Translator Disclaimer
1 April 2014 Common Loon (Gavia immer) Nesting Habitat Models for North-Central Minnesota Lakes
Author Affiliations +
Abstract

Shoreline attributes and extensive field surveys of aquatic vegetation and animal presence were used to determine probabilities of Common Loon (Gavia immer) nesting for segments of lakeshore on 35 lakes in north-central Minnesota. Model development used both a general linear mixed model and random forest classifier approach. The resulting nesting habitat models were used to predict nesting sites for a small set of independent lakes. Shoreline segments with low mean fetch and littoral slope, fewer developed shoreline parcels, and higher aquatic plant richness had higher probabilities of nesting. In addition, significantly more nesting sites were on islands than on mainland shoreline segments. The locations of predicted nesting sites on the independent lake set compared favorably to the locations of observed nests. The ability to predict suitable Common Loon nesting sites should lead to the greater protection or restoration of these valuable areas and enhance conservation efforts across the state.

The Common Loon (Gavia immer) is perhaps one of North America's most studied birds. Numerous studies have been conducted on various aspects of Common Loon life history, and research on nesting is particularly prevalent. Nesting site habitat is well-documented (Vermeer 1973; McIntyre 1988; Evers et al. 2010), and studies have identified lake-level characteristics that are associated with use by breeding Common Loons (Found et al. 2008; Evers et al. 2010; Kuhn et al. 2011). However, the ability to predict where they might nest along a given shoreline has been lacking despite the importance of this knowledge to local lake managers. In addition, several characteristics known to be important in Common Loon nest site selection, such as protection from wind and waves, have not been quantified. The objective of this study was to develop predictive models for Common Loon nesting habitat on Minnesota lakes that incorporate new data and provide spatial information about potential nesting locations.

The breeding range of the Common Loon is primarily in Canada, but a small proportion of the population breeds in the northern United States. Within the continental United States, Minnesota has the largest summer population of Common Loons. A 1989 survey estimated that approximately 12,000 adults summer and breed in the central and northern portions of the state (Strong and Baker 2000), and annual monitoring between 1994 and 2012 indicated that the population remains stable (Minnesota Department of Natural Resources 2012a).

The preferred breeding habitat of Common Loons is oligotrophic lakes with high water clarity and low anthropogenic development (Vermeer 1973; Meyer 2006; Kuhn et al. 2011). Common Loons prefer to nest near shore on vegetated hummocks, small islands, or masses of emergent vegetation. Nests are generally found in areas protected from prevailing winds and with good views to their open water territory (Yonge 1981; McIntyre 1983; Valley 1987; Evers et al. 2010). Males select nesting sites, and both males and females aggressively defend their territories (Piper et al. 2000, 2008). Suitable small lakes (< 50 ha) often have a single mated pair (Evers et al. 2010). Large lakes can have numerous pairs nesting in secluded bays and on small islands (Vermeer 1973). Common Loons show high site fidelity for sites that previously produced successful nests (Strong et al. 1987; Piper et al. 2008).

Human activities can disturb Common Loon nesting and chick rearing (Evers 2007). Meyer (2006) found that they did not reproduce on northern Wisconsin lakes when shoreline building densities exceeded 25 buildings/km; given that Wisconsin shoreline zoning regulations permit a density of 33 buildings/km, the potential for reduced Common Loon productivity in this area and others with similar zoning is substantial. Heimberger et al. (1983) reported that hatching success was a function of distance to development - nests within 150 m of lakefront homes had a 45% hatching success rate, whereas nests greater than 150 m from development had a 75% hatching rate. In addition, they noted that Common Loons appeared to avoid nesting in areas where there were five or more developments within 150 m. In a study of 98 randomly selected small to moderately sized lakes (< 300 ha) in northern Wisconsin and Michigan, Newbrey et al. (2005) found that Common Loon presence was negatively related to the number of lakefront homes.

Previously, the Minnesota Department of Natural Resources developed a model to identify sensitive lakeshore (i.e., lakeshore comprised of unique or critical ecological habitat) in central Minnesota (Minnesota Department of Natural Resources 2012b). These areas are important to a variety of wildlife species, including Common Loons, but are threatened by development and land alteration along the shoreline. Multiple biological attributes were used to identify sensitive lakeshore, including shoreline frog and vegetation-dependent fish presence, lakeshore bird richness, and aquatic plant richness. We wanted to know if these attributes, which are likely indicators of lakeshore condition, may also be associated with potential Common Loon nesting habitat.

Although Common Loons (loons) are somewhat adaptable to human activities (Titus and Van Druff 1981), efforts by lakeshore residents and lake users will help ensure that high-quality nesting habitat is retained for locally nesting loons. Actions include monitoring loon activity, alerting others to nest locations to reduce disturbance, and providing nesting rafts or platforms. In conservation plans for the Common Loon, Evers (2007) and Tischler (2011) stated the need to protect breeding habitat at a local scale and to identify site-specific habitat requirements. These goals would be facilitated by the identification of existing and potential nesting locations.

Shoreline attributes and aquatic vegetation and animal presence data were used to determine probabilities of Common Loon nesting for segments of lakeshore on north-central Minnesota lakes. We determined whether human lakeshore development influenced nest site selection, and compared important explanatory variables. The resulting nesting habitat models were used to predict nesting habitat areas for independent lakes.

