Spatial segregation between Gray-cheeked Thrush and an introduced nest predator in a managed forest landscape

,


INTRODUCTION
Extinctions, extirpations, or imperilment are frequently attributed to interactions with alien species (Bellard et al. 2016); among birds reported as having been imperiled or extirpated by the introduction of a mammalian predator, 90% were island endemics (Doherty et al. 2016).This susceptibility among island endemics occurs because they typically exist as relatively small, isolated populations that have evolved in less diverse ecosystems, and so may have reduced competitive and anti-predator abilities in the face of introduced mammals (Courchamp et al. 2003, Russell andKaiser-Bunbury 2019).
Although introduced species can affect communities and the associated ecosystem processes, native species may also suffer additive effects from other stressors such as habitat degradation or loss and climate change.Boreal forests are often subject to timber harvest, silviculture (e.g., tree thinning or planting), suppression of natural disturbances such as wildfire and outbreaks of defoliating insects, and the construction of resource roads and other corridors (Burton et al. 2003).At the same time, climate change may be leading to range shifts into more northerly ecosystems or higher elevations for many species (Lehikoinen and Virkkala 2016, Kirchman and Van Keuren 2017, Rushing et al. 2020).In combination, these factors may have detrimental influences, as synergies among stressors have been identified as complicating factors in developing conservation-based management guidelines (Brook et al. 2008).
Newfoundland, Canada is the largest island in the circumpolar boreal biome.Because of the oceanic barrier and existence of an Atlantic shelf refugium during the Pleistocene, the island supports numerous endemic subspecies of more widely distributed boreal species (Dodds 1983, Montevecchi and Tuck 1987, Dohms et al. 2017, Ralston et al. 2021).Among these is the Newfoundland Gray-cheeked Thrush (Catharus minimus minimus), which breeds primarily on Newfoundland and the archipelago of much smaller nearshore islands along its coast; as well as along the strait of Belle Isle in southern Labrador, on a small number of islands on the Atlantic coast of Nova Scotia, and historical records along the north shore of the Gulf of St. Lawrence in Quebec (Whitaker et al. 2020).The Newfoundland Gray-cheeked Thrush is genetically distinct from the Northern Gray-cheeked Thrush (C. m. aliciae;FitzGerald et al. 2017FitzGerald et al. , 2020) ) that occurs across the rest of the species' continental distribution.Breeding Bird Survey (hereafter BBS) data and localized surveys make it clear that until the 1980s, Gray-cheeked Thrushes were abundant in suitable habitat at all elevations on Newfoundland, but most abundant below 100 m and above 300 m above sea level (m.a.s.l.;Lamberton 1976a, b, Marshall 2001, Robineau-Charette et al. 2023).BBS surveys on the island during the 1970s and early 1980s detected Gray-cheeked Thrushes along 75% of survey routes, with nearly one-third of these 50-stop surveys detecting 10 or more thrushes and a maximum of 38 individuals detected during one survey (Smith et al. 2020).In comparison, since 1988 no BBS survey on Newfoundland has recorded more than three Gray-cheeked Thrushes, and from 2010 to 2019 this species was only detected during 9% of BBS surveys (Smith et al. 2020).Estimates based on BBS data indicate an overall population decline of ~95% between 1975 and 2005 within the areas covered by the BBS on the island (SSAC 2010).Along with scattered populations on nearshore islands, the most notable exception to this widespread decline is the persistence of locally abundant breeding populations in montane forests in western Newfoundland (Whitaker et al. 2015, FitzGerald et al. 2017).
Several hypotheses have been suggested to explain this population collapse and apparently restricted contemporary distribution of Gray-cheeked Thrushes (Whitaker et al. 2015).Timber harvesting has frequently been identified as having the potential to adversely affect forest bird populations (e.g., Thompson et al. 1999, Lamarre andTremblay 2021), but the coastal conifer scrub and montane forests favored by Gray-cheeked Thrushes on Newfoundland are seldom cut.Research indicates that these thrushes are also associated with regenerating (i.e., 10-30 years post-harvest; see Table 1) clear-cuts (Whitaker et al. 2015), but the extent of timber harvesting on Newfoundland has only decreased by about 50% since the early 1970s (McLaren andPollard 2009, Natural Resources Canada 2022).Likewise, Graycheeked Thrushes have also declined in large protected areas on the island like national parks, which contain the full range of elevations and habitats they traditionally occupied (e.g., Jacques Whitford Environment 1993).Declines in breeding populations of Neotropical migrants have also been linked to habitat changes in areas used during migration or winter (Newton 2004, Taylor andStutchbury 2016), and this may have contributed to the decline and altered distribution of Gray-cheeked Thrushes on Newfoundland.We have begun to address this hypothesis elsewhere (Whitaker et al. 2018;additional research underway).
Another hypothesis is that the Newfoundland Gray-cheeked Thrush population had collapsed and become restricted to higher elevations because of impacts of invasive North American red squirrels (Tamiasciurus hudsonicus; Whitaker et al. 2015).Red squirrels were introduced to Newfoundland in 1963 and 1964 (Payne 1976, Dodds 1983), and subsequent translocations and natural dispersal enabled them to become widespread across the island by the early 1990s (Whitaker 2015).Red squirrels are important nest predators in northern boreal forests (Willson et al. 2003) and can dramatically alter the reproductive success and composition of avian assemblages on islands and in isolated "insular" patches of continental forest (Martin andJoron 2003, Siepielski 2006).The introduction of red squirrels may have contributed to the imperilment of the Newfoundland subspecies of Red Crossbill (Loxia curvirostra percna) through both nest predation and competition for conifer seeds (COSEWIC 2016).In the Bicknell's Thrush (Catharus bicknelli), which is closely related to Gray-cheeked Thrush, there is evidence that predation of eggs and nestlings by red squirrels can cause nest success to fall from 14.7% to 0.44% (a 97% reduction) in years of high squirrel abundance following large cone crops, which occur every 2-3 years (McFarland et al. 2008, Hill et al. 2019).This phenomenon is mediated by upslope expansion of squirrels into Bicknell's Thrush breeding areas during these years of high abundance (McFarland et al. 2008).These facts led Whitaker et al. (2015) to predict that, though widespread and abundant across most of Newfoundland, red squirrels may be less common or absent from montane forests, offering a mechanistic explanation for the persistence of Gray-cheeked Thrushes at higher elevations.
We assessed the current distribution and habitat affiliations of the Newfoundland Gray-cheeked Thrush in the context of Scheele et al.'s (2017) niche reduction hypothesis, which suggests that declines of endemic populations may result from species becoming restricted to refugial habitats to which invaders are poorly suited.The primary goal of this study was to test the prediction of Whitaker et al. (2015) that the distribution of Graycheeked Thrushes would be negatively correlated with the occurrence of red squirrels.This is consistent with, and a first targeted examination of, the hypothesis that the introduction of red squirrels was a key factor leading to the thrush population's decline on Newfoundland.We also predicted that forest harvesting would have a positive effect on Gray-cheeked Thrush occurrence based on Whitaker et al. (2015).With data from the same study area but after a longer time since harvesting, we tested the prediction that this relationship would be strongest for stands 11-13 years versus more than 20 years post-harvest.We based this prediction on findings reported for the Bicknell's Thrush, which has similar habitat affinities as the Newfoundland Graycheeked Thrush (Chisholm and Leonard 2008, Aubry et al. 2018, Whitaker et al. 2020).Our findings for Gray-cheeked Thrushes, as well as our findings regarding red squirrel distribution and habitat associations (McDermott et al. 2020), can inform the development of management approaches for montane forests that aid the conservation and recovery of the Newfoundland Graycheeked Thrush.

