within communities across the entire growing season by altering co-flowering patterns, redistributing floral abundance across the season, as well as expanding the flowering season (Aldridge et al. 2011, CaraDonna et al. 2014). Several studies have documented the advance of first flower in grasslands, including a semiarid montane grassland in Montana in which the date 1 May 2017 Volume 8(5) Article e01819

2 of first flower advanced in response to increasing spring temperature and decreasing winter precipitation (Lesica and Kittelson 2010). A northern tallgrass prairie study on the Great Plains estimated that 5 17% of the study species showed a shift to earlier flowering relative to the previous century in response to increasing spring temperatures and a lengthening of the growing season (Dunnell and Travers 2011). There is a consensus among flowering phenology studies that the earliest flowering species are more likely to advance first flowering dates than later-flowering species (Fitter et al. 1995, Bradley et al. 1999, Post and Stenseth 1999, Fitter and Fitter 2002, Miller-Rushing et al. 2007, 2008, Lesica and Kittelson 2010, CaraDonna et al. 2014). This suggests an increase in plasticity of the first flowering date of early species compared to a less variable first flowering date of later-blooming species. However, Miller-Rushing and Primack (2008) found that early-flowering species displayed less variability from year to year than late-flowering taxa. Although temperature is a primary driver of phenological events, other variables such as soil water availability are known to influence phenology, especially in water-limited ecosystems (Yuan et al. 2007, Moore et al. 2015). In dryland systems, precipitation influences flowering commencement (Keatley et al. 2002, Lesica and Kittelson 2010, Crimmins et al. 2011). First flowering dates in a semiarid montane grassland in Montana advanced with decreases in winter precipitation and the onset of summer flowering advanced in response to increased summer rains along an elevation gradient in Arizona (Lesica and Kittelson 2010, Crimmins et al. 2011). Furthermore, the interaction between temperature and precipitation has been found to influence flowering time for some species; for example, warm and dry springs tended to advance flowering in Mediterranean ecosystems (Gordo and Sanz 2010). In a study of a North American perennial herb, increasing temperature advanced flowering, while increased precipitation delayed flowering (Matthews and Mazer 2015). The optimal flowering time for a plant is determined by genetics, biotic interactions, and abiotic environmental variables (Rathcke and Lacey 1985, Forrest and Miller-Rushing 2010). Phenological synchrony refers to the temporal overlap between two events and is most often discussed in terms of pollinator presence and open flowers. However, the concept can also be applied to the overlap of a phenological event such as flowering date and suitable temperatures or availability of a resource such as soil water (Miller-Rushing et al. 2010). It is important to recognize that in semiarid or arid ecosystems, insect-pollinated plants are often pollinated by generalists and wind-pollinated species are common or dominant. Flowering and seed set in these systems and others are often dependent on adequate resources, including soil water, being available at the time of flowering (Zeiter et al. 2016) rather than the synchrony between flowering and pollinator arrival. Studies that focus on the phenology of communities across the entire growing season are few. Several existing studies focus on community-level questions, including interaction potential (Cara- Donna et al. 2014), community-level flowering curves (Aldridge et al. 2011), or are restricted to spring and summer flowering species (Miller- Rushing et al. 2008). A small number of studies have examined the seasonal flowering dynamics of a plant community from early spring to autumn (Crimmins et al. 2011, 2013, but see Cara- Donna et al. 2014). It is well documented that spring flowering phenology is shifting in many locations, but what is not well understood is how later-blooming species respond to changes in temperature or precipitation, both of which will be influenced by climate change. Few studies have examined the seasonal flowering dynamics in grasslands and those that have are located either in a mesic northern tallgrass prairie (Dunnell and Travers 2011) or in a semiarid system that receives most of its precipitation in the form of snow (Lesica and Kittelson 2010). Some work has been done in desert ecosystems, but not on semiarid grasslands (Crimmins et al. 2011, 2013). To our knowledge, no studies to date have examined early-blooming, mid-seasonblooming, or late-blooming flowering phenology in the context of a dryland system. We intend to fill this gap in knowledge by analyzing a unique 20-yr phenology dataset. This dataset is exceptional because of the length of the study, the lack of existing data in semiarid ecosystems, and the community approach regarding plant functional groups and range of flowering times from early to late season. Here, we report on the reproductive phenology of 21 shortgrass steppe species, 2 May 2017 Volume 8(5) Article e01819