Methods

Study Area

We selected 35 lakes in north-central Minnesota for use in the models to identify suitable Common Loon nesting habitat areas (Fig. 1). Study lakes were located in Cass, Crow Wing, and Itasca Counties, and were selected based on several factors. All primary study lakes were > 202 ha in size; the three lakes under this threshold (Deep Portage, Little Boy, and Louise Lakes) were connected to larger lakes and were therefore included in the surveys (Table 1). Development on the lakes was moderate to heavy (average about seven dwellings/ km), and all study lakes were facing the threat of additional shoreline development, but still had undeveloped stretches of shoreline. All lakes were deep-water, mesotrophic, glacial lakes and were important fishery and recreational lakes in Minnesota. Cass and Crow Wing County lakes were the focus of the sensitive lakeshore identification project (Minnesota Department of Natural Resources 2012b), and were subject to aquatic vegetation, aquatic frog, near-shore fish, and lakeshore bird surveys; the only exception was Leech Lake, where the size of the lake coupled with the time frame for the study made the ability to conduct shoreline bird surveys infeasible. Itasca County lakes were not part of the earlier study so biological data were not available for these lakes.

Aquatic Vegetation and Animal Surveys

The aquatic vegetation surveys were conducted using a tiered survey approach (Minnesota Department of Natural Resources 2012b). Survey components included a lake-wide assessment of the vegetation community using a grid point-intercept method, and delineation and description of emergent and floating-leaf plant beds. Most vegetation sampling was conducted during peak growth and before plants senesced (July through early September). In lakes or bays with extensive wild rice (Zizania palustris) stands, surveys were conducted earlier (June) to minimize damage to wild rice. The grid point-intercept vegetation survey methodology followed that of Madsen (1999). We established survey points throughout the littoral (vegetated) zone on a grid using a geographic information system (GIS). The size of the littoral zone, the shape of the lake, and existing information about the plant community determined the number of points and the grid resolution used (Minnesota Department of Natural Resources 2012b). The maximum distance between survey points ranged from 30 m to 200 m. Sampling was conducted primarily from a boat, and global positioning system (GPS) units were used to navigate to each sample point. Using a double-headed, weighted garden rake attached to a rope, surveyors identified and recorded all plant taxa found within an approximate 1-m2 sample site. Plants were identified to the species level when feasible. This method produced frequency of occurrence (presence/absence) records rather than abundance estimates. Plant taxonomy followed Crow and Hellquist (2000) and nomenclature followed MNTaxa (2011). Voucher specimens were collected for most plant species (Hellquist 1993).

Figure 1.

Study area for evaluating probability of nesting by Common Loons on shoreline segments in north-central Minnesota lakes.

f01_102.jpg

Table 1.

Attributes of the 35 lakes used to develop models predicting the probability of Common Loon nesting on shoreline segments in north-central Minnesota lakes.

t01_102.gif

Surveyors mapped floating-leaf and emergent vegetation beds using several techniques. Using a GPS, surveyors delineated bulrush stands (Schoenoplectus spp.) by boating or walking around the edge of any monospecific bulrush stand or mixed emergent vegetation stand that included bulrush (Radomski et al 2011). Surveys were generally conducted during mid-day under low to moderate wind conditions (< 20 kmph). Aerial photographs were used to delineate wild rice and floating-leaf vegetation (FSA Aerial Photography 2003/04-2010). Species compositions of stands were verified and stand boundaries were re-delineated in the field. Emergent and floating-leaf plant stand surveys were conducted in August and early September.

Sampling stations for aquatic frog, fish, and bird surveys were established using GIS. The shoreline of each lake was divided into 400-m segments; this distance was the maximum detection distance for most species surveyed. Stations were established at the midpoint of each segment; the same sampling stations were used for all animal surveys. The entire shoreline of each lake was surveyed. Aquatic frog surveys were conducted between sunset and 01:00 hr from mid-June to mid-July. If rain showers or breezy conditions substantially affected hearing ability, a survey was stopped. At each station, surveyors listened for several minutes for frog and toad calls. They recorded an estimate of the abundance of frogs and a calling index for both mink (Lithobates septentrionalis) and green (L. clamitans) frogs. They also recorded the calling intensity of all other audible amphibian species.

Near-shore fish assemblages were sampled by shoreline seining, backpack electrofishing, and trap nets. Seines were 15.2 m long with a bag, and all mesh was 3.2-mm nylon. The seine was set at the shoreline and perpendicular out to the length of the seine or the maximum wading depth, and the offshore end of the seine was arced back to shore. Electrofishing was conducted using a backpack battery-based electrofisher. Surveyors conducted two shocking passes at each station sampled, one near the shoreline and one at a depth of approximately 75-100 cm. Trap nets had a 12.2-m lead approximately 1.1 m deep with two 1.5-m by 0.8-m frames and six 0.76-m hoops with an 18 cm square throat; all mesh was 6.4-mm nylon. The nets were oriented perpendicular to shore with the leader on or near the shore. Nets were set overnight and pulled the next day. After collection, surveyors identified and counted species. In places with excessive vegetation, depth, or extremely soft bottom, surveyors did not conduct seine or trap net surveys. Between one and three samples were collected at each station, and data were pooled by station for all gear used.

Two methods were used to collect data on lakeshore birds: point counts for all bird species and call-playback surveys targeting marsh birds. Surveys were conducted during the nesting season, defined as the last week of May through the first week of July. Morning point counts for all birds were conducted between sunrise and 22:00 hr at each sample station. Depending on vegetation, water depth, and wind, the survey boat was positioned 20–50 m from shore. Surveyors recorded all birds seen or heard within a 200-m radius of the sample station. Birds seen or heard in the distance or flying overhead were recorded as present at the lake but not associated with a specific sample station. For marsh birds, call-playback survey methodology was modified from Conway (2005). Surveys were conducted in the evening before sunset and focused on survey stations with appropriate (i.e., marsh/wetland) habitat. At each station, surveyors played a tape that included the calls of six marsh birds: Pied-Billed Grebe (Podilymbus podiceps), American Bittern (Botaurus lentiginosus), Least Bittern (Ixobrychus exilis), Yellow Rail (Coturnicops noveboracensis), Virginia Rail (Rallus limicola), and Sora (Porzana Carolina). Then they listened for a response. Both survey techniques were dependent on good listening conditions, and surveys were stopped if inclement conditions prevented the ability to hear bird vocalizations.

