**Assessment of Rusty Blackbird Habitat Occupancy in the Long Range Mountains of Newfoundland, Canada Using Forest Inventory Data**

**Kathleen K. E. Manson 1, Jenna P. B. McDermott 2, Luke L. Powell 3,4,5,\*, Darroch M. Whitaker <sup>6</sup> and Ian G. Warkentin 1,2,7**


Received: 13 May 2020; Accepted: 29 August 2020; Published: 4 September 2020

**Abstract:** Rusty blackbirds (*Euphagus carolinus*), once common across their boreal breeding distribution, have undergone steep, range-wide population declines. Newfoundland is home to what has been described as one of just two known subspecies (*E. c. nigrans*) and hosts some of the highest known densities of the species across its extensive breeding range. To contribute to a growing body of literature examining rusty blackbird breeding ecology, we studied habitat occupancy in Western Newfoundland. We conducted 1960 point counts across a systematic survey grid during the 2016 and 2017 breeding seasons, and modeled blackbird occupancy using forest resource inventory data. We also assessed the relationship between the presence of introduced red squirrels (*Tamiasciurus hudsonicus*), an avian nest predator, and blackbird occupancy. We evaluated 31 *a priori* models of blackbird probability of occurrence. Consistent with existing literature, the best predictors of blackbird occupancy were lakes and ponds, streams, rivers, and bogs. Red squirrels did not appear to have a strong influence on blackbird habitat occupancy. We are among the first to model rusty blackbird habitat occupancy using remotely-sensed landcover data; given the widespread availability of forest resource inventory data, this approach may be useful in conservation efforts for this and other rare but widespread boreal species. Given that Newfoundland may be a geographic stronghold for rusty blackbirds, future research should focus on this distinct population.

**Keywords:** red squirrel; boreal; wetland; *Euphagus carolinus*; point count; remotely sensed landscape data; unmarked

#### **1. Introduction**

Though the rate of decline may have eased in the last decade [1], long-term monitoring has documented range-wide declines in rusty blackbird (*Euphagus carolinus*) populations that exceed 80% over the past century, with qualitative evidence for declines dating back to the 19th century [2–4]. Migratory species such as blackbirds can be affected by stressors or threats acting during any or all phases of the annual cycle [5,6]. The factors driving this dramatic range-wide decline in the formerly wide-spread and common rusty blackbird are not fully understood, though this species may have been affected by threats manifested during at least two phases of the annual cycle. Multiple stressors

on the species' wintering grounds have likely contributed to declines, including: "Pest" control measures on agricultural land that target other species of blackbirds but also lead to incidental rusty blackbird mortality [7]; loss of up to 80% of potential wintering wetland habitat through conversion to agricultural land use and fragmentation [3]; and exposure to high levels of methyl mercury through dietary intake (during both wintering and breeding) that could cause physiological or reproductive impacts [8]. Factors acting on the species' boreal forest breeding grounds have also likely contributed to population declines [2], including wetland conversion for agricultural purposes [9,10] and habitat degradation linked to climate change [11,12]. Regarding the latter, McClure et al. [11] detected a 143 km northward shift in the southern edge of the breeding range of rusty blackbirds between 1966 and 2005, and also found a significant correlation between rusty blackbird population declines detected in Breeding Bird Survey (BBS) data and Pacific Decadal Oscillations. McClure et al. [11] suggested that warmer temperatures and drier conditions may reduce the amount of arthropod prey and change prey phenology, resulting in a temporal disconnect between breeding phenology and prey availability.

Habitat and habitat quality can play key roles in determining both distribution and productivity of forest songbirds [13,14] and boreal songbirds are known to be influenced by factors that include natural [15,16] and anthropogenic disturbance [17,18]. Broad-scale breeding season habitat associations have been described for rusty blackbirds [4] and various factors have been hypothesized as influencing them on their breeding grounds. These factors include competition with other icterids [19], timber harvesting [20,21], changes to wetland hydrology and ecology [11], and nest predator dynamics [22]. However, though most jurisdictions across the boreal biome maintain relatively standardized forest resource inventory landcover databases to support the management of forests and other natural resources, few studies have quantitatively assessed breeding habitat for rusty blackbirds using forest inventory data. Given their widespread availability and direct role in natural resource management planning, forest inventory data may be particularly useful for efficient, cost-effective monitoring, and for studying a widespread and rare species such as the Rusty Blackbird. This is especially true in the boreal forest, which is often remote or inaccessible to surveyors. Further, these landcover inventories may be used to identify key habitat for conservation and management of species across a wide geographic area, and so may prove useful when applied in the conservation of boreal species at risk.

The breeding range of rusty blackbirds extends across the boreal forest of Canada and the United States, and within this biome breeding activity is most often associated with wetland and riparian ecosystems and adjacent dense conifer stands [4]. During the breeding period, rusty blackbirds forage primarily on aquatic invertebrates along the shorelines of lakes and streams but occasionally seek terrestrial prey [22,23]. Powell et al. [24] assessed breeding site occupancy of rusty blackbirds using ground-based measurement of habitat features (i.e., they did not use remotely sensed data); occupancy was affected by various factors acting at multiple spatial scales, but was driven primarily by the availability of wetlands that afforded suitable foraging opportunities (i.e., areas of shallow water) and evidence of beaver (*Castor canadensis*) activity. The presence of dense patches of conifers in the vicinity of those wetlands was also required for nesting, while stand age and harvest history were less influential [24]. Lack of specificity in the latter matches the largely anecdotal descriptors of nest substrates reported elsewhere; rusty blackbirds appear to prefer nesting in short (less than 4.5m tall), dense conifer stands [21,22,25–27], and predominantly use black spruce (*Picea mariana*) and balsam fir (*Abies balsamea*) near wetlands [24]. They have also been reported nesting in willow thickets (*Salix* sp.; [23]). Most recently, Wohner et al. [27] assessed rusty blackbird habitat use during various periods of the breeding season. They found that streams, softwood and mixed wood sapling stands, wetlands, and areas characterized by slopes between 1% and 8% were important in predicting rusty blackbird occupancy. Streams were very important in predicting nest sites and adult occupancy, but were especially important in predicting fledgling occupancy. In contrast, fledglings and adults strongly selected wetlands, but this habitat was not strongly associated with nest sites. Wohner et al. [27] proposed that streams may be a more important source of smaller arthropod prey for nestlings, whereas wetlands may host larger prey—for example, dragonflies—valuable for adults and large dependent fledglings.

There has been limited research on the impact of nest predators on rusty blackbirds. Matsuoka et al. [25] assessed nesting success in rusty blackbirds and found that of failed nests, 89% were lost to depredation, and various studies have shown that predation risk can affect nest site selection in songbirds (e.g., [28]). In a study of rusty blackbird nest success, Luepold et al. [22] found that North American red squirrels (*Tamiasciurus hudsonicus*) were the most frequent predator of rusty blackbird nests in Maine, USA. Red squirrels are an important predator of nests and fledglings of boreal songbirds that can affect populations and community structure (e.g., [29–31]). Red squirrels were introduced to Newfoundland, Canada during the 1960s and spread rapidly; they are now the most important predator of songbird nests on the island [32–34]. Squirrels have been implicated in the imperilment of two endemic Newfoundland songbird subspecies and represent a novel threat that may affect rusty blackbird nesting success, habitat associations, and abundance on Newfoundland [35,36]. During concurrent research at our study site, McDermott et al. [37] determined that during the 2016 summer season that followed a high mast crop (G. Robineau-Charette and D. Whitaker, unpublished data), red squirrels were detected at 18% of survey points. In contrast, during the 2017 summer season which followed a low mast crop, detections occurred at only 5% of survey points. During both years there was a negative relationship between squirrel detections and elevation with no squirrels detected above 515 m in either year (see [37] for further details). Benkman [38] suggested that red squirrel populations on Newfoundland were more than double those of mainland North American populations.

Forest management has been shown to affect breeding distribution, behavior and success in many species of birds (e.g., [39–41]). By contrast, Powell et al. [24] found that recent logging in adjacent uplands did not feature among the variables retained in the best occupancy models for rusty blackbirds in Northern New England, USA. However, an assessment of nesting success in regenerating, recently harvested stands versus older, established stands at the southern edge of the rusty blackbird breeding range in Maine suggested that recently harvested areas are an ecological trap, with nests in older stands being 2.3 times more likely to fledge young than nests in stands <20 years post-harvest [20]. Conversely, Buckley [26] found that nesting success in managed forest stands was comparable to that for other cup-nesting species, while Wohner et al. [27] suggested that regenerating softwood forests provided dense cover for fledglings to hide from predators. More research is needed to examine the role of forest harvesting at territory and landscape scales on habitat occupancy by rusty blackbirds, and to assess the extent to which such findings from southern portions of the breeding range are applicable in more northerly boreal regions.

As has occurred in continental portions of the rusty blackbird breeding range, the population on the island of Newfoundland experienced a significant decline between 1970 and 2014, as estimated from Breeding Bird Survey (BBS) data (−6.33% per year, 95% credible interval −3.58 to −9.35 based on 23 routes; [1]). Despite this decline, the number of individuals encountered per BBS route on Newfoundland was substantially higher than for all other regions ([7]; mean of 2.03 birds per route on Newfoundland based on data from 1980–2005 compared to a survey-wide mean of 0.26 birds per route for data collected from 1966–2005). As is the case for many bird species found on the island of Newfoundland [42–44], it has been suggested that the rusty blackbird population breeding on Newfoundland and possibly some adjacent portions of Atlantic Canada may be a distinct subspecies (*E. c. nigrans*) from that found across the remainder of the boreal forest (*E. c. carolinus*; [45]), and so may represent a distinct conservation unit for rusty blackbirds. Thus, Newfoundland appears to remain a stronghold for rusty blackbirds and is an important element for range-wide conservation planning.

We used an occupancy modelling approach based on two years of systematic survey data from a 257 km<sup>2</sup> study area in western Newfoundland to assess how rusty blackbird occupancy is influenced by habitat. We used forest resource inventory landcover data derived from aerial photography to measure habitat availability, an uncommon approach for rusty blackbirds (but, see Bale et al. [46], Wohner et al. [27], who used aerial photography-derived habitat data). In addition, we evaluated the

influence of red squirrel presence on the probability of blackbird occupancy. We predicted that rusty blackbirds would be associated with wet environments—specifically, waterbodies, watercourses, and bogs—and coniferous stands. Furthermore, we predicted that rusty blackbird occupancy and red squirrel presence would be negatively related.

#### **2. Materials and Methods**

#### *2.1. Study Site*

We collected data in the Main River and upper Humber River watersheds, located on the eastern slope of the Long Range Mountains of western Newfoundland, Canada (49.75◦ N, 57.25◦ W; Figure 1; see also [36,37]). The 257 km2 study area spans an elevation range from 75 m to 608 m, with elevation increasing from southeast to northwest. Landcover is dominated by wet boreal forest [47] containing a mosaic of mixed and single-species stands dominated by balsam fir or black spruce along with white birch (*Betula papyrifera*), tamarack (*Larix laricina*), and white spruce (*P. glauca*). Much of the mature forest consists of a closed canopy with few, large canopy gaps, and trees at higher elevations tend to have more stunted growth forms [48]. 68% of sites had >25% cover of forest stands older than approximately 30 years of age. In a study in the northern boreal forest of Alaska and Yukon Territory, Viglas et al. [49] found that 30 year old trees had an approximately 50% chance of producing cones, and that this probability increased with age. Thus, a large proportion of our survey area may provide valuable habitat for red squirrels. Qualitative evidence suggests there was a large cone crop in 2015–2016, with a lighter cone crop during 2016–2017 (G. Robineau-Charette and D. Whitaker, unpublished data). Approximately 5% of our survey points were located above 550 m, the approximate altitude of the tree line [48]. Various forms of boreal wetlands and aquatic habitat suitable for rusty blackbirds are widespread across the study area, including bogs, fens, and the shorelines of rivers and lakes, while barrens and other natural openings also make up a proportion of landcover. Overall, landcover within our survey point buffers consisted of an average of 3% lakes, 7% bog, 29% coniferous scrub, and 48% balsam fir- and/or black spruce-dominated forest. Natural disturbances such as wildfire and outbreaks of defoliating insects are uncommon at higher elevations due to climatic conditions, leading to the development of mixed age, old growth fir forests having an abundance of canopy gaps and complex vertical structure [48,50]. Trees at this site have been aged at over 250 years old [48], and around our survey points alone, 27% of points contained 50% or more forest greater than 110 years old. Portions of the study area were harvested by clearcutting between 1990 and 2004 resulting in 19.7% of the study area being cleared in cutblocks ranging from 0.30 ha to 197.4 ha; natural regeneration of balsam fir has followed harvest at these sites. The construction of a 60 m-wide electricity transmission corridor during 2016 and 2017 (Figure 1) created a linear strip of cleared land through the study area. All lands in the study area are provincial public lands (i.e., "Crown lands").

#### *2.2. Field Methods*

We collected field data from early June through mid-July of 2016 and 2017 to span the period of peak territorial display and defense for most migratory songbirds in the region, including rusty blackbirds. Systematic surveys were carried out across a grid of points spaced 500 m apart (Figure 1), and for the 2017 season we shifted the grid 250 m north and 250 m east so that survey points fell midway between those sampled the previous year (i.e., a diagonal distance of 354 m from the points sampled the previous year). The total number of surveyed points was 991 during 2016 and 969 in 2017. Solitary observers conducted point counts; data collection included four surveyors during 2016 and five surveyors in 2017 (one individual was common to both seasons). Each surveyor sampled 5–12 adjacent points per day between 05:40 h and 14:30 h. This timeline deviated from standardized avian survey protocols such as Breeding Bird Surveys [51] and was devised as part of a survey using call broadcast originally designed to target gray-cheeked thrush (*Catharus minimus*) and red squirrels (see description below, and [37]). However, 85% of our point counts were conducted before 10am, and the probability of rusty blackbird detection did not vary substantially between hours within our survey period, despite a larger standard error after 13:00. Surveyors recorded wind strength using the Beaufort scale, and stopped field work when high winds (>5 Beaufort scale; 29 km/h) or precipitation/fog impaired visual or auditory detections of songbirds (similar to BBS protocol; [51]). We continued to operate in the presence of light drizzle and fog, as these weather conditions are frequent in this climate, particularly in the morning, and we believe that this approach did not detract from our capacity to detect individuals during surveys. Precipitation was recorded as either absent, fog, drizzle, rain, or snow. Surveyors also recorded cloud cover on a scale from 1–5 (0 = no clouds, 1 < 25% cloud cover, 2 = 26–50%, 3 = 51–75%, 4 = 76–99%, 5 = 100%).

**Figure 1.** Distribution of survey points in the Main River and upper Humber River watersheds in western Newfoundland, Canada. Each year, points in the survey grid were spaced 500 m apart, and in 2017 the grid was shifted 250 m north and 250 m east, placing the points midway between those sampled the previous year. The location of the study area on the island of Newfoundland is shown by the red box on the inset map.

Surveyors visited each survey point once, conducting an 11-min unlimited radius point count [52] that was divided into the following sequence of five sub-periods: (1) six minutes silent listening; (2) two minutes broadcast of gray-cheeked thrush (*Catharus minimus*) calls and songs (3) a one-minute

silent period; (4) one minute broadcast of red squirrel vocalizations; and (5) a final one-minute silent period. These subperiods were designed for an unrelated study that examined the relationship between gray-cheeked thrush and red squirrels. However, we recorded all bird species seen and/or heard and each red squirrel detected during each of the time blocks within the 11 min of a point count. As such, these methods provided suitable data for our study on rusty blackbirds. There is no reason to believe that the broadcast of gray-cheeked thrush vocalizations would influence Rusty Blackbird behavior or detectability. However, rusty blackbirds are known to mob potential predators (e.g., [25,53]) so might be attracted rather than deterred by the squirrel broadcasts. Surveyors used broadcast equipment (FoxPro model FX3 or Crossfire game callers; FoxPro Incorporated, Lewistown, PA 17044, USA) played at a consistent volume; when measured 1 m from the speaker, the peak volume of broadcasts was 82.6 dB.

#### *2.3. Data Analysis*

Using ArcMap (version 10.5.1; [54]), we extracted landcover data for each survey point from the provincial forest resource inventory Geographic Information System (GIS) database, which was created using high resolution (sub 10 cm pixel resolution) 3D aerial photographs taken in 2007. Landcover was mapped according to the standard forest resource inventory classification scheme used by the Province of Newfoundland and Labrador, with landcover elements assigned to cover types (e.g., forest, forest scrub, bog, barren, lakes and ponds, rivers). Forest stands were further classified according to 20-year age classes and dominant tree species composition. The provincial forest resource inventory only includes rivers >15 m wide, which are mapped as two-dimensional landcover features (i.e., polygons). However, smaller streams are likely important habitat features for rusty blackbirds [24,27], and are classified as linear features (i.e., 1-dimensional vectors) in Natural Resources Canada's CanVec geospatial database (available under the Government of Canada's open government License [https://open.canada.ca/en]). The national Canvec database is produced using several data sources and resolution varies from 1:10,000–1:50,000 scale. Consequently, we extracted two variables for moving water: (1) The extent of rivers > 15 m wide (m2) from the provincial forest resource inventory, and (2) the linear length of smaller streams (m) from the CanVec database.

We extracted landcover information within a 347 m radius around each point (i.e., a 37.8 ha circle); this approximates the rusty blackbird home range estimate of 37.5 ± 12.6 ha developed by Powell et al. [20] based on radiotracking 13 rusty blackbirds (6 males, 7 females) in Maine. We converted point count detections of rusty blackbirds into presence-absence data for each point and standardized most habitat features as the proportion of the 37.8 ha buffer circle covered by that habitat type. The only exception was for streams, which were measured as the total length (m) of streams in the 37.8 ha circle, and then re-scaled from 0–1 by dividing these values by the maximum observed stream length (2463.7 m). We assessed these raw landscape variables for collinearity using Spearman's ρ, and did not detect correlations that warranted further consideration or screening of variables (correlation coefficients were less than 0.45, which is below thresholds requiring additional consideration [55]). We aggregated balsam fir and black spruce stands into conifer stands since we believed that they would function similarly as rusty blackbird habitat [4]. Additionally, we excluded habitat features that were present in the landscape but that (1) occurred at less than 10% of the total survey points, or (2) were presumed to be unimportant for rusty blackbirds based on current understanding of the species' habitat needs. The latter habitat features included soil barrens, herbaceous soil barrens, rock barrens, sand, fens, residential land, rights-of-way, cleared land, and forests that were not dominated by either balsam fir or black spruce. Based on this approach our final analysis included seven landcover variables (Table 1).