Study location and focal species
We conducted our research across a 257 km² study area that spans an elevation gradient from 75 m to 608 m.a.s.l. in the Long Range Mountains of western Newfoundland, Canada (57°16′ W, 49°40′ N; see McDermott et al. 2020).The study area is dominated by boreal forest, and the climate in the region limits the extent of natural disturbances such as large-scale fires or outbreaks of defoliating insects, particularly at higher elevations (Thompson et al. 2003, McCarthy and Weetman 2006, Arsenault et al. 2016).
Locations below ~450 m elevation support stands of commercially productive forest dominated by balsam fir (Abies balsamea) or, to a lesser extent, black spruce (Picea mariana) within a matrix of bogs, barrens, and other natural openings (Damman 1983, McCarthy andWeetman 2006).Above 450 m elevation, factors such as increased wind, deep and late snow cover, low nutrient availability, and saturated soils lead to increasing prevalence of bogs, barrens, and scrub forest, while commercially productive forest becomes increasingly restricted to sheltered valleys (Damman 1983). Between 1990and 2004, 19.7% of the landscape in the surveyed area was logged using clear-cut harvesting approaches that created cutblocks (patches of forest that have been harvested for wood) ranging from 0.30 ha to 217.7 ha, with a mean size of 21.6 ± 31.8 ha.In addition, two ~100 ha experimental strip cuts were harvested during 2001 and 2003, and spanned 419-564 m elevation (see Whitaker et al. 2015).Recent forest clearing has been limited to the creation of a 60 m-wide electricity transmission corridor through the north of the study area during 2016 and 2017.Introduced North American red squirrels (hereafter squirrels) likely colonized the region between ~1985 and 1990~1985 and (Whitaker 2015)).
Across continental North America, Gray-cheeked Thrushes are typically found in dense regenerating stands of coniferous or deciduous saplings and shrubs, willow (Salix spp.) and alder (Alnus spp.) thickets, or old growth coniferous forest having complex understories (Whitaker et al. 2020).On Newfoundland, Gray-cheeked Thrushes are less likely to be found in deciduous shrub thickets and instead more commonly occur in coastal windswept coniferous scrub habitat and old growth (> 80 years) balsam fir forest, while avoiding second growth (40-80 years) stands (Thompson et al. 1999, Marshall 2001, Whitaker et al. 2015).Habitat use has been little studied in relation to forest harvesting, but for Bicknell's Thrush these activities provide important regenerating habitat, particularly in 11-13 year old post-harvest stands in New Brunswick (Chisholm and Leonard 2008), and in stands 20 years post-harvest or older in Quebec (Aubry et al. 2018).Whitaker et al. (2015) found an increase in occurrence of Gray-cheeked Thrushes as the local extent of harvested forest increased up to 60% cover, but that thrush occurrence decreased when the extent of clear-cuts exceeded 60%.

