Environmental and Human Influences on Trumpeter Swan Habitat Occupancy in Alaska

Abstract.

Approximately 70–80% of the entire population of the Trumpeter Swan (Cygnus buccinator) depends for reproduction on wetlands in Alaska. This makes the identification of important habitat features and the effects of human interactions important for the species' long-term management. We analyzed the swan's habitat preferences in five areas throughout the state and found that swan broods occupied some wetland types, especially larger closed-basin wetlands such as lakes and ponds, at rates much higher than they occupied other wetland types, such as shrubby or forested wetlands. We also found a negative effect of transportation infrastructure on occupancy by broods in and around the Minto Flats State Game Refuge, Kenai National Wildlife Refuge, and Tetlin National Wildlife Refuge. This finding is of particular interest because much of the Minto Flats refuge has recently been licensed for oil and gas exploration and parts of the Kenai refuge have been developed in the past. We also investigated the potential effects of the shrinkage of closed-basin ponds on habitat occupancy by nesting Trumpeter Swans. We compared nesting swans' use of ponds with changes in the ponds' size and other characteristics from 1982 to 1996 and found no relationships between occupancy and changes in pond size. However, we believe that the recent and rapid growth of Trumpeter Swan populations in Alaska may become limited by available breeding habitat, and anthropogenic and climate-induced changes to the swan's breeding habitats have the potential to limit future production.

Resumen.

Aproximadamente el 70–80% de la población total de Cygnus buccinator depende de los humedales en Alaska para la reproducción. Para el manejo a largo plazo de la especie, esto hace que sea importante identificar las principales caracteristicas ambientales y los efectos de las interacciones humanas. Analizamos las preferencias de habitat de C. buccinator en cinco areas a través del estado y encontramos que las camadas ocuparon algunos tipos de humedales, especialmente los de tipo cerrado de mayor tamaño como lagos y estanques, a tasas mucho más altas que otros tipos de humedales como los arbustivos o arbolados. También encontramos un efecto negativo de la infraestructura de transporte en la ocupación por camadas en o alrededor del Refugio Estatal de Fauna Minto Flats, del Refugio Nacional de Vida Silvestre Kenai y del Refugio Nacional de Vida Silvestre Tetlin. Este hallazgo es de particular interés porque gran parte del refugio Minto Flats ha sido recientemente habilitado para la exploración de petroleo y gas y partes del refugio Kenai han sido desarrolladas en el pasado. También investigamos los efectos potenciales de la reducción de los estanques sobre la ocupación de habitat por C. buccinator para anidar. Analizamos como los cambios en el tamaño y otras características de los estanques determinan su uso por parte de individuos anidando de C. buccinator desde 1982 a 1996, y no encontramos relación entre la ocupación y los cambios en el tamaño de los estanques. Sin embargo, creemos que el crecimiento rápide y reciente de las poblaciones de C. buccinator en Alaska puede verse limitado por la disponibilidad de habitat para anidar, y los cambios inducidos por el hombre y el clima sobre los ambientes de cria de C. buccinator tienen el potencial de limitar la producción futura.

Introduction

By the early 1900s the Trumpeter Swan (Cygnus buccinator) was nearly extinct, with only a few small populations remaining in Yellowstone National Park and at Red Rock Lakes in Centennial Valley, Montana (Bellrose 1980). In the 1950s, small numbers of Trumpeter Swans were identified in Alaska (Hansen et al. 1971), and monitoring began. In 1968, swans were surveyed across all known breeding habitats in Alaska (Caithamer 2001), and since then populations have increased substantially (Conant et al. 2002; Schmidt et al., in press).

Although numbers of the Trumpeter Swan in Alaska have increased, and continental populations have begun to recover, much of the species' ecology is still unknown, and habitat changes (natural and anthropogenic) may threaten its persistence range-wide. Its general habitat preferences are poorly described, and factors such as development and climate change may reduce available habitat. Earlier studies have identified some of the basic habitat features important for the swan's reproduction (Hansen et al. 1971, Mitchell 1994), but a larger-scale assessment of habitat occupancy by breeding swans could provide additional insight into habitat suitability. Identification of the most important habitat features may help managers anticipate the effects of climate warming and human interference on this species. The Trumpeter Swan's breeding areas in Alaska are particularly important to identify and protect because most of the population breeds there and changes in Alaska have implications for the worldwide population.

