1. Many studies have documented advancement in spring plant phenology; however, studies in dry climates, where water, rather than temperature, is the limiting factor, are rare. To better understand how plants of a water-limited environment may respond to predicted changes in climate, we used a species-rich 20-year data set collected in a semi-arid ecosystem to determine species’ relationships with precipitation and temperature for seasons coincident with and previous to flowering. Our data were collected across a 1200-m elevation gradient, allowing us to explore the consistency in relationships with climatic variables from desert scrub to pine forest. A second objective was to document evidence of changes in the onset of spring flowering over this 20-year period.
2. Onset of spring flowering for species at the lowest elevations was most commonly driven by temperature and precipitation conditions of the previous autumn. In contrast, onset of spring flowering for species in high-elevation communities was more often associated with spring temperatures, a pattern consistent with communities of higher latitudes. Despite these coarse patterns, species’ relationships to climate variables were highly variable and individualistic.
3. Approximately 10% of species showed a significant trend in changes in first flowering date over the period 1984–2003; most trends were in the direction of later onset. The decrease in autumn precipitation observed over the study period appears to explain the delay in onset observed for many of the species across the elevation gradient. Other species’ delays in spring flowering appear to be related to the slight decrease in spring temperature observed over the study period.
4.Synthesis. The south-western USA is expected to become warmer and drier. Climate relationships documented in this study suggest divergent, individualistic changes in the onset of spring flowering. Low-elevation plants may exhibit delayed spring flowering due to changes in the timing or amount of precipitation or insufficient chilling. High-elevation species may show advancement in spring flowering due to warming temperatures. The highly individualistic responses to climate change may result in significant changes in the diversity, composition and abundance of plants in flower. Variable changes in phenology such as these have major implications for species population dynamics and ecosystem functioning.
Temperatures are expected to continue to increase in the arid south-western USA between 2 and 6 °C by 2100 (Karl, Melillo & Peterson 2009). In addition, total precipitation for this area is predicted to decrease in coming decades, especially in winter and spring months, with an increase in the frequency of extreme events including droughts, heavy rain events and heat waves (Karl, Melillo & Peterson 2009). As in other parts of the world, advancement in spring flowering has been reported for low-elevation woody plants in the Sonoran Desert region concomitant with warming temperatures (Bowers 2007); however, little is known about changes expected for other life forms in low-elevation deserts or for biotic communities at higher elevations in dry climates. Species in this study area are highly individualistic in their relationships to climate variables and in their responses to changing climate conditions (Solbrig & Yang 1977; Crimmins et al. 2008; Crimmins, Crimmins & Bertelsen 2009).
The ‘sky islands’ of the arid south-western USA offer the opportunity to explore plant response to changing climate conditions across a variety of vegetation communities. ‘Sky islands’, forested mountain ranges separated by desert or grassland plains, have been likened to a latitudinal gradient stretching from the southern USA to Canada (Merriam 1894). Changes in flowering in response to precipitation and temperature patterns may manifest differently in plant communities at different elevations in water-limited environments. Divergent responses to warming temperatures have been documented along a latitudinal gradient in the eastern USA (Zhang, Tarpley & Sullivan 2007); differing changes may also be expected across elevation gradients.
The response of plants in water-limited environments to changes in precipitation and temperature patterns may also vary across life forms. For example, water-storing plants such as cacti and other succulents common in dry climates do not necessarily respond facultatively to rainfall. Flowering in these plants is cued by a large array of biotic and abiotic conditions, with cues varying widely even among congeneric species (Fleming et al. 2001; Petit 2001; McIntosh 2002; Bustamante & Búrquez 2008). Succulent plants may not be expected to be as affected by changes in climate conditions as plants that do not store water. Additionally, phenological cues among plants of dry environments vary across life forms and individual species (Beatley 1974; Bowers & Dimmitt 1994).
To better understand how plants of a water-limited environment may respond to predicted changes in climate, we used a species-rich 20-year data set collected in a semi-arid ecosystem to determine species’ relationships with precipitation and temperature for seasons coincident with and previous to flowering. Our data were collected across a 1200-m elevation gradient, allowing us to explore the consistency in relationships with climatic variables from desert scrub to pine forest. Based on the relationships with climate variables identified in these analyses, we describe the implications of predicted climate change for species in this area. A second objective was to document evidence of changes in the onset of spring flowering over this 20-year period. Because our period of record is relatively short for such a highly variable system, it may be difficult to detect statistically significant trends. Trends will also be a function of the climate conditions of this particular period, which may not represent the longer-term trajectories in climate.
