Research Paper

INTRODUCTION

The Canada Warbler (Cardellina canadensis) is a long-distance neotropical migrant listed as threatened in Canada under the Species at Risk Act (Environment Canada 2016) and as a Species of Greatest Conservation Need in nearly every U.S. state where it breeds (Reitsma et al. 2010). Canada Warbler populations have declined by an alarming 71% since 1970, with annual rates of decline reaching 5.4% in recent years (Smith et al. 2024). Their summer breeding range spans the southern boreal forest of North America, extending from northeastern USA to western Canada, and their winter range covers the eastern slopes of the Andes Mountains from Venezuela to Peru (Reitsma et al. 2010). These declines are particularly concerning given that Canada Warblers arrive late to the breeding grounds (late May to early June) and depart early (Flockhart 2007), leaving a narrow window for successful reproduction that may make them especially vulnerable to habitat disturbance during the breeding season.

Throughout their range, Canada Warblers are generally associated with mature mixedwood or deciduous-dominant forest stands ≥80 yrs old (Ball et al. 2016, Grinde and Niemi 2016, Alberta Biodiversity Monitoring Institute (ABMI) and Boreal Avian Modelling Project (BAM) 2023), although habitat associations vary regionally. In western Canada, they primarily select older deciduous and mixedwood forests with complex vertical structure, particularly near small, incised streams, and show preference for stands with high densities of large-diameter trees (≥30 cm diameter at breast height (DBH); Ball et al. 2016, Hunt et al. 2017). In the eastern portion of their range, they occupy a broader range of habitats, including regenerating forest stands with dense understory vegetation and swampy forests (Westwood et al. 2019, Bale et al. 2020). Beyond local stand characteristics, Canada Warblers are influenced by habitat characteristics at multiple spatial scales. Stand age, understory density, and proximity to water appear most influential at the local scale, whereas broader landscape-scale features like the amount and configuration of forest cover also affect their distribution (Ball et al. 2016, Grinde and Niemi 2016). Social factors such as conspecific attraction may further influence habitat selection patterns, with birds sometimes clustering their territories even when suitable habitat is available elsewhere (Hunt et al. 2017, Cupiche-Herrera et al. 2024). This pronounced regional variation in habitat use underscores the need for region-specific research to develop effective conservation strategies that address local population drivers rather than relying on range-wide generalizations.

In Alberta’s boreal forest, where Canada Warblers reach their northwestern range limit, conserving this threatened species presents unique challenges due to the cumulative impacts of multiple industrial sectors. Unlike other portions of Bird Conservation Region 6, which experience predominantly forestry and agriculture disturbances (Saskatchewan and Manitoba) or remain largely intact with localized mining impacts (Northwest Territories), Alberta’s boreal forest faces an intersecting mosaic of forestry, energy development, and transportation infrastructure that varies in extent, intensity, and persistence (Pickell et al. 2015, Fisher and Burton 2018). Energy development alone has created networks of linear features reaching densities up to 40 km/km² in oil sands regions, ranging from pipeline corridors up to 100 m wide, to narrow seismic lines 3–8 m wide (Stern et al. 2018, Filicetti at al. 2019). These linear features overlay extensive forestry operations that remove the older deciduous stands Canada Warblers prefer, potentially influencing functional connectivity and creating edge effects distinct from traditional forestry operations (Sisk and Battin 2002, Mahon et al. 2019). Understanding whether narrow linear features like seismic lines (often considered “low impact”) contribute meaningfully to habitat fragmentation for old-forest specialists remains a critical knowledge gap that has implications for conservation and restoration prioritizations.

Disentangling the relative impacts of habitat loss vs. fragmentation in Alberta’s multi-disturbance landscape requires moving beyond generic forest classifications to species-specific approaches that define habitat from the Canada Warbler’s perspective. Although the habitat amount hypothesis posits that species responses depend primarily on total habitat quantity regardless of spatial arrangement (Fahrig 2013), alternative frameworks suggest that fragmentation effects can intensify as habitat falls below critical thresholds (Andrén 1994). Yet previous research has often relied on broad forest classifications (e.g., forest vs. non-forest) that fail to capture species-specific habitat requirements and distinguish between disturbance types that may differ in their fragmentation effects (Fischer and Lindenmayer 2007, Betts et al. 2014). For Canada Warblers in Alberta, determining which industrial disturbances constitute functional barriers—and whether fragmentation effects interact with habitat amount—is essential for predicting where population declines will accelerate and for prioritizing conservation investments. Answering these questions requires moving beyond static species distribution models that provide point estimates of occurrence, to examine the underlying population dynamics. Dynamic occupancy modeling explicitly separates colonization and extinction processes while accounting for imperfect detection (MacKenzie et al. 2017), providing mechanistic insight into how demographic processes respond to Alberta’s complex and dynamic disturbance gradients (Kéry and Schaub 2012, Yackulic et al. 2015). By distinguishing whether population declines result primarily from low colonization or high extinction rates, dynamic occupancy models reveal which demographic pathway is most compromised, informing whether conservation strategies should prioritize landscape connectivity or habitat quality enhancement (Yackulic et al. 2015, MacKenzie et al. 2017).