We used the package UNMARKED [56] in Program R version 3.5.1 [57], using the function "occu" to assess relationships between rusty blackbird occurrence and the seven landcover variables (Table 1) plus year, elevation, and red squirrel presence. This program allowed us to model rusty blackbird detectability prior to running occupancy models. We explored the potential influence of six variables (cloud cover, observer, precipitation type, wind strength, time of day, and day of the

year) on likelihood of detection (Table 2) in order to predict occupancy more effectively [58]. Each of the five sub-periods of each point count was considered a site visit, for a total of 5 repeated visits (see [56]). Observer and precipitation type were fit as categorical variables. For both detectability and occupancy modeling (the latter described below), we considered the model having the lowest AICc as the best-fit model and based our conclusions primarily on this model. We then fit 31 *a priori* occupancy models including the null and global models. We used the best-fit detectability model as the base (i.e., null) model for all occupancy models. We formulated these models following the approach of Powell et al. [24] based on information presented in existing rusty blackbird habitat studies [4,24] along with anecdotal reports. Similar to Powell et al. [24], our models included habitats which reflect where one could reasonably expect to find rusty blackbirds. Candidate models included various combinations of landcover variables, including hypothesized interactions between some terms, and this resulted in models containing biologically relevant combinations of (1) nesting habitat, and (2) foraging habitat, and (3) models containing nesting and foraging habitat (Table 3). We also included elevation and red squirrel presence/absence in several of our candidate models. Rather than exploring the influence of geographic coordinates on rusty blackbird occupancy, we considered elevation to be a relevant substitute for assessing overall spatial variation in occupancy. We chose this approach because elevation increases with increased latitude, but east–west dimensions at the study site do not vary drastically. Models having a ΔAICc less than or equal to two were considered to be competing models (i.e., not measurably better than one-another), and the subset of top-ranked models having a cumulative weight of 95% were taken as the best model set [59].

**Table 1.** Landcover variables used in modeling rusty blackbird occupancy in Newfoundland, Canada, 2016 and 2017. Variables were measured as either the linear amount of the feature (streams [m]) or the proportion of landcover within a 347 m radius of each survey point (all other variables).




<sup>a</sup> AICc of best model = 2722.31.

**Table 3.** Candidate models describing occupancy (Ψ) of rusty blackbirds in Newfoundland, Canada, in 2016 and 2017 (*n* = 1960 survey points). Models in the 95% confidence set of best models are highlighted in bold. All models include a term for the effect of observer on detectability.


<sup>a</sup> AICc of best model = 2602.58.

#### **3. Results**

Observers identified rusty blackbirds at 209 of the 1960 points visited over two years (105 points in 2016, 104 points in 2017), a naïve occupancy rate of 10.7%. At 174 sites we detected only one individual, whereas at 30 sites we observed 2 individuals, and at 5 sites we observed 3 individuals. The factor that most strongly affected detectability was observer (Table 2), whereas the model including cloud cover was marginally better than the null model but performed considerably worse than the model only including observer (ΔAICc = 3.11, wi = 0.14). The model including precipitation performed similarly to the null model (ΔAICc = 4.53, wi = 0.07; Table 2). A post-hoc check of an observer + cloud model revealed that adding cloud cover improved model fit only slightly (AICc = 2721.15). Models including wind, time of day, and ordinal day were all worse than the null model, and a post-hoc assessment of the model containing time of day and observer did not prove to be important in detection probability. Based on these findings, we included observer in the base model for all subsequent occupancy models.

Four models were included in our 95% confidence set of best occupancy models, and of these the model that best predicted rusty blackbird occupancy contained lakes and ponds (β = 7.21 ± 0.94), bogs (β = 3.53 ± 0.73), streams (β = 2.06 ± 0.41), and rivers (β = 6.94 ± 3.57) (Table 3). This model included strong positive relationships for lakes and ponds, bogs, and streams, and a weaker positive relationship for rivers (Figure 2a–d). Based on this model, the mean predicted occupancy of rusty blackbirds across our study area was 12.2% (95% confidence interval = 9.4–15.7%; Figure 3). The next two models in the best model set were similar to the best model, but with rivers being replaced by either conifer scrub (β = −0.38 ± 0.44), or conifer forest (β = 0.14 ± 0.36); no strong directional relationship was observed between the probability of rusty blackbird occurrence and conifer scrub cover or conifer forest (Table 3; Figure 2e–f). The final model in our best model set was the global model. Red squirrels were more abundant in the summer of 2016 compared with the summer of 2017 (i.e., they were more abundant following a high cone production year), with 84% of squirrels detected in 2016 [37]. While red squirrel presence did not appear in our 95% confidence set of best occupancy models, adding red squirrel to the best-fit model improved it slightly (AICc = 2601.99; (β = −0.49

± 0.33). Additionally, a red squirrel by year interaction term was added to the best model post hoc; inclusion of this interaction term among the best models was not supported (AICc = 2605.17). We followed the same process with an interaction term that included red squirrel and elevation, and found that this interaction term also did not improve our best model (AICc = 2604.16). We found no evidence that squirrels meaningfully influenced rusty blackbird distribution or occupancy (Table 3; Figure 2 g).

**Figure 2.** Relationship between rusty blackbird probability of occurrence and landcover variables found in our highest-ranked occupancy models (**A**–**F**), as well as with red squirrel occurrence (**G**), in Newfoundland, Canada, in 2016 and 2017 (*n* = 1960). Landcover variables were (**A**) proportion of lakes and ponds (**B**) proportion of bogs (**C**) length of stream (**D**) proportion of rivers, (**E**) proportion of conifer scrub, (**F**) proportion of conifer forest, and (**G**) red squirrel probability of occurrence. Each variable was plotted using the best model that included that factor (Table 3). Error bars show one standard error. Note that while the vertical axis is the same for all plots, the scale and units of the horizontal axis varies.

**Figure 3.** Probability of rusty blackbird occupancy mapped across our study area in Newfoundland, Canada, in 2016 and 2017. Probability of occupancy was estimated from our best model (Table 3) based on landcover data available in the provincial forest resource inventory and publicly available national CanVec stream data.

#### **4. Discussion**

Of our 31 *a priori* models, we found that the amount of lakes and ponds, streams, bogs, and rivers best predicted rusty blackbird occurrence. Aquatic invertebrates are an important food source for rusty blackbirds [22,23], and blackbird foraging activities often focus on the edges of waterbodies with abundant shallow water [2]. Thus, the importance of these aquatic habitats in our model is not surprising and is consistent with findings from previous studies of rusty blackbird habitat occupancy [22–24]. However, the absence of conifer forest from our best model and the apparent overall weak influence of either conifer forest or conifer scrub cover on rusty blackbird occupancy initially appears counterintuitive. Powell et al. [24] suggested that blackbird habitat occupancy in Maine, where the forest has a greater component of deciduous cover, is heavily influenced by the presence of patches of suitable conifer habitat in which to nest. In contrast, our study area on Newfoundland is dominated by dense, often single species, conifer stands and consequently rusty blackbirds may not need to actively select for such a ubiquitous feature on the landscape. In particular, wet soils in riparian zones and wetlands on Newfoundland often support dense stands of conifer scrub. To retain landcover features where rusty blackbirds were most likely to occur, we only included forest types that were dominated by either black spruce or balsam fir. Since primarily deciduous stands dominated less than 5% of points, most forested points included conifer cover. Thus, within the spatial scale at which we examined occupancy, rusty blackbirds in Newfoundland appear to select sites primarily based on the availability of appropriate foraging habitat. Additional research such as radio-telemetry studies or nest searching may allow us to better describe the nesting habitat needs of rusty blackbirds in Newfoundland, as well as their space use and breeding ecology. We did not find a strong influence of red squirrel occurrence or year on rusty blackbird occupancy. This is unexpected, given the notable difference in red squirrel detections between years [37]. Red squirrels are known to reproduce later in the season following low cone crop years so that they can take better advantage of seasonal food resources [60], and this influences juvenile survivorship when compared with individuals born earlier in the year [61]. Luepold et al. [22] found that during a year when red squirrel occurrence was higher following a large mast crop, red squirrels were more frequently observed to prey on rusty blackbird nests. It is possible that red squirrels are preying upon rusty blackbird nests, but that despite this, the blackbirds continue to return to habitat where red squirrels are present due to other attractive factors. Red squirrels are prominent members of most boreal forest ecosystems and can directly affect other species through an omnivorous diet and generalist predatory behavior [62,63]. De Santo and Willson [64] found that nest predation was lowest in open wetlands and within forests, compared with both forest and wetland edges and clearcuts. Our observation that red squirrel presence was not strongly related to rusty blackbird occupancy—despite marginally improving our best model in a post-hoc test—may suggest a deviation from the typical vulnerability to nest predation commonly experienced by forest songbirds [63,65,66]. Related research at our study site indicates that squirrels are much more abundant at lower elevations [37] and in these areas they may have caused the local extirpation of breeding populations of another species, the gray-cheeked thrush (McDermott, unpublished data; see also [36,44]). Unlike gray-cheeked thrushes, we found that elevation appeared to have little or no effect on rusty blackbird occupancy. It may be that the impact of squirrels is not as strong for rusty blackbirds as for gray-cheeked thrushes because of a greater capacity to deter predators; for example, rusty blackbirds have been observed mobbing presumed threats (e.g., field crews) near nest sites [53], a trait which is known to drive nest predators away [67]. Alternatively, there may be aspects of their nest site selection that enable them to avoid strong impacts of squirrel predation. Aquatic habitat may also impede squirrel movements during summer and consequently compel squirrels and blackbirds to select different habitats.

McDermott et al. [37] found that squirrel occurrence was inversely related to surface water and ambiguously related to "open" habitats (an amalgamation of bogs, barrens, and other natural openings), which could offer some protection to rusty blackbird nests near those habitat types. Specifically, these open, wet habitats could act as barriers to red squirrel movement. However, the weak relationship between rusty blackbirds and red squirrels could also have resulted from an ecological trap (see Powell et al. [21]), as Luepold et al. [22] found red squirrels to be the most important predator of rusty blackbird nests in New England. More research into rusty blackbird nest success and predator

dynamics on Newfoundland may help to clarify the impact of the introduction of red squirrels on this and other boreal bird species.

Consistent with the findings of Powell et al. [24], clearcuts did not appear in our best model set, indicating that this habitat type had little influence on blackbird occupancy. Provincial forestry regulations require that an unharvested buffer strip at least 20-m wide be left along the shorelines of water bodies, and Whitaker and Montevecchi [68] found that, in western Newfoundland, abundance of rusty blackbirds did not differ between these buffer strips and unharvested shorelines. Thus, it may be that current forestry regulations are sufficient to safeguard blackbird habitat needs on Newfoundland. However, because it has been suggested that harvested forests can act as ecological traps for Rusty Blackbirds [21], more research into nesting success of blackbirds at this site could provide a more nuanced perspective into effects of forest management. In addition, the influence of forest management history and temporal patterns at this site is as yet unexplored. This study area provides a unique opportunity to study rusty blackbirds with an abundance of old growth forest. Given that Newfoundland remains a stronghold for rusty blackbirds, it may be worth investigating in more detail the relationship between forest characteristics (i.e., age, structure, diversity) and rusty blackbird abundance.

Our best detectability model included observer as an explanatory variable. Contrary to Powell et al. [24], wind, precipitation, cloud cover, time of day, and ordinal day did not strongly affect detectability, which may reflect differences in the sampling protocol. Variability in detection probability by the observer may reflect false negatives (species was present but was not detected) or false positives (an observer incorrectly identified a species as present; [69–71]). However, false positives are unlikely because similar-looking species (e.g., other blackbirds) are not present on Newfoundland. Intuitively, the frequency of these errors is lower if observers have higher skill levels [70]; all of our observers were skilled in the identification of local species prior to data collection. Given the naïve and predicted blackbird occupancy rates of only 10.7% and 12.2%, respectively, across the study area, differences in detectability between observers may at least in part reflect the potential for some observers to have been assigned survey areas where they were less likely to encounter rusty blackbirds (e.g., due to physical ability or back-country navigation skills). Broadcasting rusty blackbird calls would likely increase detections among all observers because of behavioural responses to conspecific bird calls [24]. In addition, ensuring that all surveyors are of a similar skill level in fitness and backcountry orienteering would improve the consistency of detection rates between observers, since it would ensure that observers cover a similar number of points across a challenging landscape.

The spatial scale at which habitat use is assessed inevitably affects the apparent habitat requirements of a given species [72]; thus, more research on the spatial ecology of rusty blackbirds may lead to improved inferences. Our habitat analysis buffer size reflected the best available home range estimate for rusty blackbirds, although values ranged from 3.8 ha to 172.8 ha [20]. Powell et al. [20] noted that the home range size of colonial rusty blackbirds was significantly larger than that of non-colonial pairs. This likely reflects the potential for birds nesting in loose colonies to share information on the location of short-lived sources of emergent insects, whereas pairs on their own may have more limited sources of food [20]. Therefore, home range size, and a bird's ability to take advantage of available resources may vary drastically based on behavioral factors between individuals. Because we only detected more than two individuals at five out of 209 occupied sites (2.4%) over the two years, it is likely that, as is typical for much of their breeding range, solitary nesting is prevalent at our study area. Concrete evidence of pair versus colonial status is another area where radio-telemetry studies may improve our understanding of rusty blackbird breeding ecology.

Our study is among the first of its kind to model rusty blackbird occupancy using information on habitat from a typical Forest Resource Inventory database that was developed based on high resolution aerial photography, as well as from other publicly available landcover data (e.g., our stream data; but, see Wohner et al. [27]). The fact that we did not undertake field habitat surveys allowed us to efficiently complete a systematic survey of over 1900 point counts across a large area having limited road access. Remote sensing resources such as aerial imagery are considered valuable tools in predicting species distributions and developing population estimates [73,74]. Further, this is the same spatial database that the province uses to plan and monitor industrial forestry and other forms of natural resource management. Consequently, it would be straightforward to use the findings of studies such as this, which are based on information contained in those spatial databases, to predict and map the distribution of rusty blackbirds across the landscape (e.g., Figure 3). This offers the opportunity to easily incorporate consideration of blackbird habitat into conservation, management, and research planning. For example, while the mean predicted occupancy across our study area was 12.2%, the 5% of points having the highest estimates had a mean predicted occupancy of 46.8% (range 32.9%–86.9%); this type of information could be of value in planning research or rapidly identifying high-value habitat during land use planning. Similar forest resource databases are available for many jurisdictions across the North American boreal forest, especially those subject to large scale extractive resource use, so this approach may be applied across much of the species' breeding range to map and protect potential blackbird habitat. Based on our findings, and the findings from other studies on Rusty Blackbird occupancy and habitat use (e.g., [21,22,24,26,27]), key Rusty Blackbird habitat with a high probability of occupancy—such as concentrations of wetlands, waterbodies and watercourses, with nearby dense conifer forest—may be identified from remotely sensed data. Once these areas are identified, prioritization of survey areas and conservation planning may proceed.

#### **5. Conclusions**

This is the first quantitative study of rusty blackbirds on the island of Newfoundland, and among the first published studies to use remotely-sensed data to predict their breeding habitat. Given that they may be genetically distinct, this population is important to the overall conservation and recovery of the species. Further, the decline of rusty blackbirds on Newfoundland has been more gradual than in most other areas of the species' breeding range, and they continue to be detected in higher numbers on Newfoundland than elsewhere in their breeding range [7]. Indeed, our naïve occupancy rate was 41.7% higher than that of Powell et al. [24] for a population in northern New England, while our predicted occupancy in the upper 5% of most preferred sites averaged more than four times the overall naïve occupancy rate.

Consistent with past research, our study indicates a strong association for breeding rusty blackbirds with aquatic habitats in the boreal forest, including lakes, ponds, streams, rivers, and bogs; these findings echo the results of Bale et al. [46], and the conservation value of these wetland environments, particularly in the face of climate change. Due to the island's cool, wet maritime climate, boreal landscapes across much of Newfoundland consist of a complex mosaic of bogs and surface water intertwined with coniferous scrub and forests. This appears to offer relatively plentiful habitat for rusty blackbirds, and presents an opportunity to study this declining species in a region where relatively high numbers persist.

**Author Contributions:** Conceptualization, K.K.E.M., L.L.P., J.P.B.M., D.M.W., I.G.W.; methodology, K.K.E.M., L.L.P., J.P.B.M., D.M.W., I.G.W.; formal analysis, K.K.E.M., L.L.P., D.M.W.; investigation, J.P.B.M., K.K.E.M.; resources, I.W., D.M.W.; data curation, K.K.E.M., J.P.B.M.; writing—original draft preparation, K.K.E.M.; writing—review and editing, K.K.E.M., L.L.P., J.P.B.M., D.M.W., I.G.W; visualization, K.K.E.M., L.L.P, J.P.B.M., D.M.W.; supervision, J.P.B.M., I.G.W., D.M.W., L.L.P.; project administration, J.P.B.M., D.M.W., I.G.W.; funding acquisition, I.G.W., D.M.W.; All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the Centre for Forest Science and Innovation (Newfoundland and Labrador Department of Natural Resources), the Natural Sciences and Engineering Research Council of Canada, and Gros Morne National Park of Canada. The Article Processing Charges were funded by Parks Canada.