3 which flower from early spring to late fall. We specifically asked: (1) Has there been a shift in the timing of flowering of shortgrass steppe species between 1995 and 2014? (2) How does the flowering phenology of shortgrass steppe species respond to the annual variability in temperature and precipitation? MATERIALS AND METHODS We conducted this study on the Central Plains Experimental Range (CPER) located 60 km east of Fort Collins, Colorado ( N, W). The climate is semiarid continental characterized by 74-yr average annual precipitation of 350 mm and average annual mean temperature of 8.3 C. Mean extreme temperatures range from 11 to 29 C in winter and summer, respectively (Fig. 1; Pielke and Doesken 2008). The majority of precipitation (70%) occurs as rain between April and September. Sixty-five percent of the average annual aboveground biomass consists of the dominant C 4 grasses Bouteloua gracilis and Bouteloua dactyloides. The remaining standing crop is composed of forbs, shrubs, C 3 grasses, and cacti (Sims et al. 1978, Liang et al. 1989, Lauenroth 2008). However, in terms of biodiversity, 71% of Fig. 1. Walter and Leith (1967)-style climate diagram for the Central Plains Experimental Range ( ; Rangleland Resources Research Unit 2014). Lined areas indicate periods of relatively humid conditions. Black bars are months with monthly mean temperatures <0 C, gray bars are months with a chance of frost, and white bars are frost-free months. the shortgrass steppe species are forbs and 18% grasses (Hazlett 1998, Lauenroth 2008). Field methods We collected 20 yr of growing season phenology data for 10 individuals of each of 21 plant species (see Table 1 for species list) located within a three-ha grazing exclosure. We sampled at oneweek intervals for early- and mid-season species and two-week intervals for late-season species. Some researchers have found that sampling frequency can influence the timing of first observations (Miller-Rushing et al. 2008). However, this influence is not significant if low sampling frequency is consistent over time. Our sampling frequency was based upon previous phenological work on the shortgrass steppe that found that a one-week interval was just as effective at capturing stages as sampling more frequently (Dickinson and Dodd 1976). Individuals were selected based on similar size compared to other individuals of the same species within the exclosure. We endeavored to track the same 10 individuals throughout the growing season and from year to year. If an individual died, we replaced it with a nearby individual such that there were always 10 individuals for each species included in the census. All taxa were selected and replaced in the same manner. The species list included the dominant species on the shortgrass steppe and common shrubs, annual grasses, early-flowering forbs, annual exotic forbs, and late-flowering forbs (Table 1). Nomenclature follows the PLANTS database (USDA 2016). We modified phenological stages from previous work (Dickinson and Dodd 1976) to facilitate data collection. The phenological stages included in this study are onset of spring, first flower, last flower, and senescence (Table 2). Some taxa were not recorded every year; for example, some annual species did not reestablish every year and were only recorded in those years when conditions were suitable for germination and establishment. Climate Abiotic data included photoperiod (day length in minutes), daily mean, maximum, and minimum air temperature ( C), first and last freeze date, and daily precipitation (mm). Daily precipitation was measured at the CPER headquarters (Rangleland Resources Research Unit 2014), located 5.1 km northeast, using a Campbell 3 May 2017 Volume 8(5) Article e01819

4 Table 1. Plant species included in phenology study from 1995 to Species Pollination method Flower time Family Common name Life form/ pathway Aristida purpurea Nutt. Wind Mid Poaceae Red three-awn Perennial grass/c 4 Artemisia frigida Willd. Insect Late Asteraceae Prairie sagewort Perennial subshrub/c 3 Bouteloua gracilis (H.B.K.) Lag. Ex Wind Late Poaceae Blue grama Perennial grass/c 4 Griffiths Carex duriuscula C.A. Mey. Wind Early Cyperaceae Needleleaf sedge Perennial sedge/c 3 Escobaria vivipara (Nutt.) Buxbaum Insect Mid Cactaceae Spiny star Cacti/CAM var. vivipara Echinocereus viridiflorus Engelm. Insect Early Cactaceae Nylon hedgehog cactus Cacti/CAM Elymus elymoides (Raf.) Swezey Wind Mid Poaceae Squirreltail Perennial grass/c 3 Ericameria nauseosa (Pall. ex Pursh) Insect Late Asteraceae Rubber rabbitbrush Perennial shrub/c 3 G.L. Nesom & Baird ssp. nauseosa var. glabrata (A. Gray) G.L. Nesom & Baird Eriogonum effusum Nutt. Insect Late Polygonaceae Spreading buckwheat Perennial subshrub/c 3 Gutierrezia sarothrae (Pursh) Britt. & Insect Late Asteraceae Broom snakeweed Perennial subshrub/c 3 Rusby Hesperostipa comata (Trin. & Rupr.) Wind Mid Poaceae Needle and thread Perennial grass/c 3 Barkworth ssp. comata Heterotheca villosa (Pursh) Shinners Insect Mid Asteraceae Goldenaster Perennial forb/c 3 Lepidium densiflorum Schrad. Mid Brassicaceae Common pepperweed Annual forb/c 3 Leucocrinum montanum Nutt. Ex Gray Insect Early Liliaceae Common star lily Perennial forb/c 3 Opuntia polyacantha Haw. Insect Mid Cactaceae Plains prickly pear Cacti/CAM Packera tridenticulata Rydb. (Rydb.) Insect Early Asteraceae Plains groundsel Perennial forb/c 3 W.A. Weber & A. L ove Pascopyrum smithii (Rydb.) A. L ove Wind Mid Poaceae Western wheatgrass Perennial grass/c 3 Plantago patagonica Jacq. Insect Mid Plantaginaceae Woolly plantain Annual forb/c 3 Salsola tragus L. Wind Late Chenopodiaceae Russian thistle Annual forb/c 3 Sphaeralcea coccinea (Nutt.) Rydb. Insect Mid Malvaceae Scarlet globemallow Perennial forb/c 3 Vulpia octoflora (Walt.) Rydb. Wind Mid Poaceae Sixweeks fescue Annual grass/c 3 Notes: Nomenclature including common name follows PLANTS National Database (USDA 2016). Flower time indicates seasonal groups: early = blooming in April May; mid = blooming in June July; and late = blooming in August or later. Pathway indicates photosynthetic pathway, and life form indicates growth form and duration. Indicates exotic