Materials and methods
The data analysed in this study are 46 783 observations of plants in flower for 428 species made by one of us (C.D.B.). These data are the results of 553 round-trip (16 km) hikes of the route to Mt. Kimball and 12 partial trips of 4.8 to 12.1 km made between the months of January–June over a 20-year period (1984–2003) [4 ± 0.02 hikes month−1 (mean ± SE)]. Days between hikes ranged from 0 to 29 and the average number of days between hikes was 11.3 (±0.02) for the period 1984–1991 and 5.7 (±0.01) for the period 1992–2003. Hikes were more frequent during months when species were expected to flower or were in flower; long periods between hikes, such as the 29-day period, occurred in months when few or no species were flowering. The route to Mt. Kimball is an 8-km hiking route located on the south slope of the Santa Catalina Mountains, approximately 7 km north of the city limits of Tucson, Arizona, USA and follows the Finger Rock and Pima Canyon trails. The Finger Rock trail climbs 7.2 km from an elevation of 945 m above sea level at the trailhead through Finger Rock Canyon to the junction with the Pima Canyon Trail at 2088 m; the Pima Canyon trail continues 0.8 km to the peak of Mt. Kimball at 2213 m. Finger Rock Canyon has a southwest–northeast orientation and is bounded by canyon walls rising 122–610 m in a series of cliffs and steep rocky slopes.
The observer recorded species observed in bloom, with the primary focus being an area within 9.1 m of either side of the trail. Observations were recorded along five trail segments approximately 1.6 km in length (Table 1) both on the ascent and descent; on the descent attention was given to verifying records and adding any taxa not already recorded. Trail segment 1 refers to the lowest elevation segment; trail segment 5 refers to the highest elevation. Blooming was defined as the presence of pollen on anthers in angiosperms; gymnosperms were considered ‘in bloom’ by the presence of airborne pollen, either seen when the plant was touched or seen against the sky when dislodged by wind gusts, in the case of tall individuals. Any species not recognized was collected for expert identification.
Table 1. Elevation range and dominant biotic communities represented within each 1.6-km trail segment along the Mt. Kimball route, Santa Catalina Mountains, Tucson, Arizona, USA
Each species was assigned to one of the following functional types: annual/biennial forbs (including annual forbs, grasses and vines and seven biennial species), herbaceous perennials (including perennial forbs, grasses and vines), woody plants (including subshrubs, shrubs and trees) and succulents (including both stem and leaf succulents). Functional type groupings were selected based on the different water use strategies employed by plants of semi-arid environments (Ehleringer et al. 1991; Flanagan, Ehelringer & Marshall 1992; Burgess 1995). Succulent plants were analysed separately from the other species due to their water storage capacity, a feature that sets them apart from the other functional types.
The climate in south-eastern Arizona is characterized by bimodally distributed precipitation, with most precipitation occurring in either the summer (July, August, September) monsoon season or winter (December, January, February, March) season. Annual average precipitation for Tucson is nearly 300 mm (WRCC 2009). Winter temperatures average 10 °C and summer maximum temperatures often exceed 40 °C at lower elevations. Mountaintops in this region are subject to average daily temperatures ranging from 4 °C to 25 °C annually and precipitation amounts exceeding 700 mm (WRCC 2009). The complex terrain results in high temperature and precipitation gradients over short distances.
Testing for trend in the start of spring
We used simple linear regression to evaluate whether significant trends in springtime first flower dates (FFD) existed for species on a particular trail segment with at least 15 observations during the study period 1984–2003. One-hundred and thirty-six unique species were tested. Several species appear on multiple trail segments; each species and trail segment combination was tested independently. In total, 236 species-segments were tested.