Data collection
To assess the distributions of thrushes and squirrels throughout our study area, we conducted surveys from early June (when thrush breeding is initiated in our study area) to mid-July in 2016 and 2017 across a grid of points spaced 500 m apart.During 2016 the grid encompassed 991 survey points; we then shifted the grid 250 m north and east during 2017, placing survey points midway between those sampled the previous year, and sampled another 969 points.Each point count location was visited once during the study.We typically alternated surveys between low, medium, and high elevation areas on consecutive days.Solitary observers (four individuals during 2016, five during 2017; one common to both years) each sampled 5-12 adjacent points per day between 05:40 h and 14:30 h.We did not conduct surveys when high winds (> 5 Beaufort scale; 29 km/h), precipitation, or fog would have impaired visual or auditory detections of thrushes or squirrels.
We followed an 11-minute unlimited distance point count protocol.This included an initial six minutes of silent listening; a two-minute broadcast of Gray-cheeked Thrush songs and calls (same broadcast as used by Whitaker et al. 2015); one minute of silence; a one-minute broadcast of red squirrel vocalizations (taken from the Macaulay Library, Cornell Lab of Ornithology: catalogue numbers ML100916 and ML136185); and one final minute of silent observation.We set broadcast units (FoxPro model FX3 or Crossfire game callers; FoxPro Incorporated, Lewistown, PA, USA) at a constant volume for all surveys; when measured 1 m from the speaker the average volume of vocalizations was 57.7 dB, with peak volumes of 82.6 dB, where the intention was to mimic natural sound production, and volumes were similar to Whitaker et al. (2015).During each of the time blocks within the 11-minute point count, we documented each thrush and red squirrel that was seen or heard.Use of conspecific playbacks during the breeding period elicits territorial responses by songbirds (Betts et al. 2005) including attracting individuals and eliciting agonistic displays and vocalizations, thereby increasing detectability by observers.During our surveys 30% of initial Gray-cheeked Thrush detections occurred during or after thrush vocalizations were broadcast, and exploratory analysis of our data indicated that detectability of thrushes was not significantly influenced by time of day or date.
We obtained land cover data from a provincial forest resource inventory Geographic Information System (GIS) database, which was developed from high resolution (sub 10 cm pixel) 3D aerial photography collected in 2007.Using ArcGIS 10.4.1 (ESRI 2002), we extracted this land cover data within a radius of 132 m and 1250 m of each survey point, which we interpreted as representing local and landscape scale habitat, respectively, for Gray-cheeked Thrushes.To our knowledge, breeding season space use has never been studied for the Gray-cheeked Thrush (Whitaker et al. 2020), so our local scale (132 m radius or 5.5 ha) approximates the maximum female core area and minimum male core area of the Bicknell's Thrush in Vermont (Collins 2007, McFarland et al. 2008), which is the focus for nesting and some foraging activities (see also Aubry et al. 2011).Our larger scale of 1250 m (490.8 ha) reflects a landscape scale that has been shown to influence space use by individual songbirds in this forest ecosystem (Leonard et al. 2008, see also Whitaker et al. 2015).At each scale, we aggregated similar cover types into generalized categories for consistency with Whitaker et al. (2015).We grouped conifer scrub into two height classes (< 6.5 m and > 6.5 m; see Table 1 for definitions).We accounted for the time between collection of the aerial photographs and our field work by adding 10 years to the age class for each forest stand, and then grouped the 20-year age classes into three broader successional categories: regenerating stands 10-30 years old; second growth stands 30-90 years old; and mature stands 90+ years old.These three age classes represent broad forest successional stages having distinct stand structure and biotic communities (Thompson et al. 2003 1).
All land cover variables were quantified as the proportion (0 to 1) of the area they accounted for within the 132 m or 1250 m local and landscape radius circles around each survey point.We also included length of shoreline within 132 m or 1250 m, which was measured as the total length in meters and was rescaled to fall within a range of 0 to 1 relative to the length of the largest observed shoreline value.Though it is unclear why amount of shoreline would be related to Gray-cheeked Thrush occurrence, it was found to be important in Whitaker et al. (2015), and so we retained it to assess the consistency of this relationship.We calculated a squirrel probability of occurrence value for each survey point using the predict function in

Data analyses (a) Comparisons between the occurrence of thrushes and squirrels
To test for an effect of red squirrels on Gray-cheeked Thrush distribution, we first assessed whether the occurrence of these species at our survey points was independent each year using a Chi-square test.We also used Kolmogorov-Smirnov tests to test for differences between the elevation distributions of thrushes and squirrels each year.These and all other statistical analyses were completed using R statistical software (version 3.3.1 and 3.6.2;R Core Team 2019).

