Resource quality and availability are strongly linked to offspring production. Grazing lawns (which are patches of grazing‐tolerant plants with high nutrient value) provide important resources for adult herbivores and their young (McNaughton ). Grazing by herbivores can enhance both the nitrogen content of food available and accelerate the primary production of grazed vegetation (McNaughton , Hik and Jefferies ). The importance of grazing lawns to herbivore populations is widely recognized because through resource regulation and repeated grazing, these high‐quality areas are maintained, thereby benefiting future generations of both plants and herbivores (Craig et al. ). Factors leading to a decrease in grazing, such as low herbivore abundance, can lead to grazing lawns reverting to taller stature plants that are relatively nutrient poor (Augustine and McNaughton , Person et al. ). Because grazing lawns benefit from a positive feedback wherein herbivores maintain high‐quality resources, any changes in grazing lawn abundance across a landscape can influence plant and herbivore population dynamics. Assessing the impacts of herbivores on forage at the landscape scale can be challenging, where researchers require long‐term data on both plants and herbivores, and vice versa.
The Yukon–Kuskokwim Delta (YKD) is one of the world's most important breeding areas for geese (Spencer et al. ). Virtually, the entire world population of emperor geese (Anser canagicus; Schmutz et al. ), a large proportion of the Pacific black brant population (Branta bernicla nigricans; hereafter brant; Sedinger et al. ), all the smallest cackling goose subspecies (Branta hutchinsii minima; King and Lensink ), and most of the Pacific Flyway population of greater white‐fronted geese (Anser albifrons frontalis; B. Conant and D. Groves, unpublished manuscript) breed there. However, populations of colony‐nesting brant on the YKD have declined by >50% since 2000 (Wilson ). On the YKD, and across much of the circumpolar arctic, these avian herbivores rely heavily on grazing lawns of a highly nutritious short sedge (Carex subspathacea; Cargill and Jefferies , Person et al. , van der Graaf et al. ). Many factors can lead to changes in grazing lawn abundance including sedimentation, erosion, and indirectly, fox predation (Jorgenson and Ely , Person et al. , Sedinger et al. ). However, moderate‐to‐heavy grazing pressure can facilitate the maintenance of this sedge (Hik et al. , Person et al. ). Grazing lawn abundance has been linked to breeding success of geese (Person et al. , Lake et al. ) and may constrain population growth (Schmutz , Sedinger et al. ). Therefore, the amount of grazing lawn likely defines the carrying capacity across the YKD (Person et al. ). Here, we use aerial imagery sampled from a long‐term dataset (>20 yr; Anthony et al. , Wilson ) to assess potential changes in grazing lawn abundance over time and identify factors influencing this abundance in a coastal subarctic tundra ecosystem. Specifically, we evaluate the role of herbivores, and predators of herbivores, on the temporal patterns of grazing lawn abundance and discuss the impacts of changes in grazing lawn abundance on an avian herbivore population on the YKD at five primary brant colonies. We hypothesized that increased fox populations would negatively impact breeding brant numbers, thereby reducing grazing lawn extent. Our results have important implications for understanding the maintenance of grazing lawns and population dynamics of both plants and herbivores.
We conducted aerial photographic surveys at five primary brant colonies on the YKD from 1993 to 2016: Baird Inlet Island (BI), Baird Inlet Peninsula (BP), Kokechik Bay (KB), Kigigak Island (KI), and Tutakoke River (TR; Fig. ). Straight‐line aerial surveys were flown in early June during incubation, with each transect ~200 m apart, which provided annual estimates of abundance of nesting brant at each colony (for additional aerial survey details, see Anthony et al. , Wilson ). Additionally, grazing lawn abundance can be readily identified through the images captured during aerial surveys (Person et al. , Lake et al. ).
Enlarge this image.Primary Pacific black brant colonies on the Yukon–Kuskokwim Delta, Alaska, where aerial photographic surveys were performed to quantify grazing lawn abundance.