Research objectives

To advance our understanding of Canada Warbler responses to habitat loss and fragmentation across Alberta’s boreal forest, we implemented a multi-scale dynamic occupancy modeling framework. Our objectives were to: (1) quantify the relative effects of habitat amount vs. fragmentation on occupancy dynamics; (2) determine whether fragmentation effects are mediated by the total amount of available habitat; (3) evaluate whether narrow linear features (seismic lines) contribute meaningfully to fragmentation effects; and (4) identify the spatial scales at which these relationships are strongest. We employed a species-specific approach to measuring both habitat amount and fragmentation to test whether these processes interact to produce non-linear effects on initial occupancy, colonization, and extinction. We predicted that fragmentation would negatively affect initial occupancy and colonization while positively affecting extinction probability, with these effects intensifying as habitat amount decreases. We tested three increasingly refined fragmentation definitions: from large polygonal disturbances alone, to including wide linear features, to including all fragmenting features, even narrow seismic lines. We predicted that even narrow seismic lines would contribute meaningfully to fragmentation effects. Finally, we conducted these analyses across multiple spatial scales (territory to landscape level), hypothesizing that effects would be strongest at intermediate scales (78 ha), corresponding to typical territory size and movement patterns. By employing dynamic occupancy models with species-specific habitat metrics across multiple spatial scales, our research will identify: (1) which disturbance types influence Canada Warblers, (2) whether habitat amount and fragmentation interact to produce threshold effects, and (3) which demographic processes (colonization vs. extinction) are most sensitive to landscape change. These insights will provide evidence-based guidance for disturbance management and conservation planning in Alberta’s multi-use landscape.

METHODS

Bird data

Canada Warbler presence/absence data were obtained from WildTrax, an online platform that stores bird survey data from multiple organizations including ABMI, BAM, Environment and Climate Change Canada, Parks Canada, the Alberta Northwest Species at Risk Committee (NWSAR), and the Bioacoustic Unit (BU) at the University of Alberta. We compiled survey data collected using both human point counts (HPC) and autonomous recording units (ARUs). Human point counts followed unlimited-distance, 3-min. point count protocols conducted by trained observers under suitable weather conditions. Three-minute ARU recordings were processed manually by trained listeners using both visual and acoustic verification of spectrograms in the WildTrax platform, with recordings during inclement weather excluded and replaced with alternative recordings from the same survey location. All surveys were conducted during peak Canada Warbler breeding season (late May through early July) and during hours of peak vocal activity (sunrise to 10:00 AM). For a complete overview of acoustic data processing in WildTrax, see Kalukapuge et al. (2024).

For site selection, we broadly targeted forested areas (deciduous, mixedwood, or coniferous) older than 60 yrs that had been surveyed across multiple years. We restricted our analysis to surveys conducted from 2010 onward to align with the temporal extent of available high-resolution human footprint and landcover data. Although our analysis spanned 2010–2023, the specific survey years varied among sites as we leveraged all available multi-yr data to maximize spatial coverage without requiring additional field data collection. Sites were included if surveyed in at least three separate years (not necessarily consecutive) with a minimum of three within-yr visits, allowing us to estimate colonization and extinction dynamics while making efficient use of existing monitoring efforts from multiple programs operating on different schedules. After applying these filtering criteria, a total of 3555 visits from 395 sampling locations across Alberta were included in the final data set for analysis (Fig. 1).

Species-specific landscape metrics

For our detailed habitat analysis, we characterized landscape structure at multiple spatial scales around each survey location to examine the relationship between Canada Warbler occupancy and habitat configuration. Using ArcGIS Pro (ESRI 2024), we first extracted landscape features within 10-km buffers of each survey location from the Alberta Vegetation Inventory (AVI) and Human Footprint Inventory (HFI) maintained by the ABMI (2024). These 10-m resolution data sets provide detailed forest attributes (species composition, stand age, height), and comprehensive coverage of human disturbances including industrial infrastructure, roads, fine-scale linear features (e.g., 3–8 m wide seismic lines), and harvested areas. Both data sets are updated annually, allowing us to match habitat conditions to the specific year of each bird survey—a temporal alignment that is particularly important in Alberta’s boreal landscape, where new disturbances are continually added, and vegetation succession is ongoing. From these data, we measured the area of suitable Canada Warbler habitat, defined as deciduous and mixedwood forests aged 80–140 yrs, based on known habitat associations of the species in this region (ABMI and BAM 2023, Ball et al. 2016).