Common Loon Nest Surveys

Data on both current and historical nesting areas were used in the analysis. Surveyors conducted Common Loon nesting area surveys between 2007 and 2012. Surveys were conducted from the water by boating slowly along the shoreline and searching for incubating loons. Surveys were conducted during May and early June when adults were still incubating eggs. To avoid disturbing incubating loons, surveyors did not approach nests and marked the general nesting area location with a GPS; the exact nest location was described on the survey form and used later in the office to modify nesting area locations after importing into GIS.

Additional nesting data were obtained from volunteers. The volunteer LoonWatcher survey is administered by the Minnesota Department of Natural Resources as a way to obtain information on nesting locations and success in the State. Volunteers are often lake residents, or spend a significant amount of time on a particular lake during the summer. As part of their report, volunteers record the locations of nesting areas on a lake map and identify the nests as either natural nests or active artificial nest platforms. These volunteer data provide some historical perspective on nesting areas (volunteer-documented nesting area locations are available back to 1978). Where available, these data were used to supplement data collected by surveyors. Although locations of nesting areas reported by volunteers were not field-verified by surveyors, locations of nesting areas identified by both surveyors and volunteers generally displayed high geographical agreement.

Each Common Loon nest was associated with a specific 400-m shoreline segment; these segments are referred to as nesting sites. Two nests were more than 100 m from shore; because it was not clear which specific shoreline segment the nest was associated with, these nests were not included in the analysis. In some cases, more than one nest was recorded per shoreline segment. Because each lake was surveyed only once within the nesting season, multiple nesting attempts were not identified. In addition, the nesting area locations represent attempted nesting sites, but do not distinguish between successful and unsuccessful nests. Natural nests and artificial platforms were treated the same in the analyses. A total of 226 shoreline segments were identified as nesting sites.

GIS Analysis

The shoreline of each study lake was buffered by 25 m, 50 m, and 100 m both lakeward and landward. Study windows were created by intersecting these buffers with the 400-m shoreline segments created earlier; therefore, the grids were 25 × 400 m, 50 × 400 m, and 100 × 400 m and reached both into the lake and onto the land. All attributes (Table 2) were analyzed within these grids with the exception of littoral zone slope, which was calculated out to the full extent of the littoral zone. Using data collected during the grid point-intercept aquatic plant survey, aquatic macrophyte richness was calculated by summing the number of taxa within each lakeward study window. Emergent and floating-leaf vegetation were analyzed by calculating the percent of the window covered by either emergent or floating-leaf vegetation. Attributes containing animal data were analyzed within the landward 100 × 400 m grid only.

Table 2.

Description of explanatory variables for predicting the probability of Common Loon nesting on shoreline segments in north-central Minnesota lakes.

t02_102.gif

We also assembled or derived GIS data from existing data sources. Littoral slope was derived using lake bathymetry data (Minnesota Department of Natural Resources 2012c). Wind fetch (m) was calculated using the U.S. Geological Survey Wind Fetch Model (Rohweder et al. 2008). This model estimates fetch based on waterbody shape and wind direction (maximum 2-min average wind direction; National Oceanic and Atmospheric Administration 2012). We used wind data collected between April and June 2007–2012 to represent the time of the primary nest surveys. Wind fetch was calculated at 10° increments using the model; individual fetch outputs were then multiplied by the percentage of wind observed from its particular direction and summed to calculate a final weighted wind fetch (Rohweder et al. 2008). Mean littoral slope and mean weighted wind fetch were calculated for each 400-m study window along the shoreline.

We analyzed shoreline development using county parcel data (collected in 2008; Minnesota Department of Natural Resources 2012c). Developed parcels were defined as those that had a building or dwelling within the parcel boundaries. We summed the number of developed parcels within each 100 × 400-m landward study window; a parcel was counted if any portion of the parcel fell within the specified buffer zone. The mean size of the developed parcels within 100 m of the shoreline was also calculated for each 100 × 400-m landward study window.

Wetland data were obtained from the National Wetlands Inventory (U.S. Fish and Wildlife Service 2012). Wetlands classified as lacustrine or occurring lakeward of the shoreline were excluded from the analysis to avoid identifying the same habitat patches documented during the aquatic vegetation surveys. All other wetland types were included in the analysis. We calculated the percent of the land area covered by wetlands within each 25 × 400-m, 50 × 400-m, and 100 × 400-m study window. All GIS analyses were conducted using Arc Map (Environmental Systems Research Institute 2011).

Statistical Analysis

Contingency analyses (Fisher's exact tests) were conducted to assess the association between nesting presence and frog, fish, and bird species presence. Significant differences between mean fetch by nest presence were tested with the Tukey-Kramer (HSD). Models were developed to predict the probability of nest occurrence within a given shoreline segment using a generalized linear mixed model (GLMM; Pinheiro and Bates 2000). All statistical analyses were conducted using R (R Development Core Team 2012) with the glmmML package (Broström and Holmberg 2012). Models were fit using restricted maximum likelihood except when comparing models of different fixed effect structure with likelihood ratio tests; then models were fit using maximum likelihood. In an effort to incorporate some lake-level attributes into the analysis, lakes were modeled as random effects. The analysis assumed that data from different lakes were statistically independent.

The model development strategy followed the suggestions of Wolfinger and Chang (1995) and Zuur et al. (2009). The influence of wetlands, aquatic vegetation, littoral slope, wind fetch, shoreline type (main lake or island), presence-absence of other animals (frogs, fish, and birds), and shoreline development were analyzed as fixed effects. After initial testing to determine significant fixed effects, eight candidate models were developed that incorporated fixed effects for the response variable (probability of a nest present on a given shoreline segment). The changes in the AIC score were used to select a preferred model, and Akaike weights were used to quantify the strength of evidence for alternative models (Burnham and Anderson 2002). To aid interpretation of model coefficients, models were also fitted to explanatory variables on a standard scale by centering and dividing each by two times their standard deviation (Gelman 2008).