During aerial surveys from 1993 to 2003, still images were captured from digital video (Anthony et al. ). During this time span, imagery from four years (1993, 1994, 1997, and 2001) was selected for analysis based on image quality and availability at all five brant colonies. Digital still images replaced video imagery during aerial surveys beginning in 2004 (Wilson ), and during this time span (2004–2016), we used imagery from 11 yr (2004, 2007–2016). Corrupted or missing images prevented additional years from being included in analyses. Captured images were at least 200 m apart, which we controlled by sampling intervals based on the estimated ground speed of the aircraft used to record the video imagery. No images were captured at KB in 1993 and KI in 1997 because of corrupted video. Ground truthing was conducted on videography, and it was determined that habitat classes could be accurately identified (Lake et al. ). Because of the improvement in imaged quality between digital stills and videography, grazing lawn became easier to identify in images, rendering costly ground truthing unnecessary.
To assess annual variability in amount of grazing lawn, we selected a random sample of images from each colony and year and estimated the proportion of grazing lawn in each image. The number of images captured using videography (1993–2001) was limited by our minimum spacing requirements and amount of video recording (mean available images per colony from 1993 to 2001 = 187). We attempted to sample over 190 images at each colony per year so that sampling intensity was similar across years (Uher‐Koch ). We modified the point‐sample technique used by Lake et al. () to estimate grazing lawn abundance within each image. Lake et al. () used videography data, while we used a combination of videography (pre‐2004) and digital stills (post‐2004). We digitally overlaid 180 systematically distributed white dots (12 × 15 dot grid) on each video and digital still image using MATLAB Image Processing Toolbox (IPT; MathWorks, Natick, Massachusetts, USA). These images were brought into Adobe Photoshop (Adobe Systems, San Jose, California, USA) to assign each dot a different color based on the habitat cover category in which it occurred (Fig. ). Dots were visually assigned to one of six habitat classes (grazing lawn, sedge meadow, slough levee, upland, water, mud, or others; Lake et al. ). The number of dots in each habitat category were then counted automatically using MATLAB IPT. Habitat categories that were not ecologically important for our study (e.g., mud and water) were included so that a similar amount of habitat was categorized across years and colonies. Because the digital still images from post‐2004 covered a much larger area per image than the pre‐2004 video images (~44 times the area; translating to 0.12 vs. 0.01 ha on the ground), we randomly sampled an area within each digital image that was equivalent to the spatial size of the video images (0.01 ha ground‐footprint). These sub‐areas were processed using the same techniques as the video images. Given our spacing between aerial survey transects (~200 m), we were only able to sample a fraction of the landscape (<1% of the habitat at each colony); however, our aerial survey methods allowed us to sample all five colonies in a systematic random fashion.
Enlarge this image.An example aerial image from the Yukon–Kuskokwim Delta, Alaska, showing the 180‐dot grid and habitat classifications used for analysis. Green dots represent grazing lawn (Carex subspathacea), blue dots are water, red dots are mud, and white dots are sedge (Carex ramenskii).
A high proportion of images did not contain any grazing lawn (zeros), and our data were over‐dispersed. Therefore, we used zero‐inflated negative binomial regression (ZINB) to assess models of variation in grazing lawn abundance. Zero‐inflated negative binomial regression models are derived from two distributions where a logistic regression is used to model the zeros (no grazing lawn) and a negative binomial regression is used for the counts (grazing lawn abundance; Zeileis et al. ). In these models, the count of total number of dots of grazing lawn was the response variable and the same effects were modeled on zeros and counts. To correct for varying sampling intensity (i.e., different number of images analyzed per year), the number of total dots analyzed per year was included as an offset variable. We used the pscl package (Zeileis et al. ) in R (R Core Team ) to perform analyses.
We evaluated variation in grazing lawn abundance in relation to several variables. Because herbivores can facilitate the maintenance of grazing lawns, we included two variables representing brant nest population size at each colony (Wilson ). We hypothesized that the number of brant nests would have a lagged effect on grazing lawn abundance (Person et al. ) with low numbers of brant nests in a given year corresponding with low grazing lawn abundance the following year (Nests1) or two years in the future (Nests2). Because of uncertainty as to the potential duration that effects of numbers of nesting brant might have on grazing lawn abundance, we only used two models of nest count data since colony estimates only go back to 1992 and the first year we quantified grazing lawn abundance was in 1993. We used a 3‐yr mean of the number of brant nests at each colony for years without nest survey data (Wilson ).