Flowering in the Sonoran Desert is characterized by two distinct flowering periods, spring (approximately March–May) and summer monsoon (mid-July–mid-October) (Solbrig & Yang 1977; Crimmins et al. 2008). This analysis focused on the spring flowering season. Within each trail segment, first flowering dates were determined by identifying the date of the first recorded bloom between 1 January and 30 June of each year. First flowering date was selected over other metrics such as mean flowering date due to the highly variable nature of flowering in plants in semi-arid environments: plants in semi-arid environments are highly responsive to environmental conditions and can commence, cease and recommence flowering several times in a single season, making estimation of mean flowering date difficult. In addition, the onset of flowering and the abundance of individual plants in flower in a season appear to be driven by separate mechanisms in dry environments (Bowers 1987, 2005; Bowers & Dimmitt 1994). Cues for initial onset of flowering in water-limited environments are typically described as an accumulation of temperature units following a triggering precipitation event, though triggers are species-specific (Beatley 1974; Kemp 1983; Bowers & Dimmitt 1994; Steyn et al. 1996; Abd El-Ghani 1997). Triggering events are described as ‘switches that break bud dormancy and start developmental processes such as leafing or flowering’ (Loomis & Connor 1992; Bowers 2007). Following the onset of flowering, the abundance of individuals of a single species in flower in dry environments appears to be a function of total seasonal precipitation (Bowers 1987, 2005).
Since the observations were not collected every day, the first flowering date was narrowed to a range of dates between the first recorded day in bloom and the previous observation in which the species was not recorded in bloom. To deal with this uncertainty in FFD, linear regressions were performed on 5000 randomly generated combinations of possible FFD dates bounded by the first observed and previous ‘not-in-bloom’ date. A distribution of regression parameters including slope coefficients and corresponding P values were then used to assess the sensitivity of the trend in FFD to sampling error. Species with mean P values less than or equal to 0.05 were considered to have significant trends in FFD robust enough to overcome uncertainty introduced through the irregular sampling intervals. Significant models were further examined to ensure that assumptions of normally distributed variables and homoscedasticity were met. Statistical tests were performed using Matlab v.R2009a (Mathworks 2009). Finally, we used Fisher’s combined test to compare the distribution of species-specific responses to that which would be expected by chance; this test was performed using stata v.9.2 (StataCorp, 2006).
Assessing potential relationships between FFDs and monthly climate variables
Stepwise multiple regression was used to evaluate potential relationships between observed FFDs and monthly climate variables. Since there was no climate monitoring station at the study site, the climate data set was created by averaging monthly precipitation and temperature anomalies calculated from the period 1970–2003 from five nearby National Weather Service Cooperative Observer sites. Multiple stations were used to capture dominant patterns in temperature and precipitation across the region and to guard against the use of single station data that represent local, microclimatic variability unrelated to the study site (Pielke et al. 2002). These stations also capture climate variability across a range of elevations representative of the study site gradient.
A forward stepwise regression was performed on each species with the observed FFD as the predictand and a pool of monthly climate variables as predictors. Climate variables included October–June monthly average precipitation and temperature anomalies from 1983 to 2003. Climate variables encompassed the autumn and winter seasons prior to flowering and concomitant spring months to capture triggering events and conditions. Correlations between monthly climate variables were examined to reduce the risk of multicollinearity in the final regression models. They were generally low and non-significant indicating that each variable could contribute as an independent predictor. The pool of potential monthly climate predictors was limited to only those months coincident with and prior to the month of the observed first flower date. This approach allowed for the examination of the role of coincident spring and antecedent autumn climate conditions on first flower date variability over the study period. Statistical tests were performed using Matlab v.R2009a (Mathworks 2009).
Trends in the onset of flowering
Of the 236 unique species-segment combinations tested for a change in the timing of the start of spring flowering, 25 species (9.8%; Table 2) exhibited a significant trend (P≤ 0.05, Fig. 1), which is significantly greater than would be expected by chance (χ2 = 607.00, P < 0.001; Fisher’s combination procedure). Five of these species’ trends were in the negative direction (toward an earlier onset of flowering); these species all occurred at the lowest elevations, in segments 1 and 2. The remaining species’ trends were in the positive direction (toward a later onset of flowering) and were observed in all five segments. Trends ranged from approximately 3 days earlier per year to nearly 5 days later per year (Fig. 1). The majority of the species exhibiting trends were perennial plants (Fig. 1).