To test hypotheses about how different anthropogenic disturbances fragment Canada Warbler habitat, we developed three increasingly refined definitions of habitat fragmentation (Fig. 2). Definition A, our most conservative approach, considered only large polygonal disturbances (e.g., wellpads, harvested areas, industrial facilities) as fragmenting features. Definition B incorporated both polygonal disturbances and wide (>8 m wide) linear features (roads, pipelines, transmission lines) as potential barriers. Definition C provided the most refined assessment of fragmentation by including all anthropogenic disturbances: polygonal disturbances, wide linear features, and narrow (4–8 m wide) linear features (seismic lines). For each habitat fragmentation definition, we dissolved the remaining suitable habitat areas into continuous patches and exported them as habitat shapefiles for further landscape metric calculations.

Using these habitat definitions, we calculated landscape metrics at three nested spatial scales around survey locations: 150 m, 500 m, and 1000 m radii. These scales were selected to capture different aspects of Canada Warbler habitat selection and space use. The 150 m radius scale (ca. 7 ha) approximates territory sizes, aligns with standard point count survey radii, and is frequently used in bird-habitat relationship models (e.g., Matsuoka et al. 2012). The 500 m radius scale (ca. 78 ha) encompasses home range and potential breeding season movements for foraging or prospecting, and the 1000 m radius scale (ca. 314 ha) represents the broader landscape context that may influence habitat selection and dispersal (Whitaker and Warkentin 2010).

For each spatial scale and fragmentation definition, we calculated the proportion of suitable habitat and edge density at the landscape level using a 2 m spatial resolution via the landscapemetrics package in R (Hesselbarth et al. 2019). The application of increasingly refined fragmentation definitions resulted in distinct changes to the landscape configuration metric (edge density), but the proportion of habitat remains relatively consistent across definitions (Fig. 3). This approach enabled examination of occupancy responses to habitat characteristics while testing whether conclusions about habitat relationships are sensitive to how fragmentation is defined.

Statistical analysis

We used dynamic occupancy models to examine the effects of habitat amount and configuration on Canada Warbler occupancy, colonization, and extinction patterns (MacKenzie et al. 2003). Dynamic occupancy models provide a valuable framework for species of conservation concern, particularly for relatively rare and difficult-to-detect species like Canada Warblers, where accounting for imperfect detection is an important consideration. Our model structure utilized detection/non-detection data collected over three primary survey periods (yrs) and three secondary survey periods (repeated surveys within yrs). The primary survey periods allow estimation of how occupancy changes over time through local colonization and extinction, whereas the secondary periods within each primary period provide the repeated surveys necessary to estimate detection probability while assuming population closure (no changes in occupancy) within each year (MacKenzie et al. 2003). The hierarchical structure enables simultaneous estimation of initial occupancy (ψ), colonization (γ), extinction (ε), and detection probability (p), providing mechanistic insights into how landscape characteristics influence each component of Canada Warbler demographic patterns while accounting for imperfect detection.

We constructed five candidate models to test our competing hypotheses about the effects of habitat amount and fragmentation on occupancy dynamics: (1) a null model with covariates on detection probability only; (2) a habitat amount model; (3) a fragmentation model; (4) an additive model combining habitat amount and fragmentation; and (5) an interactive model between habitat amount and fragmentation. All models included potential detection covariates (time of day, Julian date, and survey method—ARU vs. HPC) to account for imperfect detection. For models 2–5, we incorporated the habitat and fragmentation metrics as covariates on all occupancy parameters (initial occupancy, colonization, and extinction), as we had no a priori reason to expect differential effects across these parameters. We selected edge density as our primary fragmentation metric because it: (1) effectively captures the unique landscape context created by linear features; (2) demonstrates clear distinctions between our three fragmentation definitions. To account for inconsistent time intervals between primary surveys, we initially included “delta years” as a fixed effect on colonization and extinction parameters, representing the number of years elapsed between successive surveys at each site, however preliminary analysis revealed little effect of this covariate, so for the sake of minimizing model complexity, we removed it.