**Acknowledgments:** Many thanks to field surveyors Elora Grahame, Brendan Kelly, Noah Korne, Anna Rodgers, Meaghan Tearle, and Benjamin West. Matthew Brooks and Andrew Curtis also provided critical field support, while Jake Burton (Parks Canada) offered geomatics assistance and Dylan Harding helped with data curation. The Newfoundland and Labrador Department of Fisheries and Land Resources, Forestry and Wildlife Branch provided use of a cabin for field surveys and other logistical support. Research was conducted under a scientific research permit from the Newfoundland and Labrador Department of Environment and Conservation, Parks and Natural

Areas Division, as well as a research permit from the Department of Fisheries and Land Resources, Forestry and Wildlife Branch, and animal care approval (16-16-IW) from the Memorial University of Newfoundland Institutional Animal Care Committee.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Rusty Blackbird (***Euphagus carolinus***) Foraging Habitat and Prey Availability in New England: Implications for Conservation of a Declining Boreal Bird Species**

**Amanda Pachomski 1,\*, Stacy McNulty 2,\*, Carol Foss 3, Jonathan Cohen <sup>1</sup> and Shannon Farrell <sup>1</sup>**


**Abstract:** The Rusty Blackbird (*Euphagus carolinus*) is an imperiled migratory songbird that breeds in and near the boreal wetlands of North America. Our objective was to investigate factors associated with Rusty Blackbird wetland use, including aquatic invertebrate prey and landscape features, to better understand the birds' habitat use. Using single-season occupancy modeling, we assessed breeding Rusty Blackbird use of both active and inactive beaver-influenced wetlands in New Hampshire and Maine, USA. We conducted timed, unlimited-radius point counts of Rusty Blackbirds at 60 sites from May to July 2014. Following each point count, we sampled aquatic invertebrates and surveyed habitat characteristics including percent mud cover, puddle presence/absence, and current beaver activity. We calculated wetland size using aerial imagery and calculated percent conifer cover within a 500 m buffer of each site using the National Land Cover Database 2011. Percent mud cover and invertebrate abundance best predicted Rusty Blackbird use of wetlands. Rusty Blackbirds were more likely to be found in sites with lower percent mud cover and higher aquatic invertebrate abundance. Sites with Rusty Blackbird detections had significantly higher abundances of known or likely prey items in the orders Amphipoda, Coleoptera, Diptera, Odonata, and Trichoptera. The probability of Rusty Blackbird detection was 0.589 ± 0.06 SE. This study provides new information that will inform habitat conservation for this imperiled species in a beaver-influenced landscape.

**Keywords:** Rusty Blackbird; *Euphagus carolinus*; boreal wetlands; aquatic macroinvertebrates; foraging ecology; occupancy modeling

#### **1. Introduction**

The Rusty Blackbird (*Euphagus carolinus*) is a migratory songbird that breeds in and near wetlands of the boreal forests of Canada and Alaska as well as in the northern regions of New York and the Acadian Forest (New England and the Canadian Maritime Provinces). The Rusty Blackbird is representative of global boreal avian species declines and has experienced the worst population loss of thirteen boreal-breeding species [1,2]. Although the Rusty Blackbird was once common, the species has declined by an estimated 90% since the 1960s [3]. The species was estimated to have declined by 5.1% per year from over 13 million birds in 1965–1966 to roughly 2 million birds in 2002–03 based on modeling from standardized winter counts [1]. Furthermore, the southeastern limits of the bird's breeding range appear to have retreated northward coincident with the population decline [4]. The US Fish and Wildlife Service has listed the Rusty Blackbird as a Focal Species of Birds of Management Concern [5]; the IUCN Red List considers the species to be Vulnerable [6]. The cause of the Rusty Blackbird's decline is not fully understood; climate change [4], mercury

**Citation:** Pachomski, A.; McNulty, S.; Foss, C.; Cohen, J.; Farrell, S. Rusty Blackbird (*Euphagus carolinus*) Foraging Habitat and Prey Availability in New England: Implications for Conservation of a Declining Boreal Bird Species. *Diversity* **2021**, *13*, 99. https:// doi.org/10.3390/d13020099

Academic Editor: Michael Wink

Received: 31 December 2020 Accepted: 18 February 2021 Published: 23 February 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

contamination [7], hematozoa infections [8], timber harvesting on breeding grounds [9], and winter habitat loss [10] have been suggested as possible factors.

As with all species, suitable foraging and nesting habitat are key components for Rusty Blackbird persistence. In New England, Rusty Blackbirds select nest sites with minimal canopy cover, high basal area of young conifers (<1.5 m height) [9,11], and patches of forest adjacent to open areas, including wetlands [11]. They build cup nests in live trees, usually red spruce (*Picea rubens*), black spruce (*Picea mariana*), or balsam fir (*Abies balsamea*), surrounded by other young conifers, and occasionally in speckled alder (*Alnus incana*) swamps, in snags, or in isolated conifers in open areas [11]. Rusty Blackbird habitat can be created or improved by forest management or natural disturbance, especially by the American beaver (*Castor canadensis*). The beaver, an ecosystem engineer, creates impoundments of water by damming streams [12], forming wetlands utilized by a diversity of wildlife species. Because deciduous trees and shrubs are a preferred food of the beaver [13], they selectively harvest several woody species in proximity to their impoundments, thereby increasing the percent cover of conifers [14], which is desirable for Rusty Blackbird nesting. Beavers have a long-lasting impact on the landscape; by digging channels and creating dams, beavers increase both the depth and area of wetlands [15]. Individual beaver ponds may be active for one to many years at a time as the animals move to find new sources of food, resulting in a matrix of different-aged ponds, meadows, and streams in a wetland complex [16]. Beavers increase the diversity of and shift the macroinvertebrate assemblage [17,18] as well as increase macroinvertebrate abundance [18] within impounded wetlands and streams. Thus, beavers may create an ideal habitat for Rusty Blackbirds, with flooded, macroinvertebrate-rich wetlands for foraging and clumps of nearby conifers for nesting.

It is critical to monitor Rusty Blackbird populations and habitat use in order to make and evaluate the result of management decisions. Previous research within the Rusty Blackbird's breeding range has focused on demography, nesting ecology, and possible causes of decline. Information about the species' diet and foraging site preferences are scant. Rusty Blackbirds are more insectivorous than other Icterids, based on their skull and bill anatomy [19] and analysis of stomach contents [19–22]. They forage aerially, from perches, and while walking along the water's edge. A breeding Rusty Blackbird diet consists mostly of aquatic macroinvertebrates, such as beetle adults and larvae [22,23], Odonate (dragonfly and damselfly) larvae [24], Trichoptera (caddisfly) larvae and emergent adults [25], and Tipulid (crane fly) larvae [25], but they also hunt aerial prey such as mosquitoes [26]. Breeding Rusty Blackbirds also forage for a variety of terrestrial and volant invertebrates, including snails [27], grasshoppers [20], caterpillars [20], spiders [20,27,28], adult dragonflies [25,29], adult mayflies [29], ants, centipedes, and crustaceans [22]. Furthermore, although Rusty Blackbirds are mostly insectivorous during the breeding season [30], the species has been known to eat some vertebrates such as small fish [27,28] and salamanders [27]. In addition to consumption of various prey items during the breeding season, Rusty Blackbirds exhibit diet plasticity by switching to a more generalized diet of seeds, fruit, acorns, grains, and insects during autumn and winter [22,23]. Overall, summer feeding observations and stomach specimens are very limited in number, and little is known about their foraging habitat requirements. Understanding wetland prey availability during the breeding season in key patches will enable scientists and land managers to identify high-quality foraging sites and may eventually suggest mechanisms behind the Rusty Blackbird decline and potential for recovery.

Rusty Blackbirds are rare and difficult to detect, especially within their remote and hard-to-access breeding grounds; thus, traditional, short duration (five or ten minutes) avian point-counts are not sufficient for accurately detecting breeding Rusty Blackbirds [3]. We used a single-season site occupancy modeling approach [31] to account for imperfect detection and model Rusty Blackbird use of 60 wetland sites as a function of habitat covariates, with a focus on foraging habitat and food availability. We chose multiple a priori habitat covariates that we expected to be biologically important for Rusty Blackbirds

based on previous studies and our own experience. We hypothesized that the probability of Rusty Blackbird wetland use: (1) increases with current beaver activity, (2) increases with the presence of puddles, and (3) increases with increasing conifer cover. We hypothesized that the probability of detecting a present Rusty Blackbird (1) is not affected by the time of day during daylight hours, (2) decreases with increasing wind speed, (3) is highest during the chick-rearing period, and (4) decreases with increasing wetland size. This study is the first to assess Rusty Blackbird foraging habitat use in New England and the first to include prey availability as a covariate in Rusty Blackbird occupancy modeling.

#### **2. Materials and Methods**

#### *2.1. Study Area*

We surveyed breeding Rusty Blackbird use of both active and inactive beaver- influenced wetlands in Coos County, New Hampshire, and Oxford County, Maine (Figure 1). Sites were located either on federal land at Umbagog National Wildlife Refuge (44.832344, −71.075496) or were privately owned and managed by Wagner Forest Management, Ltd. The study area has a mean precipitation of 102.9 cm per year, a mean annual high temperature of 11.67 ◦C, and a mean low temperature of −2.22 ◦C (measured in Colebrook, NH) [32]. The mean elevation of surveyed wetlands was 473.3 m (range = 110 m to 780 m). Beavers have been modifying wetland hydrology and upland vegetation in the region for decades; survey sites were categorized as either active (impounded/modified in the past year and hosting a resident beaver colony) or inactive (previously impounded, but not currently occupied by beavers) wetlands. The forests in this remote area of New England are extensively managed, with active logging operations occurring near most of our study sites.

**Figure 1.** Land cover classes of the study area and digitized polygons of wetlands where Rusty Blackbirds were detected (black triangles) or were not detected (black circles) in northern New England, USA, with 500 m buffer circles showing habitat in 2014.

#### *2.2. Methods*

To evaluate Rusty Blackbirds' use of wetlands, we conducted timed, 30 min long, unlimited-radius point counts at 60 sites in 2014 within a 25 km radius of the town center of Errol, NH. We selected count locations from a pool of 263 discrete (e.g., above a dam) beaver-influenced impoundments within 500 m of a road, which we identified using satellite imagery and expert knowledge of the study area. We used ArcMap to randomly select 39 wetland sites and then systemically selected 21 additional, nearby wetlands to maximize our sample size. Bird surveys were conducted from a fixed, marked point located at the southern edge of each wetland. Some sites were located relatively close together (100–300 m apart), so to avoid double counting, pairs of observers concurrently surveyed adjacent sites, cross-checked the timing of field observations field post-survey, and excluded possible double counts of birds. Due to time and logistical constraints, we surveyed sites in pairs based on their spatial proximity; thus, sites were not randomly visited.

We surveyed each point three times, once during each of three two-week intervals (14 May to 27 May, 28 May to 10 June, and 11 June to 24 June 2014). Due to site access restrictions, two of the 60 sites were only surveyed twice. We timed the survey intervals to align with Rusty Blackbird incubation, nestling rearing, and fledging stages of the breeding season because we suspected that detectability would vary based on the birds' nesting behavior.

We conducted surveys between 8:00 and 18:00. We chose not to limit our surveys to early morning hours in order to maximize our sample size and because Rusty Blackbirds are known to sing throughout the day [33]. Through pilot surveys, we found that a point count duration of 30 min was the most effective yet efficient count duration for detecting Rusty Blackbirds in our study area [33]. During each point count, two independent observers recorded visual and/or auditory detections of Rusty Blackbirds and time to first detection (if detected) without the use of any recorded playback. We also recorded wind speed, temperature, precipitation, and time of day (Table 1).

**Table 1.** Site and field survey covariates used to model detectability and site occupancy of Rusty Blackbirds in northern New England in 2014; the data collection method was by remote sensing (GIS) or at the site (Field).


We collected field data on both vegetation and land cover variables, including percent spruce and balsam fir cover around the wetland, percent cover of open water, percent cover of mud, depth of open water, as well as evidence of recent beaver activity (Table 1). We used the perimeter-based cover estimate method [34] to measure the percent exposed mud and percent open water within the wetland during each survey occasion and then averaged the habitat data across the three occasions.

We collected aquatic invertebrate samples at the point count site (southern edge of each wetland's standing pool of water) with ten sweeps of a D-frame dip net, probing along the water's edge. Macroinvertebrate samples were stored in 70% ethanol and identified to family when possible, otherwise to order or subclass [35]. We used the macroinvertebrate count as a proxy for abundance of Rusty Blackbird prey within a wetland. We averaged the counts for each of the sites' three invertebrate samples and included this abundance as a site covariate in our wetland use models. We also included the total invertebrate family richness for all of each site's samples. For taxa that we were only able to be identified to subclass or order, such as leeches, we assumed that one family was observed for each subclass or order.

We used ArcMap to calculate wetland size and quantify land cover type as habitat characteristics that might drive Rusty Blackbird foraging site selection. Because Rusty Blackbirds forage within multiple wetland types, we chose not to use a wetland classification system and instead delineated all wetland features within a site as a single polygon, using Google Earth satellite imagery (dated 18 September 2013) and field experience as a guide. We then calculated the area of each wetland polygon and used the National Land Cover Database 2011 (NLCD) [36] to calculate the percent conifer cover within a 500 m radius of each wetland, which is the approximate size of a typical Rusty Blackbird breeding home range [11]. We recorded the elevation of each wetland survey point in the field using a GPS unit.

#### *2.3. Analysis*

We used the protocol developed by MacKenzie et al. [31] to model Rusty Blackbird use of wetlands using single-species occupancy modeling based on our detection histories and field and geospatial data (Table 1). Occupancy probability (Ψ) and detection probability (p) were modeled as linear functions of covariates using the logit link to constrain predicted values between 0 and 1. These models assume a closed population between surveys, and where that assumption is violated, Ψ is interpreted as habitat use rather than occupancy. Because Rusty Blackbirds may forage among multiple wetlands within a large area, defining a site as a single wetland, as we did, may violate the assumption of independence of observations. Furthermore, concurrent Rusty Blackbird productivity research revealed that Rusty Blackbirds nested near many of our study sites. However, breeding Rusty Blackbirds in New England have been found to nest up to 400 m away from wetlands, and fledglings can move over 1 km away from their nests within a few weeks of fledging [37], so we were not able to assume that site occupancy was constant throughout the study period. Thus, we considered sites with at least one positive Rusty Blackbird detection to be "used" rather than "occupied" [38].

We performed this analysis using Package unmarked [39] in Program R [40]. Our candidate set of models included biologically plausible variables known or thought to affect Rusty Blackbird habitat suitability. We first modeled survey-specific covariates affecting p (date, precipitation, temperature, time, visit, wetland size, and wind) while modeling Ψ as constant. Then, we chose the model with the lowest Akaike Information Criterion (AIC) score as the best-fit detectability model and used its survey covariate(s) in our base Ψ model, following the approach of Powell et al., 2014 [41]. Next, we created a candidate set of models with one or more site covariates (beaver, invertebrate abundance, invertebrate richness, open water, mud, puddles, water depth, percent conifer, young conifer, elevation, and wetland size). We avoided including significantly correlated (*p* < 0.05) covariates (calculated using Spearman Rho and Pearson Chi-Square tests for continuous and categorical covariates, respectively) within the same wetland use models. We used Package AICcmodavg [42] to estimate c-hat, the overdispersion parameter, to adjust for overdispersion as needed, and to assess model fit. We used the MacKenzie and Bailey Goodness-of-fit Test [43] to test the fit of our global wetland use model, which contained all covariates included in our candidate set of models, with 1000 bootstraps.

To assess whether Rusty Blackbirds select foraging sites based on the abundance of specific aquatic invertebrate prey types, we used the two-sample Poisson rate test to test for a difference between the maximum invertebrate abundance per order at sites with and without Rusty Blackbird detections. We included underrepresented invertebrate groups in abundance totals in order to reflect total prey availability at each site.

#### **3. Results**

We detected Rusty Blackbirds during 66 of 178 surveys. Our base detectability model (without covariates) yielded a detection probability (p) of 0.589 ± 0.06 SE (95% CI: −0.10, 0.82). Detection probability was best predicted by visit (survey period), which was the only covariate in the top model (Table 2). The probability of detection was highest in visit 2 and lowest in visit 1. Back-transformed parameter estimates on the probability scale for visit yielded *p* = 0.416 ± 0.09 SE for visit 1, *p* = 0.765 ± 0.08 SE for visit 2, and *p* = 0.742 ± 0.09 SE for visit 3.

**Table 2.** Model selection for detectability of Rusty Blackbirds in northern New England in 2014.


<sup>a</sup> Number of parameters; <sup>b</sup> Akaike's Information Criterion; <sup>c</sup> Difference in the model's AIC from that of the top model; <sup>d</sup> Akaike weight. (.) Indicates that the parameter Ψ was held constant.

We detected Rusty Blackbirds at 35 out of 60 sites. Based on the null occupancy model without site covariates, Rusty Blackbirds used 0.629 ± 0.07 SE of the study sites. With α = 0.05 and a c-hat value of less than 3, we concluded that our global wetland model had an acceptable fit [44]. However, because a c-hat value greater than 1 suggests overdispersion, we adjusted standard error estimates for each wetland use model by a factor of c-hat [44] and ranked models based on Quasi Akaike's Information Criterion (QAIC) scores.

The top model (number of parameters k = 7, −2 log-likelihood = 188.231, QAIC = 123.4366), included the survey covariate "visit" in the detection probability linear predictor and the site covariates "invertebrate abundance" and "mud" in the occupancy linear predictor (Table 3). Rusty Blackbirds preferred sites with higher aquatic invertebrate abundance (Figure 2a) and lower percent mud cover (Figure 2b). This model accounted for over 60% of the adjusted model weight. Because the second model was not within four delta QAIC units of the top model, we did not model average parameter estimates across all of the models included in the candidate set of wetland use models [45].

**Figure 2.** Relationship between probability of Rusty Blackbird occupancy and site covariates, percent cover of aquatic invertebrate abundance (**a**) and mud (**b**), with 95% confidence bands.

Aquatic invertebrate richness (across all taxa) was not a predictor of Rusty Blackbird wetland use. However, it is worth noting the extent to which each taxon varied in abundance across our study sites as well as the difference (or lack thereof) between taxa at sites with and without Rusty Blackbird detections. Survey sites had a mean of 6.92 insect families (±0.37 SE; range = 1 to 14) and a mean invertebrate count of 49.98 specimens (±6.70 SE; range = 6 to 205.5) combined from three samples in 2014. Sites with Rusty Blackbird detections had higher maximum invertebrate abundance of Amphipoda (*p* < 0.001), Coleoptera (*p* = 0.002), Collembola (*p* < 0.001), Diptera (*p* < 0.001), Hemiptera (*p* = 0.018), Odonata (*p* < 0.001), Oligochaeta (*p* = 0.013), Plecoptera (*p* = 0.004), and Trichoptera (*p* = 0.033) than did sites with no detections (Table 4).