species. Scientific TE525 (Logan, Utah, USA) tipping bucket rain gauge. Daily mean, maximum, and minimum air temperatures were measured inside the exclosure. We did not use the heat unit approach (growing degree days), because it does not consider factors such as precipitation and soil moisture, which are important drivers of phenology in a semiarid grassland (Wang 1960, Moore et al. 2015). Additionally, monthly temperatures are simpler to consider and have been reported to be important explanatory variables in other studies (Abu-Asab et al. 2001, Gordo and Sanz 2010, Lesica and Kittelson 2010). To capture how precipitation and temperature over the 20-yr observation period compared to Table 2. Phenological stage descriptions recorded between 1995 and 2014 for 21 species on the shortgrass steppe. Stages Name Description 3 Onset of spring Early growth in which the green leaves are below the height of the previous year s growth 9 First flower Open flowers may have buds present, but this is the first sign of open flowers 11 Last flower Indicator of end of flowering for plants. No flowering structures visible; i.e., no persistent petals, the anthers, and stigmas are shriveled. No buds and flowers shriveled 14 Senescence Plants are fully brown and dead. Beginning of dormancy Note: Stages were modified from Dickinson and Dodd (1976). Denotes stages that are recoded once during a growing season. 4 May 2017 Volume 8(5) Article e01819

5 historical observations, we calculated monthly anomalies from a 74-yr dataset ( ) for precipitation and temperature. Anomaly is a dimensionless value with mean = 0 and standard deviation = 1. It was calculated by subtracting the long-term mean from the observation and dividing by the long-term standard deviation. We considered monthly anomalies 1 standard deviation below or above to be important deviations from the 74-yr long-term mean. Statistical analysis Phenology: response over time. For all statistical analyses, we used the earliest first flowering day within a year rather than mean flowering day. Similarly, we used the latest last flowering day within a year. We used linear regression models to look for shifts in flowering phenology over the 20-yr period by assessing both individual species and species grouped by seasonal flower time (Table 1). We grouped species as early season (blooming in April May), mid-season (blooming in June July), and late season (blooming in August or later). Seasonal grouping was based upon previous phenological work by Dickinson and Dodd (1976), the seasonal timing of precipitation (see Fig. 1), and the mean flowering date over the 20-yr study period for each species. To determine whether flowering time has advanced over the 20-yr period, we regressed the first flowering day (stage 9, Table 2) over year for both individual species and seasonal groups. To assess whether individual species or seasonal groups were flowering later, we regressed the last flowering day (stage 11, Table 2) over year. We considered shifts in phenology significant when the P-value is <0.05. To assess an overall shift in flowering time for all species and all years, we used a one-sample t- test to determine whether the distribution of regression slopes across all species was greater than a distribution with a mean of zero. To determine whether earlier-flowering species showed a stronger tendency for advanced flowering, we regressed the model coefficients for first flowering day on the mean first flowering date calculated for all species across all years. Similarly, we tested whether later-flowering species showed a tendency to delay by regressing the model coefficients for last flowering day on the mean last flowering date (Lesica and Kittelson 2010). To determine whether shifts in flowering dates are related to climate variables on the shortgrass steppe, we regressed the average first and last flowering date for all species for all years on the average maximum, mean, and minimum March September temperatures. Phenology: response to environment. Exploratory data analysis showed that previous year s precipitation and temperature resulted in no significant relationship with flowering date for any species or seasonal group. Based on climate data on the shortgrass steppe (Fig. 1), we reasoned that the cold dry months of January and February would have little or no effect upon first flowering day. The onset of spring on the shortgrass steppe is a function of combined temperature and precipitation and occurs on average during the month of March defined using normalized difference vegetation index (Moore et al. 2015); therefore, we reasoned that climate variables from March to September would likely have an influence on flowering phenology. To investigate how the timing of precipitation and temperature influenced first and last flowering days, we used an information theoretic approach (Anderson 2008, Burnham and Anderson 2013). We developed a number of biologically relevant candidate models specific to each species or seasonal group (Appendix S1: Table S1). We used a two-step process of model selection. First, we identified the best precipitation-only and temperature-only models. We tested cumulative precipitation periods occurring from one to three months before and summed to either the first flowering day or the last flowering day, such that only precipitation events that occurred prior to and/or including the date of first or last flower were considered. Then, we tested monthly mean temperature from one to three months before the date of first and last flowering for each species and seasonal group (Appendix S1: Table S1). The best precipitation-only and temperature-only models were selected based upon the following criteria; the best model had the lowest Akaike Information Criterion corrected for small sample sizes and a delta value <4, where the delta value is an evidence ratio, estimated as the Kullback-Leibler distance between each of the models and the best one (Anderson 2008). Second, the precipitation-only and temperature-only models that were identified in the selection process were combined 5 May 2017 Volume 8(5) Article e01819