Models were fit using the “colext” function in the unmarked package (Fiske and Chandler 2011) in R version 4.4.2 (R Core Team 2024). Prior to analysis, we standardized landscape variables (mean = 0, SD = 1) to facilitate direct comparison of effect sizes across parameters. We examined the distribution of habitat and fragmentation metrics using histograms and quantile-quantile plots to verify approximate normality and identify potential outliers. We also calculated Pearson correlation coefficients between landscape variables to assess potential collinearity. No strong correlations (|r| > 0.7) were found between metrics included together in the same models, indicating that strong multicollinearity was unlikely to affect parameter estimates or their interpretation.

For each combination of fragmentation definition (A, B, C) and spatial scale (7, 78, 314 ha), we compared the five candidate models using Akaike’s Information Criterion (AIC). Models with ΔAIC < 2 were considered competitive (Burnham and Anderson 2002). For reporting and comparing coefficient estimates across scales, we used model averaging weighted by AIC weights when multiple competitive models existed. This approach allowed us to account for model selection uncertainty in parameter estimation. For visualization of ecological relationships and derived occupancy estimates over time, we used predictions from the single top-ranked model from the best performing scale and fragmentation definition combination. We verified that the ecological interpretations remained consistent between model-averaged coefficients and the top model predictions. To evaluate our hypotheses about fragmentation definitions and scale of effects, we systematically compared model performance metrics and effect sizes across our three fragmentation quantification approaches and spatial scales.

Model fit and diagnostics

We assessed the fit of our top models using multiple approaches. First, we performed MacKenzie-Bailey goodness-of-fit tests with 1000 parametric bootstrap simulations (MacKenzie and Bailey 2004) using the “mb.gof.test” function in the AICcmodavg package (Mazerolle 2025). This test generates a chi-square statistic and an estimate of the variance inflation factor (ĉ), where values close to 1.0 indicate adequate fit with no overdispersion. Additionally, we evaluated the predictive performance of our top models using receiver operating characteristic (ROC) curves and the area under the curve (AUC) statistic. For each primary survey period, we extracted model-predicted occupancy probabilities and compared them with observed detection histories. Receiver operating characteristic curves plot the true positive rate against the false positive rate across different threshold values. The resulting AUC values range from 0.5 (no discriminatory ability) to 1.0 (perfect discrimination), with values > 0.7 considered acceptable and > 0.8 considered excellent (Hosmer et al. 2013). For each model, we also calculated the percentage of sites correctly classified as occupied or unoccupied using a probability threshold of 0.5. To test for spatial autocorrelation, we used Moran’s I test with a k-nearest neighbor approach (k=5) implemented in the spdep package (Pebesma and Bivand 2023) in R. For each sampling season (primary period), we calculated Moran's I and assessed statistical significance using randomization tests with 1000 permutations. Significant positive values would indicate that model residuals were spatially clustered, suggesting unmodeled spatial processes may exist.

RESULTS

Occupancy patterns and population dynamics

Canada Warblers were relatively uncommon across our study sites, with a raw detection rate of 6.0% (214 of 3555 visits). Of the 395 survey locations, 113 (28.6%) had at least one Canada Warbler detection over the study period. When detected, observations typically consisted of single individuals, with multiple individuals recorded in only 26 visits (0.7% of total visits).

In our overall best-fit dynamic occupancy model, the estimated initial occupancy rate was 24.5% (95% confidence interval (CI): 12.9 to 36.9%). We observed an overall declining trend in Canada Warbler occupancy over the three primary study periods, from an estimated occupancy probability of 30.0% (95% CI: 22.1 to 39.0%) of sites in primary period (yr) 1 to 19.2% (95% CI: 10.1 to 31.0%) in primary period 3 (Fig. 4). Over the study period, the local extinction rate of 42.3% (95% CI: 24.1 to 62.2%) was substantially higher than the colonization rate of 8.2% (95% CI: 3.8 to 15.5%), although with much greater uncertainty.

Detection probability

After accounting for imperfect detection, the estimated per-visit detection probability was 30.9% (95% CI: 22.6 to 40.1%), indicating that the species was frequently present but undetected during surveys. We found a consistent negative effect of Julian date on detection probability (β = -0.29, 95% CI: -0.50 to -0.09), with confidence intervals not overlapping zero. In practical terms, for each standard deviation increase in Julian date (approximately 10 d), detection probability decreased by approximately 7%. Similarly, detection probability was consistently lower when HPCs were used as the survey method (β = -0.58, 95% CI: -1.09 to -0.07), compared with ARU as the survey method. Time of day also showed a slight negative effect on detection probabilities (β = -0.18, 95% CI: -0.42 to 0.07), although the confidence intervals slightly overlapped zero, suggesting more uncertainty in this relationship.