One important consideration is the amount and degree of human development and recreational activity in and around breeding areas during the reproductive season. Swans are sensitive to human disturbance (Hansen et al. 1971, Henson and Grant 1991), and increasing the numbers of roads, pipelines, and other types of infrastructure could reduce the suitability of habitats. These birds are strongly territorial and seldom breed in high densities (Hansen et al. 1971, Bellrose 1980), so if current breeding areas are degraded through development, fewer swans will be able to breed successfully in a given year. This possibility has important implications for areas such as the Minto Flats State Game Refuge (MFSGR), where large numbers of Trumpeter Swans breed annually. Most of this area has been licensed for oil and gas exploration (Alaska Department of Natural Resources 2002), and the associated increase in human activity could reduce the suitability of large portions of this important area if swans respond negatively to these disturbances.

Environmental variables may also play a role in the swans' occupancy of breeding areas. Wildfires are quite common in interior Alaska and could change drainage, food availability, or cover in wetlands (Schindler et al. 1996). Rising temperatures in Alaska (Hartmann and Wendler 2005, Sagarin and Micheli 2001, Stafford et al. 2000) may increase the number of ice-free days, allowing swans to breed in areas where they have not bred previously (Schmidt 2008), at higher altitudes or latitudes. The length of the breeding season is also an important factor in determining habitat suitability (Hansen et al. 1971), and we expected that spring conditions, such as temperature and precipitation, might explain some variation in habitat occupancy.

Another aspect of climate change that may influence Trumpeter Swans is the reduction and loss of ponds from drying. Much of Alaska has been experiencing a reduction in the number and extent of closed-basin ponds (Klein et al. 2005, Riordan et al. 2006), and over the long term this loss could result in a permanent reduction of breeding habitat for many birds, including the Trumpeter Swan. Others have suggested that these changes in climate could affect populations of breeding waterfowl by altering or eliminating ponds (Sorenson et al. 1998, Murphy-Klassen et al. 2005, Forcey et al. 2007). Pond shrinkage could have some positive effects in the short term, such as increasing available foraging areas as ponds become shallower. Depending on the magnitude of these effects and the extent of change, populations of breeding swans and many other wetland obligates could be affected over a majority of their range.

To investigate the relationships between habitat and occupancy by breeding swans, we conducted two separate analyses. The first was a general habitat analysis that related occupancy of wetlands of all types [as classified by the U.S. Fish and Wildlife Service (USFWS) National Wetlands Inventory (NWI)] by Trumpeter Swan broods in five regions throughout Alaska to the anthropogenic and ecological factors discussed above. The second combined historical imagery with data on nesting swans to investigate variation in occupancy of closedbasin wetlands (i.e., ponds) relative to changes in pond area. Our goals were to describe the preferred habitats of breeding swans and to identify how habitat changes may affect the future abundance of Trumpeter Swans in Alaska.

Methods

Habitat

We surveyed five regions throughout Alaska—the Cordova area, areas in and around Tetlin National Wildlife Refuge (TNWR), the core MFSGR, the greater MFSGR area, and the Kenai National Wildlife Refuge (KNWR) (Fig. 1)—for portions of the period 1957–2005. Survey frequency depended on the region. For the MFSGR area we undertook two separate analyses to explore differences between the core area of the MFSGR and the greater area around and including the MFSGR because habitat and survey frequency in the two areas differed. More frequent surveys of the core area allowed us to examine its habitat characteristics more closely. Fall (late August/early September) brood surveys were conducted in 28 years (1978–2005) for Cordova, 11 years (1985–1995) for TNWR, 16 years (1980–1986, 1988, 1990, 1991, 1994–1996, 2000, 2003, 2005) for MFSGR, 8 years (1980, 1985, 1988, 1990, 1994, 1995, 2000, 2005) for greater MFSGR, and 43 years (1957–1975, 1978–1980, 1982–1987, 1989–2002, 2005) for KNWR. We surveyed the areas from the air by using a modified strip-transect design, with transects ∼1.6 km apart. We recorded the numbers of swans observed within 800 m of either side of the aircraft. Survey units were based on 1:63 360 quadrangle maps of areas containing potential swan-breeding habitat. If necessary, when a swan was sighted, the plane circled so the birds could be counted and the location could be recorded accurately. Our goal was to cover all potential habitats completely. We recorded information either directly onto a topographic map or directly into a laptop computer through the use of a touch-screen mounted in the cockpit (Schmidt et al., in press). We used these survey data to examine relationships between habitat occupancy by swans and human-related and environmental variables. An occupied wetland was defined as a wetland containing at least one brood. Each polygon in the NWI database that was not labeled as “upland” was considered to be a wetland.

