Biological consequences of earlier snowmelt from desert dust deposition in alpine landscapes

Abstract

Dust deposition to mountain snow cover, which has increased since the late 19^th century, accelerates the rate of snowmelt by increasing the solar radiation absorbed by the snowpack. Snowmelt occurs earlier, but is decoupled from seasonal warming. Climate warming advances the timing of snowmelt and early season phenological events (e.g., the onset of greening and flowering); however, earlier snowmelt without warmer temperatures may have a different effect on phenology. Here, we report the results of a set of snowmelt manipulations in which radiation-absorbing fabric and the addition and removal of dust from the surface of the snowpack advanced or delayed snowmelt in the alpine tundra. These changes in the timing of snowmelt were superimposed on a system where the timing of snowmelt varies with topography and has been affected by increased dust loading. At the community level, phenology exhibited a threshold response to the timing of snowmelt. Greening and flowering were delayed before seasonal warming, after which there was a linear relationship between the date of snowmelt and the timing of phenological events. Consequently, the effects of earlier snowmelt on phenology differed in relation to topography, which resulted in increasing synchronicity in phenology across the alpine landscape with increasingly earlier snowmelt. The consequences of earlier snowmelt from increased dust deposition differ from climate warming and include delayed phenology, leading to synchronized growth and flowering across the landscape and the opportunity for altered species interactions, landscape-scale gene flow via pollination, and nutrient cycling.

The transfer of dust from arid and semiarid lands to snow-covered landscapes takes place around the world (1–3). While this process has occurred historically, the increasing area of arid lands and the utilization of these lands by expanding human populations have increased dust loads (4–6). For example, during settlement of the western U.S., the intensification of human activities such as agriculture, grazing, and resource exploration in semiarid landscapes led to 500% greater dust deposition in the adjacent mountains (7). Furthermore, dust deposition in many regions could increase as a result of the increasing extent of arid lands and greater human activity in these areas (8). Dust deposited in winter and spring decreases the albedo of the snow surface at the time it is deposited and again, when the buried dust layers reemerge during spring snowmelt. Through this change in surface energy exchange, dust can advance the timing of snowmelt by more than a month (9).

Seasonal snow cover has a substantial effect on ecosystem function where freezing temperatures over winter facilitate the formation and retention of a snowpack (10–12). It protects and sustains plant and soil communities by moderating temperatures during winter and later by supplying a source of water to fuel plant growth at the start of the growing season (12–14). The timing of snowmelt also regulates the timing of early season phenological events and can affect reproductive output (15–17). Experimental studies that have altered the timing of snowmelt through infrared warming lamps or by changing snow depth have demonstrated that a corresponding shift in early season phenological events occurs for most species (18–20). Similarly, the onset of greening and flowering occur earlier in years when air temperatures warm and snow melts earlier (21–23).

Since dust-driven advances in the timing of snowmelt result from a change in the radiation balance of the snowpack, they can occur without a corresponding change in air temperatures or a decrease in snow inputs (9). A 1-month shift in the timing of snowmelt is likely sufficient for locations to become snowfree when mean daily air temperatures are still <0 °C. Consequently, we hypothesized that earlier snowmelt from dust deposition would not shift the onset of greening and flowering, but we also considered 2 alternative hypotheses. The onset of greening and flowering may be a linear function of when snowmelt occurs or, alternatively, may initially be delayed then shift to a linear function after the earliest day that the onset of greening or flowering can occur (i.e., a threshold response). Support for either the first or the latter hypothesis would suggest that earlier snowmelt from dust deposition has a different effect on phenology from that of climate warming.