Model selection and fragmentation definitions

Across all spatial scales, models incorporating both habitat amount and fragmentation (edge density) metrics substantially outperformed null models (Table 1). Notably, across all fragmentation definitions, habitat-only models consistently performed worse than null models, indicating a lack of predictive power of habitat amount alone on Canada Warbler occupancy dynamics. The least refined fragmentation definition (Definition A), which included only larger polygonal disturbances, provided the best model fit across all three spatial scales, although this advantage was most pronounced at the 78-ha scale and least distinct at the 314-ha scale, where the top models using each method were competitive (ΔAIC < 2; Append. 1, Table A1).

At the 7-ha scale, the additive Habitat+Edge model was the top performing (AIC = 1404.84, AICwt = 0.51), with the Edge model also competitive (AIC = 1405.66, ΔAIC = 0.81, AICwt = 0.34). Together, these models accounted for approximately 85% of model weight. At the 78-ha scale, the Habitat×Edge interaction model was the top-performing (AIC = 1403.37, AICwt = 0.52), with the Edge model competitive (AIC = 1403.89, ΔAIC = 0.53, AICwt = 0.40). These two models accounted for approximately 92% of model weight. And finally, at the 314-ha scale, the Habitat×Edge interaction model performed best (AIC = 1405.36, AICwt = 0.52), with the Edge model (AIC = 1406.99, ΔAIC = 1.63, AICwt = 0.23) remaining competitive. The overall best-performing model was the Habitat×Edge interaction model at the 78-ha scale, although the difference in AIC values across scales was relatively small (ΔAIC < 2 between best models).

Scale-dependent effects on dynamic occupancy parameters

Initial occupancy (ψ)

Edge density exhibited negative effects on initial occupancy probability across all spatial scales, with stronger effects at the 78-ha and 314-ha scales (Fig. 5). The model-averaged edge density coefficient at the 7-ha scale was -0.40 (95% CI: -0.67 to -0.13), increasing in magnitude to -0.71 (95% CI: -1.12 to -0.27) at the 78-ha scale. At the 314-ha scale, edge density maintained a strong negative effect (model-averaged β = -0.65, 95% CI: -1.05 to -0.25). These confidence intervals consistently excluded zero, indicating robust negative relationships. In ecological terms, at the 78-ha scale, a site with edge density at the 75th percentile (69.64 m/ha) was approximately 46% less likely to be initially occupied than a site with edge density at the 25th percentile (46.16 m/ha), holding habitat amount constant at its mean value.

The effect of habitat amount on initial occupancy varied with scale, showing a positive effect at the 7-ha scale (β = 0.36, 95% CI: -0.12 to 0.84) and negative effects at the 78-ha (β = -0.21, 95% CI: -0.60 to 0.17) and 314-ha scales (β = -0.17, 95% CI: -0.57 to 0.22). In all cases, confidence intervals overlapped zero, suggesting uncertainty in these relationships.

The top models at the 78-ha and 314-ha scales both included negative interactions between habitat amount and edge density (78 ha: β = -0.50, 95% CI: -0.88 to -0.12; 314 ha: β = -0.40, 95% CI: -0.73 to -0.07), with confidence intervals excluding zero. This indicates that fragmentation effects intensify as habitat amount increases (Fig. 6). Initial occupancy probability was highest in landscapes with high habitat amounts and low edge density (0.8), but the strength of the negative relationship between edge density and occupancy varied with habitat amount, being strongest in high-habitat landscapes and weaker in low-habitat landscapes (Fig. 6).

Colonization probability (γ)

Edge density showed consistent negative effects on colonization probability across all scales, although with varying precision (Fig. 5). At the 7-ha scale, the model-averaged coefficient for edge density was -0.44 (95% CI: -0.86 to -0.01), with confidence intervals just excluding zero, indicating a consistent effect. At the 78-ha scale, edge density maintained a negative effect (β = -0.19, 95% CI: -0.74 to 0.36), although with wider confidence intervals overlapping zero. The effect was most pronounced at the 314-ha scale (β = -0.60, 95% CI: -1.14 to -0.06), with confidence intervals excluding zero. In other words, at the 314-ha scale, a site with edge density at the 75th percentile (57.84 m/ha) was approximately 49% less likely to be colonized than a site with edge density at the 25th percentile (40.19 m/ha), holding habitat amount constant at its mean value.