We considered occupancy data for all surveyed wetlands to be true presence/absence data (i.e., detection probability = 1) because swans are highly detectable and each area was fully covered. In most situations, before occupancy can be estimated, multiple surveys are needed for detection rates to be estimated because the assumptions of constant and high rates of detection through time are not usually met by singlesample surveys (MacKenzie et al. 2006). We did not have the option of collecting multiple samples in this case, and it is likely that some broods were missed during each survey, but we believe that the number missed is small and constant enough for wetlands where a brood was not sighted to be treated as unoccupied for that year. Swans are highly visible from the air, and differences in detection rates from observer to observer do not appear to be large (Schmidt et al., in press). Imperfect detection would bias our occupancy estimates low.

We explored variation in the probability of occupancy relative to wetland size, wetland type, year, elevation, fire history, mean temperature and snow depth in April, summer precipitation, presence of transportation infrastructure (roads, railroads, airstrips), and presence of oil/gas wells. We also considered a possible interaction between wetland size and type to assess whether size effects vary by type. We determined each wetland's type and area from wetlandclassification maps produced by the USFWS NWI. We used the wetland classification to assign each wetland to one of nine major categories: lacustrine, palustrine unconsolidated bottom (PUB), palustrine aquatic bed (PAB), palustrine emergent, palustrine scrub shrub, palustrine unconsolidated shore, palustrine forested, estuarine, and riverine (not all regions contained all types). We obtained a historical fire map from the Bureau of Land Management Alaska Fire Service (http://agdc.usgs.gov/data/blm/fire) to identify the boundaries of most of the large wildfires since 1950. To provide a comparison between burned and unburned habitats, we considered wetlands located within a fire perimeter to have been burned. We determined the average elevation of each wetland with a digital elevation model of the state. We included the April mean temperature and snow depth as indices of the spring weather conditions. We also included the amount of summer precipitation because we expected precipitation to affect habitat suitability or brood survival. We acquired weather data for each area from the weather stations of the National Climatic Data Center. Weather stations were either within the study regions or immediately adjacent to them. Finally, we used data produced by the Alaska Department of Natural Resources (AKDNR) Division of Oil and Gas to identify the location of oil and gas wells in our study area and the timing of their construction. This information, combined with a GIS coverage of transportation infrastructure produced by the AKDNR's Land Records Information Section, allowed us to determine if a wetland was within 402 m of either potential source of human disturbance during a given year. Transportation infrastructure was added through time in all areas except TNWR, and proximity was not limited to any specific wetland type. We used a buffer of 402 m because the AKDNR, Division of Oil and Gas Nenana Basin Exploration License: Final Finding of the Director, states in its mitigation section that no permanent facilities, including roads, may be located within 1/4 mile (402 m) of known Trumpeter Swan nesting sites (AKDNR 2002). Because this distance is used in current regulations, we thought it appropriate for assessing the effects of infrastructure on swans' habitat use during the breeding season.