Table 2. Count of significant spring onset climate models and significant spring onset trends for species recorded along the route to Mt. Kimball, Santa Catalina Mountains, Tucson, Arizona, USA
Significant relationship with more than one climate variables
Species showing significant trend
Of the 25 species exhibiting significant trends in the change of FFD over this period, 18 species showed significant relationships with one or more climate variables (Fig. 1). October and November precipitation occurred most frequently in these models (six and five models, respectively). April temperature occurred in four models (segments 1–5). Adjusted R2 values ranged from 0.18 to 0.63.
One species, Erigeron divergens Torr. & A. Gray (spreading fleabane), demonstrated significant trends in segments 1, 2 and 3, and exhibited a significant relationship with November precipitation in segments 2 and 3. Phacelia ramosissima Douglas ex Lehm. (branching phaceila) showed significant trends on both segments 1 and 4 (Fig. 1). Anecdotal evidence indicates that these species have not shown fluctuations in abundance over the period of record (C.D.B., pers. obs.), so trends in FFD are not due to increases or decreases in population sizes (Miller-Rushing, Inouye & Primack 2008).
Climate variable results
Over the period of record, temperatures varied seasonally (Table 3). The variability of monthly temperatures from year to year during the study period was relatively low with SD values ranging from 1.0 in December to 1.8 in April.
Table 3. Summary of monthly temperature and precipitation variables averaged across five weather stations in proximity to Mt. Kimball, Santa Catalina Mountains, Tucson, Arizona, USA
Precipitation also varied seasonally with relatively high values during the winter months of December, January and February and low values in late spring (May and June). Variability in monthly precipitation was quite high, with all months exhibiting SD values above 10 (Table 3). The highest SD values generally occurred in the autumn and winter months (October, December and January). Overall these values indicate the presence of large swings in monthly total precipitation from year to year during the study period of 1984–2003. Each monthly time series of precipitation and temperature was also examined for trends over the study period. Most months showed weak and non-significant trends of varying signs in their temperature data. November had the strongest trend towards warming (P = 0.16) while April had the strongest trend towards cooler temperatures (P = 0.30; Table 3). All monthly precipitation time series exhibited a trend towards dryer conditions, albeit generally weak and not significant. The strongest trends occurred in the autumn (November, P = 0.15) and early winter (December, P = 0.21) with much weaker trends occurring through the remaining winter and spring months.
Relationships with climatic variables
Many species exhibited relationships with climate variables but did not show trends in FFD. Of 236 species-segments tested, 99 showed a significant relationship with a single climate variable, and 91 showed a significant relationship with two or more climate variables (Table S1, Supporting information). The fit (adjusted R2) of the regression equations varied from 0.16 to 0.92. In general, the models performed best for woody plants and model fits were lowest for succulent species (Fig. S1, Supporting information). Models generally performed best in segment 1 (lowest elevation) and poorest in segment 5 (highest elevation; Fig. S1, Supporting information).
Of the species showing significant relationships with a single climate variable, the variables appearing most frequently in models for annual, herbaceous perennial and woody plants were October and November precipitation (Fig. 2), November temperature and spring (March, April and May) temperature (Fig. 3). These variables also appeared most commonly in models with two or more climate variables. December precipitation and January temperature also appeared in several models with more than one climate variable.
October or November precipitation was the sole predictor in 27 of the models for annual, herbaceous perennial and woody plants (12% of annual, herbaceous perennial and woody plant species tested), primarily in segments 1, 2 and 3 and in all three functional types (Fig. 2). Similarly, October or November precipitation appeared in 51 (23% of species-segments tested) models with more than one variable. In nearly all cases the relationship was negative, indicating that a wet autumn is associated with earlier blooming.
November temperature was the sole predictor in 17 of the models for annual, herbaceous perennial and woody plants (8% of species-segments tested), primarily in segments 1 and 2 and most frequently in annual plants (Fig. 3). Likewise, November temperature occurred in 21 models (10% of species-segments tested). In every case, the relationship was positive, indicating that cool autumn temperatures are associated with earlier flowering.