(b) Red squirrel and habitat influences on Graycheeked Thrush occurrence
During our surveys we did not detect Gray-cheeked Thrushes below 317 m.Thrushes were historically common in lower elevation habitat on Newfoundland (see Introduction), and we are unaware of any substantive, widespread habitat changes in such areas since that time.Consequently, we did not expect elevation per se to be a proximate or causal factor affecting the suitability of habitat for thrushes.Nor did we expect that habitats occurring at elevations below their current distributional limit would be inherently unsuitable given that this species was historically common throughout Newfoundland at elevations down to sea level (Lamberton 1976a, SSAC 2010, Fitzgerald et al. 2017).Thus, we truncated our dataset to include only points above this elevation (n = 1670 points) to lessen any confounding effect of correlations with elevation on our results.All candidate variables were tested for independence using a Spearman correlation matrix (i.e., r < 0.6 indicating independence; Dormann et al. 2013).Elevation and squirrel probability of occurrence were highly correlated (Spearman's rho = -0.69,p < 0.001), which is logical given that elevation was important in the model developed by McDermott et al. (2020) to predict squirrel probability of occurrence.To resolve this, we dropped elevation as a candidate variable and retained squirrel probability of occurrence as it was of greater biological interest.Land cover variables were eliminated for each spatial scale if they were present at < 5% of survey points.
Using the variables listed in Table 1, which had been identified as being correlated with Gray-cheeked Thrush occurrence by Whitaker et al. (2015), we created nine candidate linear models to explain the occurrence of thrushes across our study area.All models included year, red squirrel probability of occurrence, plus different combinations of landcover types at the local and landscape scales (Table S1 in Appendix 1; Table 2).These combinations reflected the potential influences at the two spatial scales of forest composition, or successional stage, or composite variables combining composition and successional stage, as well as other non-forest cover types, and timber harvesting.Before fitting our candidate models, we examined each land cover variable in a univariate Generalized Additive Model (GAM) using the function "gam" (R package mgcv version 1.8-33; Wood 2011) with occurrence of thrushes as the response variable.This was done to assess the likely nature (shape) of any relationship between thrush occurrence and that variable.In these exploratory models we fitted the land cover variables as smoothed nonparametric splines to allow for nonlinear relations, and we specified a clog-log link function, which typically performs better when the ratio of presence to absence is skewed (Zuur et al. 2009).
Land cover variables that appeared to be non-linear were generally parabolic, and so were also given a second order polynomial term in each of the candidate models where they were specified.Within each candidate model only one variable was included from any pair of correlated variables (Booth et al. 1994; see Table S2 of Appendix 1).
We fit each of our nine candidate models using a Generalized Linear Model (GLM) having a binomial error distribution, a cloglog link function, and thrush presence/absence as the response variable using the function "glm" (R package lme4 version 1.1-26; Bates et al. 2015).We simplified each of these nine models through manual backwards step-wise selection to eliminate unimportant variables.At each step, the explanatory variable having the lowest explanatory power was assessed, and if removing it yielded an improvement in model fit (p < 0.05 or a reduction in Akaike's Information Criterion adjusted for small sample sizes [ΔAICc] > 2), then it was dropped from the model.Once all unimportant variables had been eliminated, these resulting nine best models were compared using AICc (R package MuMIn version 1.43.15;Bartoń 2020) to decide which of the nine combinations of squirrel probability of occurrence, successional or compositional aspects of the forests, and other land cover variables, had the most explanatory power.The model having the lowest AICc was assumed to be the most parsimonious model explaining the distribution of Gray-cheeked Thrushes, and any other models having a ΔAICc < 2 were considered competing models (Burnham andAnderson 2002, Symonds andMousalli 2011).
(c) Silviculture analyses: age and proportion of harvested forest Whitaker et al. (2015) noted a broad pattern of increasing occurrence of Gray-cheeked Thrushes as the extent of harvested forest in the surrounding landscape increased up to 60% cover; however, they collected their data within five years of forest harvesting and so were unable to assess the effects of these disturbances at the temporal scale that has been found to be most relevant for the closely related Bicknell's Thrush.To address this, we conducted a separate analysis to assess relationships between thrush occurrence and the age and proportion of harvested forest within a 132 m radius circle of each survey point.For this, we retained all points at elevations above 317 m that had a cutblock present within 132 m (n = 664).Age (years since harvesting; date of harvest known from the forest resource inventory) was calculated for each of these cutblocks based on the year (2016 or 2017) when we visited that survey point.Where multiple cutblocks of different ages were present within 132 m, we took a weighted average of the ages of harvested stands to calculate a representative cut age for the point.We used this approach for 67 of 664 points, and most (72%) of these had cuts that were only one or two years apart in age.Proportion of harvested forest within a 132 m radius and age of cutblocks were fit as continuous variables.We created five candidate GLMs (R package lme4 version 1.1-26; Bates et al. 2015) having all possible univariate, additive, and multiplicative combinations of cutblock age and proportion harvested, along with a null model containing no explanatory variables.We fit these models using a binomial error distribution and a clog-log link.As above, we used R package MuMIn (Bartoń 2020) to rank the models based on AICc, as well as to identify the most parsimonious and any competing models (Burnham andAnderson 2002, Symonds andMousalli 2011.Cutblocks ≥ 19 years old were only present below 428 m, an elevation range where thrushes were less frequently detected in all landcover classes (Fig. S1C in Appendix 1), raising the concern that the observed relation between cutblock age and thrush occurrence was an artefact of this non-random distribution of older cutblocks.To explore this possibility, we also carried out a post hoc analysis using our best model but in which we truncated the range of age of cuts to ≤18 years-old.