Habitat amount had positive but uncertain effects on colonization across all scales (7 ha: β = 0.41, 95% CI: -0.51 to 1.32; 78 ha: β = 0.35, 95% CI: -0.43 to 1.14; 314 ha: β = 0.32, 95% CI: -0.48 to 1.13), with confidence intervals overlapping zero in all cases. The interaction between habitat amount and edge density was negligible at the 78-ha scale (β = 0.00, 95% CI: -0.58 to 0.58), and there was a weak negative interactive effect at the 314-ha scale (β = -0.25, 95% CI: -0.77 to 0.27), with confidence intervals overlapping zero.

Extinction probability (ε)

Extinction parameter estimates showed substantial uncertainty across all scales and models (Fig. 5). At the 7-ha scale, edge density had a positive effect on extinction probability (model-averaged β = 0.43, 95% CI: -0.07 to 0.93), whereas habitat amount had a weak negative but uncertain effect (β = -0.32, 95% CI: -1.07 to 0.44).

At broader scales, the relationships showed even greater variability, with confidence intervals widely overlapping zero for edge density (78 ha: β = 0.03, 95% CI: -0.90 to 0.95; 314 ha: β = -0.23, 95% CI: -0.78 to 0.31), habitat amount (78 ha: β = 0.54, 95% CI: -0.21 to 1.29; 314 ha: β = 0.44, 95% CI: -0.17 to 1.04), or their interaction (78 ha: β = 0.38, 95% CI: -0.38 to 1.13; 314 ha: β = 0.00, 95% CI: -0.47 to 0.46).

Model fit and diagnostics

The MacKenzie-Bailey goodness-of-fit tests for our top models at the 78-ha scale indicated adequate fit. The Habitat×Edge interaction model showed no evidence of lack of fit (χ² = 25.05, P = 0.51) with no overdispersion (ĉ = 0.96). Similarly, the Edge-only model demonstrated good fit (χ² = 24.99, P = 0.532, ĉ = 0.94). Model fit statistics for top models at 7-ha and 314-ha scales showed similar patterns of adequate fit (χ² p values > 0.4, ĉ < 1.0).

For predictive performance, our top Habitat×Edge model showed moderate discrimination ability across the three primary survey periods (AUC values = 0.66, 0.56, and 0.59, respectively). The Edge-only model showed similar performance (AUC values = 0.66, 0.46, and 0.61). Although these AUC values suggest moderate discriminatory ability, they are typical for rare species occupancy models. The high percentage of sites correctly classified (80–90%) primarily reflects the models’ ability to predict absences in this system with relatively low occupancy rates.

Moran’s I tests on model residuals revealed varying spatial autocorrelation patterns across sampling seasons. For the Habitat×Edge model, season 1 showed no significant spatial autocorrelation (I = 0.04, p = 0.07), season 2 exhibited significant spatial autocorrelation (I = 0.16, p < 0.001), and season 3 showed no evidence of spatial autocorrelation (I = -0.01, p = 0.55). The Edge-only model displayed similar patterns, with significant spatial autocorrelation in season 1 (I = 0.06, p = 0.02) and season 2 (I = 0.17, p < 0.001), but none in season 3 (I = 0.01, p = 0.33). This spatial structure in model residuals suggests some unmeasured spatially correlated factors may have influenced occupancy dynamics, particularly in season 2.

DISCUSSION

Canada Warbler occupancy dynamics in Alberta’s boreal forest are strongly influenced by landscape configuration, with edge density consistently emerging as the dominant predictor across spatial scales and fragmentation definitions. To our knowledge, this represents the first species-specific dynamic occupancy analysis for Canada Warblers, providing novel insights into how their colonization–extinction dynamics respond to industrial disturbance patterns. Models incorporating edge density substantially outperformed habitat-amount-only models. This contradicts the habitat amount hypothesis, which states that fragmentation effects are negligible once habitat quantity is properly accounted for (Fahrig 2013). More surprisingly, fragmentation effects intensified in landscapes with greater habitat availability rather than diminishing below a certain habitat threshold—a pattern opposite to theoretical predictions from the fragmentation threshold hypothesis (Andrén 1994). These configuration effects were strongest at scales encompassing multiple territories (78–314 ha), indicating that conservation strategies focused solely on maximizing forest cover within small management units are unlikely to reverse population declines. Maintaining low-edge forest configurations appears equally critical as preserving total forest cover for Canada Warbler persistence in heavily industrialized boreal landscapes.