6 into a new multiple regression model (full model) where estimates were calculated for each species and seasonal group. The resulting models were refined using null hypothesis significance testing by the general linear F-test approach (Kutner et al. 2003) where the full model with the most temperature and precipitation variables was tested against models with fewer variables (reduced model) until a model where all variables were foundtobesignificant (P < 0.05) was obtained (Kutner et al. 2003). In a few cases, predictor variables were included with a (P < 0.1) because the overall fit (R 2 value) was improved by including the predictor. Growing season length To determine how first and last flowering dates were influenced by growing season length, we calculated growing season length based on the first and last freeze (0 C) dates. We used simple linear regression models to test whether the growing season on the shortgrass steppe has lengthened over the last 20 yr based on first and last freeze dates. To assess the relationship between first flowering day and growing season length for individual species and seasonal groups, we regressed the first flowering day against date of last freeze. To determine whether later-flowering periods were related to growing season length, we regressed the last flowering day to the date of first freeze. We also compared growing season length based on phenological observations calculated as the average date of earliest onset of spring (stage 3, Table 2) among all non-cacti species. The end of season date was calculated as the average date of senescence (stage 14, Table 2) among all species. We used linear regression model to determine whether the growing season length has extended over the last 20 yr based on phenological observations. To assess whether growing season length is influenced by the first flowering day, we regressed the first flowering day against the date of onset of spring (stage 3, Table 2). To determine whether growing season length was related to later-flowering periods, we regressed the last flowering day against the date of end of season (stage 14, Table 2). We used R statistical software to run all analyses (R Development Core Team 2016) using the packages MuMIn (Barton 2016) and stargazer (Hlavac 2015). RESULTS Climate Historical March September mean temperature anomalies from 1940 to 2014 show four periods above the 74-yr mean, , , , and A five-year moving average showed that the longest period above the 74-yr mean is from 2000 to 2014 (Appendix S1: Fig. S1). Annual March September maximum temperature significantly increased between 1995 and 2014 on the shortgrass steppe (0.133 C per year; R 2 = 0.30; P = 0.013). Annual March September mean temperature also significantly increased over the study period (0.07 C peryear;r 2 = 0.21; P = 0.043). There was no trend in March September minimum temperature. March September mean temperature anomalies from 1995 to 2014 were above the 74-yr mean for 13 of the 20 yr (Fig. 2A). Anomalies for March September precipitation over the 20-yr period significantly decreased between 1995 and 2014 on the shortgrass steppe ( 8.65 mm per year; R 2 = 0.30; P = 0.011). March September total precipitation anomalies from 1995 to 2014 were below the 74-yr mean for 13 of the 20 yr (Fig. 2B). For specific monthly maximum, minimum, and mean temperature and precipitation anomalies for , see Appendix S1: Figs. S2 and S3. Phenology: response over time The overall trend for first flowering date of all shortgrass steppe species showed an advance of 0.53 d per year or 10.6 d over the 20-yr period (one-sample t-test, t= 3.34, df = 20, P = 0.003; lower confidence limit = 0.85; upper confidence limit = 0.20). First flowering day for all species for all years showed an advance in response to increasing March September maximum temperature ( 4.83 d per year; R 2 = 0.29; P = 0.018). A weak relationship was found between first flowering day for all species and March September mean temperature ( 5.68 d per year; R 2 = 0.19; P = 0.066). No relationship was found with March September minimum temperature. There was no significant trend for an overall change in the last flowering day for all species combined. Of the 21 species observed, two species showed a statistically significant change in first flowering 6 May 2017 Volume 8(5) Article e01819

7 Fig. 2. Growing season (March September) climate anomalies on the shortgrass steppe from 1995 to (A) Mean temperature and (B) precipitation. Refer to Appendix S1: Figs. S3 and S4 for detailed monthly temperature and precipitation anomalies. Line indicates 1 standard deviation. day, Carex duriuscula and Lepidium densiflorum (Table 3, Fig. 3). Two species showed a shift in last flowering day. One showed a significant advance (Pascopyrum smithii) and one species displayed a significant delay in last flowering date (Eriogonum effusum)(table3,fig.3). Analysis of seasonal groups showed that the early-seasonal group showed a significant advance in both first flowering date and last flowering date. The mid-seasonal group showed a significant advance in first flowering date. The late-seasonal group exhibited a significant delay in last flowering date (Table 3, Fig. 3). The regression of the early-flowering slope and mean first flower day for each species showed that earlier-flowering species exhibited a trend toward earlier flowering (R 2 = 0.29, P = 0.012). The regression of the late-flowering slope and mean last flower day for each species showed that late-flowering species showed a tendency to end flowering later (R 2 = 0.39, P = 0.002). Phenology: response to environment In the analysis of the effect of environment on first and last flower days, model selection yielded a majority of models consisting of a single explanatory variable. First flowering date of 12 species was best explained by a single variable; first flowering date of two species had no significant relationship to specific periods of temperature or precipitation (C. duriuscula and Sphaeralcea coccinea), and seven species were best 7 May 2017 Volume 8(5) Article e01819