(d) Silviculture analyses: harvesting method
We assessed the importance of harvesting method (clear-cut or modified [strip] cut) on thrush occurrence.Modified cuts were present only between 419 and 564 m elevation (1039 points), so we restricted the comparison to points in this elevation range.We fit four GAMs (R package mgcv version 1.8-33; Wood 2011) including a null model and all additive combinations of the proportions of clear-cut and modified cut within a 132 m radius circle around each point.Models were fit using binomial error distributions and clog-log link functions, and continuous variables were fit as splined, non-parametric terms."k" values of each term, which are akin to the maximum number of knots (turning points) in the curve, were specified a posteriori as instructed by Wood (2011).We compared models based on AICc as above using R package MuMIn (Bartoń 2020) to identify the best and competing models (Burnham andAnderson 2002, Symonds andMousalli 2011).

RESULTS
During 2016 we saw or heard a total of 142 Gray-cheeked Thrushes at 117 (11.8%) of 991 survey points, whereas in 2017 we detected a total of 123 thrushes at 100 (10.3%) of 969 survey points.During 2016, we saw or heard 241 squirrels at a total of 184 survey points (18.6%), whereas during 2017, only 47 squirrels were detected at 46 points (4.7% of point counts).In 2016 no Gray-cheeked Thrushes were observed at elevations below 340 m (representing 18.5% of points), whereas in 2017 only one thrush was observed below 317 m (representing 14.9% of points).The proportion of survey points within 50-m elevation increments where thrushes were observed steadily increased above these elevations and peaked at 525-575 m, where thrushes were detected at 29% and 22% of points during 2016 and 2017, respectively (Fig. 1).Conversely, in both years the proportion of survey points where squirrels were observed was highest below 200 m and then decreased steadily to an upper elevation limit at ~500 m (Fig. 1).
In 2016 we were 13 times more likely to observe a thrush at points where squirrels were not seen compared to those where squirrels were seen (χ² = 23.69,p < 0.001), while in 2017 after a large drop in squirrel numbers (McDermott et al. 2020), we were only 1.7 times more likely to see a thrush at points without squirrels (χ² = 0.02, p = 0.902).Although the elevation distribution of squirrels and thrushes overlapped from 340-515 m in 2016 and 317-513 m in 2017, the distributions of squirrels and Gray-cheeked Thrushes were significantly different in both years (2016: D = 0.81, p < 0.001; 2017: D = 0.68, p < 0.001).The two species were only detected at the same points on six occasions across the two years: two in 2016 and four in 2017.These sites of co-occurrence fell between 370-470 m elevation, near the lower elevation limit of thrushes and the upper limit of squirrels.Each of the nine best thrush occurrence models retained squirrel probability of occurrence as a predictor, along with a suite of land cover variables at the local and landscape scales (see Table 2 for model summaries).When we ranked these simplified models according to AICc, Model 2 performed better than the rest (Table 2) with ΔAICc values of 3.6 or greater for other models.This best model indicated that year was important, with lower thrush occurrence in 2017 than 2016, and that squirrel probability of occurrence had a strong negative association with thrush occurrence when all covariates were taken into account (Table 3, Fig. 2); predicted probability of thrush occurrence fell to zero when the probability of squirrel occurrence exceeded 40%.At the local scale, Gray-cheeked Thrush presence was positively related to both the proportion of tall scrub (Fig. 3A) and proportion of harvested forest (Fig. 3B).Shoreline length had a quadratic relationship with predicted thrush occurrence at the local scale, peaking when ~495 m of shoreline was present within the 132 m radius (Fig. 3C).Thrush presence was also related to local-scale forest composition through positive association with proportion of coniferous forest (Fig. 3D); predicted thrush presence also increased with up to 30% cover of mixed forest, but then was weakly negatively associated with higher proportions of mixed forest (Fig. 3E).At the landscape scale, thrush presence was negatively associated with tall scrub, regenerating forest, and second growth forest (Fig. 3F-H).
With respect to forest harvesting, for survey points above 317 m elevation that contained a cutblock (n = 664) an additive model including both cutblock age and proportion of harvested forest was the most informative in predicting thrush presence when compared to univariate models or a model that included an interaction between the two variables (Table 4).The multiplicative model was within 2 AICc of the best model but had a nearly  identical log likelihood and was more complex, so the interaction between cutblock age and proportion of harvested forest is likely uninformative based on the criteria outlined by Leroux (2019).In the best model, thrush occurrence was negatively related to time since harvesting and positively associated with the proportion of harvested forest (Figs.S1A and S1B in Appendix 1).However, our post hoc analysis using the best model but with a truncated range of age of cuts indicated that thrush occurrence was not related to time since harvesting for 12-18 year-old cutblocks (Fig. S1D in Appendix 1).
For the harvesting method analysis, which was restricted to elevations between 419 and 564 m, the model including additive effects of both clear-cuts and modified strip cuts had the most power to explain thrush occurrence (Table 5).Occurrence of thrushes was positively influenced by the amount of both clear-cuts and modified (strip) cuts in a linear manner (Fig. 4).The slope of the relationship was steeper for modified cuts than for clear-cuts, suggesting a stronger positive effect (Fig. 4).However, model fit was weaker for modified cuts, as reflected by broader confidence intervals, likely due in part to the relatively low number of survey points having modified cut.