To determine the support for these alternative hypotheses, we conducted a set of snowmelt manipulations on a southeast- and northeast-facing hillslope in the Senator Beck Basin Study Area (SBBSA) at Red Mountain Pass in the San Juan Mountains, CO. We aimed to advance the timing of snowmelt using radiation-absorbing fabric and dust additions of finely ground local rock to delay snowmelt by removing background deposits of desert dust from the snow surface. In early May 2008, on a SE-facing hillslope, we set up 2 fabric, 2 dust addition, and 2 control plots (each 8 × 12 m). We also set up 1 fabric, 2 dust addition, 2 dust removal, and 2 control plots on a nearby NE-facing hillslope. The snowpack was approximately 1 and 2 m deep on the SE- and NE-facing hillslopes, respectively. In early June, when the first plots became snowfree, we began to observe when the first leaf had fully expanded and the first flower had fully opened for each of 12 common alpine species, 5 of which were common to all plots. Observations were made every 2–3 days until mid-August. We averaged the data across species in each plot to determine the community-level response for all species observed and for the 5 species common to all plots. Constant, linear, and threshold (i.e., a piece-wise linear) models were fit to the data by maximizing the log likelihood and calculating Akaike's Information Criteria (AIC). We used the AIC values to quantify the weight of support (w_r) for each model (values of w_r range from 0 to 1.0 or 0 to 100% support and sum to 1.0), which enabled us to identify which model among the 3 competing models had the most support (24).

Results and Discussion

The control plots were snowfree on June 14 [day of year (DOY) 166] and June 27 (DOY 179) on the SE- and NE-facing hillslopes, respectively. Altering the energy balance of the snowpack through radiation-absorbing fabric or adding or removing dust from the snow surface in late winter resulted in earlier or delayed snowmelt relative to control plots (Fig. 1). On average, snow melt occurred 11 days later when dust was removed and 7 and 13 days earlier than in control plots for the dust addition and fabric covered plots, respectively. The effect of the snow manipulations was similar on the SE- and NE-facing hillslopes. Topography and the experimental treatments affected the date that plots became snowfree (P < 0.05 for both, see Table S1). As a result, snowmelt occurred approximately 12 days earlier on the SE hillslope than on the NE hillslope within each treatment (Fig. 1). Across all plots, the timing of snowmelt varied by 44 days from June 1 (DOY 153) to July 15 (DOY 197).

Fig. 1.

The effects of the snowmelt manipulations and topography on the timing of snowmelt. (Upper) The effect of the treatments relative to the controls is shown, where positive values indicate earlier snowmelt and negative values indicate snowmelt was delayed. (Lower) Treatments are arranged in order of increasingly earlier snowmelt, which occurs with greater dust loads. Data are means ± 1 SE, which for some treatments was zero. P values compare aspects within each treatment (n = 1 or 2).

Over this range, plant phenology exhibited a threshold response to the timing of snowmelt, which was characterized by the piece-wise linear model (Fig. 2). Greening and flowering were initially delayed, after which their response to snowmelt date was linear. The weight of support (w_r) for the threshold model was overwhelming, 96% and 100% for the onset of greening and flowering, respectively, when considering all species observed (Table 1). The data also supported the threshold model when considering only the 5 species common to all plots, although support was somewhat reduced (Fig. 2). Because variation in species composition across the plots did not influence model selection, we focused on the community-level response based on all of the species observed.

Fig. 2.

The onset of greening and flowering in relation to the timing of snowmelt. Dates for the onset of greening and flowering are shown for all species observed within a plot (Left) and for the 5 species common to all plots (Right). Statistics report the weight of support for each model (w_r) and the proportion of variation explained (R²) by the model. The dashed line is the 1:1 line. Data are means ± 1 SE (n = 12). Overlapping values were offset by up to 1.5 days for clarity.

Table 1.

Weight of support (w_r) and the proportion of variation explained (R²) by alternative models of plant phenological response to the timing of snowmelt

Data set	Variable	Models	AICc*	ΔR†	Likelihood	wr	R2
All species observed	Onset of greening	Constant	95.1	31.1	0.00	0.00	n.a. ‡
		Linear	70.5	6.5	0.04	0.04	0.90
		Threshold§	64.0	0.0	1.00	0.96	0.96
	Onset of flowering	Constant	90.3	35.7	0.00	0.00	n.a.
		Linear	65.4	10.8	0.00	0.00	0.90
		Threshold	54.6	0.0	1.00	1.00	0.97
Species common to all plots	Onset of greening	Constant	95.2	29.7	0.00	0.00	n.a.
		Linear	68.7	3.2	0.20	0.17	0.91
		Threshold	65.5	0.0	1.00	0.83	0.95
	Onset of flowering	Constant	94.8	25.8	0.00	0.00	n.a.
		Linear	70.1	1.1	0.58	0.37	0.90
		Threshold	69	0.0	1.00	0.63	0.94

*Lower values of Akaike's Information Criteria (AIC) indicate greater support in the data for the model.