8 Table 3. Overall phenological shifts for first flowering date (FFD) and last flowering date (LFD) for 21 species and seasonal groups from 1995 to Species Seasonal group No. years flower FFD LFD Coefficient R 2 P-value Coefficient R 2 P-value Leucocrinum montanum E Carex duriuscula E * Packera tridenticulata E Echinocereus viridiflorus E Vulpia octoflora M Plantago patagonica M Lepidium densiflorum M * Sphaeralcea coccinea M Opuntia polyacantha M Hesperostipa comata M Escobaria vivipara M Elymus elymoides M Aristida purpurea M Pascopyrum smithii M * Heterotheca villosa M Eriogonum effusum L * Bouteloua gracilis L Gutierrezia sarothrae L Salsola tragus L Ericameria nauseosa L Artemisia frigida L Early Season Mid-Season Late Season Notes: E, early season; M, mid-season; L, late season. Species are listed by sequence of flowering from earliest to latest. Includes regression coefficients (days/year 1 SE), R 2, and P-value. Boldface indicates species with a significant change in phenology over the observation period (*P < 0.05). Negative coefficient denotes an advance; a positive coefficient denotes a delay. Denotes exclusion due to sampling error. explained by a combination of temperature and precipitation (Table 4). Warmer March and April mean temperatures were predictors for earlier first flowering day for five species (Table 4, Fig. 4). Warmer May average temperature advanced first flowering day for three species and delayed first flowering day for two species, warmer June mean temperature delayed first flowering day for one species, and various periods of precipitation suggest a later first flowering day for the majority of species (Table 4, Fig. 4). Of the seasonal groups, only the early-season group showed a significant relationship between first flowering day and any predictor variable; March temperature advanced first flowering date and May precipitation delayed first flowering date (Table 4, Fig. 4). In the last flowering date analysis, single parameter models were significant in 19 out of 21 species, and outperformed multi-parameter models in all cases (Table 5). Last flowering date was earlier in three species in response to warmer June or July mean temperature. All other species exhibited a later last flower date in response to various periods of increased precipitation. All seasonal groups demonstrated a later last flowering date in response to wetter conditions (Table 5, Fig. 5). Two species (S. coccinea and Opuntia polyacantha) exhibited no relationship between timing of last flower and temperature and precipitation. The seasonal group analysis showed that the early-species group bloomed earlier and stopped flowering earlier in response to warmer spring temperatures. All species flowered later in response to a wet spring. Mid- and late-blooming species delayed both first flower and last flower in response to wetter conditions (Tables 4 and 5, Figs. 4 and 5). The majority of species responded to the period of temperature and precipitation immediately preceding and including the month of flower time (Tables 4 and 5). A wetter July, 8 May 2017 Volume 8(5) Article e01819

9 Fig. 3. Shifts in flowering phenology over the 20-yr observation period for individual species, seasonal groups, 9 May 2017 Volume 8(5) Article e01819

10 (Fig. 3. Continued) and all species combined. Species are listed from early season to late season. Each bar for individual species and seasonal groups is the simple linear regression slope of the day of year for first flower (blue) and last flower (red) by year from Table 3. Error bars are 1 SE from Table 3. All species combined is the mean from one-sample t-test, and the error bars are the 95% confidence interval. Cross-hatched bars represent significant shifts (P < 0.05). August, and September delayed the first flowering date of three late-blooming species (B. gracilis, Salsola tragus, and Artemisia frigida). High total precipitation for July, August, and September delayed the last flowering date of two mid-season grasses (Hesperostipa comata ssp. comata and Elymus elymoides) as well as S. tragus (Table 5). Growing season length We found no trend for a longer growing season based on first and last freeze date, nor did we find a relationship between last spring freeze date and first flowering date or between first fall freeze date and last flowering date. We found that the onset of spring based on phenological observations has advanced 24.4 d over the 20-yr period ( 1.22 d per year, R 2 = 0.36; P < 0.006). We were unable to determine whether the growing season length has changed over the sample period because late fall observations were terminated when the field crew ended the observation period during the first week of October from 1995 to 2008; this resulted in incomplete senescence data for several late-season species. Table 4. First flowering day regression coefficients (parentheses) for explanatory variables (boldface) for all 21 species for all