**Table 3.** Model selection for wetland use of Rusty Blackbirds in northern New England in 2014, using site occupancy analysis with AIC scores adjusted for overdispersion (c-hat = 1.78).

<sup>a</sup> Number of parameters. <sup>b</sup> Quasi Akaike's Information Criterion. <sup>c</sup> Difference in the model's QAIC from that of the top model. <sup>d</sup> Quasi Akaike weight.

**Table 4.** Summary statistics and results of a 2-sample Poisson rate test for differences between maximum invertebrate specimen abundance per survey per order from three aquatic macroinvertebrate surveys per site for sites with and without Rusty Blackbird (RUBL) detections in northern New England in 2014.


<sup>a</sup> Considered used if at least one Rusty Blackbird was detected at least once during three surveys. <sup>b</sup> Considered undetected if no Rusty Blackbirds were detected during any of the three surveys. <sup>c</sup> Rate for RUBL-detected – RUBL-undetected sites. <sup>d</sup> Significant *p*-values (α = 0.05) are bolded. <sup>e</sup> The normal approximation may be inaccurate for small total number of occurrences. <sup>f</sup> Subclass, rather than order.

#### **4. Discussion**

#### *4.1. Wetland Use and Aquatic Macroinvertebrate Prey*

Food availability appears to be important to breeding Rusty Blackbird wetland use in northern New England. According to our model ranking, the best predictors of Rusty Blackbird use of wetlands in northern New Hampshire and western Maine are aquatic invertebrate abundance (positively related) and percent cover of mud (negatively related). Contrary to our hypotheses, current beaver activity, presence of puddles, and conifer cover were not important factors. Our data suggest that adult Rusty Blackbirds choose foraging sites based on aquatic invertebrate prey abundance. Abundance of amphipods, beetles, true flies, dragonflies, and caddis flies were higher in sites with Rusty Blackbird detections. This is expected because young birds cannot easily digest dry seeds [30], and invertebrates contain protein needed for chicks' growing skeletons. Aquatic insects are high-quality prey items, in part because many emergents (e.g., Odonata) are soft-bodied in the teneral or subimago stage, vulnerable to capture, and easily digested.

Our model results suggest that invertebrate abundance is more important than invertebrate richness in predicting Rusty Blackbird wetland use. Given the spatial and temporal variability of beaver-influenced wetland food resources within the landscape, it makes sense that Rusty Blackbirds would choose sites with large numbers of prey items rather than sites with a diverse selection of aquatic invertebrate taxa. From 2010 to 2012, the mean distance of Rusty Blackbird nests in this area to nearest wetland was 409.65 m (+/− 46.15 SE) and the maximum distance was 1347 m [46]. Because the foraging habitat in the study area is patchy and nests can be 400 m or more from foraging sites [37], we would expect Rusty Blackbirds to seek out sites with abundant food resources to minimize time spent foraging and provisioning. Home ranges of 13 telemetered adult Rusty Blackbirds in Maine averaged 37.5 ha and ranged from 4–179 ha [9], suggesting that foraging strategies in this heterogeneous environment must be flexible but efficient for successfully raising chicks.

We observed Rusty Blackbirds catching and provisioning multiple prey items at once. This behavior has also been documented in a camera trap study [47] in our survey area as well as in other field observations in Maine and New Hampshire [48] and nearby Vermont [25]. Optimal Foraging Theory predicts how an animal may decide where to forage, what to forage for, and for how long to forage based on maximizing energy gained from food items while minimizing searching and handling time [49]. In addition to nesting hundreds of meters from foraging sites, Rusty Blackbirds often forage in multiple wetlands throughout their home range [9]. They may provision more food items at a time with increasing distance from the nest to the foraging site to maximize energy gained versus that expended, as has been observed in related Icterids [30]. Regional differences in breeding Rusty Blackbird foraging strategies may exist; an Alaskan study found that adult Rusty Blackbirds usually fed chicks one large (>2 cm long) prey item at a time [50]. Thus, it is possible that Rusty Blackbirds in Alaska may operate under a different foraging strategy based on finding high-quality prey rather than minimizing energy spent foraging.

Although invertebrate richness was not a strong predictor of Rusty Blackbird site use in our study, Rusty Blackbirds were more likely to use sites with higher aquatic invertebrate abundance. We modeled a total count of all aquatic invertebrates per site rather than abundance broken down by order due to small sample sizes. Because we suspected that some orders are more important than others, we conducted an exploratory analysis of Rusty Blackbird site use and invertebrate abundance by order. The total number of individual Coleopterans, Collembolans, Dipterans, and Odonates was three times as high at wetlands used by Rusty Blackbirds, four times as high for Amphipods, and twice as high for Trichopterans (Table 4). However, as noted, sites used by Rusty Blackbirds also had higher Coleoptera richness, underscoring the likely importance of aquatic beetles in the bird's diet. Although summer diet information is scarce for Rusty Blackbirds, aquatic beetles can make up 10% or more of their diet in spring and may exceed 25% in some regions of the US [20].

Prey availability is not consistent over time due to differences in insect life histories, weather changes, and other factors. Rusty Blackbirds forage aerially for Odonates and other flying insects as well as hunt at the water's edge for aquatic prey. Orians [30] suggested that invertebrate prey availability is higher in warm, dry weather due to higher insect emergence rates. In 2014, the weather in our study area was generally favorable for Odonate emergence, as mean temperature was 20.8 ◦C ± 0.39 SE and it rained during just 11% of surveys. Odonate emergence rates are highest during mid to late morning [30]; however, we opportunistically observed breeding Rusty Blackbirds foraging throughout the day. This observation suggests they may be choosing other prey later in the day, although we did not quantify the relative foraging time budgets for aerial, emergent, and aquatic prey. Future studies should consider using a combination of aerial and aquatic sampling methods to target different invertebrate orders and life stages.

#### *4.2. Wetland Characteristics*

Mud was a variable in our top wetland use model, but puddles, conifer cover, and current beaver activity were not important predictors of Rusty Blackbird wetland use. In contrast, Powell et al. found that Rusty Blackbird occupancy in New England was best explained by the presence of puddles (i.e., shallow pools of standing water), conifer cover greater than 70%, and evidence of current beaver activity [41]. Although that study did not find strong support for the mud cover survey covariate, our survey methods differed; Powell et al. used a binary measure of mud presence or absence within a site [41], whereas we estimated the percent cover of mud at the wetland.

Our research indicates that Rusty Blackbirds forage in wetlands with abundant aquatic invertebrates and low percent cover of mud. Percent cover of mud is conversely related to that of open water, in which macroinvertebrates located between the water surface and the substrate are accessible to prey-seeking birds. Furthermore, the Rusty Blackbird bill ranges from 17.5 to 19 mm in length [51] and is not morphologically designed to probe for prey in deep mud. Rusty Blackbirds foraging in mud are likely procuring invertebrates on the surface. Studies [52,53] have also suggested that wintering Rusty Blackbirds prefer sites with shallow water. In addition, Wright et al. [54] found that during migration, the birds use forest edges with leaf litter and shallow water, likely to take advantage of the proximity of both arthropods and perches.

Microhabitat features are likely important; we observed Rusty Blackbirds foraging from the surface of deep (>1 m) water while standing on emergent debris. A Vermont study noted that Rusty Blackbirds forage in the water from debris or logs [25]. Other studies in New England suggested an unclear relationship between shallow water extent and Rusty Blackbird occupancy [55]. Breeding Rusty Blackbirds forage in fens and wet meadows [46], yet bird use and prey communities in these shallow-water ecosystems remain understudied. Future breeding-season wetland surveys should note the presence of emergent substrates in deep standing water, as such microhabitats give Rusty Blackbirds access to otherwise inaccessible foraging areas. Additionally, future researchers should assess the heterogeneity of invertebrate food availability within each wetland, including within mud as well as around edges of emergent substrates within deeper water.

Current beaver activity in wetlands did not strongly influence wetland use by Rusty Blackbirds in our study area, as the model with wetland use as a function of beaver occupancy ranked lower than the null model. All of our study sites had been modified by beavers, which may have affected model performance. While the relationship between current beaver activity and breeding Rusty Blackbird wetland use is still unclear, beavers create both breeding and foraging habitat by increasing conifer cover and by making ponds [13], many of which persist for years to decades [56]. Furthermore, beaver-impounded streams contain greater numbers of Odonates [18], and Anisoptera nymphs prefer dams of woody debris over other habitat types [57], so beavers may increase the abundance of preferred food for breeding Rusty Blackbirds. Previous research found that the presence of current beaver activity increased the probability of Rusty Blackbird occupancy [41], which

is expected given that beavers are associated with improved habitat for aquatic invertebrates [58,59]. Furthermore, aquatic invertebrate availability is likely related to water depth and vegetation cover along the wetland edge, which are factors that beavers influence indirectly, rather than to the actual presence of beavers [57]. The relationship between beaver occupancy and Rusty Blackbird wetland use is worthy of further study and refinement.

#### *4.3. Landscape Factors*

Glennon [60] suggested that climate change and habitat modification are the main contributors to declines of several boreal bird species including the Rusty Blackbird. Furthermore, climate change is affecting the hydrology and invertebrate communities of North American boreal wetlands [61]. Sánchez-Bayo et al. [62] noted that species in several aquatic taxa known to be prey for Rusty Blackbirds (e.g., Odonata) have declined or disappeared from many sites in North America. While we did not study insect loss or forested wetland change, the implication for the Rusty Blackbird's breeding habitat and prey base at the species' southern range limit, and perhaps across its North American summer range, is sobering. Although current beaver occupancy was not significant in the model, the influence of beavers on wetland hydrology, heterogeneity, and aquatic invertebrate assemblages is strong [58]. Land managers within the Rusty Blackbird's breeding range should, to the degree feasible, continue to manage the boreal forest landscape by allowing beaver populations to persist and include a mixture of forest stand ages in planning. It is important to allow beaver populations to continue to create impoundments in order to help invertebrate-rich wetlands persist in a changing climate [63].

#### *4.4. Probability of Detection*

We found that the most important predictor of detection probability for breeding Rusty Blackbirds was the visit (survey) period. As we hypothesized, the probability of detection given wetland use was highest during the second visit, when parents were rearing nestlings (28 May–10 June 2014). Rusty Blackbirds tend to be highly secretive and hard to detect during nest-building, egg-laying, and incubation. Once eggs hatch, adult Rusty Blackbirds become more vocal and more obvious as they frequently forage for food and rear their young. Soon after fledging, Rusty Blackbird broods tend to move away from their nesting areas and towards wetlands [37]. Thus, we designed our study to capture differences in breeding season behavior by surveying for Rusty Blackbirds in three survey periods that coincide with their breeding stages. To maximize breeding season detection, future studies could focus sampling effort on the chick-rearing period.

Time of day, date, wind, temperature, precipitation, and wetland size were not important predictors of Rusty Blackbird detectability. Additional factors, including vegetation cover within a wetland, noise created by running water, and anthropogenic noise, could have impacted detectability. During our study, a few sites were within earshot of logging operations, but most surveys were not noticeably impacted by anthropogenic sounds. There is also a need to compare detectability among multiple wetland types. No information exists on Rusty Blackbird occupancy of fens or wet meadows, yet the birds often forage in these shallow-water ecosystems [46]. Such information would better prepare land managers to survey areas that have not been previously studied. Lastly, although we defined a site as a wetland, our actual unit of measurement is the distance over which we were able to detect Rusty Blackbirds; however, we were unable to accurately quantify the distance at which we could hear Rusty Blackbird calls or songs at each site.

#### *4.5. Considerations*

Our single-season occupancy analysis provides a snapshot of Rusty Blackbird use of wetlands in New England; the study was designed to characterize differences between wetlands used by Rusty Blackbirds and wetlands that were unlikely to have hosted foraging birds. Due to limited time and resources, our study scope was defined as wetlands within 500 m of a road, which could have caused bias. Between-year variation in prey availability and habitat features were not examined in this study but are likely important, especially given changes in hydrology through time resulting from shifting beaver occupancy and precipitation patterns.

We sampled a small area (approximately 1 m2) for aquatic invertebrates at the edge of each wetland because the entire wetland perimeter was not accessible due to flooding or areas of downed trees. With multiple invertebrate surveys in the same marked area of each site, we were able to compare temporal changes in invertebrate food availability within a site as well as compare food availability for Rusty Blackbirds across a range of wetlands. However, because we did not sample all Rusty Blackbird prey species (such as snails and spiders) or all life stages of prey species, our surveys provide a useful but incomplete picture of each site's invertebrate community structure.

#### **5. Conclusions**

Our research suggests that Rusty Blackbirds forage in wetlands with abundant aquatic invertebrates and low percent cover of mud, using sites with more open water and emergent vegetation. Conservation of Rusty Blackbird populations and the diverse invertebrate communities upon which these birds depend will require land managers and biologists to explore uncharted territory. Because habitat change, mercury pollution, and climate change are regional to global issues that are difficult to address, we recommend that decision-makers within the breeding range focus on maintaining and improving nesting and foraging habitat. It is important that land managers retain existing beaver populations and manage hydrology in the face of climate change [63]. We recommend continued bird population and wetland monitoring especially because the relationships between water level and prey availability are mediated by climate and will likely experience greater variance over time. If this region becomes more drought-prone, land managers could experiment with managing wetland hydrology to support adequate soil moisture during the growing season by increasing the size of existing wetlands and creating new ones, mimicking the work of beavers.

Because much of the Canadian breeding range has not been surveyed for Rusty Blackbirds, US and Canadian researchers can collaborate to fill information gaps and identify key areas in need of protection. We recommend long-term monitoring of wetland habitat and aquatic invertebrates in the Acadian forest. Land managers, both public and private, have an exciting opportunity to help maintain and improve breeding habitats for Rusty Blackbirds and other imperiled boreal species. Conservationists should expand on education and engagement initiatives, such as the Rusty Blackbird Migration Blitz, to increase the general public's awareness of and concern for this species.

**Author Contributions:** Conceptualization, S.M., A.P., C.F., J.C., and S.F.; methodology, A.P., S.M., J.C. and C.F.; software, A.P. and J.C.; validation, A.P., S.M., C.F., and J.C.; formal analysis, A.P. and J.C.; investigation, A.P.; resources, S.M. and C.F.; data curation, A.P.; writing—original draft preparation, A.P. and S.M.; writing—review and editing, A.P., S.M., C.F., J.C., and S.F.; visualization, A.P. and S.M.; supervision, S.M., A.P., and C.F.; project administration, S.M. and A.P.; funding acquisition, S.M. and A.P. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by Umbagog National Wildlife Refuge and the SUNY College of Environmental Science and Forestry Graduate Student Association Research Grant. The initial pilot study in 2013 was supported by the Northeastern States Research Cooperative through funding made available by the USDA Forest Service. The conclusions and opinions in this paper are those of the authors and not the NSRC, the Forest Service, or the USDA.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Data presented in this study are available on request from the corresponding authors.

**Acknowledgments:** Sean Flint, Ian Drew, and other US Fish and Wildlife Service personnel from Umbagog National Wildlife Refuge provided field housing, site access, and administrative support. Dan Hudnut of Wagner Forest Management, Ltd. provided site access. Devon Cote, Kelsey Schumacher, Amasa Fiske-White, and Thomas Ruland conducted surveys as field technicians. Donald Arthur processed and identified aquatic invertebrate samples. Patricia Wohner provided field research support. Luke Losada Powell provided statistical and technical support.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


## *Article* **Rusty Blackbird Habitat Selection and Survivorship during Nesting and Post-Fledging**

#### **Patricia J. Wohner 1,\*, Carol R. Foss <sup>1</sup> and Robert J. Cooper <sup>2</sup>**


Received: 13 May 2020; Accepted: 29 May 2020; Published: 2 June 2020

**Abstract:** Rusty blackbird (*Euphagus carolinus*) populations have declined dramatically since the 1970s and the cause of decline is still unclear. As is the case for many passerines, most research on rusty blackbirds occurs during the nesting period. Nest success is relatively high in most of the rusty blackbird's range, but survival during the post-fledging period, when fledgling songbirds are particularly vulnerable, has not been studied. We assessed fledgling and adult survivorship and nest success in northern New Hampshire from May to August in 2010 to 2012. We also assessed fledgling and adult post-fledging habitat selection and nest-site selection. The likelihood of rusty blackbirds nesting in a given area increased with an increasing proportion of softwood/mixed-wood sapling stands and decreasing distances to first to sixth order streams. Wetlands were not selected for nest sites, but both adults and fledglings selected wetlands for post-fledging habitat. Fledglings and adults selected similar habitat post-fledging, but fledglings were much more likely to be found in habitat with an increasing proportion of softwood/mixed-wood sapling stands and were more likely to be closer to streams than adults. No habitat variables selected during nesting or post-fledging influenced daily survival rates, which were relatively low for adults over the 60-day study periods (males 0.996, females 0.998). Fledgling survival rates (0.89) were much higher than reported for species of similar size.

**Keywords:** *Euphagus carolinus*; nest success; post-fledging; rusty blackbird; survivorship; streams; wetlands

#### **1. Introduction**

Songbird breeding productivity is often measured by nest success and associated habitat characteristics [1]. However, mortality during the post-fledging period can also result in low breeding productivity and limit populations even when nest success is high [2,3]. Juvenile songbird survivorship is typically estimated to be 50–60% of adult survivorship [4,5]. However, actual juvenile survival rate estimates are quite variable, ranging from 0.19 in hooded warblers (*Setophaga citrina*) to >0.65 in ovenbird (*Seiurus aurocapilla*) and worm-eating warbler (*Helmitheros vermivorum*; [1,3]. Despite this potential variability in survivorship, the post-fledging period remains an understudied component of avian demographics for many species in part due to difficulties of following individuals after fledging [5–9].

Differences in food requirements and vulnerability to predators between the nesting and post-fledging periods may lead to differences in habitat selection between these stages of the breeding season [7,10,11]. For example, ovenbird, worm-eating warbler, hooded warbler, Acadian flycatcher (*Empidonax virescens*), cerulean warbler, red-eyed vireos (*Vireo olivaceus*), and Swainson's thrush (*Catharus ustulatus*) families all move to stands with greater structural complexity following fledging [1,2,11–15]. High mortality rates are associated with this movement as young birds seek better-suited post-fledging

habitat [6,7]. Therefore, full assessment of breeding habitat quality requires evaluation of post-fledging habitat selection and survival [16].