^†The difference in the AICc value as compared to the best model.

^‡n.a. indicates not applicable.

^§Bold type indicates the model that had the most support in the data in each set.

The threshold model explained 96% of the variation in the onset of greening and 97% of the variation in the onset of flowering as a function of the timing of snowmelt (Fig. 2 and Table 1). However, if snowmelt occurred on or before June 7 [DOY 159, the parameter estimate for the threshold, which corresponded to 33 thaw degree days from snow melt (TDD_sm)], then the timing of snowmelt did not directly determine the onset of greening (see Fig. 3 for SE of parameter estimates). Instead, early season phenology was characterized by a delay in leaf expansion at the community-level until June 18 (DOY 170, 94 TDD_sm), the estimate for the parameter that defines the day greening first began. Likewise, the onset of flowering was delayed until June 28 (DOY 180, 182 TDD_sm). Shifting the timing of snowmelt before the threshold specific to flowering, June 12 (DOY 164, 44 TDD_sm), did not affect the timing of flowering.

Fig. 3.

Parameter estimates made when the threshold model was fit to the data for all species observed within a plot and seasonal patterns of air temperature and solar irradiance beginning on June 1 (DOY 153), the day the first plots became snowfree. Parameter estimates are means ± 1 SE and are shown for the onset of greening (OG) and the onset of flowering (OF). Minimum and mean daily air temperatures from hourly observations are shown. Thaw degree days are the sum of mean daily temperatures >0 °C since June 1, when the first plots were snowfree. Mean daily and cumulative solar irradiance since June 1 is shown. The dashed gray lines mark 0 °C and the last day on which the mean temperature was 0 °C (DOY 164).

In addition to snowmelt (17), temperature and day length can also have a significant effect on plant phenology (18, 25, 26). To identify whether 1 of these environmental factors could account for the delay in greening and flowering, we plotted the seasonal pattern of air temperature and solar irradiance in relation to the parameter values from the threshold model (Fig. 3). June 12 (DOY 164) is the last day on which the mean daily air temperature was at or <0 °C. Soon after June 12, mean and minimum air temperatures increased and stabilized well above freezing. Greening started 6 days after the mean air temperature was last 0 °C and flowering began 16 days later. This time period around June 12 also corresponds with a dramatic shift in the seasonal pattern of warming as characterized by thaw degree days (Fig. 3). Mean solar irradiance is relatively high during this time period, and day length varies by <10 min at this latitude during the month of June. The seasonal pattern of solar irradiance shifts somewhat later in the summer (DOY 174), after greening began but before flowering began, and did not exhibit a threshold change around June 12. Thus, the initial delay in the onset of greening and flowering is more likely controlled by air temperatures shifting across the phase change for water than by irradiance. Control over the timing of phenological events shifts from air temperature to snowmelt on the threshold dates for greening and flowering, which precede or correspond to when mean daily air temperatures increased above melting. When snow cover persisted beyond the threshold date, leaf expansion (greening) occurred 9 days after snowmelt and the first flowers opened 14 days after snowmelt (Fig. 2).