Pond Shrinkage

In the Cordova region, from 1978 to 2005, nests were surveyed in spring by same method used for broods. We considered using data on broods, but their movements introduced uncertainties in associating a specific pond with a specific brood. Ponds with swan nests could be identified easily, and we expected these areas to provide similar information because swans often use them for brood-rearing as well. Infrared imagery at sufficiently fine resolution was available for the Cordova region during the survey period, and it provided information on changes in pond size. To avoid complications associated with annual variability in water levels and swan occupancy, we chose only years with data from both nest surveys and habitat imagery. This selection resulted in two years with corresponding nest and imagery data, 1982 and 1996. Using the survey data, we identified ponds (n = 79) associated with a nest in either of these years. We then digitized margins of open water in each year with ArcGIS 9.1 to estimate any change in pond area between the two periods. To provide a comparative sample of unoccupied ponds from across the landscape, we randomly selected 79 additional ponds that we considered to be potential swan habitat but were not occupied during either year.

Statistical Analyses

Habitat. We used the program R 2.6.1 (R Development Core Team 2006) and the package Ime4 (Bates 2007) to fit generalized linear mixed models to the occupancy data from all five regions. This package uses Laplace's approximation method (see Bates 2007) to provide an approximate solution to complex likelihoods containing random effects that cannot be maximized easily. To constrain occupancy probabilities between 0 and 1 we used the logit-link function, a common choice for binomial data (Williams et al. 2002). We began with simple models, then added additional parameters successively according to our a priori ranking of their potential importance. On the basis of prior studies (Hansen et al. 1971) we believed that combinations of year, habitat type, habitat size, interaction between type and size, and disturbance from transportation would be the most important variables and considered them first. Next we considered the effect of elevation and an interaction between elevation and year to investigate perceived changes in the use of higher-elevation wetlands during later years (J. King, USFWS, pers. comm.). The remaining variables, burned/unburned, oil wells (mostly unused during the study), and weather, were expected to have less influence on swans' habitat use, so we added them last. We used Akaike's information criterion (AIC) to select among competing models (Burnham and Andersen 2002), and if a parameter did not improve the model based on AIC comparisons, we removed it before adding additional parameters. After considering all fixed effects, we added random effects of individual wetland, year, and an individual trend for each wetland to help explain any remaining variation in the data. These random terms could help explain additional differences in habitat quality, differences by year, and allow trends to vary by site, respectively. Including a site-specific trend allows the trend to vary by site and helps account for changes in suitability at the site level through time that would not be addressed with only a fixed-effect trend parameter. We again used AIC to determine whether the additional complexity due to the random effects was justified. Parameter estimates are presented as means ± SE.

Pond shrinkage. We used R to fit generalized linear occupancy models relating swan occupancy to changes in pond size. We believed that for identifying responses to changes in pond size occupancy by nesting swans is a better choice than broods because nests can be more definitively assigned to a single pond. The models contained combinations of the following independent variables: differences in pond size between survey years, pond size in both years, and rate of change between the two years. We again used AIC to select among competing models (n = 5).

Results

Habitat

The Cordova region was surveyed annually from 1978 to 2005. The area covered included 11 quad maps: Cordova A1_, A2, B1-B5, and C2-C5. During this period, 1326 wetland occupancies by swans were observed. The TNWR area, covering 9 quad maps (Nabesna B1, B2, and Cl and Tanancross A2-A4 and B4-B6) was surveyed annually from 1985 to 1995, and 207 wetland occupancies were recorded. Portions of the MFSGR and surrounding areas were surveyed from 1980 to 2005. The core area included 6 quad maps covering most of the MFSGR: Fairbanks C4, C5, D4, and D5 and Livengood A4 and A5. This core area was surveyed for 16 years of the study period, and 1830 wetland occupancies were observed. Surveys of MFSGR and the surrounding area included the quad maps from the core analysis as well as Fairbanks B1-B4, C2, C3, and D6 and Livengood B4. Broods were surveyed over this entire area 8 times during the study period, and 1287 wetland occupancies were observed. Portions of the KNWR, including quad maps Kenai B1-B4, C1-C4, and D1-D4 and Tyonek Al and A2 (Fig. 1) were surveyed in 43 of the years from 1957 to 2005.

Model-selection results for each of the five areas showed many similarities. For each, the best approximating model included similar combinations of year, wetland size and type, proximity to transportation infrastructure, elevation, trend through time, and random effects of individual wetland and rate of change in occupancy through time (Table 1). Models for all areas except the greater MFSGR and KNWR included random-effects terms. Calculations of random-effects terms for these areas were not possible because sample sizes were extremely large. Cordova was the only region where an influence of one or more spring-weather variables was not supported. Precipitation during the breeding season did not appear to affect habitat occupancy in any of the areas (Tables land 2).