Determination of model adequacy consisted of reviews of quantile-quantile plots (Landwehr et al. 1984), partial residual plots, and measurements of agreement between nest observations on the modeled lakes (i.e., model residuals, in-sample error, or resubstitution accuracy). We used the area under the curve (AUC) of the receiver operating characteristic (ROC) and Cohen’s kappa statistic to evaluate the predictive accuracy of the models (Fielding and Bell 1997; Pearce and Ferrier 2000; Manel et al. 2001). The AUC value ranges from a value of 0.5, indicating no discrimination ability, to 1.0 for models with perfect discrimination capability. Youden's index and the maximum value of kappa were used to select a probability threshold for classifying shoreline segments with likely suitable nesting habitat.

The random forest classifier, a type of recursive partition method for constructing classification trees, was used to challenge the analyses of the structured model-building approach (Breiman 2001). We used the randomForest package (Liaw and Wiener 2002). Based on trials to minimize error with Common Loon nesting, the random subset of variables at each node of a tree was set at three with 500-classification tree forests constructed. Given that nesting presence constituted a small fraction of the data, each tree in the forest was constructed from a balanced sample of presences and absences by using all of the shoreline segments with nests present and an equal number of shoreline segments without nests. The measures of variable importance from the random forest calculations, computed by the mean decrease in the Gini index for each variable over all trees in the forest, were compared to the explanatory variables from the GLMMs. Patterns in partial dependence plots, which are plots of the marginal effect of a predictor variable when other variables are held constant, were compared to the a priori linear responses of GLMMs (of particular interest was the plot area between the first and tenth deciles as the patterns at both of the margins are influenced by few data). In addition, the out-sample predictive abilities of the random forest approach were tested by using a two-fold random cross-validation where the training dataset was created by randomly withholding 25% of the shoreline segments (i.e., random shoreline segments were selected from random lakes). The with-held sampled segments were then used as a test dataset for determining the AUC out-sample accuracy measure.

Finally, we used a small independent lake set to compare the location and number of predicted Common Loon nesting sites to the location and number of nests documented by volunteers. The three independent lakes were located in Crow Wing County and were representative of the lake size and nest abundances found throughout the study area.

Results

Common Loon nesting prevalence on north-central Minnesota lakes was fairly low. Surveys documented 258 nests on 35 lakes, and 226 nesting sites (shoreline segments with nests) were identified within 3,359 shoreline segments (Table 1). Eighty-three percent of the nesting sites had a natural nest present at some time, and many artificial platforms were located near or within areas previously used by naturally nesting Common Loons. Thirty-seven of the nesting sites contained artificial platforms only. The density of known nesting sites on lakes with nests ranged from 0.03 to 0.75 nesting sites per shoreline km (mean = 0.36 nesting sites/ km, coefficient of variation = 77%).

Observed nesting sites were not randomly distributed. Significantly more nesting sites were on islands than on mainland shoreline segments (Fisher's exact test, P = 0.000); Common Loons were 4.2 times more likely to nest on island shorelines than mainland shorelines. Average mean wind fetch differed significantly between shoreline segments with nests present and those without (Tukey-Kramer HSD test, P < 0.05). Nesting sites were also associated with green and mink frog breeding locations (Fisher’s exact test, P = 0.000 for green frog presence and P = 0.0001 for mink frog presence). Common Loons were 4.7 times more likely to nest in shoreline segments with green frogs present and 2.0 times more likely to nest in segments with mink frogs present than in shoreline segments without frogs. There was also a slight association between Common Loon nesting sites and number of vegetation-dependent fish species present at a shoreline segment (Fisher's exact test, P = 0.049). Nest presence did not appear to be strongly associated with shoreline development class (Fisher's exact test, P = 0.1846).

Table 3.

Suite of candidate generalized linear mixed models used to understand the relative influence of variables on Common Loon use of shoreline segments for nesting in north-central Minnesota lakes. Fixed effects included mean fetch, shoreline type, mean slope, shoreline development, frog class, and bird community composition (proportion of bird species that were forest habitat species and proportion of bird species that were fragmented habitat species). Akaike Information Criterion (AIC) values were estimated by maximum likelihood. Models are ranked by increasing AIC and decreasing area under the curve of the receiver operating characteristic (AUC) based on in-sample validation. Cohen's kappa statistic summarizes the confusion matrix.

t03_102.gif

Initial testing of GLMMs suggested that littoral slope, shoreline development, frog class, and bird community composition might be important variables to predict probable nesting sites. The highest ranked model for lakes where bird surveys were conducted included shoreline type, mean fetch, frog class, bird composition, and mean littoral slope (Table 3). Based on the quantile-quantile plots, the models appear to fit the data with no major departures from model assumptions. The estimates of random lake effect were correlated with nest density.

Several GLMMs were of moderate to high quality in prediction of nesting sites. AUC values generally exceeded 0.8, indicating that the models were able to distinguish correctly between nesting presence and absence over 80% of the time after accounting for all variables. Cohen's kappa values indicate fair model performance (Table 3). In order of importance, the probability of nesting on a shoreline segment increased with island shorelines, lower mean fetch, the presence of both aquatic frog species (green and mink frogs), and lower shoreline development (the number of developed shoreline parcels with-in 100 m of the shoreline) (Table 4). Other important explanatory variables to predict nesting sites included mean littoral slope and bird composition. For Model 1, the predicted probability threshold using Youden's index was 0.1, meaning that values greater than 0.1 may be classified as nesting habitat and values below 0.1 may be considered likely unsuitable habitat in the modeled lakes. This threshold resulted in a model prediction sensitivity, or proportion of correctly predicted positive observations, of about 70%, and it predicted about three times more suitable nesting sites than observed nesting sites (Range = 1.0-5.3 times observed); whereas, based on the maximum value of kappa, Model 1 predicted about 1.4 times more suitable nesting sites than observed (Range = 0.1–2.0).