The rusty blackbird (*Euphagus carolinus*) is a continental migrant that breeds primarily in the boreal forests of Canada and Alaska, with populations in Acadian forests of northern New England and eastern Canada [17] and northern New York [18,19]. The species has experienced an estimated 85–95% continent-wide population decline with accelerated decline rates beginning in the early 1970s [20–24]. It has been declared a species of conservation concern by the United States Fish and Wildlife Service [25] and is considered vulnerable to extinction by the International Union for the Conservation of Nature [26]. Reasons for the decline remain unclear despite extensive study on both the breeding and wintering grounds. However, habitat use and survivorship during the post-fledging period have not been studied.

Nest success is relatively high where studied in Alaska (x = 0.56 [27]) and New England (x = 0.62 [28], x = 0.53 [29]), and rusty blackbirds do not seem to be suffering from chronically low rates of nest success. Rusty blackbird nest success estimates are much higher than the 0.30–0.39 estimated for red-winged (*Agelaius phoeniceus*), yellow-headed (*Xanthocephalus xanthocephalus*), and Brewer's (*Euphagus cyanocephalus*) blackbirds, which have not experienced dramatic declines [30].

Rusty blackbirds typically nest in young spruce-fir stands with high stem densities and forage for aquatic invertebrates such as Ephemeroptera, Odonata, Plecoptera, Diptera, and Tricoptera [17,31] in shallow wetlands or along stream edges [17,28,29,31]. Previous studies have identified percent cover of young softwood stands to be the most important variable selected for nest sites at multiple scales and associated with higher survival [27–29]. However, wetlands were not selected for nesting at the 5 m or 500 m radius scale even though the species is well-known as a wetland obligate [17].

Understanding the roles of both wetlands and young softwood stands could have important consequences for research and management. Dense spruce-fir sapling stands are selected for concealment during nesting and the first few days after fledging. Similar to many songbird species, young rusty blackbird fledglings fly poorly and lack the ability to escape predation in the first week or so after leaving the nest [8]. Before they become capable of sustained flight, fledglings spend much of their time on or near the ground, where they are easy prey for predators [2,3,7,10,32]. As energetic demands increase [33] and fledglings become more mobile, proximity to wetlands and streams with abundant and diverse invertebrate resources becomes important as fledglings learn to forage independently and fulfill their high energetic demands [6,7,31,33].

Our objectives were to 1) determine what forest stand types are important to rusty blackbirds during nesting and post-fledging, 2) assess nest site and post-fledging habitat selection, 3) analyze the influence of selected habitat characteristics on nest success and post-fledging survivorship of fledglings and adults, 4) provide forest management recommendations for incorporating post-fledging habitat into management plans for rusty blackbirds, and 5) advocate a paradigm shift from wetland-centric to landscape-centric for future rusty blackbird researchers in New England, New York, and Maritime Canada.

#### **2. Materials and Methods**

#### *2.1. Study Area*

We studied rusty blackbirds during 2010–2012 in the upper Androscoggin River watershed in northern New Hampshire. Ecosystem classification places this watershed within the White Mountains Section of the New England-Adirondack Province [34]. The landscape is mountainous, with most of the area at elevations between 460 to 800 m.a.s.l., valleys between 305 to 460 m.a.s.l., and a few peaks and ridges exceeding 800 m.a.s.l. [35]. Surface waters include the 6th order Androscoggin and Magalloway rivers, 4th order Clear Stream and Swift Diamond River, numerous lower order streams, Umbagog Lake (3177 ha), and several ponds of 10 to 125 ha. Climatic conditions include cold winters and warm summers, with mean monthly lows ranging from −15 ◦C to 13 ◦C and highs from −3.3 ◦C to

25.6 ◦C in January and July, respectively; mean annual precipitation is 105 cm, with a monthly means ranging from 5.9 cm in February to 11.1 cm in October, and 198 cm of snowfall [36].

The area is heavily forested in a patchwork of northern hardwood, Acadian spruce-fir, and mixed northern hardwood-spruce-fir. Balsam fir (*Abies balsamea*) and red spruce (*Picea rubens*) dominate softwood stands; eastern white pine (*Pinus strobus*), black spruce (*Picea mariana*), northern white cedar (*Thuja occidentalis*), and tamarack (*Larix laricina*) occur at lower densities. Major hardwood species include sugar maple (*Acer saccarum*), red maple (*Acer rubrum*), yellow birch (*Betula alleghaniensis*), white birch (*Betula papyrifera*), quaking aspen (*Populus tremuloides*), bigtooth aspen (*Populus grandidentata*), balsam poplar (*Populus balsamifera*), American beech (*Fagus grandifolia*), black cherry (*Prunus serotina*), pin cherry (*Prunus pensylvanica*), American mountain-ash (*Sorbus americana*), striped maple (*Acer pensylvanicum*), white ash (*Fraxinus americana*), and black ash (*Fraxinus nigra*). Primary stand disturbance agents include timber harvesting, wind throw, breakage from ice and snow loads, beavers (*Castor canadensis*), and insect outbreaks, notably spruce budworm (*Choristoneura* spp.) [35].

We conducted our study in three drainages: 1) Swift Diamond River valley (SWDI), 2) Mollidgewock (MOLL), and 3) Interior (INTE). SWDI was located on Wagner Forest Management Ltd. lands (hereafter Wagner) and INTE and MOLL were located on the Umbagog National Wildlife Refuge (hereafter Umbagog; Figure 1).

**Figure 1.** Drainages and their locations relative to the nearest township of Errol, Coos County, New Hampshire within the USA. Symbols indicate approximate site locations from 2010−2012 and include Swift Diamond (SWDI), Interior (INTE), and Mollidgewock (MOLL).

#### *2.2. Field Procedures*

During our pilot study in 2010, we conducted presence-absence surveys within 100 m of wetlands by observing passively for 3 min, broadcasting male rusty blackbird songs and calls for 38 s, and observing passively for another 5 min [28]. We discovered that these surveys failed to detect previously located rusty blackbird pairs that were nesting > 100 m from wetlands. Thus, we changed our survey protocol for 2011 and 2012 to include regenerating softwood stands up to approximately 0.5 km from wetlands during 30 min to 2 h passive surveys [27]. We identified survey locations using a combination of forest industry maps, topographic maps, and Google Earth imagery. We began nest-searching in early May in areas of rusty blackbird activity and continued through mid-June. Once located, nests were monitored approximately every 5 days until they either fledged young or failed. When nests failed during incubation, we attempted to locate re-nests.

To collect spatial and survival data on fledgling and adult rusty blackbirds, we deployed radio-transmitters on a subset of nestling and adult rusty blackbirds captured at nest sites; not all nests received transmitters. We attached VHF transmitters to nestlings when they were approximately 7–10 days old. We captured adult rusty blackbirds near active nests using 60-mm mesh size, 6 m long mist nests, and a broadcast call. We used transmitters from Blackburn Transmitters (Nacogdoches, TX, USA), Advanced Telemetry Systems (Isanti, MN, USA), and Holohil Systems, Ltd. (Carp, ONT, CA). Battery life ranged from 30 days in 2010 to 60 days in 2012. Transmitters were attached via a synsacrum harness with a degradable 1 mm stretchy jelly cord [37]. In 2012, we designed a harness with a weak link because birds returned with harnesses from previous years that had become embedded in their skin.

Transmitter weight with harness varied from 1.8% (70 g bird) to 2.3% (55 g bird) of adult bird mass. Transmitters on nestlings weighed from 5.6 to 8% of mass at time of attachment but were lower than adult ratios once nestlings fledged because transmitters used on fledglings were lighter (0.9 g) than those for adults (2.7 g). Harnesses were fitted loosely on nestlings and were the largest size to enable growth. Each adult and nestling blackbird received a US Geological Survey (USGS) federal band and adults received a unique color-band combination for identification (USGS BBL Permit # 22665). This study plan was approved by the University of Georgia IACUC (AUP# A2009 1-003).

Observers located tagged rusty blackbirds 3–5 times per week using ATS R2000 and R2100 receivers in vehicles with roof-mounted dipole antennas and hand-held R410 receivers with three-element Yagi antennas. In > 94% of locations, we determined coordinates with a Garmin GPS by following a radio signal until we saw or heard the bird. In remaining cases, we used triangulation of bearings taken from multiple points on logging roads. We conducted observations between sunrise and sunset and avoided periods of excessive wind and rain.

#### *2.3. Data Analyses*

#### 2.3.1. Habitat Selection

We used binomial logistic regression to determine habitat characteristics that rusty blackbirds used out of proportion to availability [38]. We conducted separate analyses for nests, fledgling, and adult telemetry locations. Nests or telemetry locations were considered use points and we generated random availability points within rusty blackbird areas of use with ARCMAP 10.4. We then assigned a value of 1 to use points and 0 to availability points to be used as response variables. If multiple tagged birds were documented at the same location at the same time, we only used a location once for a use point and prioritized triangulated points for removal.

To generate availability points, we created a 90% kernel density background for each of the three analyses and the three drainages separately. For the post-fledging analysis, the background area of use included area outside nesting sites because fledglings moved increasing distances (>1 km) from the nest as they became more mobile. We generated roughly the same number of availability points as use points. We constrained availability points to be at least 50 m apart for telemetry data and 300 m apart for nests. Availability points within large bodies of open water or other unsuitable habitats (e.g., paved road surfaces, buildings) were deleted and not replaced.

We created predictor habitat variables based on published literature [24–26,38], our field observations, habitat characteristics used in our revised survey protocol, and relevance to forest management. We generated predictor variables for SWDI from layers we created in ARCMAP 10.4 using forest stand maps (Wagner), and soil, river, and wetland maps from New Hampshire Geographically Referenced Analysis and Information Transfer System (NH GRANIT, hereafter GRANIT; Table 1). For MOLL and INTE, where forest stand maps were unavailable, we used aerial photography to digitize forest stand types within a 30 m radius around use and paired availability points for each habitat variable for every other telemetry point/bird/day.


**Table 1.** Descriptions of habitat variables used for generating logistic regression models for rusty blackbird nests and post-fledging adult and fledgling locations in northern New Hampshire from 2010−2012. Buffer is a 30 m radius circle around location points.

<sup>1</sup> swsaplings were stands <10 yrs old or with saplings 1.3–14 cm DBH.

Once we created predictor variables, we extracted raster data for every other point/day/bird (due to time and computing constraints and to match Umbagog data) for the 13 variables (Table 1). We calculated values for predictor variables as the mean of raster cell values in a 30 m radius buffer around use and availability points. The 30 m scale is roughly an order of magnitude between the 5 and 500 m scales used as buffers for variables in previous research [28,29]. To summarize variables, we calculated the mean and standard deviation of use and availability points for nests and blackbird locations.

We used generalized linear mixed models (GLMM; R package lme4) in program R 3.4.3 [39] to run logistic nest site and post-fledging habitat selection analyses [40]. We determined Pearson correlations on all pairs of predictor variables prior to modeling and avoided including variables correlated at r > 0.40 in the same model. We used Akaike information criterion (AICc) to compare model fit between nested models with and without year as a random effect and determined that year did not account for significant variation. For post-fledging telemetry data, we included individual birds nested within drainage (SWDI, INTE, MOLL) as a random effect in each model to control for spatial autocorrelation [40]. Availability points were assigned to individual birds based on location. We did not consider pseudoreplication to be a problem because blackbirds did not maintain a consistent home range post-fledging. We did not include fledglings and adults from the same nest in the same analysis, randomly omitting one.

We used model selection to evaluate the relative plausibility of habitat selection models with combinations of the 13 predictor variables [41,42]. To avoid over-fitting models with parameters, we first reduced our set of 13 variables by comparing correlated sets individually with the second-order AICc. We retained variables for further analysis that had the lowest AICc scores and were better than the null model. We included one stream, wetland, and forest stand type variable in a model at a time to avoid correlation and then compared 17 candidate models. The top habitat selection models for males and females included the same variables with similar weights, and parameter estimates were similar, so we pooled males and females into adults for habitat selection analyses.

We assessed the relative fit of each candidate model by calculating Akaike weights (ωi; [42]. We report a confidence set of models that included only those candidate models with Akaike weights that were within 10% of the largest weight [43] for evaluating strength of evidence. If more than 3 models were within 10% of the model with the greatest ωi, we report the subset of models with ωi > 0.01. To choose the best model in analyses that generated several top models, we chose the model with the fewest parameters within ΔAICc < 2 of the top model. We chose the top model if the level of support (ωi > 0.6) indicated that it was far better than other models.

We used the chi-square G statistic to evaluate the strength of our top models in explaining goodness-of-fit (*p* < 0.05; [44]). We assessed the predictive ability of our top models by estimating the Area Under the Curve (AUC) of the Receiver Operating Characteristic (ROC) curve, i.e., the rate of true positives (sensitivity) to false positives (1-specificity). We plotted ROC curves with the R package "pROC". We used an AUC ≥ 0.85 to indicate good predictive ability of our models.

We assessed the precision of model-averaged coefficients by calculating 95% confidence intervals. Confidence intervals including 1 indicated inconclusive estimates because we could not determine the nature of the relationships (i.e., whether positive or negative) due to imprecision in parameter estimates [45]. We report scaled, model-averaged coefficients from the top models (ωi > 0.6). The scalars correspond to biologically relevant unit changes in predictors.

#### 2.3.2. Nest Success and Adult and Fledgling Survivorship

We used the R package RMark 2.2.4 [46] to estimate daily survival for nests, fledglings, and adults. We used nest success models appropriate for use with telemetry data when data are not collected at a consistent time interval and the exact date of fates are not known [47,48]. For nest success, we considered the first day we observed eggs, incubation behavior, or nestlings to be the date the nest was found. We assumed a 29-day exposure time when calculating nest success; 5 days for laying, 12 days for incubation, and 12 days for nestlings [28]. We considered the date of fledging to be the "date found" in RMark for both adults and fledglings. Tagged nestlings that were depredated before they fledged were not included.

Covariates tested included variation in survival among years (2010, 2011, and 2012), among drainages (SWDI, MOLL, INTE), and with number of days since nest initiation (nests). To determine fledgling survival in the first week, we assessed seasonal survival of fledglings (season). We did not include adult age as a covariate because only two adults could be aged beyond the after-hatch-year class. We included habitat variables from top models from nest and post-fledging habitat selection analyses. We used overall means to interpolate habitat variables for 9 blackbirds that were missing habitat data to enable daily survival rate calculations for those individuals. To account for classes with no mortality, i.e., survivorship = 1 and variance cannot be estimated, we added a fictitious bird to the dataset in that class and assumed the bird died on the last occasion [28]. We assigned mean habitat covariate values to this bird. We calculated the mean of all use locations for each habitat variable to include in the corresponding survival record for each bird and assumed the mean was representative of variables leading up to predation or survival. We used a model selection approach to determine the best model for nest success and blackbird survival, such as the methods described above for habitat selection.

#### **3. Results**

#### *3.1. General*

We found 59 rusty blackbird nests during the 2010–2012 breeding seasons, the majority of which (92%) were in previously clear-cut, regenerating softwood-dominated stands 5–15 years post-harvest. These stands averaged 5.4 ha and ranged from 0.36–13 ha. Of the 59 nests in our study, 31 (53%) were within 75 m and 10 (17%) were > 200 m from a stream or wetland.

We used 37 nest locations that were > 100 m apart and generated 57 random availability points to use in logistic regression. We attached 36 radio-transmitters to adults (13 female, 23 male) and 25 to nestlings. After removing adult birds tagged from the same nests, we were left with points from 11 females and 18 males. Of the 25 nestlings tagged, 8 died before fledging. In the seven cases where two nestlings from a nest had transmitters, one of the two was omitted from the fledgling analysis, leaving 10 fledglings in the habitat selection analysis. We documented a total of 318 use locations for 20 blackbirds in 2010, 993 use locations for 24 birds in 2011, and 626 use locations for 17 birds in 2012 (Appendix A). We omitted location points for one female who died from apparently capture-related causes. We omitted 70 location points of adults and fledglings documented when fledglings were still

in the nest to eliminate data from the nestling period. After refining the telemetry data, 771 use points remained, and we generated 806 availability points.

New fledglings flew weakly and most remained within 100 m of the nest for the first few days after fledging. Once fledglings moved away from the nesting area, we observed pairs dividing their fledglings between them, joining other family groups, and feeding fledglings of other pairs. Most fledglings depended on adults for food for about 3 weeks. Some fledglings traveled alone after becoming independent, while others remained in family groups or joined up with fledglings from their own or other families. During our study, all tagged individuals that fledged before 8 June had been detected more than 1 km from the nest by the end of the month.

#### *3.2. Habitat Selection*

The top model for habitat selection included similar variables for all 3 groups tested (i.e., nest sites, fledglings, and adults): distance to streams, proportion of softwood/mixed-wood sapling stands, wetlands, and low slope class (Table 2). In each case, the top model received strong support (ωi > 0.62; Table 2). However, parameter estimates differed substantially among the groups, justifying not pooling adults and fledglings in the same analysis (Table 3).

**Table 2.** High ranking candidate models for rusty blackbird **nesting**, **fledgling**, and **adult** habitat selection in Northern New Hampshire from 2010−2012. k = number parameters, AICc = Akaike information criterion for small sample sizes, ΔAICc is the difference between the model and top model, ωi = weight, cum., ωi = cumulative weight, and LL = log likelihood. See Table 1 for variable definitions.


Distance to 1st–6th order streams was an important variable in all 3 analyses, and had the largest effect for fledglings (Figure 2, Table 3). Fledglings were 4.6 times more likely to occur with every 50 m decrease in distance from streams, compared to 1.2 times more likely for adults and 1.3 times for nest sites. Nests, fledglings, and adults were all 1.2 times more likely to occur with every 10% increase in softwood/mixed-wood sapling stands. Only nests were influenced by low slope and were 1.1 times more likely to be observed with a 10% increase in area with low slope. While the proportion of wetlands in an area was important for fledgling and adult habitat use, (blackbirds were 1.3 times more likely with each 10% increase), wetlands were not important for nest-site selection (Figure 2).

**Table 3.** Adjusted, scaled back-transformed log-odds estimates for rusty blackbird **fledglings** (n = 10), **adults** (n = 29), and **nests** (n = 37) in New Hampshire from 2010–2012 including estimates, SE, and 95% upper and lower confidence intervals. Scaled log odds estimate reads: it is 1.3 times more likely a rusty blackbird will be present with every 10% increase in the proportion of sapling stands in a 30 m radius buffer. Confidence limits including 1 indicate no effect and are denoted "–".