years. Species or seasonal group Explanatory variable (Coefficient) No. obs. R 2 Adj. R 2 Leucocrinum montanum March T ( 4.415***) Packera tridenticulata May T ( 1.651) + May PPT (0.106**) Carex duriuscula NR 12 Echinocereus viridiflorus March T ( 4.011***) + April T ( 2.148*) + June PPT (0.292***) Vulpia octoflora June PPT (0.297*) Plantago patagonica May T (1.022*) + May June PPT (0.137*) Lepidium densiflorum March April T ( 8.419*) Sphaeralcea coccinea NR 15 Opuntia polyacantha May T ( 0.904) + June July PPT (0.174***) Hesperostipa comata June T (3.674**) + May June PPT (0.105**) Escobaria vivipara April T ( 3.113**) + June July PPT (0.098*) Elymus elymoides June July PPT (0.333***) Aristida purpurea July PPT (0.477***) Pascopyrum smithii June July PPT (0.211***) Heterotheca villosa July August PPT (0.269***) Eriogonum effusum May T (2.95*) + July August PPT (0.258***) Bouteloua gracilis July August September PPT (0.314***) Gutierrezia sarothrae August PPT (0.209***) Salsola tragus July August September PPT (0.449***) Ericameria nauseosa September PPT (0.597***) Artemisia frigida July August September PPT (0.295***) Early Seasonal March T ( 2.484**) + May PPT (0.087*) Mid-Seasonal NR 191 Late Seasonal NR 76 Notes: Species are listed in flowering order from earliest to latest. Positive coefficients indicate a delay in first flowering day, and negative coefficients indicate an advance. Average temperature ( C) = T. Precipitation (PPT) is totaled for each month or multiple months. R 2 is reported for simple regression models, and adjusted R 2 is reported for multiple regression models. NR = no relationship to timing of temperature or precipitation. No. Obs. = No. of observations. P < 0.1; P < 0.05; P < May 2017 Volume 8(5) Article e01819

11 Fig. 4. Explanatory variables for first flowering day for all 21 species including both significant and non-significant shifts. Bars indicate the number of species either delayed or advanced in response to explanatory variable. Average temperature ( C) = T. Precipitation (PPT) is totaled for each month or multiple months. Blue = earlyseasonal group; red = mid-seasonal group; and green = late-seasonal group. DISCUSSION Our first objective was to determine whether the flowering phenology of shortgrass steppe plants has shifted over a 20-yr period. We identified clear responses to increases in temperature through time. We documented an advance of first flowering date of 10.6 d over the course of the study; this advance is significantly related to a trend in warming of March September mean temperatures at a rate of 7.5 d earlier for every 1 C increase. Increasing growing season temperatures over time are an indication of climate change; therefore, it is likely that the shift in first flowering dates over time is also related to climate change. We identified species with statistically significant shifts in first and last flowering dates. Table 4 shows several species that were borderline significant (P < 0.1). We bring attention to these species because some of these taxa were significant with the removal of one year. It is important to identify taxa that may be candidates for larger shifts in the future and these species warrant continued monitoring. Our second objective was to examine how phenology responds to the environment in a 11 May 2017 Volume 8(5) Article e01819

12 Table 5. Last flowering day regression coefficients for explanatory variables for all 21 species. Species or seasonal group Coefficient Explanatory variable No. obs. R 2 Leucocrinum montanum 0.180*** May June PPT Carex duriuscula 0.185*** April May June PPT Packera tridenticulata 6.388*** June T Echinocereus viridiflorus 4.384** June T Vulpia octoflora 0.160** June July PPT Plantago patagonica 0.425*** July PPT Lepidium densiflorum 7.213*** July T Sphaeralcea coccinea NR NR 15 NR Opuntia polyacantha NR NR 20 NR Hesperostipa comata 0.286*** July August September PPT Escobaria vivipara 0.411*** August September PPT Elymus elymoides 0.388*** July August September PPT Aristida purpurea 0.302*** August September PPT Pascopyrum smithii 0.288*** August September PPT Heterotheca villosa 0.550*** September October PPT Eriogonum effusum 1.517*** October PPT Bouteloua gracilis 0.510*** September PPT Gutierrezia sarothrae 0.545*** October PPT Salsola tragus 0.387*** July August September PPT Ericameria nauseosa 0.662*** October PP Artemisia frigida 0.906*** October PP Early Seasonal 0.249*** June July PPT Mid-Seasonal 0.407*** July August September PPT Late Seasonal 0.308*** September October PPT Notes: Species are listed in flowering order from earliest to latest. Positive coefficients indicate a delay in last flowering day, and negative coefficients indicate an advance. Average temperature ( C) = T. Precipitation (PPT) is totaled for each month or multiple months. NR = no relationship to timing of temperature or precipitation. No. Obs. = No. of observations. *P < 0.1; **P < 0.05; ***P < semiarid steppe. Our results show that both temperature and precipitation are critical in assessing how first and last flowering dates respond to the environment in a semiarid steppe. Warm temperatures are associated with early-season flowering advances in early-blooming taxa, whereas precipitation is associated with delays in last flowering dates in all seasonal groups. Phenology: response over time Much current phenological research is directed toward understanding the effects of climate change, but it can be difficult to attribute shifts in phenology to climate change (Rosenzweig et al. 2008, Menzel 2013). Researchers have suggested that the minimum time series length in which adetectionofachangeinphenologycanbeidentified is 20 yr (Sparks and Menzel 2002); however, the