Table 4.

A summary of the generalized linear mixed models for Common Loon use of shoreline segments for nesting in north-central Minnesota lakes. Models are described in Table 3. Fixed effects and their coefficients are presented. To aid interpretation of model coefficients, models were also fitted to explanatory numeric variables on a standard scale by centering and dividing each by two times their standard deviation (standardized coefficients, SC). Lakes were modeled as random effects.

t04_102.gif

The random forest classifier generated similar results to the GLMMs in that the majority of the top predictor variables identified by both analyses were the same. The top four continuous predictor response variables identified by the random forest classifier for the all shoreline segment dataset were mean fetch, mean littoral slope, shoreline development, and aquatic macrophyte richness. The partial dependence plots showed negative, mostly linear trends in nesting site occurrence for mean fetch, mean littoral slope, and shoreline development for the interdecile range of the data (Fig. 2). The random forest classifier also noted the importance of aquatic macrophyte richness, which was not a significant predictor variable in the GLMMs. The partial dependence plot shows a linear relationship between aquatic macrophyte richness and the logit of predicted probability of nesting occurrence. The random forest classifier also had high accuracy as estimated with out-sample AUC (random forest classifiers on the all shoreline segments and those shoreline segments with bird surveys datasets both had an AUC of 0.8). Maps depicting the probability of Common Loon nesting on shoreline segments in two lakes (selected to represent the range of observed nesting site densities) are shown in Fig. 3.

Figure 2.

Partial dependence plots for variables of a random forest classifier of Common Loon nesting sites. Small ticks on the x-axis indicate deciles of the variables. The y-axis is one-half the logit of the occurrence probability. See Table 2 for variable descriptions.

f02_102.jpg

Figure 3.

Probability of Common Loon nesting habitat predicted by the generalized linear mixed-effects Model 1 (1a, 2a) and a random forest classifier (1b, 2b) for two Cass County study lakes chosen to represent the range of observed nest densities (high nest density — Pleasant Lake (1a, 1b); low nest density — Wabedo Lake (2a, 2b)). Darker colors indicate increased probability. Locations of observed nests are denoted by triangles.

f03_102.jpg

The location of predicted Common Loon nesting sites on the independent lake set compared favorably to the location of nests observed during volunteer surveys (Fig. 4). Most of the observed nesting locations were concordant with higher probabilities of nesting habitat as predicted by the GLMM.

Discussion

We found strong evidence that Common Loon nesting sites were related to various geomorphological and biological attributes. In particular, we determined that mean fetch was an important variable for predicting nesting sites within lakes. Although several studies have noted that Common Loons often nest in sheltered areas such as bays (Olson and Marshall 1952; Vermeer 1973; McIntyre 1975), this attribute had been previously unquantified. We found that shoreline segments with low mean fetch had higher probabilities of nesting. On larger lakes, in particular, nests are highly susceptible to washout caused by large waves crossing unobstructed expanses of water. Nesting sites protected from wave and wind action are necessary for nesting success (Yonge 1981). Young chicks are also vulnerable to wind and waves, and successful nurseries will need protection from these elements (McIntyre 1983). Bays and other low wind fetch shoreline areas, such as pockets or points, may also experience reduced boat traffic. Finally, Titus and VanDruff (1981) noted that nests were more difficult to find when located within or near shoreline features such as bays or peninsulas, and documented that less visible nests were significantly more successful than nests with greater visibility. In addition, we found that shoreline segments with low littoral slope also had increased nesting probabilities. Although Common Loons sometimes use sites with steep drop-offs to allow them to approach and exit their nests underwater (McIntyre 1988), this is not a consistent predictor of nest location. In an earlier study on the Whitefish Chain of Lakes, Valley (1987) found that many nests were located in shallow areas with a low-sloping lake bottom. Mean fetch and littoral slope likely jointly represent small, shallow bays or coves that provide nesting loons protection from wind and recreational lake use. Our analyses also supported previous work that documented a strong preference for nesting on islands (Olson and Marshall 1952; Vermeer 1973; Yonge 1981). The simplest GLMM we found had mean fetch and shoreline type (island or mainland) as the most important variables (Model 4). Along with littoral slope and shoreline development data, this model along with a random forest classifier provided reasonable means to predict suitable nesting habitat. In more complex models, frog presence, aquatic macrophyte richness, and bird community composition provided some gain in predictive performance.

Figure 4.

Probability of Common Loon nesting habitat predicted by the generalized linear mixed-effects Model 4 on three independent lakes (top to bottom: Horseshoe, Borden, Ossawinnamakee) in Crow Wing County. Darker colors indicate increased probability. Locations of observed nests are denoted by triangles.

f04_102.jpg

We also found evidence that nesting site location was negatively related to shoreline development. As the number of developed shoreline parcels within 100 m of the shoreline increased, the chance of nesting within a shoreline segment decreased. This is consistent with the findings of other studies. Found et al. (2008) noted that breeding Common Loon presence was negatively related to human activity and disturbed shorelines, and occurrence was positively related to the proportion of forested shoreline. To predict the presence of Common Loons on northern Wisconsin lakes, Newbrey et al. (2005) used a multiple logistic model with three variables including the number of houses on the lake; they found that loon presence was positively related to lakes with forested riparian areas and negatively associated with high levels of human habitation. Human activities along shore degrade habitat (Christensen et al. 1996; Radomski 2006; Radomski et al. 2010) and water quality (Carpenter et al. 1998), which may indirectly affect Common Loon abundance. Human activity may also decrease nesting success and increase chick and adult mortality via disturbance, impact with watercraft, and ingestion of lead fishing tackle (Titus and Van Druff 1981; DeSorbo et al. 2007; Rattner et al. 2008). Common Loons appear to select areas away from intense human activity for nesting, perhaps as a mechanism for increasing nesting success.