Twenty-four annual, herbaceous perennial and woody plant species (11% of species-segments tested) exhibited relationships with only a single spring (March, April or May) temperature variable. Spring temperature variables occurred in another 42 models (19% of species-segments tested) with another variable. These relationships were very commonly strongly negative, indicating that warm spring temperatures are associated with earlier spring flowering. Spring temperature variables occurred in models in all five trail segments. In the upper segments (segments 3, 4 and 5), many of the significant models were comprised of a single spring temperature variable and models incorporating spring temperature accounted for most significant models in these segments (Fig. S1, Supporting information; Table 2).
Of the 17 succulent species tested, 10 exhibited significant relationships with a single climate variable and six showed significant relationships with two or more climate variables. Succulent species showing significant relationships with climate variables were mainly concentrated in segments 1, 2 and 3. March and April temperature variables occurred most frequently in these models (four models each, 24% of species-segments tested for each variable), almost always in a negative relationship, indicating that warm spring temperatures are associated with earlier spring flowering in these species (see Fig. S2, Supporting information). December precipitation occurred in three models (18% of species-segments tested) in a positive relationship, indicating that wetter Decembers are associated with earlier spring flowering for these species (Fig. S2, Supporting information).
Common patterns in climate variables
Several models of annual, herbaceous perennial and woody plant species exhibited similar patterns in the variables included (see Fig. S3, Supporting information). November precipitation occurred in many models with January temperature, March temperature or April temperature (see Fig. S3, Supporting information). In these models, both variables were often negative, indicating wet autumns and warm springs are associated with earlier spring flowering. In some models, April temperature exhibited a positive relationship with flowering time; these models were primarily for species in segments 1 and 2. This suggests that for these species, wet autumns and cool Aprils are associated with earlier spring flowering.
November temperature occurred in many models with January precipitation, January temperature, or March or April temperature. In these models, November temperature and January precipitation exhibited positive coefficients and as above, January, March and April temperatures exhibited a negative coefficient. This indicates that these species flower earlier in the spring under combinations of either cool autumn and dry winter or cool autumn and warm winter or spring temperatures. The species flowering earlier under cool autumn and dry winter conditions occurred almost solely in segments 1 and 2 and were primarily annual plants.
Model regression coefficients
The variability in the regression coefficients was higher for some climate variables than others (Table S1, Supporting information). November precipitation, a variable occurring in many models, ranged from −2.2 to −0.3, indicating that depending on the species, an increase in autumn precipitation of 1 mm could translate to advancement in the FFD from 0.3 to 2.2 days per year. November temperature, another commonly occurring variable, also exhibited relatively large coefficients, ranging from 2.7 to 14.2, translating to a delay in FFD of nearly 3–14 days per year for each 1 °C increase in temperature. Similarly, the coefficients for spring (March–May) temperature variables, also common in many models, were on the order of 2–10 days (Table S1, Supporting information).
Consistent with other studies of climate drivers of phenological events (Peñuelas et al. 2004; Cleland et al. 2006; Sherry et al. 2007), responses in this study were highly species-specific. However, some general patterns were apparent across the elevation gradient. Although a combination of temperature and precipitation variables played a key role in predicting flowering date in species of all elevations, autumn conditions were much more important to low-elevation species and spring variables played a greater role in flowering of high-elevation species (Figs 2 and 3). These patterns are consistent with climate drivers reported for the diversity of species in flower of this study area (Crimmins et al. 2008) and also with climate constraints along a north–south latitude gradient (e.g. Bowers & Dimmitt 1994; Kullman 2002).
High autumn precipitation associated with earlier spring flowering
The high number of species demonstrating a strong relationship with autumn precipitation variables at all elevations (Fig. 2) reinforces the importance of antecedent season moisture conditions on the timing of spring flowering in plants of semi-arid environments (Beatley 1974; Kemp 1983; Fox 1990; Bowers & Dimmitt 1994; Friedel et al. 1994; Abd El-Ghani 1997; Borchert et al. 2004). Nearly half of the species exhibiting a strong relationship with November precipitation in this study were annual plants, known to depend on autumn moisture to germinate (Beatley 1974; Mulroy & Rundel 1977). These results support previous findings that the timing of an autumn rain event of a critical size can influence the onset of flowering of many species in the following spring.