We predicted that there might be lagged effects of predation on grazing lawn abundance because high numbers of Arctic fox (Vulpes lagopus) can lead to lower numbers of brant nests and decreased productivity (Sedinger et al. ). We used three measures of the yearly proportion of random plots with fox sign (observed fox, scat, fur, tracks, and/or active dens) across the YKD as variables (Fischer et al. ). These variables represented fox abundance in the current year (Fox), the previous year (Fox1), and 2 yr previous (Fox2). We did not include fox abundance and nest abundance variables in the same model since we predicted that they would be related. We also predicted that phenology might influence grazing lawn abundance. If the timing of green‐up and goose hatch are not synchronous, grazing lawn abundance could be reduced due to decreased grazing effort (i.e., lawn reverted to lower forage quality), thereby also decreasing goose productivity (Brook et al. ). We used the mean initiation dates for each year from across the YKD to create the Initiation Date variable (range 16 May–3 June; Fischer et al. ).
Finally, we examined whether changes in grazing lawn abundance were consistent across the YKD or whether they were specific to a certain brant colony which would suggest variation in per capita food availability. We included two‐way interactions between the colonies and other variables (Initiation Date, Fox2, Nests2) to identify whether the influence of those variables differed among colonies. We did not include a year effect since we had multiple years with no grazing lawn and some year–colony combinations were missing (1993 KB and 2007 KI).
To investigate influences on grazing lawn abundance, we evaluated a candidate set of 12 models and included an intercept‐only model (constant). We used an information theoretic approach to quantify and interpret the influences on grazing lawn abundance (Burnham and Anderson ). Models were ranked using Akaike's information criterion adjusted for small sample size (AICc) by comparing ∆AICc scores and Akaike weights (wi).
We analyzed 15 yr of aerial images (mean number of images analyzed per year at each colony = 197, range = 42–239) to quantify changes in grazing lawn abundance (1993, 1994, 1997, 2001, 2004, 2007–2016). Grazing lawn abundance in our study varied from 0.0% to 10.4% (0–99.7 ha per colony) of the sampled landscape across years and colonies. We found higher grazing lawn abundance in the 1990s decreasing to almost no grazing lawn post‐2007 (Fig. ). We found no images containing grazing lawn in 26.0% of year–colony combinations (19/73), and no grazing lawn was identified within sampled images at any colony in 2012. Grazing lawn abundance did not differ among the 5 brant colonies (∆AICc = 20.03, wi = 0.00; Table ).
Enlarge this image.Proportion of grazing lawn habitat (solid line) across all five primary brant colonies on the Yukon–Kuskokwim Delta in western Alaska (95% confidence interval) and the number of brant nests two years prior (dashed line; from Wilson ). Note that not every year was sampled for grazing lawn from 1993 to 2007 and the different axes’ scales and units.
| Model | ∆AICc | w i | K | Deviance |
| Nests2 | 0.00 | 0.81 | 5 | 859.10 |
| Fox2 | 3.80 | 0.12 | 5 | 862.90 |
| Fox1 | 6.25 | 0.04 | 5 | 865.34 |
| Colony + Initiation Date + Fox2 + Nests2 | 8.42 | 0.01 | 17 | 833.28 |
| Intercept only | 9.53 | 0.01 | 3 | 873.18 |
| Initiation date | 9.70 | 0.01 | 5 | 868.80 |
| Nests1 | 12.79 | 0.00 | 5 | 871.88 |
| Fox | 14.06 | 0.00 | 5 | 873.16 |
| Colony | 20.03 | 0.00 | 11 | 863.70 |
| Colony × Initiation date | 27.45 | 0.00 | 21 | 837.32 |
| Colony × Nests2 | 40.30 | 0.00 | 21 | 850.18 |
| Colony × Fox2 | 43.17 | 0.00 | 21 | 853.04 |
Models were ranked based on difference in Akaike's information criterion (∆AICc) and model weight (wi). K is the number of model parameters. The same effects were modeled on the counts (grazing lawn abundance) and zeros (no grazing lawn). The Fox variable represents fox abundance in the current year, Fox1 represents the previous year, and Fox2 represents 2 yr previous. Similarly, the Nests1 variable represents the number of nesting brant the previous year, and Nests2 represents 2 yr previous. AICc for the best supported model = 869.99.