Although several studies have attempted to predict Common Loon occurrence on lakes (e.g., Blair 1992; Found et al. 2008), there have been few studies predicting nesting habitat across a range of lakes. There are several challenges in predicting nesting habitat. First, it is difficult to predict low probability events, and predictions of these rare events are susceptible to model errors. Given that Common Loon nesting has low prevalence, the selection of threshold criteria on nesting probability predictions to create potential nesting habitat maps can be subjective. Freeman and Moisen (2008) recommend that if the goal is to identify all potential habitats, then use of a threshold that gives the highest possible specificity while meeting a user-defined required sensitivity is most appropriate. We used a threshold that minimized the mean error rate of positive observations and the error rate for negative observations (i.e., Youden's threshold) and a threshold that results in a maximum value of kappa. While these criteria are independent of prevalence, they have a tendency to overpredict rare events.

In addition, because Common Loons are territorial, all suitable nesting habitat will likely not be used in a given year or even across a period of years. However, our goal was to identify potential nesting sites and not necessarily to determine whether or not a Common Loon is actually going to nest on a given shoreline segment. As shoreline development and lake use change over time, protection of these potential nesting sites may be as important as protecting current nesting sites. As others have recommended, managers should use probability maps to identify potential habitat areas (Lobo et al. 2007; Freeman and Moisen 2008). Using probability maps for nesting habitat, including providing advice on artificial platform placement on a range of lakes, may be helpful to lake residents interested in promoting the conservation of Common Loons.

The ability to predict suitable Common Loon nesting areas should lead to the greater protection or restoration of these valuable areas. In addition, it will aid in the identification of potential areas for placement of nesting platforms. Kuhn et al. (2011) developed multiscale models to identify explanatory variables at the catchment, riparian, and nesting scale to predict breeding Common Loon presence and nesting locations. They found that road densities were lower and open water areas were greater within a 150-m buffer of the nest (the latter may have been due to Common Loon preference for nesting on islands). These models allowed them to make specific recommendations at the watershed and lake scale. For a single New Hampshire lake, McCarthy and Destefano (2011) found that areas were more likely to be used for nesting as the distance to buildings and boat launching areas increased. Based on our analysis and the work of others, we can provide both general and lake-specific recommendations for identifying and improving nesting habitat. One such recommendation is the preservation of island shorelines, particularly those shoreline segments that are not exposed to wind and waves. In addition, those portions of the mainland shoreline protected from wind and waves with gentle littoral slopes and sustaining frog populations should be identified as valuable habitat and treated as such. Our models also provided probabilities of nesting suitability for specific lakes, and application to an independent lake set showed potential to provide management advice across an even wider range of lakes.

As lakes continue to be developed and human disturbance forces Common Loons to nest in less optimal locations, the development of simple predictive models to identify potential areas for placement of nesting platforms will provide wildlife managers with valuable information for Common Loon conservation. Wildlife managers will also have to address Common Loon territoriality when providing this advice. The appropriate placement of artificial nesting platforms may offset increasing human disturbance and habitat degradation. The use of artificial nesting platforms may also increase nesting success, especially in lakes with fluctuating water levels (Piper et al. 2002; DeSorbo et al. 2007). Analysis of long-term trends in Common Loon abundance in places like Minnesota may provide evidence of the merits of these mitigation measures. Prediction of nesting or other important habitats for other species may also be applicable to future conservation efforts.

Acknowledgments

We thank Stephanie Simon and Donna Perleberg for providing data on aquatic plant surveys for the study lakes, and Pam Perry and Kent Montgomery for their assistance in conducting bird surveys. We also thank the anonymous reviewers for their valuable comments on this manuscript.

Literature cited

1.

R. B. Blair 1992. Lake features, water quality and summer distribution of Common Loons in New Hampshire. Journal of Field Ornithology 63: 1–9. Google Scholar

2.

L. Breiman 2001. Random forests. Machine Learning 45: 5–32. Google Scholar

3.

G. Broström and H. Holmberg . 2012. Generalized linear models with clustering. R package, v. 0.82–2. R Foundation for Statistical Computing, Vienna, Austria.  http://cran.r-project.org/web/packages/glmmML/glmmML.pdf, accessed 16 November 2012. Google Scholar

4.

K. P. Burnham and D. R. Anderson . 2002. Model selection and multimodel inference: a practical information-theoretic approach. Springer, New York, New York. Google Scholar

5.

S. R. Carpenter , N. F. Caraco , D. L. Correll , R. W. Howarth , A. N. Sharpley and V. H. Smith . 1998. Nonpoint pollution of surface waters with phosphorus and nitrogen. Ecological Applications 8: 559–568. Google Scholar

6.

D. L. Christensen , B. R. Herwig , D. E. Schindler and S. R. Carpenter . 1996. Impacts of lakeshore residential development on coarse woody debris in north temperate lakes. Ecological Applications 6: 1143–1149. Google Scholar

7.

C. J. Conway 2005. Standardized North American marsh bird monitoring protocols. Wildlife Research Report #2005-04. U.S. Department of the Interior, Geological Survey, Arizona Cooperative Fish and Wildlife Research Unit, Tucson, Arizona. Google Scholar

8.

G. E. Crow and C. B. Hellquist . 2000. Aquatic and wetland plants of northeastern North America (). University of Wisconsin Press, Madison, Wisconsin. Google Scholar

9.