The results of our study are consistent with others in demonstrating that air temperatures after snowmelt affect plant phenology (10, 18, 27), however, this study reports a threshold response of phenology to an experimental manipulation in the timing of snowmelt. A threshold response has only been reported in a 33-year study of plant phenology in a subalpine meadow that included data from several dry years when snow inputs were low and snowmelt occurred at least 15 days earlier than most other years (21). The response of phenology to warming-induced early snowmelt has been linear or linearity has been assumed in model fitting (18, 25). Studies that have manipulated snow depth to shift the timing of snowmelt have included multiple treatments or multiple sites as ours did, but have not fit the data to models that relate phenology to the timing of snowmelt (19, 27, 28). Instead, the effects of snowmelt on phenology were inferred from treatment effects through ANOVA. The study by Dunne et al. (2003) is a notable exception, in which they found that the onset of flowering could best be predicted by multiple linear regression models that included snowmelt date and temperature indexes, which we suggest is indicative of a threshold response. The integration of experimental and gradient methods in our study (29) or long-term monitoring studies (21), which both expand the range of dates when plots become snowfree, may be essential to detect this response. By decoupling snowmelt from seasonal patterns of warming, dust-induced early snowmelt may increase the frequency of years in which phenology is delayed after snowmelt.

Since phenology is often controlled by multiple environmental factors (30–32), a threshold or nonlinear response to environmental change should not be unexpected, especially at the community-level. For example, the onset of flowering in an annual grassland occurred earlier if air temperatures were increased experimentally, but this response was dampened by other global environmental changes (30). The reproductive phenology of late-season species was delayed midseason by warming in a temperate grassland, while for early-season species phenology was advanced, creating a midseason gap in phenology at the community level (31). And across a large latitudinal gradient, the effect of warming on phenology was contingent on accumulated chilling days in winter, advancing the onset of greening in cool climates but delaying it in moderate climates (32). As in other systems, the strategies of alpine plants vary, influencing the responsiveness of alpine plant communities to environmental change (33). A natural delay in phenology after snowmelt can occur where snowpack is shallow and when spring air temperatures are relatively low after snowmelt, such as in dry years (10, 21, 23).

However, our data suggest that the threshold response of phenology to the timing of snowmelt may be more characteristic of alpine landscapes where winter and spring dust deposition leads to earlier snowmelt. Although our experiment was only for 1 year, the data encompass much of the historical (pre-20^th century) and current interannual range of when snowmelt occurs. Modeled estimates of when snowmelt would have occurred in the absence of dust place the timing of snowmelt in mid-June to early July (9). Low rates of dust deposition are likely more typical of alpine landscapes in the past before increased human settlement of up-wind arid lands (7). Therefore, later snowmelt in the year of our study due to an above average snowpack and dust removal created conditions more typical of the past, while other treatments were more typical of a landscape where winter and spring dust loads are greater. The timing of the thresholds for greening and flowering in early June precede the historical timing of snowmelt in most years. We expect that a delay in phenology after snowmelt occurs markedly more often now than before increased winter and spring dust deposition.

Our study also demonstrated that climate affected key events in the life history of alpine plants in a similar way. A pattern of delayed phenology was characterized at the community-level for the onset of greening and flowering. Analysis of sequenced phenological events can show which events constrain plant response to climate change (34). Plant growth may constrain reproduction, if reproduction cannot begin until after sufficient new growth has occurred (35, 36). Climatic constraints on leaf expansion may thereby constrain plant reproduction. Community-level analyses in other systems have shown a similar dependence of the timing of reproduction on the timing of growth, but individual species responses can vary (18, 31, 37). Species responses can include hastened, constant, or delayed flowering after phenological events that characterize early season growth (34). We found that flowering was delayed longer than greening, including a later threshold for when snowmelt date began to control the timing of flowering (Fig. 3). Reproductive development in alpine plants may be more conservative than for plant canopy development to protect flowers from freezing temperatures and frost. As environmental conditions change over time, a conservative reproductive life history may minimize damage, whereas for subalpine species frost damage increased when snowmelt occurred earlier (21).