Of the nine habitat types identified, in all study areas swans were most likely to use larger lacustrine, palustrine aquatic bed (PAB), and palustrine unconsolidated bottom (PUB) habitats; they used forested and riverine wetland habitats much less often (Table 2). Wetland size generally had a positive effect on occupancy, but the effect varied in magnitude by type. In all five areas the additional adjustment for wetland area had a much larger positive effect on occupancy for PAB and PUB habitats than for other wetland types (Table 2), and occupancy probability increased dramatically as wetland size increased (Fig. 2). Wetlands at higher elevations had lower probabilities of occupancy in all five areas (core MFSGR, β_elve = -0.035 ± 0.006; greater MFSGR area, β_elev = -0.014 ± 0.003; TNWR, β_elev =-0.017 ± 0.007; Cordova, β_elve= -0.011 ± 0.002; KNWR, β_elve= -0.015 ± 0.002), and there is some evidence that this relationship was changing through time, but the magnitude and direction are uncertain (Table 2). There were no fires in the Cordova area, but for three of the four remaining areas fire was positively associated with occupancy probability (core MFSGR, β_fire = 0.120 ± 0.068; greater MFSGR area, β_fire = 0.291 ± 0.073; TNWR, β_fire = 0.489 ± 0.266), although the estimates were imprecise. The presence of transportation infrastructure within 402 m of wetlands had a negative effect on the probability of occupancy (Fig. 2) in the core MFSGR (β_trans = -1.13 ± 0.35), the greater MFSGR area (β_trans= -1.32 ± 0.44),TNWR (β_trans = -1.84±0.76), and KNWR(β_trans=-0.51± 0.11), while there was a positive effect in the Cordova area (β_trans= 0.85 ± 0.10). The inclusion of a covariate representing the presence of oil wells (a majority were plugged and abandoned) was not supported in the Cordova area (AIC increased by 2 units), but there was a small positive effect of the presence (β_wellpres = 0.03 ± 0.02) and number of wells (β_wellnum = 0.02 ± 0.01) on occupancy in the KNWR (Table 2).

Pond Shrinkage

Change in pond size from 1982 to 1996 was variable, as both decreases and increases were common. On average, however, the ponds in our sample decreased by 0.32 ha (SD = 1.73) from 1982 to 1996. The average wetland covered 10.24 ha, but wetlands' sizes were highly variable (SD = 30.13). The only parameter the model selection supported was an intercept adjustment for 1996 (3.1 AIC units lower than the constant model), and it indicated that occupancy probability was greater (β₁₉₉₆ = 0.565, SE = 0.253) in that year. There was little support for an effect of the pond size (ΔAIC = 3.3) or change (ΔAIC = 4.2) on the probability of nesting Trumpeter Swans occupying a pond.

Discussion

Habitat

Our results confirmed several habitat associations recognized by past research and identified additional factors affecting wetland occupancy by Trumpeter Swan broods in five important breeding areas in Alaska. We found that swans occupied some wetland types, larger lakes and ponds (PAB and PUB types) in particular, at higher rates than others. The probability of swans' occupying wetlands within historical fire perimeters was also increased, which could be important because of the recent large fires in many areas of Alaska. Spring weather conditions were important predictors of broods' habitat occupancy, probably indicating negative effects of cold weather early in the breeding season on breeding probability and brood survival. The effects of higher elevations were similar to the weather effects and appeared to change in later years, possibly indicating a response to climate change lengthening the season. Finally, one of the more important findings was a negative influence of transportation infrastructure on wetland occupancy by broods. This relationship could have implications for management decisions involving future development.