DISCUSSION
Our findings are consistent with the hypothesis of Whitaker et al. (2015) that the introduction of red squirrels constrains the contemporary distribution of the Newfoundland Gray-cheeked Thrush.As predicted, we found nearly allopatric distributions of the two species: thrushes at high elevations, red squirrels at low elevations, with a zone of overlap at intermediate elevations in which both species are rare.These findings are also consistent with BBS data from 2001-2015 that showed thrush occurrence near zero at all but the highest elevations surveyed by BBS on the island (Robineau-Charette et al. 2023).Although we have no data on the distribution of thrushes in our study area prior to the arrival of squirrels, BBS data from across the island of Newfoundland, as well as other reports, indicate that Gray-cheeked Thrushes were common songbirds at lower elevations (i.e., below 400 m.a.s.l.) throughout Newfoundland in the 1970s and 1980s (Lamberton 1976a, b, Marshall 2001, SSAC 2010, Robineau-Charette et al. 2023).Historical survey data for higher elevation areas on the island of Newfoundland are limited, but Lamberton (1976a) also reported that thrushes were abundant in the highlands of the Long Range Mountains in Gros Morne National Park, just ~15 km west of our study area; this is consistent with their occurrence in montane boreal habitats across the southern portion of the species' range (Whitaker et al. 2020).
Red squirrel population ecology may have important consequences for the current and future distribution of Gray-cheeked Thrushes.We found no difference between years in the elevation range distribution of squirrels, but did observe a four-fold change in squirrel detections between 2016 and 2017.This is consistent with squirrel population dynamics elsewhere and is largely driven by year-to-year variation in the masting of conifer trees (Wheatley et al. 2002).Large annual fluctuations in red squirrel populations (Martin andJoron 2003, Boutin et al. 2006) mean that the predation pressure that they exert on Gray-cheeked Thrushes and other songbirds (e.g., Darveau et al. 1997, Martin and Joron 2003, Mahon and Martin 2006) varies substantially from year-to-year.In agreement with this, the apparent decline in thrush detections that we observed in 2017 may have reflected poor recruitment during 2016, when squirrel detections were four times higher.Similar population fluctuations caused by cyclical fluctuations in squirrel abundance and nest predation have been reported in several other species including Bicknell's Thrush (McFarland et al. 2008, Hill et al. 2019), Brown Creeper (Certhia americana;Poulin et al. 2010), and American Redstart (Setophaga ruticilla; Sherry et al. 2015).
Our study compared species distributions but did not assess fitness consequences of squirrels on Gray-cheeked Thrushes.Therefore, this study does not provide direct evidence that squirrels caused the collapse of the Newfoundland thrush population or precipitated the restriction of thrushes to their contemporary montane range.However, there was a close temporal correlation between the rapid colonization of Newfoundland by squirrels during the 1970s and 1980s (Whitaker 2015) and collapse of the Gray-cheeked Thrush population beginning in the 1980s (Jacques Whitford Environment 1993, SSAC 2010).Unfortunately, there is a dearth of monitoring data between the mid-1980s, when thrushes were still common in many areas, and the early 2000s, by which time they had disappeared from much of the island (SSAC 2010).This makes it difficult to relate thrush declines to the precise timing of squirrel colonization, though one local example comes from Gros Morne National Park: during 1974 and 1975 an average of 24 thrushes was counted along a 50-stop BBS route across a low elevation area of the park (BBS Route 57021; SSAC 2010, Smith et al. 2020).The first squirrel was reported in the park in 1975 (Minty 1976), when Gray-cheeked Thrushes were still abundant in these lowland regions of the park (Lamberton 1976a), and as elsewhere on Newfoundland, squirrels quickly became abundant throughout lowland forests in the park (Whitaker 2015;G. Robineau-Charette and D. M. Whitaker, unpublished data).
Average counts on that BBS route dropped by 68% from 1981 to 1985 (mean = 7.75 thrushes per year), and thrushes virtually disappeared from park lowlands by 1992 (Jacques Whitford Environment 1993).An average of just one thrush per year was detected along this same BBS route from 1992 to 1997 and none have been detected since 1997 (Pardieck et al. 2020, Smith et al. 2020).
Although understanding the likely causes of range restriction for the Gray-cheeked Thrush is clearly important for focusing conservation efforts, a second important finding of this study regards the species' use of managed forests.As with Whitaker et al. (2015), we found that Gray-cheeked Thrushes were strongly associated with regenerating clear-cuts at the local scale.But here we also were able to show that although thrush occurrence was positively associated with the proportion of forest affected by harvesting, thrush occurrence did not vary with clear-cut age from 12 to 18 years post-harvest.The overall pattern we observed is reminiscent of the habitat associations of Bicknell's Thrush, which make extensive use of regenerating stands (Chisholm and Leonard 2008).
Our findings and past research (Whitaker et al. 2015) suggest that the levels of timber harvest and silvicultural techniques used in our study area may benefit Gray-cheeked Thrushes.However, although increased harvesting at the local scale led to greater likelihood of thrush occurrence, the amount of habitat available at the landscape scale that was regenerating from either harvesting or natural disturbances was negatively related to thrush occurrence.This suggests that there is a limit in the extent to which the modified habitat created by harvesting across the broader landscape benefits Gray-cheeked Thrushes.A closer look at point count locations associated with harvested forest suggested that thrushes were more common in larger, 12-year-old regenerating cutblocks that were either clear-cut or produced through modified strip cuts, but that the value of harvested areas declined linearly with age.Our post-hoc test on a truncated dataset controlled for the skewed elevation distribution of older cutblocks, but only included 12-18-year-old cuts.This indicated that cut age was unimportant, suggesting that cuts ranging in age from 12 to 18 years post-harvest were equally used by thrushes.Further study is needed to assess the value of younger (< 12 years) and older (> 18 years) cuts for Gray-cheeked Thrushes.Likewise, our finding that modified (strip) cuts may support greater occurrence of thrushes than clear-cuts requires further investigation, as the limited amount of this type of harvesting in our study area precluded strong inference.
Aspects of the broader habitat associations of Gray-cheeked Thrushes in our study area appear linked to avoidance of nest predation balanced by foraging needs.For example, Graycheeked Thrush nests are often on the ground or relatively low in trees (< 2 m; Whitaker et al. 2020), which could place them at high risk of squirrel depredation (Pelech 1999, Martin andJoron 2003).Our findings suggest a positive association between thrush occurrence and the amount of tall conifer scrub at the local scale, but a negative association at the landscape scale.For this reclusive species, tall conifer scrub, which squirrels avoid (McDermott et al. 2020), may provide a safe location to nest and display because of the tight, dense weave of branches.But at the landscape scale, tall scrub may be less useful for thrushes as this low productivity habitat lacks the arthropod populations that are the foundation for thrush diets during the breeding season (Whitaker et al. 2020).
The negative relationship with increasing proportion of regenerating forest and second growth forest at the landscape scale has also been seen in work by Thompson et al. (1999) and Whitaker et al. (2015).This may indicate that a landscape ) are often more focused on nesting and territorial defense, but home ranges are often much larger than the defended territory and at this broader landscape scale activities typically include foraging and extra-pair mate acquisition (Leonard et al. 2008, Whitaker andWarkentin 2010).Thus Gray-cheeked Thrushes may select habitat at the landscape scale for food availability, and at the local scale for food availability and, additionally, safety of nests from predators, the latter of which may be increasingly important in the context of introduced squirrels.However, it should be noted that red squirrels are more likely to use second growth (30-70 year old) forests in Newfoundland, where cone production is higher (Thompson andCurran 1995, McDermott et al. 2020).Thus, reduced thrush occurrence in landscapes having more forest of this age may also be a consequence of greater suitability for squirrels.