**Figure 2.** Mean and 95% CI of variables from top models of (**a**) nest use (black circles; n = 37) relative to availability (white circles; n = 57), (**b**) fledgling habitat use (black circles; n = 198) relative to availability (white circles; n = 196), and (**c**) adult use (black circles; n = 537) relative to availability (white circles; n = 609) in N. New Hampshire from 2010–2012. Proportion of lowslope, wetlands, and softwood/mixed-wood saplings are on the left *y*-axis, and distance (m) of points from allstreams is on the right *x*-axis. Lowslope = % of 30 m buffer in soil class poorly or very poorly drained, wetlands = % of 30 m buffer in any vegetated palustrine wetland, saplings = % of 30 m buffer in softwood/mixed-wood sapling stands, and allstreams = mean distance of buffer to 1st–6th order streams.

#### *3.3. Nest Success*

We included 37 nests with known outcomes that were > 100 m apart in the nest success analysis. The null model for nest survival received most of the support (Table 4); candidate models including year, drainage, and trend through time (season) received little support (Table 4). Daily nest survival for all years combined was 0.975, SE = 0.26, resulting in an overall nest success rate with 29-day exposure of 0.48 (Table 4). Although the top model for nest success included wetlands (Table 4), the confidence limits for the estimate of this variable included 0. We therefore chose the intercept only model as the top model and report estimates from this model (Table 5).

**Table 4.** High-ranking candidate models for rusty blackbird **fledgling**, **adult**, and **nest** survival in northern New Hampshire. k = number of parameters, AICc = Akaike information criterion adjusted for small sample sizes, ΔAICc is the difference between the model and the top model, ωi = weight, and cum. ωi = cumulative weight; see Table 1 for habitat variable descriptions. nestage is the age of the nest when first found, S = survival, time = time from first day of study, hy = fledgling, SWDI = Swift Diamond drainage.


**Table 5.** Parameter estimates, standard error (SE), and lower (LCI) and upper (UCI) confidence intervals for daily survivorship and overall survival; 29 days for nests (n = 37), and 60 days for fledglings (n = 18) and adult male (n = 20) and female (n = 11) rusty blackbird survival in northern New Hampshire from 2010–2012.


<sup>1</sup> Female survivorship was estimated with a hypothetical record. See text for details.

Goodness-of-fit values for top habitat selection models were very close to or exceeded our threshold of 0.85 for nest sites (AUC = 0.87), fledglings (AUC = 0.87), and adults (AUC = 0.84). The Hosmer–Lemeshow test for nests (chi-squared = 9.0, *p* = 0.3) and fledglings (chi-squared = 15.8, *p* = 0.051) indicated no lack of fit. However, tests for adults indicated a difference between observed and expected proportions (chi squared = 28.1, *p* ≤ 0.001). In the adult analysis, confidence intervals for estimates of two variables in the top model (softwood/mixed-wood sapling stands and low slope)

overlapped one and thus were not useful for prediction; estimates for distance to streams in the nest analysis and for low slope in the fledgling analysis were similarly ambiguous (Table 3). The simplified adult model improved the chi-squared to 18.8, *p* = 0.02. Normal probability and residual plots indicated that all models satisfied assumptions of normality.

#### *3.4. Blackbird Survival*

Of the 18 fledglings, 20 adult males, and 11 adult females with transmitters in the survival analysis, 7 fledglings (39%), 5 males (25%), and 0 females succumbed to mortality during the life of their transmitter. Only one fledgling and one adult blackbird failed to survive the first week after fledging. There were no differences between years, drainages, or seasonal survival. Because the season variable was not important for explaining variation in survival of fledglings, including the period one week after fledging, males, females, and fledglings were analyzed together using age/sex dummy covariates; fledgling sex was unknown. The top model for blackbird survivorship included age and sex categories, with fledglings having the lowest survival rates (0.988, SE = 0.34), males with 0.996, SE = 0.44, and females with 0.998, SE = 0.63 (Table 5).

#### **4. Discussion**

#### *4.1. Survival and Habitat Selection*

Rusty blackbirds in our study selected (at a 30 m radius scale) a combination of habitat conditions for nesting and post-fledging, including shallow wetlands and low slope with softwood/mixed-wood sapling stands. This was the first study to assess the importance of distance to 1st to 6th order streams and found that this feature was selected for both nesting and post-fledging. Availability of streams in addition to wetlands may increase foraging opportunities because streams and wetlands have different availability of prey types throughout the season [31,49–52]. Early nestling development may be synchronized with the availability of small items typical of streams, such as Ephemeroptera, Plecoptera, and Trichoptera larvae, while dependent fledglings may require larger Odonata larvae, which are more abundant in wetlands (P. Wohner unpublished data). A diverse network of interconnected channels and impoundments, such as those created by beavers, may provide ideal foraging opportunities for rusty blackbirds [52].

We found that while habitat selection was similar between nest sites and post-fledging locations at the 30 m scale, the stages had one important difference: wetland cover. Likewise, in a study that included the same drainages, rusty blackbirds did not select wetlands for nesting at the 5 m scale [29]. At the 500 m scale, rusty blackbirds were reported more likely to select nesting habitat with increases in wetland cover in New Hampshire [29]. However, the confidence interval for the wetland estimate ranged from −2 to 13, overlapping 1, and is therefore inconclusive [29]. We concur that rusty blackbirds select nest sites independent of wetland cover at the 5, 30, or 500 m scale. However, wetlands are important to adults and fledglings at the 30 m scale. Concealment from predators may be more important than proximity to foraging areas during nesting and early post-fledging, while foraging opportunities quickly become a priority after fledging [9].

It is likely that proximity to wetlands is important for nest-site selection at a larger scale than has been studied thus far, i.e., >500 m [53]. Rusty blackbirds are highly mobile, traveling up to 2 km between wetlands and nesting areas, and have large home ranges (3.8–172.8 ha in Maine, equivalent to a circle with radius 35–741 m [54]). It is possible that cues for selection of nesting and foraging habitat are decoupled at different spatial scales, and rusty blackbirds select wetlands at a 1–5 km scale for foraging, and nesting habitat at a much smaller scale (0.05–0.5 km [53]). We propose that future nesting habitat selection studies that use multiple spatial scales including the 1–5 km scale would help to determine the density and size of wetlands important for rusty blackbirds at the home range scale and provide better predictive ability for rusty blackbirds [55].

Finding the appropriate scale could have important consequences for research and management [55]. For example, many surveys for nesting rusty blackbirds have been wetland-centric and typically only detect songbirds within 100–200 m of a point. In our study, 35 percent of nests were located >200 m from the nearest wetland. If rusty blackbird pairs forage at multiple wetlands and streams and nest in relatively distant uplands, wetland-based surveys likely underestimate occupancy and may result in biases [56]. The low detectability documented in other studies that only surveyed wetlands could be a consequence of survey design [28]. Other researchers have noted that point count-based surveys are not well suited for estimating the abundance or habitat use of breeding rusty blackbirds, even when broadcast calls are used [19].

Our nest success estimates (0.48) are well within the range reported by other rusty blackbird researchers (0.21–0.75; [27–29]) and are much higher than those reported for red-winged, yellow-headed, and Brewer's blackbirds (0.30–0.39; [30,57]). Regenerating clear-cuts were suspected ecological traps for nesting in rusty blackbirds in Maine [28]. However, we found evidence that regenerating clear-cuts support successful nesting by rusty blackbirds. Although we did not explicitly test stand origin, nest success did not decrease with increases in softwood sapling stands, 92% of which originated from even-aged forest management. This result is consistent with other research that found rusty blackbird nest success was independent of recent harvesting [26]. We agree with other researchers that nest success is unlikely to be limiting rusty blackbird populations [24,26].

As in other New England rusty blackbird breeding studies, where nest selection was assessed at 5 and 500 m radius scales [27,29], rusty blackbirds in our study selected young (5–15 years old) or stunted softwood/mixed wood stands at the 30 m radius scale for nesting. Many of the New England nesting stands were created by even-aged management and are composed of regenerating softwoods. For example, 75% of nests in Maine [28], 88% in Maine and New Hampshire [29], and 92% in this study were located in young, even-aged softwood stands. While rusty blackbirds seem to have an affinity for regenerating clear-cuts at the 5, 30, and 500 m scales, there may be a threshold for increasing proportions of regenerating softwood stands at larger scales than 500 m. Blackpoll warblers (*Setophaga striata*) were found to be positively associated with large proportions of clear-cut at the 115 m scale [6]. However, at the 1250 m scale, they were positively associated with clear-cuts only when < 5% of cover was clear-cut, above which the relationship was negative [6]. Thus, while even-aged forest management may be beneficial for rusty blackbirds at smaller scales, e.g., 5, 30, and 500 m radius, researchers should consider a scale larger than 500 m for future studies.

We observed low mortality rates for rusty blackbird fledglings in the first week after fledging (3%) compared to those reported for other songbirds, e.g., 21–81% for ovenbird [2], 11% for worm-eating warbler [8], and 90% for rose-breasted grosbeak (*Pheucticus ludovicianus*) fledglings [7]. All observed fledgling mortality took place in the first four days of a yellow warbler (*Setophaga petechi*) study [58]. Our high rusty blackbird fledgling survival contrasts with virtually every other songbird post-fledging study. We suggest that dense regenerating softwood stands approximately 5–15 years post-harvest, may afford < 1-week-old fledglings protection from potential predators while they are most vulnerable. Fledglings of songbirds that nest in mature hardwood stands with relatively open understories have sparse protective cover before moving to early successional habitats with dense vegetation [1,2,6,12]. Our overall fledgling survival during the 60-day post-fledging period was also relatively high (0.49) and is likely not an overwhelming factor contributing to population decline. However, we did not study survivorship after fledglings were completely independent from adults, nor during migration when young birds could experience substantial mortality [9]. Populations of fledgling barn swallows (*Hirundo rustica erythrogaster*) were found to be limited by the pre-migration phase [9].

While few studies have estimated adult survival during either nesting or post-fledging, adult barn swallow survival over 60 days was 0.92, SE = 0.11 for males and females together [59], compared to our estimates of 0.89 for females and 0.79 for males over the same time frame. Rusty blackbird male survival may truly be relatively low during the post-fledging period. However, the confidence intervals around our estimates are wide (0.54–1), likely due to small sample sizes. Transmitter harnesses may reduce survival, resulting in a survival estimate that is lower than actual survival in the general population. We regularly observed rusty blackbirds picking at their harnesses which could distract birds and expose them to higher predation. We have also recaptured blackbirds with transmitters embedded in their skin. Transmitter harnesses have been found to affect survival in sensitive species such as pileated woodpeckers (*Dryocopus pileatus*; [60]). Why harnesses would have a disproportionate effect on males over females is unknown but could be due to morphological differences between the sexes. High mortality of adult rusty blackbird males during the breeding season seems unusual, but if true, could itself be a key factor in population decline. Male survival during the breeding season warrants future study.

#### *4.2. Recommendations*

Nest success and adult and fledgling survival were not affected by any of the habitat variables we analyzed, which suggests that something other than breeding ground habitat may be limiting rusty blackbird populations. Thus, we expect that current forest management practices continue to create suitable landscapes for successful nesting and post-breeding survival. Our study adds support to a study in Nova Scotia, that found rusty blackbird habitat remains relatively abundant and well-distributed and is often located in wet lowlands which is a climate-resilient topographic landform [61]. Targeted habitat management for rusty blackbirds is likely to be unnecessary in many areas due to the species' use of regenerating softwood stands that are created by a variety of harvesting practices. We do recommend that harvest plans ensure the availability of at least one softwood stand 5 to 15 years post-harvest within 300 m of streams and shallow wetlands over time. In the Acadian forest, on ownerships where wildlife habitat is the focus of forest management, prioritizing harvests on poorly drained sites where trees grow more slowly could provide rusty blackbird nesting habitat for longer periods of time (≥40 years) [62]).

To aid in finding high priority areas for research or conservation, we recommend overlaying the regression equation from our top models for nests and fledglings in the raster calculator in ARCMAP (i.e., for nests: −0.87–5.2 (distance to streams) + 0.01 (prop slope) + 0.02 (prop softwood/mixed-wood sapling stands). The nests and fledgling models had high goodness of fit, and together, could identify rusty blackbird habitat at the landscape level. We expect our regression equations will be applicable in the southeastern portion of the rusty blackbird's range in Acadian Forest, i.e., New Brunswick, Nova Scotia, northern New England, and the Adirondacks.

Finally, we hope that rusty blackbird researchers will move from traditional survey protocols [63,64] like presence-absence surveys to protocols such as those used for western yellow-billed cuckoos (*Coccyzus americanus occidentalis*), which also have large home ranges (16–91 ha) [65,66]. These surveys use broadcast calls every 100 m along transects in appropriate habitat [67,68]. (Spring surveys to locate pairs prior to nest searching for intensive research should avoid using broadcast calls, however). Future rusty blackbird research on the breeding ground should incorporate a landscape perspective and include multiple habitat types, including but not limited to dense young softwood stands, streams, seepages, and wetlands. Studying the species at a much larger scale than previously (e.g., >500m), may shed new light on wetland requirements during nesting.

**Author Contributions:** Conceptualization, C.R.F. and P.J.W.; methodology, C.R.F. and P.J.W.; formal analysis, P.J.W. and R.J.C.; investigation, C.R.F. and P.J.W.; resources, C.R.F. and R.J.C.; data curation, P.J.W. and C.R.F.; writing—original draft preparation, P.J.W.; writing—review and editing, P.J.W., C.R.F., R.J.C.; visualization, C.R.F. and P.J.W.; supervision, C.R.F.; project administration, C.R.F. and R.J.C.; funding acquisition, C.R.F., R.J.C., P.J.W. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by The US Fish and Wildlife Service, New Hampshire Audubon, The Eastern Bird Banding Association, The Smithsonian Institution, The Natural Sciences and Engineering Research Council of Canada, and The University of Georgia.

**Acknowledgments:** A special thanks to all field assistants who spent considerable time in the clear-cuts and wetlands of New Hampshire from 2010−2012. They include S. Hribal, E. Prohl, R. Rabinovitz, H. Batcheller, E. Dancer, C. Ross, and J. Cosentino. Other supporters include those who granted access to rusty blackbirds on

their lands including Umbagog National Wildlife Refuge, Wagner Forest Management, Conner, and Plum Creek. S. Edmonds and L. Powell contributed field expertise, on-the-ground assistance, and advice. Members of the International Rusty Blackbird Technical Working Group including L. Powell, S. Edmonds, S. Matsuoka, and D. Tessler provided technical advice on rusty blackbird surveying, nest searching, and capture. V. Jones from New Hampshire Audubon and N. Nibbelink from the University of Georgia assisted with GIS spatial analysis.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **Appendix A**

Summary of rusty blackbirds tagged by year and drainage in northern New Hampshire, USA from 2010−2012; includes total birds and number of points collected, and the subset of blackbirds and telemetry points retained in the analysis by adults (M = male and F = female) and fledglings. SWDI = Swift Diamond, MOLL = Mollidgewock, and INTE = Interior.


#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Flock Size Predicts Niche Breadth and Focal Wintering Regions for a Rapidly Declining Boreal-Breeding Passerine, the Rusty Blackbird**

**Brian S. Evans 1,2, Luke L. Powell 1,3,4,\*, Dean W. Demarest 5, Sinéad M. Borchert <sup>6</sup> and Russell S. Greenberg 1,†**


**Abstract:** Once exceptionally abundant, the Rusty Blackbird (*Euphagus carolinus*) has declined precipitously over at least the last century. The species breeds across the Boreal forest, where it is so thinly distributed across such remote areas that it is extremely challenging to monitor or research, hindering informed conservation. As such, we employed a targeted citizen science effort on the species' wintering grounds in the more (human) populated southeast United States: the Rusty Blackbird Winter Blitz. Using a MaxEnt machine learning framework, we modeled patterns of occurrence of small, medium, and large flocks (<20, 20–99, and >99 individuals, respectively) in environmental space using both Blitz and eBird data. Our primary objective was to determine environmental variables that best predict Rusty Blackbird occurrence, with emphasis on (1) examining differences in key environmental predictors across flock sizes, (2) testing whether environmental niche breadth decreased with flock size, and (3) identifying regions with higher predicted occurrence (hotspots). The distribution of flocks varied across environmental predictors, with average minimum temperature (~2 ◦C for medium and large flocks) and proportional coverage of floodplain forest having the largest influence on occurrence. Environmental niche breadth decreased with increasing flock size, suggesting an increasingly restrictive range of environmental conditions capable of supporting larger flocks. We identified large hotspots in floodplain forests in the Lower Mississippi Alluvial Valley, the South Atlantic Coastal Plain, and the Black Belt Prairie.

**Keywords:** Black Belt Prairie; citizen science; conservation; machine learning; niche modeling; group size; habitat use; species distribution models

#### **1. Introduction**

The Rusty Blackbird (*Euphagus carolinus*) is a Nearctic icterid that exhibits near exclusive reliance on forested wetland habitats in boreal, transitional and deciduous biomes throughout its annual cycle [1–3]. The species is among the most steadily and precipitously declining temperate North American landbirds, [4–6] and is recognized as "Vulnerable" by the International Union for the Conservation of Nature. There is some anecdotal evidence that factors in boreal breeding regions of northeast North America may be limiting the population (timber management and nest success: [7]; mercury: [8]). However, demographic rates for Rusty Blackbirds across the Boreal are similar to those of other songbirds [1]), suggesting that the overall population is likely limited by factors outside the breeding grounds.