length of time needed will depend on the signal-to-noise ratio specific to the area and time period under study. Previous research has linked advancement of phenology with increasing spring temperatures that were attributable to climate change (Beaubien and Freeland 2000, Abu-Asab et al. 2001, Ahas et al. 2002, Scheifinger et al. 2002, Sparks and Menzel 2002, Menzel 2013). Studies from mesic ecosystems, such as those cited above, have shown a higher number of species with significant shifts in phenology compared to the semiarid shortgrass steppe in this study. The reason for this is likely due to the interactions between temperature and moisture in this system (Moore et al. 2015, Moore and Lauenroth 2016). Some of these studies showed a response in the 1990s, which at the time of publication of these studies was the warmest decade on record (Abu- Asab et al. 2001, Ahas et al. 2002, Scheifinger et al. 2002). We recorded observations for the last half of the 1990s; however, the climate on the shortgrass steppe during this five-year period was relatively cool and moist (Fig. 2A, B). This suggests that it is difficult to link phenological change on the shortgrass steppe with other locations during this period due to differences in regional 12 May 2017 Volume 8(5) Article e01819

13 Fig. 5. Explanatory variables for last flowering day for all 21 species including both significant and non-significant shifts. Bars indicate the number of species either delayed or advanced in response to explanatory variable. Average temperature ( C) = T. Precipitation (PPT) is totaled for each month or multiple months. Blue = earlyseasonal group; red = mid-seasonal group; and green = late-seasonal group. climate. We identified a significant increase in March September mean and maximum temperature from 1995 to 2014, which is consistent with increases in temperature anomalies across North America during the same period and months (Settele et al. 2014, Redmond and Abatzoglou 2014 their figures 2.9 and 2.10). Our study showed that this warming trend on the shortgrass steppe is significantly related to changes in phenology. The 21 species included in our study advanced the date of first flowering at the rate of 7.5 d C 1 change in March September mean temperature. The rate of change in March September mean temperature was small (0.07 C per year), but the 13 May 2017 Volume 8(5) Article e01819

14 rate of change in the March September maximum temperature over our study period was greater (0.133 C per year). Given that there was no significant trend for March September minimum temperature, the phenological responses we found are likely related to warmer days and not nights. This is supported by current research; a recent study has documented that bud-break response to daytime warming occurred in Picea mariana under controlled conditions (Rossi and Isabel 2017). Phenology: response to environment Our results show that the timing of specific temperature and precipitation conditions drives the phenology of specific groups of plants or individual species. We found that first flowering dates of early- and mid-season species and groups were influenced by the specific timing of warmer and wetter conditions. For example, the first flowering date for the early-season group occurred earlier in response to warmer March temperatures, but delayed in response to a wet May (Table 4, Fig. 4). This may seem counterintuitive; however, if a dry, warm spring occurs, followed by heavy rain in May, the cooler temperatures often associated with spring precipitation events may delay flowering. The important message is that in a semiarid system, the combination of temperature and precipitation are important controls. For example, shortgrass steppe canopy development requires the interaction between warm temperatures and adequate moisture (Moore et al. 2015). This contrasts with results found in Washington D.C. where increasing nighttime temperatures were related to advanced first flowering dates (Abu-Asab et al. 2001). We found that mean temperature and precipitation were important and minimum temperatures (a proxy for nighttime temperatures) were not significant. Some species did not flower every year; during the drought years of 2002 and 2012, which were in the top ten driest of the 74- yr climate dataset, several species did not flower those years or the following year. It is likely that environmental conditions during these non-flowering years are as important in understanding phenological response on the shortgrass steppe as are years where flowering was successful. Our results are unique in the documentation of the extension of flowering time through the delay of last flower by the late-seasonal group and one mid-season species. These delays were related to increases in late-season precipitation (Table 5 and Fig. 5). Most species in this study utilize the C 3 photosynthetic pathway. B. gracilis is one of the dominant species on the shortgrass steppe, in part because of the match between its root system and the water available during the growing season (Sala et al. 1992). Thus, fitness is not linked tightly to the reproductive process. Flowering phenology in grasses can be approached as an indicator of resource availability. Grasses are wind-pollinated, which eliminates the need to synchronize anthesis with pollinator arrival. However, synchrony is important in a resource context. B. gracilis did not flower during the extreme dry years (2002 and 2012), which is noteworthy because B. gracilis can rapidly recover from drought (Lauenroth et al. 1980, 1987) and respond quickly to small rain events (Sala and Lauenroth 1982, Heisler-White et al. 2008), which can be numerous in September. Our site experienced