Colonization-extinction imbalance

The declining occupancy trend observed across our study region, from 30.0% of sites occupied in the first survey period to 19.2% in the third, suggests ongoing population challenges partially attributable to forest fragmentation. Although these patterns don’t represent continuous annual trends as surveys encompass non-consecutive years (2010–2023), this decrease within suitable habitat aligns with documented range-wide population declines (Environment Canada 2016). The substantial imbalance between local extinction (42.3%) and colonization (8.2%) probabilities reveals the demographic mechanism underlying regional population declines: Canada Warblers are abandoning sites faster than they establish new territories. The five-fold difference indicates a population disequilibrium, where local extinctions compound across the landscape without sufficient recruitment to compensate.

Edge density negatively affected colonization probability particularly strongly at broader scales, with a 49% reduction in colonization between low-edge (25th percentile) and high-edge (75th percentile) landscapes, representing a substantial barrier to population recovery. In contrast, extinction probability showed weaker and more variable relationships with landscape metrics, suggesting that once established, individuals may persist despite fragmentation, but edge effects prevent new settlement and range expansion. Although climate-driven range shifts have been projected for boreal songbirds, including Canada Warblers (Stralberg et al. 2015), our observed extinction–colonization imbalance more likely reflects habitat-driven dynamics rather than northward expansion. Canada Warblers exhibit high breeding site fidelity (Hallworth et al. 2008), and the strong negative effects of edge density specifically on colonization (rather than extinction) indicate that landscape configuration constrains territory establishment independent of range-wide distributional changes.

Scale-dependent configuration effects

The scale-dependency of fragmentation effects provides insights into Canada Warbler habitat selection and movement patterns. Edge density effects on initial occupancy were strongest at broader spatial scales (78 ha and 314 ha) rather than at the territory scale (7 ha), indicating that Canada Warblers evaluate landscape configuration beyond individual territories during settlement. With individual territories typically ranging from 0.2–4 ha (Reitsma et al. 2010), the 78-ha scale encompasses multiple potential territories and may represent the scale at which individual dispersal and recolonization events shape population structure. This finding aligns with previous research demonstrating multiscale habitat selection in Canada Warblers, from microhabitat features like understory structure, to landscape-level patterns of forest cover (Ball et al. 2016, Grinde and Niemi 2016). For conservation, this scale-dependence indicates that management units must encompass scales relevant to population processes (> 78 ha) rather than individual territory sizes. Current forest management planning in Alberta often operates at cutblock scales (20–40 ha), which our findings suggest are too small to capture the landscape configurations that influence Canada Warbler occupancy dynamics.

Habitat amount-fragmentation interaction

The negative interaction between habitat amount and edge density contradicts the fragmentation threshold hypothesis (Andrén 1994). Although our results do indicate a non-linear relationship between habitat amount and fragmentation, rather than fragmentation effects emerging only below a certain habitat threshold, we found fragmentation effects were strongest in areas with more habitat. Conspecific attraction offers the most plausible explanation for this counterintuitive pattern. Canada Warblers exhibit social aggregation behavior, with individuals preferentially settling near conspecifics even when suitable habitat is available elsewhere (Hunt et al. 2017, Cupiche-Herrera et al. 2024). If birds select habitat based on the presence of other individuals rather than habitat quality alone, clustering in specific patches could lead to heightened sensitivity to fragmentation even in areas with high habitat availability (Bourque and Desrochers 2006, Fletcher 2006). In high-habitat landscapes, where multiple patches exist, social cueing could result in birds concentrating in a subset of these patches, increasing their sensitivity to edge effects in those patches. Conversely, in low-habitat landscapes where few patches remain, birds may occupy available habitat regardless of configuration because alternatives are absent, potentially masking fragmentation effects beyond simple habitat loss. This tendency toward social aggregation, particularly in recently disturbed areas (Cupiche-Herrera et al. 2024), could create a situation where landscapes with higher amounts of suitable habitat experience higher extinction rates or lower occupancy, when the rates actually reflect social information use rather than habitat structure preferences (Nocera et al. 2006, Betts et al. 2008).

The social aggregation mechanism would also explain the unexpected positive effect of habitat amount on extinction probability: if conspecific attraction drives clustering, those patches may experience density-dependent effects or higher social turnover that increase local extinction risk. Complex social dynamics notwithstanding, the robust negative effect of edge density across all models underscores that landscape configuration influences Canada Warbler distribution independently of habitat amount, adding to growing evidence against habitat-amount-only frameworks (Haddad et al. 2015, Fletcher et al. 2018).

Influence of fragmentation definition

The dominance of Definition A (large polygonal disturbances only) at finer spatial scales (7 and 78 ha) contradicts our prediction that the most refined definition including narrow seismic lines would best capture fragmentation effects. For Canada Warblers in Alberta’s boreal forest, larger disturbances (harvested areas, wellpads, and permanent clearings) appear more influential than narrow linear features in driving edge effects at territory and local scales. This suggests that habitat configuration resulting from these disturbances, measured as edge density, is an important driver beyond simple habitat area reduction.