Warm spring temperatures associated with earlier spring flowering
The influence of spring temperatures on spring flowering was apparent in many species and at all elevations (Fig. 3). This relationship is consistent with many other studies that have documented advancement in the onset of flowering following conditions of high temperatures (Bradley et al. 1999; Schwartz & Reiter 2000; Fitter & Fitter 2002; Miller-Rushing & Primack 2008). Accumulated temperature is a likely explanation for this consistent relationship; if necessary temperatures are accumulated more quickly following a triggering event or threshold, plants may respond by flowering earlier in the season. In annual plants, another possible explanation may be that flowering is triggered by drought stress (Aronson et al. 1992; however see Fox 1990) that is brought on by rapidly increasing temperatures.
A relationship with a combination of warm spring and wet November conditions was exhibited by several annual and herbaceous perennial species at low and intermediate elevations. These conditions are a combination of two of the relationships most frequently associated with species in this study and point to the importance of both autumn rainfall events of a critical size and accumulated temperatures for the timing of spring flowering, as suggested in other studies of this semi-arid environment (Beatley 1974; Bowers & Dimmitt 1994).
Cool November temperatures associated with earlier spring flowering
Many species at low and intermediate elevations flowered earlier following cool autumns (Fig. 3). A positive relationship between autumn temperatures and flowering time has been observed in other studies as well (Fitter et al. 1995; Sparks & Carey 1995; Sparks, Jeffree & Jeffree 2000). Colder autumn conditions could enable plants with a chill requirement to meet these requirements sooner and therefore begin flowering earlier in the spring. In desert environments, low winter soil temperatures have been associated with mass germinations of annual plants (Mulroy & Rundel 1977; Baskin & Baskin 1985); a large proportion of the plants exhibiting this relationship in the present study were annual species. Several models indicated a combination of cool autumn temperatures and warm spring temperatures associated with earlier spring flowering. These models suggest the importance of a chilling requirement followed by accumulated warm temperature to trigger flowering, a set of conditions experienced by species in other environments (Baldocchi & Wong 2008; Luedeling, Zhang & Girvetz 2009). Several other models, primarily for annual plants at low elevations, indicated a combination of cool autumn temperatures and dry January conditions. This relationship may reflect the importance of factors not captured in this study, such as the sequencing of precipitation events through the winter months.
Spring temperature appears to play the biggest role in influencing the timing of spring flowering in succulent species in this study area (Fig. S2, Supporting information) where warm springs are associated with earlier flowering. This pattern has been observed elsewhere; Fleming et al. (2001) documented that cold spring temperatures delayed the onset of spring flowering in some species of cacti in the Sonoran Desert. Bustamante & Búrquez (2008) noted a strong relationship between time of spring flowering and winter temperature for organ pipe cactus, perhaps documenting a similar phenomenon. Winter and spring precipitation variables were also relatively common in models of succulent species, indicating that wet winter or spring conditions are associated with earlier flowering. Fleming et al. (2001) similarly noted that dry conditions delayed onset of flowering in cacti. Succulent plants store water and are therefore buffered from drought; this may explain why temperature variables were more common than precipitation variables in these plants’ models.
Observed trends and future expectations
Trends in FFD reported in this analysis may underestimate changes that are occurring. First, our period of record is relatively short for such a highly variable system. Second, the variable sampling interval may also mask trends in FFD. It does not appear that trends are due to the more frequent hikes later in the record, as more frequent hikes coupled with constant FFDs would yield negative trends and most of the trends reported in this analysis were positive. Moreover, trends are a function of the climate conditions of this particular period, which may not represent the longer-term trajectories in climate. Accordingly, the relationships with climate variables identified in this study may be the most useful findings for predicting future trajectories of these systems under changing climate conditions.
At all elevations, many of the species showing trends toward later onset of spring flowering were influenced by precipitation of the previous autumn in that dry autumn conditions were associated with later spring flowering. The later spring flowering observed in these species may be explained by the slight decrease in autumn precipitation over the period of study (Table 3). As precipitation is expected to continue to decrease in the future, species at all elevations may be driven toward later spring flowering, perhaps on the order of many days to weeks. Additionally, if autumns become so dry that triggering precipitation amounts are not met, some species may not flower at all in the following spring (Fisser 1986), a phenomenon observed in the Finger Rock Canyon following particularly dry autumn and winter conditions (C.D.B., pers. obs.). As different life forms employ diverse strategies for handling precipitation (Noy-Meir 1973; Burgess 1995), decreased precipitation and increased variability in rainfall events may also significantly impact the diversity, composition and abundance of plants in flower.