**Citation:** Evans, B.S.; Powell, L.L.; Demarest, D.W.; Borchert, S.M.; Greenberg, R.S. Flock Size Predicts Niche Breadth and Focal Wintering Regions for a Rapidly Declining Boreal-Breeding Passerine, the Rusty Blackbird. *Diversity* **2021**, *13*, 62. https://doi.org/10.3390/d13020062

Academic Editor: Michael Wink Received: 30 November 2020 Accepted: 29 January 2021 Published: 4 February 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

Specifically, habitat modification on the species' wintering grounds has been implicated as a likely driver of the decline [5,9], stressing the importance of studying this period of the species' life cycle. Further, population monitoring and broad assessments of habitat selection across the boreal breeding range are extremely challenging given the general remoteness and the very low density of roads, humans, and Rusty Blackbirds themselves. The Breeding Bird Survey, for example, detects so few Rusty Blackbirds that with the exception of more developed parts of Alaska it is not possible to use these data to monitor population trends. All of these factors coupled with the species' considerably more social (flocking) nature on the wintering grounds and the high densities of citizen scientists (birders) in the southeast United States point to the advantages of studying the species during winter. The wintering grounds are not without their challenges, however; diffuse and localized distribution, nomadism, and difficulty in accessing often ephemeral wetland habitats constrain our ability to dependably study wintering Rusty Blackbirds [9]. Conventional means of investigation, such as use of mark-recapture (e.g., to evaluate survival as a function of winter habitat metrics), may not be viable due to high vagility and low site fidelity between years (e.g., [10,11].

Once reported in wintering flocks numbering tens of thousands [4], Rusty Blackbirds are now occasionally observed in flocks approaching a few thousand individuals, with much smaller groups being more common [12,13]. Still, flocking behavior in Rusty Blackbirds may provide a proximate cue allowing researchers to identify and characterize winter habitat (e.g., key sites, condition, relative quality or suitability for foraging). Foraging theory predicts that abundance and distribution of food resources influence group size, with more resource rich environments favorable to flocks of larger size (sensu [14–18]). Relationships between group size and resource availability have been shown in a wide variety of taxa, including primates (e.g., *Lemur catta* [19]), fish (e.g., *Fundulus diaphanous* [20]), and birds (e.g., *Junco phaeonotus* [21]). Where food resources are less available, interference competition can moderate flock size as individuals seek to optimize foraging efficiency beyond the confines of competitive groups [15]. Demographic studies in a number of bird species support that the size of conspecific aggregations is positively associated with higher quality habitats [22,23].

Because flock size is expected to vary as a function of habitat quality or compatibility, evaluating environmental features associated with flocks of different sizes may provide evidence for conditions and locations most favorable for greater numbers of wintering Rusty Blackbirds. Assessing distributions in environmental space, as opposed to geographic space, is necessary for the species as wintering individuals often travel long distances across the landscape, and thus site level habitat characteristics have not consistently predicted occupancy [2]. An understanding of such relationships could offer immediate relevance in informing conservation programs and facilitating further research and monitoring supporting full annual cycle stewardship of this species, as well as yield new insights regarding niche dynamics in flocks of different sizes.

Here, we evaluate the relationships of climatic and landcover attributes in predicting the occurrence of Rusty Blackbirds across their wintering range and ask whether these relationships vary with flock size. Previous studies have observed that, at local scales, large flocks of Rusty Blackbirds are constrained to a narrow set of environmental conditions and geographic areas, whereas smaller flocks are comparatively widespread (e.g., [12,24,25]). We sought to quantify these observations at the spatial extent of winter range of the species, predicting that (1) large flocks of Rusty Blackbirds (>100 observed individuals) occupy a different environmental niche than small or medium flocks (<20, 20–99, respectively) and (2) large flocks occupy a narrower environmental niche breadth than small or medium flocks [26]. We define niche breadth as the multidimensional range of specific biotic and abiotic conditions that a group of organisms occupies [27–29]. We used occurrence data from two citizen science programs, the Rusty Blackbird Winter Blitz (Blitz; http: //rustyblackbird.org/outreach/migration-blitz/) and eBird [30], to test these predictions by modeling the distributions of Rusty Blackbird flock size classes in environmental space. In a closely related project, we performed a three-year "Spring Blitz" for migrating Rusty Blackbirds, but those results are addressed elsewhere [31]. We compared biotic and abiotic attributes that best predicted occurrences of flock size classes and evaluated differences in environmental niche breadth among classes. We suggest focal regions for wintering Rusty Blackbirds (hotspots), defined as regions with higher predicted occurrence, and discuss implications to future research and conservation. Finally, given the proliferation of citizen science monitoring projects, we evaluated whether the Blitz, a targeted volunteer effort requiring considerable investment and coordination, yielded observational data that improved predictive strength of relationships relative to use of comparable eBird data alone.

#### **2. Methods**

We used a maximum entropy (MaxEnt) modeling approach to examine habitat suitability across flock size classes and observational methods. In species distribution modeling, habitat suitability is defined as the likelihood of occurrence of a species in association with environmental variables ([32], but see [33]). MaxEnt is a machine-learning method that compares occurrence data with that of background samples in environmental space (see Supplementary File S1 1.1; [34]). In this context, training samples represent subsets of locations in which Rusty Blackbirds were observed. Background samples represent subsets of locations that were sampled but no Rusty Blackbirds were observed.

We used Rusty Blackbird observations submitted from the Blitz, as well as those collected through traditional eBird [30] protocols. The Blitz was a coordinated effort among citizen scientists to search for Rusty Blackbirds across the species' wintering range by conducting traveling counts of 8 km or less during target dates of 1 January–28 February 2009–2011. Blitz observations were submitted using a special eBird portal (https://ebird.org/home) and recorded date, location, species, numbers, effort and other attributes (see [35]). To supplement Blitz data, we used traditional eBird checklists that reported Rusty Blackbirds on traveling counts conducted in 2009–2011 during corresponding Blitz periods [30]. We considered each eBird and Blitz checklist as an independent observation, omitting duplicates (i.e., observations submitted by multiple persons). We georeferenced observation locations with a 4-km resolution raster grid (see below). To avoid pseudoreplication, if a grid cell contained >1 independent observation we retained only the observation recording the highest Rusty Blackbird count.

We classified Rusty Blackbird observations by flock size (small: <20; medium: 20–99; large: >99 individuals), and sampling protocol (Blitz or eBird). Flock size classes were determined by the frequency of observations of a given number of individuals and approximate the lower (small flocks) and upper (large flocks) quantiles of observations within the combined pool of eBird and Blitz samples. While we use the term "flock" to designate the number of birds observed, both Blitz and traditional eBird observations were reported as traveling counts. Thus, it is possible that in some cases, aggregates designated as "large flocks" may be comprised of several small or medium-sized true flocks.

Given that flocks aggregate in common roosts at night and may be observed traveling in smaller groups during the day, we examined whether the time of observation biased our results. We determined the time of observation for each checklist used in our analysis (i.e., the maximum observed Rusty Blackbirds within a given raster cell). All time values were standardized to Coordinated Universal Time as a function of their geographic coordinates. Given that day length varies by geographic location and over the two-month period of this study, we used the R package *activity* [36] to transform the times of observation to solar times, which are defined here as radian time values relative to civil sunrise times for a given date and location [37,38]. We then fit Von Mises circular kernel density distributions to detections within each flock size class [36]. Following Ridout and Linkie [39], we calculated the degree of overlap (Δˆ 4) between fitted distributions and used randomization (with replacement; *n* = 10,000 iterations) to test the null hypothesis that detections come for the same distribution (α = 0.05). For each of the flock size class pairs, we failed to reject the

null hypothesis (File S1), providing supportive evidence that the time of day in which an observation occurred was unlikely to introduce bias into our results.

A limitation to occurrence-only modeling methods such as MaxEnt is that occurrence data, especially data collected opportunistically, are often biased toward areas of higher sampling effort (e.g., as a function of accessibility, convenience, or availability of human observers; see [40]. This confounds inferences regarding distribution, habitat use and environmental niche, and may cause inflated measures of model performance [41,42]. To mitigate this potential bias in our models, we generated background data by summarizing eBird checklists (2009–2011; *n* = 13,218 checklists) across each 4 km grid cell. In doing so, this method assumes that the sampling bias associated with checklists observing Rusty Blackbirds is similar to that of eBird checklists where Rusty Blackbirds were not observed [43,44]. By using random eBird sampling locations as background points, rather than random locations from the entire study extent, we were able to construct models that evaluate the occurrence of Rusty Blackbird observations in environmental space relative to points on the landscape presumed to be representative of any geographic biases in our Rusty Blackbird observations (see [45]). Additionally, we limited background point selection, and thus the extent of our distribution models [46], to the geographic extent that contained 99% of Rusty Blackbird observations, which corresponded to the land area of the Eastern United States, bounded in the west at a longitude of −100◦.

We constructed models with up to 14 environmental covariates to predict Rusty Blackbird habitat suitability, including two climatic and 12 land cover covariates (see Table S1). We used the *raster* package in Program R [47,48] for all environmental layer processing. We obtained mean precipitation and minimum temperature data for the months of January and February 2009–2011 (4-km resolution, [49]). Though there was considerable annual variation in average minimum temperature and precipitation within our study area, small annual sample sizes of Rusty Blackbird observations necessitated averaging these covariates across the three winters. We obtained 30-m-resolution land cover data from the US Gap Analysis project [50] and reclassified Gap classes into the following 12 categories: floodplain forest, hardwood forest, plantation hardwood forest, upland forest, mixed forest, woodland, shrub, wetland, grassland, pasture, row crops, and developed land (Figure 1, Table S1). We calculated the mean proportional cover of each category within a 4 km resolution raster grid. This spatial scale is roughly equivalent to winter home range sizes observed for Rusty Blackbirds [12]. The values of all environmental variables were extracted to the spatial locations of Rusty Blackbird observations and eBird background samples.

Models were constructed using MaxEnt 3.3.3 [34] and implemented in the R package *dismo* [51]. We randomly partitioned observations into five replicates (k-fold partitioning with cross-validation), with 80% of observations for each replicate used to fit the MaxEnt model (i.e., training samples) and the remaining 20% of the observations held aside to evaluate model performance (i.e., test samples). To minimize the number of features used in model construction and maximize the interpretability of individual covariate effects, we used linear feature constraints of environmental covariates in model construction [52]. However, because Rusty Blackbirds may exhibit a non-linear distribution in response to minimum winter temperatures, we included a quadratic form of the minimum temperature model covariate. To limit model over-fitting, we calibrated models by selecting the most parsimonious models for each flock size class (see Supplementary File S1 1.2).

**Figure 1.** Processing steps from classified land cover layer to distribution modeling.

#### **3. Model Evaluation**

We evaluated model performance by comparing model sensitivity (the true positive rate—the proportion of correctly identified samples at a given threshold of habitat suitability) and specificity (the false positive rate—the proportional predicted area for the model) for each flock size class and observational method. To do so, we assessed the area under the receiver operator curve (AUC) for test samples. AUC describes the sensitivity and specificity of a model at a given threshold of suitability and represents how well the model predicts the Rusty Blackbird observations (see Figure 2). AUC values of 0.5 represent equivalent model performance relative to random, 0.5–0.6 "poor" performance, 0.6–0.7 "fair" performance, 0.7–0.8 "good" performance, 0.8–0.9 "very good" performance, and 0.9–1.0 "excellent" model performance [53,54]. We tested whether the predictive capacity of models varied by flock size by comparing training AUC across folds for a given flock size class against a null distribution developed by permuting two flock size classes (i.e., suitability models developed by shuffling large and small flock observations). To determine whether Blitz data improved model performance, we compared observed AUC against a null distribution developed by randomizing eBird and Blitz samples.

**Figure 2.** Habitat suitability estimates for small, medium, and large flocks of Rusty Blackbirds (<20, 20–99, and >99 individuals, respectively). Across all flock size classes, maps suggest focal wintering regions within the Lower Mississippi Alluvial Valley, Black Belt Prairie, and South Atlantic Coastal Plain. These are also available in Supplementary File S2 as spatially-referenced kmz files suitable for Google Earth.

#### **4. Influence of Environmental Variables**

To determine which environmental variables best predict Rusty Blackbird occurrence for each flock size class, we compared the influence of variable inclusion or removal on model performance. The coefficient for each variable (λ) describes the influence of that variable on suitability estimates, with positive values representing enhanced suitability and negative values representing lower suitability. To determine contribution of a variable to model fit, we evaluated jackknife estimates of the increase and decrease in AUC with the inclusion and removal of a given variable, averaged across replicate runs [55]. Additionally, we examined the difference in the occurrence of environmental covariates between flock sizes, observation protocols, and background data by permuting covariate values between classes. We compared empirical and null distributions of each environmental variable with a one-tailed test of the null hypothesis that the distributions are not statistically different (α = 0.05) using the R package *permute* (*n =* 10,000, [56]).

#### **5. Difference in Environmental Niche by Flock Size**

To assess differences in Rusty Blackbird environmental niche breadth among flock size classes, we evaluated the predicted niche breadth using ENMTools [57]. This metric follows Levins' [29] definition of niche breadth, which describes the degree of uniformity of resource states within a distribution of individuals. Niche breadth is standardized to range between 0 and 1, where 0 represents a habitat specialist and 1 represents a habitat generalist. To evaluate our prediction that niche breadth will vary by flock size, we compared the empirical niche breadth metric for each size class with species distribution maps developed by permuting flock size class assignments. We further explored Rusty Blackbird use of niche space by assessing model prevalence, which is the MaxEnt estimate of the average probability of presence at background sites [58], and fractional predicted area of suitable habitat. We calculated fractional predicted area at the logistic threshold of equal model sensitivity and specificity. We statistically evaluated differences in prevalence and fractional predicted area between flock size classes by comparing the observed values for a given flock with a null distribution developed by permuting flock size class labels (α = 0.05).

To test whether realized ecological niches differed among flock size classes, we evaluated the suitability distributions using niche equivalency analysis and assessed the distribution of samples across individual environmental variables. Niche equivalency between flock size classes was determined by calculating the degree of similarity (modified Hellinger distances, *I*) between habitat suitability maps [59] and implemented in the R package *phyloclim* [60]. *I* ranges in value from 0, in which there is no overlap in the environmental niche between classes, to 1, in which the two classes have identical distributions [59]. We compared observed similarity against a null distribution developed by randomly permuting flock class labels (*n* = 99 permutations). We assessed statistical significance (α = 0.05) by comparing the observed and null distributions, with a one-tailed test of the null hypothesis that environmental niche models are equivalent. We compared the observed distributions of flock size classes of Rusty Blackbirds for each environmental variable using two-tail Kolmogorov–Smirnov tests of the null hypothesis that samples were drawn from the same distribution.

#### **6. Results**

The Blitz produced 678 independent checklists east of −100◦ longitude. eBird provided an additional 1429 traditional checklists with Rusty Blackbird observations corresponding to the same area and time periods. Limiting checklists to one per 4-km raster grid cell resulted in 495 Blitz and 714 traditional eBird checklists (Table 1). Predicted suitability maps (Figure 2) across all flock size classes suggested hotspots within the Lower Mississippi Alluvial Valley (LMAV), Black Belt Prairie (BBP), and South Atlantic Coastal Plain (SACP). For small flocks, there were additional broad areas of moderately high to high predicted suitability, whereas for large flocks the extent of high predicted suitability

was more clearly limited to the aforementioned areas albeit much more restricted within the LMAV.

**Table 1.** Number of checklists used in this analysis by flock size class and observation method. The sample size reflects the number of samples used after subsetting samples to one observation per raster grid cell.


#### **7. Model Performance and Environmental Correlates by Flock Size**

Model performance increased with flock size, with fair performance for small flocks (AUC: 73.7 across samples) and good performance for medium and large flocks (AUCs 83.2 and 88.0 across samples, respectively; Figure 3B). Compared to models using eBird data alone, models augmented with Blitz data showed improved performance for medium and large flocks as suggested by higher AUC values (0.73 ± 0.01 and 0.72 ± 0.01 vs. 0.83 ± 0.02 and 0.88 ± 0.02, respectively; Figure 3B).

**Figure 3.** Receiver operator curves (**A**) and Area Under the curve (AUC; **B**) for small, medium, and large flocks of Rusty Blackbirds using Blitz data, eBird data alone, or Blitz and eBird data ("combined"). Solid lines (**A**) represent receiver operator curves for replicate runs across sampling methods and flock size classes. AUC values of 0.5 represent equivalent MaxEnt model performance relative to random, 0.5–0.6 "poor" performance, 0.6–0.7 "fair", 0.7–0.8 "good", 0.8–0.9 "very good", and 0.9–1.0 "excellent".

Among the environmental covariates, average minimum temperature most strongly predicted habitat suitability, contributing >60% of the models' predictive capacity for each flock size class (Figures 4 and 5A). Larger flock sizes were associated with higher average minimum temperatures (small: −1.74 ◦C, 95% CI [−1.74, −1.56]; medium: −0.40 ◦C, CI [−0.52, −0.27]; large: 0.05 ◦C, CI [−0.09, 0.19]), with peak densities of small, medium and large flocks occurring at −3.81, 1.91, and 1.92 ◦C, respectively (Figure 5A). Average

minimum temperatures associated with the species' occurrence were considerably higher than that of background points (−3.59 ◦C, CI [−3.62, −3.56]).

**Figure 4.** The relative contribution of environmental covariates to predicted Rusty Blackbird occurrence. Variables are subset to those with the greatest explanatory power, with the bar graph on the right excluding minimum temperature (i.e., showing landcover variables only). Legend symbols (+,−) represent whether Rusty Blackbirds were positively or negatively associated with the variable. Rusty Blackbird observations peaked at intermediate minimum temperatures, and is thus denoted with +/−. For a complete list of variable contributions, see Table S1.

**Figure 5.** The gaussian kernel density distributions of average minimum temperature and floodplain forest (**A**,**B**) at occurrence points for each flock size class. Plot (**C**) describes the proportional density distribution of floodplain forest, which is the ratio of the kernel density of observations relative to that of the background points.

Among land cover covariates (Figures 4 and 5, Table 2), proportion of floodplain forest had the greatest influence on predicted habitat suitability. The variable contribution of floodplain forest to large flock distributions was nearly twice as that of small and medium flocks. Other landcover variables that contributed to predicted habitat suitability across all flock size classes were proportion of row crops and pasture, both of which were positive influences (Table 2). Proportion of highly developed land was negatively associated with large flock distributions but was not predictive of small or medium-sized flocks. Likewise, woodland and shrub land cover classes were negatively associated with small and mediumsized flocks but were not predictive of large flocks. Surprisingly, the data provided only limited support for a positive association between habitat suitability and non-floodplain wetlands, with emergent wetlands only associated with medium-sized flock distributions and wooded wetlands for only small flocks.