CONCLUSION
Our findings add to the evidence suggesting that the introduction of red squirrels to Newfoundland was a key stressor that led to the near extirpation of the Newfoundland Gray-cheeked Thrush from lower elevations.Matching the niche reduction hypothesis of Scheele et al. (2017), higher elevation areas and those nearshore islands without squirrels may now act as refugia, allowing thrushes to persist under reduced threat from red squirrels.Terrain above ~400 m.a.s.l., in which Gray-cheeked Thrushes still regularly occur, is restricted to the western portion of Newfoundland and represents ~11% of the island.However, much of this high elevation landscape is covered by arctic/alpine barrens and bogs where Gray-cheeked Thrushes do not occur, leaving relatively little area for this thrush to persist.Their contemporary elevation distribution may also now match that of Gray-cheeked Thrushes along some southern portions of their continental range, which overlaps the northern portion of the range of red squirrels and where the thrushes appear to be restricted to montane forests (Höhn andMarklevitz 1974, Di Corrado 2015).This may suggest a broader vulnerability of Gray-cheeked Thrushes to squirrels, but no information is available on the codistribution of these species on continental portions of the thrush's breeding range.
Within western Newfoundland montane forests, regenerating clear-cuts and strip cuts < 19 years old support high numbers of thrushes, possibly because they offer a preferred combination of dense cover and high productivity (e.g., of foods such as insects and fruit).However, we cannot rule out that these may also be ecological sinks.If future circumstances arise that allow squirrels to colonize these montane forests, Gray-cheeked Thrush populations could be placed at further risk.This could happen as natural regeneration after forest harvesting matures into 30-70year-old stands, which are favored by red squirrels (McDermott et al. 2020), but which are currently rare at higher elevations in our study area (McCarthy and Weetman 2006).Similarly, climate change may enable expansion of boreal vegetation further upslope (Harsch et al. 2009, Myers-Smith andHik 2018) or largescale disturbances to occur, such as insect outbreaks, which are currently limited to lower elevations by climate (McCarthy andWeetman 2006, Arsenault et al. 2016).Although these changes may provide short-term benefit to montane Gray-cheeked Thrush populations through the creation of early-successional forest, there may also be a longer term impact if post-disturbance forests later allow squirrels to increase in abundance at higher elevations.
Management of these montane forests must take into consideration the impact any changes to forest structure may have on the potential for further upslope expansion by red squirrels.
Likewise, introduction of squirrels to additional nearshore islands must be prevented.
1 Spatial segregation between Gray-cheeked Thrush and an introduced nest predator in a managed forest landscape APPENDIX 1 -SUPPLEMENTAL MATERIALS Table S1 Table S2