increased precipitation in September above the 74-yr mean (33.1 mm) in 13 out of 20 yr; the average September precipitation was 41 mm (Appendix S1: Fig. S3C). Because B. gracilis can rapidly respond to water inputs, warm fall temperatures and increased precipitation provide a favorable environment for producing reproductive tillers. Growing season length Past studies on northern tallgrass prairie (Dunnell and Travers 2011) and on a maritime peninsula (Taylor and Garbary 2003) have found both advances and delays in first flowering date associated with growing season length as defined by first and last freeze dates. Advances in flowering date and advances in the onset of spring as defined here (stage 3, Table 2) are linked. Both early- and mid-season species have a weak relationship between first flowering day and onset of spring date. The advance of first flowering date of early-season species in response to an earlieronset date is well documented in the literature and we report similar results here (Fitter et al. 1995, Bradley et al. 1999, Menzel and Fabian 1999, Post and Stenseth 1999, Fitter and Fitter 2002; Miller-Rushing et al. 2007, 2008, Lesica and Kittelson 2010, CaraDonna et al. 2014). Due to sampling problems at the end of season, we were unable to calculate a growing season length based on phenological observations over the study 14 May 2017 Volume 8(5) Article e01819

15 period. Other work has examined the shortgrass steppe for trends in growing season length; for example, recent research using near-surface remote sensing found no significant trend in change of growing season length over a 12-yr period (Moore and Lauenroth 2016). The onset of vegetative growth on the shortgrass steppe is dependent upon adequate temperature and water (Moore et al. 2015). Recent experiments have shown that increased CO 2 resulted in extending the growing season by counteracting the negative effect of warming through conserving water in the soil. This effect was significant for B. gracilis vegetative growth, but not for flowering phenology (Reyes-Fox et al. 2014). Our study showed two clear results. The first is that shortgrass steppe plants are flowering earlier over time in response to increasing growing season temperatures. Phenology has the capacity to detect effects of climate change (IPCC 2013). The changes in phenology identified in this study are likely related to climate change. Second, our study contributes important information about how flowering phenology responds to the environment. In temperate grasslands, ecosystem processes are governed by variables that influence productivity, for example, water, light, or nitrogen (Burke et al. 1998). Our results suggest that, in a water-limited ecosystem such as the shortgrass steppe, earlier flowering is occurring primarily in response to warming spring temperatures for early-flowering taxa and late-blooming species are flowering later primarily in response to precipitation. Under future climate change scenarios, insect-pollinated species on the shortgrass steppe will need to adapt to both resource availability (abiotic synchrony) and pollinator presence (biotic synchrony). Our study shows that shifts by wind-pollinated and insectpollinated species may have the ability to adjust the timing of reproduction in response to temperature and precipitation and maintain abiotic synchrony. ACKNOWLEDGMENTS We thank the National Science Foundation Grant NSF DEB ; M. Lindquist site manager of the SGS LTER, N. Kaplan information manager of the SGS LTER, K. Meierbachtol field crew leader, T. Martyn assistant field crew leader, the SGS LTER field crew from 1995 to 2008 for data collection; Dana Blumenthal, Julie Kray, and Julie Bushey for 2014 data collection and quality assurance and quality control; and anonymous reviewers for help improving the manuscript. LITERATURE CITED Abu-Asab, M. S., P. M. Peterson, and S. G. Shetler Earlier plant flowering in spring as a response to global warming in the Washington, DC, area. Biodiversity and Conservation 10: Ahas, R., A. Aasa, A. Menzel, V. G. Fedotova, and H. Scheifinger Changes in European spring phenology. International Journal of Climatology 22: Aldridge, G., D. W. Inouye, J. R. K. Forrest, W. A. Barr, and A. J. Miller-Rushing Emergence of a mid-season period of low floral resources in a montane meadow ecosystem associated with climate change. Journal of Ecology 99: Anderson, D. R Model-based inference in the life sciences: a primer on evidence. Springer Science & Business Media LLC, New York, New York, USA. Barton, K MuMIn: multi-model Inference. R package version package=mumin Beaubien, E., and H. Freeland Spring phenology trends in Alberta, Canada: links to ocean temperature. International Journal of Biometeorology 44: Bradley, N., A. Leopold, J. J. Ross, and W. Huffaker Phenological changes reflect climate change in Wisconsin. Proceedings of the National Academy of Sciences of the USA 96: Burke, I. C., W. K. Lauenroth, M. A. Vinton, P. B. Hook, R. H. Kelly, H. E. Epstein, M. R. Aguiar, M. D. Robles, M. O. Aguilera, and K. L. Murphy Plant-soil interactions in temperate grasslands. Biogeochemistry 42: Burnham, K. P., and D. R. Anderson Model selection and inference. Springer Science & Business Media, New York, New York, USA. CaraDonna, P. J., A. M. Iler, and D. W. Inouye Shifts in flowering phenology reshape a subalpine plant community. Proceedings of the National Academy of Sciences of the USA 111: Craine, J. M., E. M. Wolkovich, E. G. Towne, and S. W. Kembel Flowering phenology as a functional trait in a tallgrass prairie. New Phytologist. Crimmins, T. M., M. A. Crimmins, and C. D. Bertelsen Onset of summer flowering in a Sky Island 15 May 2017 Volume 8(5) Article e01819