The higher probabilities of swans' occupying some bodies of water such as lakes and PAB and PUB habitats (the two most common types of closed-basin ponds) over other wetland types was not surprising because swans feed primarily on submerged aquatic and emergent plants (Mitchell 1994). These plants are more abundant in these habitats than in forested or shrub-dominated wetlands. These results agree with those of others who have previously identified lakes and ponds as the swans' important breeding areas (e.g., Hansen et al. 1971). We also found that the probability of occupancy of both PAB and PUB habitats increased dramatically as their size increased, indicating that larger ponds of these types are particularly important for brood rearing. Earnst and Rothe (2004) identified a similar positive effect of pond size on occupancy by the Tundra Swan (Cygnus columbianus). This finding suggests that priority should be given to lake and ponds, especially larger ones, when management or development in areas where swans breed is considered.

The effects of spring weather on breeding swans were similar to those found for other large waterfowl (e.g., Bluhm 1992, Lindberg et al. 1997). The positive effect of April temperature and the negative effect of April snow depth are likely related to breeding probability and cygnet survival rates until the fall survey period. Others have shown that spring conditions can delay nesting in other waterfowl (Johnson et al. 1992), and these conditions could be particularly important for Trumpeter Swans, which in some areas are limited by the number of ice-free days in the season (Hansen et al. 1971, Schmidt 2008). In years with cold springs, some pairs may forgo breeding in response to the shortened season. This failure to breed would be reflected in lower occupancy rates in the fall brood-survey data. The lack of these effects in the Cordova region probably reflects the longer breeding season and the moderating influence of the ocean on air temperature in the area. Elevation likely had a negative effect for similar reasons; higher elevations have shorter nesting seasons and colder weather. A change in this relationship through time was indicated by the support for an interaction between elevation and year, suggesting that changes observed in the arctic climate (Hinzman et al. 2005) may be affecting habitat suitability; however, a small sample of wetlands at higher altitudes resulted in imprecise estimates. Most of the habitats studied are in lowlands with little variation in elevation. If climate warming continues, however, we would expect the occupancy rates of high-elevation wetlands to increase.

A positive association between occupancy and fire history for the two MFSGR areas and TNWR suggests that wildfires may improve the quality of brood-rearing habitats in some areas. Carignan et al. (2000) have shown that boreal lakes have substantially higher levels of phosphorous, nitrogen, and potassium after fires, thus wetlands in and around burned areas may be more productive as a result of increased nutrients, which may improve the quality and quantity of the forage for both adults and cygnets. The frequency of extreme fires in Alaska has increased (Soja et al. 2007), and this change could improve the long-term suitability of some wetland habitats for swans in the state.

Unlike Hansen et al. (1971), we did not find a negative effect of oil wells on swan occupancy in the Cordova region, but we caution the interpretation of this result because most of the wells in the area were not in use during the study period. The slight positive effect of wells in the KNWR was somewhat unexpected but might be explained by a reduction in human visitation to those areas. Much of the KNWR is designated wilderness and is used for many forms of nonmotorized recreation. We suspect that the presence of oil wells in this situation might actually decrease disturbance because people avoid those areas; however, we did not have appropriate data to determine the cause of this relationship.

On the basis of previous research (Hansen et al. 1971, Henson and Grant 1991), we expected roads, railroads, and airstrips to decrease habitat use because of increased noise and direct interference associated with increased access for recreation and other human activities. The Cordova area supported a small positive effect, but we suspect this could be due to much of the infrastructure getting little or no use. This area has a low human population and is not connected to the rest of Alaska's road system, so disturbance levels are likely to be much lower than in the other areas. The negative effects of actively used transportation infrastructure in the other four areas support the findings of past studies (Hansen et al. 1971, Henson and Grant 1991) and are of special concern for the MFSGR because most of the area has been licensed for oil and gas exploration (AKDNR 2002). A network of roads and pipelines would likely increase access and make large portions of this important area less suitable for breeding swans. This area is also important for the breeding of many other species of waterfowl which could be similarly affected. To address these concerns, additional monitoring in any newly developed areas within the MFSGR may be required for the effects of these disturbances to be evaluated.