Model selection provided evidence for a 2‐yr lagged effect of nesting goose numbers (Nests2; βnests2 = 1.30, 95% confidence interval [CI]: 0.07, 2.52) on grazing lawn abundance (AICc wi = 0.81, Table ) with higher nesting goose numbers resulting in higher grazing lawn abundance (Fig. ). The 1‐yr lagged effect of goose nest numbers did not affect grazing lawn abundance (∆AICc = 12.79, wi = 0.00). The 2‐yr lagged effect of fox abundance (Fox2; βfox2 = −3.08, 95% CI: −4.92, −1.26) received moderate support from the data (∆AICc = 3.80, wi = 0.12). The third best‐fitting model (∆AICc = 6.25, wi = 0.04) was the 1‐yr lagged effect of fox abundance (Fox1; βfox1 = −4.34, 95% CI: −7.85, −0.84). The mean proportion of plots containing fox sign was 0.48 (range = 0.16–0.89 plots). No other models received even moderate support from the data (Table ), and no variables influenced the presence/absence of grazing lawns (i.e., the zero‐inflated portion of the ZINB models).
This study provides evidence of avian herbivores exerting strong impacts on plant communities and driving vegetation abundance. Our results suggest that herbivore populations can influence the amount of their preferred forage habitat, wherein a lack of continuous grazing pressure (due to low herbivore abundance) can lead to a decrease in grazing lawn abundance. This inference is perhaps counterintuitive because an increase in herbivore populations can lead to a decrease in resources due to density dependence in some ecosystems (Fowler ). In one of the colonies in our study, low nest numbers in the mid‐2000s coincided with low productivity and an almost complete lack of grazing lawn (Sedinger et al. ). If herbivores are not abundant (e.g., due to predation by foxes), then grazing lawn grows into taller stature plants that are relatively nutrient poor (Person et al. ).
Our results suggest that factors influencing herbivore populations would also likely influence grazing lawn abundance. While we did not evaluate the specific mechanism leading to decreases in nesting brant numbers, we also found a relationship between fox numbers and grazing lawn abundance, which is not surprising because nest and fox abundance are likely correlated. Declines in nesting brant numbers have been attributed to fox abundance and predation by Arctic foxes on goose nests (Anthony et al. , Sedinger et al. , ). Our grazing lawn abundance estimates correspond to previous estimates that also found higher abundance during the 1990s (Person et al. , Sedinger et al. ). Surface area of grazing lawns increased on the YKD throughout the 1990s (Person et al. ) and appeared stable between 1999 and 2004 (Lake et al. ). This increase was interpreted as recovery from a period of high fox predation and low nest success in the 1980s (Person et al. ). Changes in the number of nesting brant have also been linked to variation in abundance and availability of their primary food source (eelgrass), during the non‐breeding period (Ward et al. , Sedinger et al. ) and local subsistence harvest (Sedinger et al. ). Factors outside of herbivore numbers can moderate grazing lawns, such as other herbivores (e.g., sympatrically nesting species), sedimentation, soil salinity, and storm surges (Srivastava and Jefferies , Jorgenson and Ely , Hupp et al. ), and future research efforts should be directed at identifying these physical and environmental factors. From a population management perspective, if any of these factors affect the amount of grazing lawn habitat for geese, they would strongly influence breeding success and the maximum size of goose populations (Person et al. ).