C. R. DeSorbo , K. M. Taylor , D. E. Kramar , J. Fair , J. H. Colley Jr ., D. C. Evers , W. Hanson , H. S. Vogel and J. L. Atwood . 2007. Reproductive advantages for Common Loons using rafts. Journal of Wildlife Management 71: 1206–1213. Google Scholar

10.

Environmental Systems Research Institute (ESRI). 2011. ArcGIS Desktop, v. 10. ESRI Redlands, California. Google Scholar

11.

D. C. Evers 2007. Status assessment and conservation plan for the Common Loon (Gavia immer) in North America. U.S. Department of the Interior, Fish and Wildlife Service, Hadley, Massachusetts. Google Scholar

12.

D. C. Evers , J. D. Paruk , J. W. McIntyre , and J. F. Barr . 2010. Common Loon (Gavia immer). No. 313 in The Birds of North America Online ( A. Poole , Ed.). Cornell Lab of Ornithology, Ithaca, New York,  http:// bna.birds.cornell.edu/bna/species/313, accessed 17 April 2013. Google Scholar

13.

A. H. Fielding and J. F. Bell . 1997. A review of methods for the assessment of prediction errors in conservation presence/absence models. Environmental Conservation 24: 38–49. Google Scholar

14.

C. Found , S. M. Webb and M. S. Boyce . 2008. Selection of lake habitats by waterbirds in the boreal transition zone of northeastern Alberta. Canadian Journal of Zoology 86: 277–285. Google Scholar

15.

E. A. Freeman and G. G. Moisen . 2008. A comparison of the performance of threshold criteria for binary classification in terms of predicted prevalence and kappa. Ecological Modeling 217: 48–58. Google Scholar

16.

A. Gelman 2008. Scaling regression inputs by dividing by two standard deviations. Statistics in Medicine 27: 2865–2873. Google Scholar

17.

M. Heimberger , D. Euler and J. Barr . 1983. The impact of cottage development on Common Loon reproductive success in central Ontario. Wilson Bulletin 95: 431–439. Google Scholar

18.

C. B. Hellquist 1993. Taxonomic considerations in aquatic vegetation assessments. Lake and Reservoir Management 7: 175–183. Google Scholar

19.

A. Kuhn , J. Copeland , J. Colley , H. Vogel , K. Taylor , D. Nacci and P. August . 2011. Modeling habitat associations for the Common Loon (Gavia immer) at multiple scales in northeastern North America. Avian Conservation and Ecology 6: 4. Google Scholar

20.

J. M. Landwehr , D. Pregibon and A. C. Shoemaker . 1984. Graphical methods for assessing logistic regression models. Journal of American Statistical Association 79: 61–71. Google Scholar

21.

A. Liaw and M. Wiener . 2002. Classification and regression by randomForest. R News 2: 18–22. Google Scholar

22.

J. M. Lobo , A. Jiménez-Valverde and R. Real . 2007. AUC: a misleading measure of the performance of predictive distribution models. Global Ecology and Biogeography 17: 145–151. Google Scholar

23.

J. D. Madsen 1999. Point intercept and line intercept methods for aquatic plant management. Aquatic Plant Control Research Program Technical Notes Collection (TN APCRP-M1-02). U.S. Army Corps of Engineers, Engineer Research and Development Center, Vicksburg, Mississippi. Google Scholar

24.

S. Manel , H. C. Williams and S. J. Ormerod . 2001. Evaluating presence-absence models in ecology: the need to account for prevalence. Journal of Applied Ecology 38: 921–931. Google Scholar

25.

K. P. McCarthy and S. Destefano . 2011. Effects of spatial disturbance on Common Loon nest site selection and territory success. Journal of Wildlife Management 75: 289–296. Google Scholar

26.

J. W. McIntyre 1975. Biology and behavior of the Common Loon (Gavia immer) with reference to its adaptability in a man-altered environment. Ph.D. Dissertation, University of Minnesota, Minneapolis. Google Scholar

27.

J. W. McIntyre 1983. Nurseries: a consideration of habitat requirements during the early chick-rearing period in Common Loons. Journal of Field Ornithology 54: 247–253. Google Scholar

28.

J. W. McIntyre 1988. The Common Loon: spirit of northern lakes. University of Minnesota Press, Minneapolis, Minnesota. Google Scholar

29.

M. W. Meyer 2006. Evaluating the impact of multiple stressors on Common Loon population demographics-an integrated laboratory and field approach. EPA STAR Co-operative Agreement Number: R82-9085. Environmental Protection Agency, Washington, D.C. Google Scholar

30.

Minnesota Department of Natural Resources (Minnesota DNR). 2012a. Minnesota Loon Monitoring Program - 2012. Minnesota DNR, Nongame Wildlife Program, Division of Ecological and Water Resources, St. Paul, Minnesota. Google Scholar

31.

Minnesota Department of Natural Resources (Minnesota DNR). 2012b. Minnesota's sensitive lakeshore identification manual: a conservation strategy for Minnesota lakeshores (v. 3). Minnesota DNR, Division of Ecological and Water Resources, St. Paul, Minnesota. Google Scholar

32.

Minnesota Department of Natural Resources (Minnesota DNR). 2012c. The GISDNR Data Deli. Minnesota Department of Natural Resources, St. Paul, Minnesota,  http://deli.dnr.state.mn.us/, accessed 30 April 2012. Google Scholar

33.

MNTaxa . 2011. Minnesota State checklist of vascular plants. Minnesota Department of Natural Resources, Division of Ecological and Water Resources. St. Paul, Minnesota. Google Scholar

34.

National Oceanic and Atmospheric Administration. 2012. Local Climatological Data. National Oceanic and Atmospheric Administration, Asheville, North Carolina,  http://www.ncdc.noaa.gov/IPS/lcd/lcd.html, accessed 28 March 2012. Google Scholar

35.