The negative associations between swans and human infrastructure indicate that mitigation measures used by management agencies such as the AKDNR may not protect populations of breeding swans from development adequately. An AKDNR report on the Nenana Basin exploration license, including much of the MFSGR, states that “permanent facilities, including roads, material sites, storage areas, powerlines, and aboveground pipelines may be prohibited within 1/4-mile of known nesting sites” in the area (2002). We believe that brood-rearing habitats should be included in this prohibition as well because our observations show negative effects of this new infrastructure on brood-habitat occupancy. As shown in Figure 2 and Table 2, the presence of infrastructure can reduce the probability of occupancy for some wetlands substantially. Because of the large number of current and historic brood-rearing areas in the MFSGR the building of this infrastructure could reduce the swans' habitat use and productivity substantially. We suggest that future development in potential breeding areas be restricted to at least 402 m from all lakes and ponds, not nest sites alone, if minimizing effects on Trumpeter Swans is considered important.

Our occupancy estimates should be considered conservative because small errors associated with the NWI classifications, discrepancies between swan locations and the topographic maps, and variable movements of swan broods can result in smaller wetlands being “unoccupied” during a survey. NWI classifications were created from one or two years' imagery, and annual variation in water levels could have resulted in a brood being associated with a nearby habitat type during the surveys. In our analysis, swan locations were not buffered by habitat type to prevent an arbitrary bias for one wetland type over another. Because of the small size of many wetlands, especially ponds, in some cases small amounts of error probably resulted in broods being associated with the surrounding wetland types. For this reason, we believe that closed-basin wetlands (lakes and ponds) are probably even more important than our results indicate.

Pond Shrinkage

We did not detect any differences in habitat occupancy by Trumpeter Swans relative to variation in pond size, despite the change in the size of many ponds in our sample from 1982 to 1996. We originally suspected that these changes would affect swans breeding in our study area, but, on the basis of the available data, this does not appear to be the case. Our results suggest that swans may not respond to changes in pond size or that the level of change we observed was not great enough to elicit a response.

Because previous habitat analysis indicated wetland size was an important predictor of occupancy by broods, it seemed reasonable to expect that shrinking pond size would affect occupancy negatively. Johnson et al. (2005) also suggested that climate warming could have negative consequences, including wetland shrinkage, for waterfowl on a large scale. If swans are affected also, understanding their response would be important for their future management in Alaska. Some ponds we studied shrank substantially, but others grew, and neither scenario appeared to affect swan occupancy. Thus swans may not respond to pond size alone, and other factors may be more important in the selection of nesting habitats. It is also possible that most of the ponds in the sample were large enough to prevent any size-dependent response. There could be some minimum pond size necessary to provide adequate room for feeding and take-off, and if the ponds sampled did not shrink below that minimum, then reductions in occupancy would not be expected.

Overall, we believe that the lack of evidence for a response to pond shrinkage by breeding Trumpeter Swans should be viewed with caution because of the potentially large reductions in pond size and number as climate warming continues (Klein et al. 2005, Riordan et al. 2006). Our sample was relatively small, and it is possible that over longer periods larger changes in size might be observed. In some areas of Alaska (e.g., Riordan et al. 2006), some bodies of water have completely disappeared over longer periods, and changes of this degree would have obvious effects on habitat occupancy by breeding swans. The time scale for which we have data also may not be sufficient to detect responses at the population level. In the previous analysis, habitat occupancy by broods was found to vary from year to year, suggesting that some unexplained annual variation may be influencing swans' use of individual ponds. This additional annual variability would make it more difficult to identify the effects of a reduction in pond size if present. The metric we used, nesting swans, also may be too coarse for the levels of change that are occurring. Much larger changes may be needed to make a pond completely unsuitable for nesting, but there is some evidence that other waterfowl experience reduced clutch sizes or lowered survival of nests or young when habitat quality is diminished (Johnson et al. 1992). Larger samples or additional years of both survey and imagery data will likely be necessary to accurately identify the Trumpeter Swan's long-term responses to climate change.

Acknowledgments

Support for the first author was provided by the Alaska Department of Fish and Game, State Wildlife Grant, the Alaska EPScOR program, and the University of Alaska, Fairbanks, Department of Biology and Wildlife and Institute of Arctic Biology. USFWS Migratory Birds Management Division collected and graciously provided most of the data used in this manuscript. Staff at the Kenai NWR, J. Morton in particular, helped us investigate long-term data from that area. We thank D. Verbyla for comments on previous versions of this manuscript.

Literature Cited