The effects of earlier snowmelt on phenology differed in relation to topography, decreasing the difference in the timing of greening and flowering across the landscape. Our experiment was superimposed on a system where the timing of snowmelt varies with topography and has been affected by increased dust loading. Topography, which affects snow depth and the rate of snowmelt, had a significant effect on the timing of snowmelt (Fig. 1) and influenced when snowmelt occurred in relation to the threshold and the seasonal pattern of warming. On the SE-facing hillslope, all plots were snowfree before or soon after June 12, when mean air temperatures were last <0°C. As a result, phenology was delayed after snowmelt on the SE-facing hillslope, resulting in a similar date for the onset of greening and flowering for all experimental treatments (Fig. 4, Table S2). In contrast, on the NE-facing hillslope, the first plot, a fabric treatment plot, became snowfree on June 13 and the final plot, one where dust was removed, was snowfree on July 15. Thus, earlier snowmelt led to earlier greening and flowering after snowmelt on the NE-facing hillslope (Fig. 4). The differential effect of earlier snowmelt on phenology in relation to topography synchronized the timing of greening and flowering across the tundra landscape. Although phenology was not synchronized in the control plots, only a 1-week shift in the timing of snowmelt (i.e., dust addition plots), which is more typical of recent years with less snow, led to similar flowering times across the hillslopes. Greening times were synchronized when snowmelt occurred ≈2 weeks earlier (i.e., the fabric addition plots). A 1-month shift in the date of snowmelt from winter and spring dust deposition (9) may therefore be sufficient to cause phenological events to occur concurrently across the tundra in certain years.

Fig. 4.

The effects of the snowmelt manipulations and topography on the onset of greening and flowering. Treatments are arranged in order of increasingly earlier snowmelt, which occurs with greater dust loads. Data are means ± 1 SE. P values compare aspects within each treatment (n = 1 or 2).

Synchronization of phenology decreases complementarity across the landscape, which could increase nutrient losses to aquatic ecosystems before greening and alter species interactions. Typically, complementarity is considered at the community level, where variation in the timing of species' life histories can promote coexistence by reducing competition for resources (38, 39). It is also one of the primary mechanisms by which species composition affects nutrient cycling through seasonal asynchrony in resource use (40). Synchronized life histories across a landscape could decrease nutrient retention by reducing temporal variation in nutrient demand among topographic positions. In particular, delayed phenology after snowmelt would postpone plant demand for resources (41), leading to decreased nutrient retention when nutrient availability is high (42). During the growing season, concurrent growth and flowering across the landscape could alter species interactions, increasing competition for limiting resources and pollinators and changing landscape-scale gene flow via pollination. Decreased spatial variation in plant phenology can also reduce foraging success by large herbivores with consequences for offspring production (43). Thus, the atmospheric transport of desert dust is a process that links human activities in desert ecosystems (7) to changes in phenology in alpine landscapes, which could affect biotic interactions and nutrient cycling by synchronizing phenology across the tundra.

Methods

Study Site.

The study was conducted in the 290 ha Senator Beck Basin Study Area at Red Mountain Pass (SBBSA), operated by the Center for Snow and Avalanche Studies (CSAS). SBBSA lies at coordinates N 37° 54′ and W 107° 43′ in the San Juan Mountains of southwest Colorado. The basin has a generally east-facing orientation and ranges in elevation from 3,362 to 4,118 m (11,040 to 13,510 ft). Water year (WY, Oct. 1–Sept. 30) precipitation totals measured in the SBBSA in recent years have been 1,090 mm (2004), 1,180 mm (2005), 1,214 mm (2006), 1,233 mm (2007), and were 1,238 mm in WY 2008, the year of our study. Seasonal maximum snow accumulations in SBBSA generally occur during April. Maximum rates of snow melt in SBBSA occur in May. Mean air temperatures measured in SBBSA during recent years were −9.1 °C in winter (Dec.–Feb.) and 8.3 °C in summer (June–Aug.). Vegetation in the upper two-thirds of the basin is alpine tundra. Climate data are from a climate station in the basin at 3,718 m that is maintained by the CSAS.

Experimental Design.