J. L. Newbrey , M. A. Bozek and N. D. Niemuth . 2005. Effects of lake characteristics and human disturbance on the presence of piscivorous birds in northern Wisconsin, USA. Waterbirds 28: 478–486. Google Scholar

36.

S. T. Olson and W. M. Marshall . 1952. The Common Loon in Minnesota. Occasional Papers of the Minnesota Museum of Natural History 5: 1–77. Google Scholar

37.

J. Pearce and S. Ferrier . 2000. Evaluating the predictive performance of habitat models developed using logistic regression. Ecological Modeling 133: 225–245. Google Scholar

38.

J. C. Pinheiro and D. M. Bates . 2000. Mixed-effects models in S and S-PLUS. Springer-Verlag, New York, New York. Google Scholar

39.

W. H. Piper , K. B. Tischler and M. Klich . 2000. Territory acquisition in loons: the importance of take-over. Animal Behaviour 59: 385–394. Google Scholar

40.

W. H. Piper , C. Walcott , J. N. Mager and F. Spilker . 2008. Nest site selection by male loons leads to sexbiased site familiarity. Journal of Animal Ecology 77: 205–210. Google Scholar

41.

W. H. Piper , M. W. Meyer , M. Klich , K. B. Tischler and A. Dolsen . 2002. Floating platforms increase reproductive success of Common Loons. Biological Conservation 104: 199–203. Google Scholar

42.

R Development Core Team. 2012. R: a language and environment for statistical computing, v. 2.14.3. R Foundation for Statistical Computing, Vienna, Austria,  http://www.R-project.org/, accessed 13 May 2012. Google Scholar

43.

P. Radomski 2006. Historical changes in abundance of floating-leaf and emergent vegetation in Minnesota lakes. North American Journal of Fisheries Management 26: 932–940. Google Scholar

44.

P. Radomski , L. A. Bergquist , M. Duval and A. Williquett . 2010. Potential impacts of docks on littoral habitats in Minnesota lakes. Fisheries 35: 489–495. Google Scholar

45.

P. Radomski , K. Woizeschke , K. Carlson and D. Perleberg . 2011. Reproducibility of emergent plant mapping on lakes. North American Journal of Fisheries Management 31: 144–150. Google Scholar

46.

B. A. Rattner , J. C. Franson , S. R. Sheffield , C. I. Goddard , N. J. Leonard , D. Stang and P. J. Wingate . 2008. Sources and implications of lead-based ammunition and fishing tackle to natural resources. Wildlife Society Technical Review 08–01, The Wildlife Society, Bethesda, Maryland. Google Scholar

47.

J. Rohweder , J. T. Rogala , B. L. Johnson , D. Anderson , S. Clark , F. Chamberlin and K. Runyon . 2008. Application of wind fetch and wave models for habitat rehabilitation and enhancement projects. U.S. Department of the Interior, Geological Survey, Open-File Report 2008–1200, Washington, D.C. Google Scholar

48.

P. I. V. Strong and R. J. Baker . 2000. An estimate of Minnesota's summer population of adult common loons. Pages 5–13 in The Common Loon: Population Status and Fall Migration in Minnesota ( P. H. Svingen and A. X. Hertzel , Eds.). Minnesota Ornithologists' Union Occasional Papers Number 3, Minneapolis, Minnesota. Google Scholar

49.

P. I. V. Strong , J. A. Bissonette and J. S. Fair . 1987. Reuse of nesting and nursery areas by Common Loons. Journal of Wildlife Management 51: 123–127. Google Scholar

50.

K. B. Tischler 2011. Conservation assessment for the Common Loon (Gavia immer) in the upper Great Lakes. U.S. Department of the Interior, Fish and Wildlife Service, I ron wood, Michigan. Google Scholar

51.

J. R. Titus and L. W. Van Druff . 1981. Response of the Common Loon to recreational pressure in the Boundary Waters Canoe Area, northeastern Minnesota. Wildlife Monographs 79: 6–58. Google Scholar

52.

U.S. Fish and Wildlife Service. 2012. National Wetlands Inventory. U.S. Department of the Interior, Fish and Wildlife Service, Washington, D.C.  http://www.fws.gov/wetlands/, accessed 1 May 2012. Google Scholar

53.

P. J. Valley 1987. Common Loon productivity and nesting requirements on the Whitefish Chain of Lakes in north-central Minnesota. Loon 59: 3–11. Google Scholar

54.

K. Vermeer 1973. Some aspects of the nesting requirements of the Common Loon in Alberta. Wilson Bulletin 85: 429–435. Google Scholar

55.

R. D. Wolfinger and M. Chang . 1995. Comparing the SAS GLM and MIXED procedures for repeated measures. Pages 1–11 in Proceedings of the Twentieth Annual SAS Users Group International Conference. SAS Institute, Inc., Cary, North Carolina. Google Scholar

56.

K. S. Yonge 1981. The breeding cycle and annual production of the Common Loon (Gavia immer) in the boreal forest region.M.S. Thesis, University of Manitoba, Winnipeg, Canada. Google Scholar

57.

A. F. Zuur , E. N. Ieno , N. J. Walker , A. A. Saveliev and G. M. Smith . 2009. Mixed effects models and extensions in ecology with R. Springer, New York, New York. Google Scholar
Paul J. Radomski, Kristin Carlson, and Kevin Woizeschke "Common Loon (Gavia immer) Nesting Habitat Models for North-Central Minnesota Lakes," Waterbirds 37(sp1), 102-117, (1 April 2014). https://doi.org/10.1675/063.037.sp113
Received: 10 February 2013; Accepted: 1 May 2013; Published: 1 April 2014
JOURNAL ARTICLE
16 PAGES


SHARE
ARTICLE IMPACT
Back to Top