Most adult brant undergo a molt migration to higher latitudes (King and Hodges , Lewis et al. , Sedinger et al. ) following predation on their nests. This migration, combined with lower gosling production, can further reduce grazing pressure in years of high fox predation. Gosling growth rates (Sedinger et al. , Fondell et al. , Hupp et al. ) are directly associated with biomass on grazing lawns, and first‐year survival (Sedinger and Chelgren ) and recruitment into the breeding population (Sedinger et al. ) are largely determined by growth rates during the first summer. Consequently, differences in brant population trajectory between the YKD and northern Alaska are at least in part likely a result of differences in grazing lawn abundance, which are in turn influenced by predation (Sedinger et al. , Hupp et al. ).
Although we found an almost complete lack of preferred brood‐rearing habitat in some years of our study, grazing lawns are dynamic and our results suggest that grazing lawn abundance recently increased (e.g., highest abundance in 2016 since 2007). It is unknown how quickly grazing lawns can recover, but we suspect a substantial time lag (up to a decade; Person et al. ) is required to fully restore grazing lawns following catastrophic predation events and low herbivore numbers (Maron and Crone ). Since brant do not reach peak breeding probability until 4 yr of age, it is reasonable to conclude that the number of brant nests from >2 yr ago could influence grazing lawn (Sedinger et al. ). Recent observations suggest large increases in grazing lawn abundance in other areas of the YKD, where failed and non‐breeding brant have been observed (e.g., Punorat Point). These observations combined with our results showing recent slight increases in grazing lawn abundance suggest that unfavorable or deteriorating habitat conditions may not be unidirectional. However, decreasing nest numbers (Wilson ) and low recruitment rates (Sedinger et al. ) indicate that colonial brant populations are declining rapidly in this ecosystem, thereby suggesting that grazing lawn in these areas will continue to disappear if recent predator–herbivore–vegetation dynamics remain unchanged. These data demonstrate the importance of identifying positive feedback loops, such as observed on the YKD, where predation reduces the number of offspring, thereby reducing potential grazing by herbivores and grazing lawn extent, which in turn reduces offspring growth and recruitment, ultimately resulting in fewer nests in the future. These complex tritrophic interactions between predators, herbivores, and plants have important implications for population dynamics (e.g., potential for continued population declines of nesting brant on the YKD). Moreover, continued monitoring of these ecosystem‐scale processes will be useful for further insight, particularly as climate change alters these dynamics (Schedlbauer et al. ).
J. A. Schmutz and T. F. Fondell conceived the idea. B. D. Uher‐Koch, R. M. Anthony, and T. L. Day performed the analyses. H. M. Wilson and R. M. Anthony collected the data. B. T. Person and J. S. Sedinger provided critical context from their many years working in this ecosystem. B. D. Uher‐Koch drafted the manuscript, and all authors read and commented on the manuscript. We thank J. Pearce, T. Riecke, S. Dobson, and two anonymous reviewers who provided helpful comments on earlier versions of this manuscript. We also thank the U.S. Fish and Wildlife Service, Migratory Bird Management Office, and Yukon Delta National Wildlife Refuge for assistance with aerial surveys and covariates. The U.S. Geological Survey Alaska Science Center, Alaska Climate Adaptation Science Center, and Western Alaska Landscape Conservation Cooperative provided funding. Data used in this study are publicly available via the U.S. Geological Survey Alaska Science Center (Uher‐Koch ). Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.
Copyright John Wiley & Sons, Inc. Jun 2019
Details
; Schmutz, Joel A 1 ; Wilson, Heather M 2 ; Anthony, R Michael 1 ; Day, Thomas L 3 ; Fondell, Thomas F 1 ; Person, Brian T 4 ; Sedinger, James S 5 1 U.S. Geological Survey Alaska Science Center, Anchorage, Alaska, USA
2 Migratory Bird Management, U.S. Fish and Wildlife Service, Anchorage, Alaska, USA
3 Institute of Culture and Environment, Alaska Pacific University, Anchorage, Alaska, USA
4 Department of Wildlife Management, North Slope Borough, Barrow, Alaska, USA
5 Department of Natural Resources and Environmental Science, University of Nevada Reno, Reno, Nevada, USA