Fig. 1 .
Fig. 1.Red squirrels (Tamiasciurus hudsonicus) were more common at lower elevations of the Long Range Mountains, Newfoundland, whereas Gray-cheeked Thrushes (Catharus minimus minimus) subspecies were almost exclusively detected at elevations higher than 375 m for both 2016 and 2017.Data labels above x-axis indicate number of point counts (total n = 1960) in each elevation category.

Fig. 2 .
Fig. 2. Strong negative relationship between predicted probability of Gray-cheeked Thrush (Catharus minimus minimus) subspecies presence in the Long Range Mountains, Newfoundland in 2016 and 2017 and red squirrel (Tamiasciurus hudsonicus) probability of occurrence.Plot was generated from fitted values while holding all other variables constant at their mean.Dotted lines indicate 95% confidence intervals and individual observations (n = 1670) are shown as presences and absences (1 or 0).

Fig. 3 .
Fig. 3. Linear and curvilinear relationships between predicted probability of Gray-cheeked Thrush (Catharus minimus minimus) subspecies presence in the Long Range Mountains, Newfoundland in 2016 and 2017 for factors retained in the best model.Each plot was generated from fitted values while holding all other variables constant at their mean.Dotted lines indicate 95% confidence intervals and individual observations (n = 1670) are shown as presences and absences (1 or 0).All variables except for tall scrub (132 m) are statistically significant.

Fig. 4 .
Fig. 4. Positive predicted probability of Gray-cheeked Thrush (Catharus minimus minimus) subspecies presence in the Long Range Mountains, Newfoundland in 2016 and 2017 influenced by the proportion of clear-cut or modified cut within 132 m.Data were restricted to point counts at elevations between 419 and 564 m (n = 1039).Each plot was generated from fitted values while holding the other variable constant at its mean.Dotted lines indicate 95% confidence intervals and individual observations are shown as presences and absences (1 or 0).

Figure S1 A
Figure S1 A and B. Predicted probability of Newfoundland Gray-cheeked Thrush, minimus subspecies presence at elevations above 317 m in the Long Range Mountains, Newfoundland as a function of the age of harvested areas and proportion of habitat within 132 m that was cut, respectively (n = 664).Plots were generated from predicted values while holding the other variable constant at its mean; observations are shown as presences and absences (1 or 0) while dotted lines indicate 95% confidence intervals.C. Distribution of cut age classes as a function of elevation.D. Post-hoc analysis showing probability of thrush presence as a function of the age of cut areas, where data were truncated to only include points where 12 to 18 year-old cuts were present (n = 601).

Table 1 .
Variables included in models to explain occurrence of Gray-cheeked Thrush (Catharus minimus minimus) subspecies in the Long Range Mountains, Newfoundland in 2016 and 2017.All variables occurred at the local (132 m) and landscape (1250 m) scale, except mixed mature forest (mt3), which only occurred at the landscape scale.
We then created additional composite successional stage/ composition variables by merging age class with forest cover type, resulting in nine composite variables that occur at both spatial scales.Four of these variables were eliminated because of rarity (see below), and the five successional stage/composition variables that were retained for model selection are listed in Table1.Other land cover classes included open habitat (bogs, barrens, and other natural openings), and harvested forest (clear-cuts and modified strip cuts; Table

Table 2 .
One model (model 2) of nine land cover and red squirrel models was clearly more effective for explaining Gray-cheeked Thrush (Catharus minimus minimus) subspecies occurrence in 2016 and 2017 in the Long Range Mountains, Newfoundland (n = 1670).Variables retained in each model have a coefficient under the model column, and otherwise were dropped from the model.See Table1for variable name descriptions.Local and Landscape indicate variable scale at radii of 132 m and 1250 m, respectively.Composition, Successional stage, Composition/successional stage, and Other indicate variable groupings within each variable scale.

Table 3 .
Year, land cover variables, and red squirrel (Tamiasciurus hudsonicus) predicted occurrence were among the parameter estimates for the best model explaining Gray-cheeked Thrush (Catharus minimus minimus) subspecies occurrence in the Long Range Mountains, Newfoundland in 2016 and 2017 (n = 1670; see Table2).See Table1for variable descriptions.

Table 5 .
Comparison of Generalized Additive Models investigating the influence of proportion of modified cut (MC) and clear-cut (CC) on Gray-cheeked Thrush (Catharus minimus minimus) subspecies occurrence in the Long Range Mountains, Newfoundland in 2016 and 2017 (n = 1039).