**Table 2.** Land cover variable coefficients (λ) and contribution of each variable to model performance and small, medium, and large flocks. Variables represent the proportional cover of the land use category within a 4 km grid cell. Empty cells represent variables that were excluded from the candidate model sets due to lack of statistical support, as evaluated with AICc.


#### **8. Environmental Niche**

Niche breadth decreased with increasing flock size. Predicted niche breadth for small flocks was 0.67, medium flocks were 0.47, and large flocks 0.37. Prevalence (i.e., average probability of presence at background sites) was inversely associated with flock size, with small, medium, and large flocks having predicted prevalence values of, 0.44, 0.28, and 0.20, respectively. The observed prevalence for large flocks was significantly lower than that of small flocks (*p* < 0.001) but statistically indistinguishable from medium-sized flocks (*p* = 0.773). Observed prevalence for medium flocks was significantly lower than that of small flocks (*p* = 0.005). Likewise, the fractional predicted area of suitable habitat increased with flock size, ranging from 33 percent for small flocks, 24 percent for medium-sized flocks, and 20 percent for large flocks. Fractional predicted area of suitable habitat was significantly lower for large and medium-sized flocks relative to small flocks (*p* < 0.001) but was statistically indistinguishable for large and medium flocks (*p* = 0.72).

The three flock size classes occupied similar but statistically distinct realized environmental niches. Across niche axes, we found significant differences between the small flock size class and the medium (*i* = 0.939, *p* < 0.001) and large flock size classes (*i* = 0.906, *p* < 0.001) but only marginal differences between the medium and large flock size classes (*i* = 0.979, *p* = 0.102). Among individual environmental variables, we found strong evidence of differences between the distribution of samples across flock size classes, most notably regarding the distribution about minimum temperatures and floodplain forests (Table 3). Whereas the modes of the temperature distributions for all three flock size classes were similar, medium and large flocks were observed in a narrower range of temperatures than small flocks (Figure 5B). Likewise, while all flock size classes were positively associated with floodplain forest (Figure 5B), the degree of association to floodplain forest increased with flock size (Figure 5C).


**Table 3.** The difference (D) in the distribution of each environmental covariate between Rusty Blackbird flock size classes. *p*-values below 0.05, in bold, suggest that the flocks occupy a different environmental space.

#### **9. Discussion**

We used novel targeted citizen science data from the Blitz and eBird to identify and describe suitable habitat for a widespread, highly vagile, and declining species, the Rusty Blackbird. Across two winters, occurrence was most strongly linked to minimum temperatures and the proportion of floodplain forest in the surrounding landscape. Although all flock sizes were associated with proportion of floodplain forest, pasture, and row crops, and the convex quadratic effect of average minimum temperature, we found considerable differences among flock size distributions. Predictably, large and medium-sized flocks were more similar in their environmental distributions relative to small flocks. Large and medium flocks had narrower environmental niches than small flocks, showing a greater preference for floodplain forest. We identified Rusty Blackbird hotspots in the LMAV, the Black Belt region of Alabama and Mississippi and the Southeast Coastal Plain. The Blitz was ultimately successful: adding Blitz data to eBird data significantly increased our predictive power.

#### **10. Environmental Predictors of Occurrence**

Rusty Blackbirds exhibited a strong association with floodplain forest, with the degree of influence of this variable positively associated with flock size (Table 3, Figure 5). Given that floodplain forest is expected to represent optimal foraging habitat for Rusty Blackbirds, this provides supportive evidence that large flocks are representative of higher quality habitats as a consequence of coarse-level local enhancement (e.g., [22,23]). Winter floodplains in the southeastern United States historically provided vast areas of shallow water crucial for migratory birds [61]. These environments, which represent seasonally flooded forested habitats, provide an abundance of invertebrates for Rusty Blackbirds, which spend much of the winter foraging in, or adjacent to, shallow water [1,62]. In the southeastern United States, floodplain forests have experienced extraordinary rates of direct landscape modification, historically due primarily to agricultural conversion and more recently as a result of urban development [9,61,63,64]. Moreover, hydrologic alteration has greatly reduced the extent and integrity of floodplain wetlands throughout the region [65]. Combined, these pressures have resulted in a 75–85% loss of the historic extent of floodplain forests in the Southeast [66,67] and the remnant patches of this habitat are highly fragmented [68,69]. Within our study extent, the LMAV hosts the largest area of floodplain forest and this region yielded the highest predicted habitat suitability for Rusty Blackbirds across flock size classes (Figure 2). Christmas Bird Count data suggest that the LMAV, considered to

be the core of the species' range, has experienced the steepest population declines and thus, maintenance of floodplain forests in this region is likely critical for Rusty Blackbird conservation [70].

Surprisingly, neither emergent nor woody wetlands had substantial influence on habitat suitability for Rusty Blackbirds. Although these land cover classes appear to provide important habitat features for the species [71], they may have been underrepresented by Blitz and eBird samples. Our ability to detect an influence of these land cover classes may have also been limited by the coarse spatial grain of our analysis. For example, though Rusty Blackbirds often forage on wet lawns in suburban areas [25] it would not be feasible to detect these habitat features at our scale of analysis, especially when using traveling count data. Sampling constraints also limited our ability to assess the effects of pecan orchards on Rusty Blackbird occurrences. Lipid-rich tree mast from pecan orchards may help the birds prepare for cold weather events [25], and wintering individuals that utilized pecan groves in Mississippi (mostly adult males), were in better body condition than those in wet forest or along creeks [24]. Despite the expected importance of this anthropogenic resource, there was not adequate representation of this land cover class in background nor Rusty Blackbird samples to include the variable in our analysis.

The influence of human land use on habitat suitability varied with flock size. Large flocks, unlike medium or small flocks, were negatively associated with high intensity development; i.e., when the proportional cover of high intensity development exceeded 70%, the probability of encountering a large flock approached zero. These results are consistent with those of Mettke-Hofmann et al. [24], Newell [10], and Newell Wohner et al. [25] who observed that smaller groups of Rusty Blackbirds can use certain agricultural and suburban areas, taking advantage of abundant pecans and earthworms, respectively. The species' positive (though relatively weak) association with pasture and row crops is more puzzling, as it is not generally associated with these habitats (pecans are not classified as a row crop; [1]). It is possible that the low-lying, nutrient-rich floodplain landscapes that the species favors also make good croplands and pasture for farmers, and as such, both floodplain forest that Rusty Blackbirds presumably prefer, and that these agricultural habitats were often included in the 4-km grid cells we employed. Combined, our findings suggest that, though suburban and agricultural areas appear to offer resources for smaller groups of Rusty Blackbirds, larger tracts of floodplain forest will have to be maintained to support the persistence of large flocks.

Average minimum temperature was the most important variable driving habitat suitability for all flock sizes, with medium and large flocks predicted to occur at higher temperatures than small flocks. Habitat suitability estimates for medium and large flocks peaked where minimum mean temperatures occurred at means slightly higher than the freezing point of water: 1.91 ◦C and 1.92 ◦C, respectively. Suitability estimates for small flocks peaked at −3.8 ◦C—nearly 6 ◦C less than medium and large flocks. The warmer temperatures associated with medium and large flocks likely reflect the foraging behavior of Rusty Blackbirds, which spend much of their time wading in shallow water searching for aquatic prey [1]. Shallow water tends to freeze before deeper water, which would quickly prohibit birds from foraging in the substrate, making aquatic prey unavailable. Moreover, wetland invertebrate abundance is positively associated with water temperature [72–74]. Conversely, the colder temperatures associated with small flocks may represent marginal habitats in which birds utilize terrestrial resources in freezing temperatures (see [24,25]).

Because temperatures were averaged across years of the study, we were unable to assess the annual occurrences of Rusty Blackbirds. The species varies widely in its wintering distribution [9] and site fidelity [10], most likely because of differences in winter temperatures and water levels across years. Rusty Blackbirds' association with temperatures just above freezing supports our theory that the species, which is a facultative migrant not tightly linked to day length [1], migrates south from the Boreal forest to avoid freezing water. Birds may then spend many weeks in the northern United States during stopover periods in fall [75,76], only to move south again when frozen foraging habitat

pushes them further. Future work should assess whether carryover effects due to variable wintering conditions impact the birds across their full annual cycle [77,78], including on their breeding grounds in the Boreal forest.

#### **11. Larger Flocks Have a Narrower Niche Breadth**

We found strong evidence that large and medium flocks have narrower environmental niches than small flocks, as measured by Levins' [29] niche breadth metric and, consequently, prevalence and fractional predicted area of suitable habitat was higher for small flocks than medium and large flock size classes. Whereas such differences in niche breadth have been observed across a variety of taxa using environmental niche modeling (e.g., [79]), this represents among precious few examples of the relationship between flock size and niche [26]. This pattern is abundantly clear across large scales, as small flocks have a much broader geographic range of suitable habitat than large and medium flocks. Conversely, the area of suitable habitat (Figure 2) for large flocks is much smaller and is largely constrained within the center of the species range. Again, this highlights the relative vulnerability of larger flocks of Rusty Blackbirds and the relative flexibility of smaller flocks. Small flocks may be able to persist in heterogenous landscapes with small patches of ephemeral resources (e.g., fruiting pecan orchards, lawns with surfacing earthworms), whereas larger flocks may be limited to the highest quality habitats. Future studies (e.g., using tracking technology) should examine space use of large vs. small flocks to understand how medium (i.e., <4 km) and small-scale habitat features affect fitness and niche space.

The use of flock size herein to approximate habitat quality represents a limitation to the inference of our results. Blitz and traditional eBird observations were traveling counts, thus in some instances, birds observed on a given count could have comprised one or several true flocks. As such, it was not technically possible to distinguish between a single large flock or a higher density of smaller flocks; however, given our own observations of the species (e.g., [12]), we believe the vast majority of the counts probably represented single flocks, so we are comfortable interpreting results based on this assumption. Moreover, traveling counts limit the spatial grain of the analysis, as counts may occur at any point along the observer's route. To address this, we highly recommend the use of stationary count data to allow for assessing the distribution of flocks as it relates to land cover data (e.g., evaluating resource dispersion hypothesis as described by [80]). Even without these limitations, density itself is not a clear indicator of habitat quality [11,81]. Further research (e.g., survival) is therefore needed to confirm that large flocks are representative of environments of higher quality habitat [33]. Sex or age-related segregation of the observed flocks may provide additional evidence of habitat quality [82]. Dominance hierarchies can indicate habitat quality in a number of bird species (e.g., the American Redstart, *Setophaga ruticilla*, [83]). For example, in Rusty Blackbirds, Mettke-Hofmann et al. [24] observed that adult males maintained a higher body condition and occupied habitats with higher pecan mast production than females and immature birds. Unfortunately, age and sex data from the Blitz were not adequate to address this issue in the current study.

Ultimately, a demographic response, most appropriately overwinter survival, is necessary to truly ascertain whether flock size is an appropriate indication of winter habitat quality for Rusty Blackbirds [84]. In particular, despite some location-specific evidence hinting at population limitation on the breeding grounds in the Boreal forest [7,8], a full-annual cycle population model suggests that wintering juvenile survival is the demographic parameter most tied to the species' rate of population change (Clark Rushing, Steve Matsuoka, Luke L. Powell et al. unpublished analysis). Whereas estimating overwinter survival may be very challenging given the species' low site fidelity, alternative proximate habitat quality metrics that can be utilized in combination for Rusty Blackbird flocks include but are not limited to: departure time (e.g., [85]), body condition (e.g., [86,87], but see [84,88]), and telomere length (e.g., [89]).

#### **12. Regional Hotspots and Conservation**

The LMAV was a particularly suitable area for all flock sizes (Figure 2, [5]). Though we also found a broad swath across the southeast coastal plain that was generally suitable, with small hotspots in north coastal South Carolina [25], the majority of the East Coast of the United States appears less suitable than the LMAV. Stable isotope data [90] and light level geolocators [75] indicate that Rusty Blackbirds wintering in the LMAV primarily migrate to Alaska and western Canada—the Western Boreal—whereas those from the southeast coastal plain breed in northeast North America—the Southeastern Boreal. It remains unclear whether flock sizes in the southeast coastal plain are constrained by anthropogenic disturbance or if it is the quality and extent of floodplain forests of the LMAV that leads to larger flock sizes in that region.

A novel finding from our study was the emergence of the Black Belt region (Ecoregion 65, [91–93]) as a hotspot across all Rusty Blackbird flock size classes. The Black Belt, which arcs along the boundary of the piedmont and coastal plain in Mississippi and Alabama, was historically a mosaic of prairie and forested habitat [94]. Whereas most of the region's native vegetation was cleared for cotton-based agriculture [94–97], the Black Belt remains embedded within a vast matrix of floodplain forest. To date, we know of no Rusty Blackbird research projects within 500 km of the Black Belt—thus, our results highlight a need for targeted field studies to further qualify the use and importance of this region to Rusty Blackbirds.

The emergence of the Black Belt region as a hotspot underscores the benefits and limitations of species distribution modeling. Whereas comparably few eBird and Blitz samples were recorded within the region, the MaxEnt model, as implemented here, is carried out in environmental space [52], which allowed us to generate predictions for the Black Belt as a function of its environmental characteristics [98]. As a cautionary note, however, this approach is limited in its ability to directly inform conservation efforts, as it yields information about the type of environment an organism uses rather than occurrences per se [99]. Moreover, our models are undoubtedly imperfect—for example, some areas of high predicted suitability for large flocks were of low predicted suitability for medium-sized flocks (e.g., north-central LMAV). This was likely driven by the difference in habitat preferences of large vs. medium flocks (e.g., large flocks had higher preference for floodplain forest, Table 3) as well as by stochasticity in where birds were detected. As such, we suggest that systematic sampling of predicted Rusty Blackbird hotspots—particularly in the Black Belt region—is a crucial step towards linking our results with management efforts.

#### **13. Did the Blitz Help Relative to eBird Alone?**

The Blitz significantly improved the predictive power of spatial models compared to using eBird alone (Figure 3), highlighting the value of this novel citizen science approach to improve understanding of a wide-ranging species of conservation concern. Further, the addition of the Blitz especially improved our predictive power for large flocks (Figure 3B), as we effectively doubled the number of large flocks detected (Table 1). This is important, as it allows us to precisely concentrate targeted studies on sites at which we predict only the very highest flock sizes of Rusty Blackbirds. We also utilized a somewhat novel way to address biases associated with modeling eBird data by using background samples from eBird lists across taxa: our background data were generated with equivalent bias to our occurrence data. Our comparison of background and occurrence data therefore provides conservative estimates of Rusty Blackbirds in environmental space. An important caveat to our Blitz effort is that participants were searching for Rusty Blackbirds in habitats that were expected to be suitable for the species–this has the potential to bias suitability predictions towards environments that are pre-determined to be suitable. A preliminary analysis of our results, however, produced estimates from eBird and Blitz data that were similar, suggesting that any such bias was nominal. In addition to the quantitative benefits of the Blitz, it was clear the Blitz raised considerable awareness for the plight of the Rusty Blackbird (e.g., Audubon magazine, eBird website, conservation partners participating

in the Blitz, etc.). Given that citizen scientists submitted 678 checklists specifically for the Blitz, the program was clearly successful in engaging birders with ornithological research and conservation concerns to which they may not previously have been exposed.

Taken together, the data provided by citizen science participants of the Blitz and eBird program have enhanced our understanding of the occurrence of this Boreal-breeding species throughout its wintering areas. Whereas wintering Rusty Blackbirds can be found across most of the southeastern United States, only a narrow subset of the region was found to be suitable for large flocks of the species. By targeting these regions for future research, we can maximize sampling efficiency by searching areas predicted to provide high quality habitat. Targeted research is crucial given the limitations associated with presenceonly observations. Using the MaxEnt framework, with presence-only data, locations where birds were not observed (i.e., background data) cannot be treated as true absences (see [100,101]). This is especially problematic when detection probability varies and cannot be incorporated into the model [102] and imparts a limitation to how citizen science data can be used to inform conservation in this context [103]. The use of presence-absence data to model occupancy (e.g., [2]), can also allow researchers to address temporal variation in the presence of Rusty Blackbird flocks, which may yield further insight into habitat quality (e.g., abundance-occupancy relationships, [104]). Moreover, identifying and counting individual flocks from fixed points in the landscape and evaluating these observations at multiple spatial scales would provide researchers with the ability to assess alternative explanations for flock size distributions, such as the resource dispersion hypothesis (reviewed in [80]). Future work that utilizes systematic sampling to estimate the geographic distribution of Rusty Blackbird flocks and assess the relationship between flock size and habitat quality is therefore a critical next step towards conserving this imperiled species.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/1424-281 8/13/2/62/s1: Table S1: rubl\_gap\_reclassification.csv, Supplementary File S1: Additional methods and results, Supplementary File S2: rubl\_distribution\_maps.zip.

**Author Contributions:** Conceptualization, R.S.G.; methodology, B.S.E.; formal analysis, B.S.E.; data curation, B.S.E.; writing—original draft preparation, B.S.E., L.L.P., D.W.D., and S.M.B.; supervision, R.S.G.; project administration, R.S.G.; funding acquisition, D.W.D. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research and the associated APC were supported by the U.S. Fish & Wildlife Service/USFWS.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Publicly available datasets were analyzed in this study. This data can be found here: https://github.com/SMBC-NZP/rubl.

**Acknowledgments:** Russ Greenberg conceived of, advertised, and led the execution of the Blitz; our hope is that this manuscript honors the memory of a brilliant ornithologist. This paper was written in association with the International Rusty Blackbird Working Group (RustyBlackbird.org; founded and led by Russ). We thank the United States Fish and Wildlife Service for funding. We deeply thank the regional coordinators and many citizen scientists that contributed to eBird and to the Blitz, and we thank the Cornell Lab of Ornithology and eBird for creating a portal for Blitz data entry. The findings and conclusions in this article are those of the author(s) and do not necessarily represent the views of the U.S. Fish and Wildlife Service. Any mention of trade names is purely coincidental and does not represent endorsement by the author(s) and / or their respective organization.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


MDPI St. Alban-Anlage 66 4052 Basel Switzerland Tel. +41 61 683 77 34 Fax +41 61 302 89 18 www.mdpi.com

*Diversity* Editorial Office E-mail: diversity@mdpi.com www.mdpi.com/journal/diversity

MDPI St. Alban-Anlage 66 4052 Basel Switzerland

Tel: +41 61 683 77 34 Fax: +41 61 302 89 18

www.mdpi.com