A small number of high-elevation species that showed trends toward later spring flowering exhibited a consistent negative relationship with April temperature, where cooler Aprils are associated with later onset. The later onsets observed in these species may indeed be the result of the slight cooling trend in April observed over this study period (Table 3). As temperatures are expected to warm substantially in coming decades, we may expect to see these species to advance their flowering, as in other parts of the world (e.g. Fitter & Fitter 2002; Miller-Rushing et al. 2007). The same prediction of advanced FFD may be true for the many species across all elevations in this study showing a negative relationship with spring temperatures but not exhibiting a significant trend over the study period.
Many species at low elevations not showing change in flowering onset during the period of record are most heavily influenced by autumn temperature, where cooler autumn temperatures were associated with earlier spring flowering, perhaps reflecting a chill requirement. Warmer autumn temperatures in the future could lengthen the time to meet chill requirements and result in later flowering in plants dependent upon these cues. Fitter et al. (1995) noted that warm temperatures several months prior to spring flowering retarded the onset of flowering. If autumn temperatures increase beyond plants’ thresholds, chill requirements could fail to be met, possibly precluding species with these requirements from flowering. Recent work has documented changes in the elevation range at which species flower of approximately 25% of the species tested in the study area (Crimmins, Crimmins & Bertelsen 2009), perhaps evidence of species already responding to changing climate conditions.
Several species in the lowest elevation segment that exhibited significant trends showed no correlation with climate variables; these species apparently are influenced by variables not captured in these analyses. These could include photoperiod, documented to be important in a single Sonoran Desert plant (Bowers & Dimmitt 1994), the presence of pollinators, also suggested to be important for some Sonoran Desert species (Waser 1979; Bowers & Dimmitt 1994) or perhaps the seasonal sequence, duration or intensity of precipitation events or accumulated temperatures. The trends toward earlier flowering documented at the lowest elevations are in agreement with patterns documented by Bowers (2007) for woody plants in the Sonoran Desert and in other parts of the globe, but because the drivers for these changes are not clear, it is difficult to predict these species’ responses to future changes in seasonal precipitation and temperature.
The divergent predictions described in this study illustrate that species can be expected to shift their flowering phenologies individualistically, rather than at consistent rates in the same direction. At all elevations, many species may be expected to flower later in the spring under future drier conditions. Conversely, many other species across the elevation gradient may be prompted to flower earlier with warmer temperatures. Advances or delays in phenological events of only a few days can have significant and direct impacts on species population dynamics and ecosystem functioning. The highly variable and individualistic responses documented in this study could lead to changes in abundance or community composition. Sherry et al. (2007) documented a warming-induced divergence of flowering and fruiting in a tallgrass prairie ecosystem, resulting in a gap in the staggered progression of flowering and fruiting in the community during the middle of the growing season. Small advances or delays in phenological events also have the potential to lead to the decoupling of synchronous events and changes in trophic interactions (Stenseth & Mysterud 2002; Hegland et al. 2009). In the Sonoran Desert, the timing of flowering is adapted to coincide with the arrival of pollinators; a change in flowering time could lead to miscues between pollinators and their major food sources, with effects cascading throughout the ecosystem. Seemingly minor changes in the timing of phenological events have the potential to increase competition for resources (Veresoglou & Fitter 1984; Rathcke & Lacey 1985), affect flower size, number and seed set (Kudo & Suzuki 2002; Saavedra et al. 2003) and allow invasion by opportunistic non-native species (Dech & Nosko 2004). Advancement in the onset of flowering is likely associated with advancement in fruit and seed ripening, exposing seeds to longer periods of predation and harsh conditions before conditions are suitable for germination. The patterns recorded here indicate that changes to species’ phenologies and the cascading impacts to community characteristics can be expected to continue as conditions become warmer and drier.
We are very thankful to R. Wilson and P. Jenkins at the University of Arizona Herbarium, and the late J. Reeder and C. Reeder, Gramineae specialists, for assistance with plant identification. C. Williams played a key support role. Two anonymous referees provided valuable suggestions that improved the manuscript.