The 8 × 12 m experimental plots were located on gently sloping SE-facing and NE-facing hillslopes at 3,720 m and 3,767 m, respectively. To increase energy absorption by the snowpack and shift snowmelt earlier, we darkened the snow surface with black shade fabric and artificially applied dust on May 3 on the SE hillslope and on May 4th (DOY = 125) on the NE hillslope. We added approximately 50 g/m² of dust, a load well in excess of the 5–10 g/m² natural dust loading rates observed during winter and spring in the past 4 years, to ensure snowmelt occurred earlier. Ground local rock from SBBSA was used for our dust additions. At this time, we also marked the areas for the control and dust removal plots. Plots were approximately 10 m apart and were placed midslope to minimize the effects of neighboring plots and upslope areas.

Snowfall subsequent to the installation of the fabric and added dust manipulations resulted in the natural dust layers being buried for most of May and delayed the removal of the natural dust (NE plots only) until it had reemerged at the snowpack surface in early June, just as the fabric and added dust manipulations were also reemerging. When scraping the natural dust from the surface of the snowpack, only 1–3% of the snowpack or ≈5 cm of snow was removed. Compression of the snow was minimal, because the snow was dense and travel on the plot was on skis over surfaces where snow had been removed. A total of 13 plots was established, 6 on the SE hillslope (2 each of all but the dust removal treatment) and 7 on the NE hillslope (1 fabric plot and 2 each of all other treatments). Dust removal on the SE hillslope was not possible, because dust layers were not exposed long enough before snowmelt. A 1-m² intensive monitoring plot, where we observed plant phenology, was established in each experimental plot. These monitoring plots were randomly located within the lower one-third of each plot and were at least 2 m from the edges of the 8 × 12 m area. High rock cover and a steep slope prevented placement of a monitoring plot in one of the dust addition plots on the SE hillslope.

Phenological Observations.

We initiated observations of the date of leaf emergence, the first leaf fully expanded, and the first flower fully opened on June 1 when the first plots became snowfree. Observations were, consistently, made every 2–3 days until Aug. 11. We spent no more than 3 min scanning the 1-m² plots for the occurrence of these events for each of 12 common alpine species. The species observed included: Acomastylis rossii, Deschampsia caespitosa, Artemesia scopulorum, Sibbaldia procumbens, Saxifragra rhomboidea (in 3 plots, we observed the phenology for this species within 1 m of the monitoring plot), Bistorta bistortoides, Castelljia occidentalis, Potentilla diversifolia, Caltha leptosepala, Trollius laxus, Rydbergia grandiflora, and Carex scopulorum. Due to the patchy presence of plant species in natural plant communities, all species were not present in all plots. At least 5 and up to 9 species were observed in each monitoring plot, and the first 5 species listed were common to all plots. All but 2 species, D. caespitosa and C. scopulorum, are forbs. Observations of multiple species within each monitoring plot were used to characterize phenology at the community level.

Data Analysis.

Data were analyzed in SAS version 9.1 (SAS Institute). The effect of snowmelt manipulations and topography on the timing of snowmelt and plant phenology were analyzed by a 2-way ANOVA in PROC MIXED. Phenological data were analyzed at the community level. Plot was included as a random effect. Models of phenological response to the timing of snowmelt, included a constant, linear and threshold (i.e., a piece-wise linear) model as follows: Constant model: y = DOY_b; linear model: y = s * DOY_sf + DOY_b; threshold model: if DOY_sf < DOY_sm, then y = DOY_b; if DOY_sf ≥ DOY_sm, then y = s * (DOY_sf – DOY_sm) + DOY_b, where DOY_b is the DOY the phenological event first began, s is the slope, DOY_sf is the DOY a site was snowfree and is the independent variable, and DOY_sm is the DOY snowmelt began to control the timing of a phenological event. The constant, linear, and threshold models, thus have 1, 2, or 3 parameters (italicized), respectively. Models were fit to the data through maximizing the log likelihood in PROC NLMIXED, which then reports the AIC_c value (AIC_c adjusts the AIC value for small sample sizes). The weight of support (w_r) for each model was calculated from the AIC_c values and was used to select the best model (i.e., greatest w_r) (24). Maximum likelihood estimates of model parameters and confidence limits were determined through model fitting. Thaw degree days and cumulative solar irradiance were calculated from June 1, when the first plots were snowfree.