When we expanded to Definition C by including narrow seismic lines, which can reach densities of 40 km/km² in some areas of Alberta (Stern et al. 2018, Filicetti et al. 2019), model performance decreased at fine scales. Canada Warbler’s apparent tolerance of narrow linear features likely reflects their documented use of partially vegetated seismic lines and ability to incorporate small gaps into territories (Gregoire et al. 2022). Unlike completely cleared areas such as wellpads or new harvest cutblocks, partially regenerated seismic lines may provide some habitat value, particularly understory structure, or at least fail to create edge effects strong enough to disrupt territory establishment (Gregoire et al. 2022). This contrasts with responses documented for other old-forest specialists in the same region: Black-throated Green Warbler (Setophaga virens) persistence was negatively affected by proximity to seismic lines (Hart et al. 2025), and Ovenbirds (Seiurus aurocapilla) and Bay-breasted Warblers (Setophaga castanea) showed decreased abundance near seismic lines (Leston et al. 2023), highlighting the species-specific nature of these responses.

At the broadest spatial scale (314 ha), all three fragmentation definitions performed equivalently, suggesting that cumulative landscape configuration, regardless of feature type, influences occupancy at scales relevant to population processes. This aligns with findings that boreal forest specialists respond to combined disturbance effects rather than individual feature types at landscape scales (Mahon et al. 2019). Although Canada Warblers may tolerate individual narrow features at territory scales, the cumulative effects of higher linear feature density degrade habitat quality at broader extents, underscoring the importance of multi-scale analyses in heterogeneous industrial landscapes (Crosby et al. 2023).

Limitations and future directions

Several factors constrain the inference of our results and suggest directions for future research. First, Canada Warblers exhibit nuanced habitat preferences that are challenging to capture in landscape-scale analyses, including preferences for fine-scale habitat features like understory structure, shrub density, and specific moisture regimes (Reitsma et al. 2010, Ball et al. 2016). This unmeasured heterogeneity likely contributed to unexplained model variation and may partially explain the weaker-than-expected habitat amount effects. Future research incorporating LiDAR-derived understory structure metrics or field-validated habitat quality indices would further disentangle the complex drivers of Canada Warbler population dynamics. Second, the weak and uncertain extinction effects probably reflect both limited sample size for this parameter and the possibility that abandonment is driven by fine-scale factors (nest predation, food availability) not captured by landscape metrics. Additionally, spatial autocorrelation detected in only the second survey period suggests unmodeled spatial processes, potentially related to social information use, that warrant investigation through spatially explicit models or experimental approaches examining conspecific attraction directly (Cupiche-Herrera et al. 2024). Longer-term data sets with repeated annual surveys would provide more robust demographic estimates and enable distinguishing true declines from normal temporal variability (MacKenzie et al. 2005).

CONCLUSION

Reversing Canada Warbler population declines in Alberta’s industrial boreal forest requires forest managers to explicitly manage landscape configuration alongside habitat retention. Our findings demonstrate that maximizing forest cover alone is insufficient—edge density consistently emerged as the dominant predictor of occupancy dynamics, with particularly strong negative effects on colonization that prevent population recovery. The substantial colonization–extinction imbalance revealed by our dynamic occupancy modeling approach indicates that recovery strategies must prioritize landscape connectivity at scales facilitating dispersal and territory establishment. Forest planners should minimize edge density at broad spatial scales (78–314 ha) that encompass multiple territories rather than focusing on individual cutblock configurations. In practice, this means spatially aggregating disturbances to preserve large, intact forest blocks, rather than dispersing lower-intensity disturbances across the landscape. This approach is especially important in areas with high amounts of suitable Canada Warbler habitat, where conspecific attraction most likely concentrates territories, making populations more vulnerable to fragmentation effects.

Conservation efforts should prioritize identifying and protecting population centers where birds aggregate, then actively managing fragmentation in surrounding areas to facilitate colonization and population expansion. Although narrow linear features like seismic lines had weaker effects at territory scales, all disturbance types contributed equally to cumulative edge density at landscape scales, meaning comprehensive restoration must address total landscape configuration rather than selectively targeting individual feature types or assuming narrow disturbances are inconsequential. For restoration efforts with limited budgets, prioritizing large polygonal disturbances (harvest cutblocks, wellpads) over narrow linear features provides an efficient starting point, but the overarching priority remains reducing cumulative edge density around identified population centers and maintaining large tracts of intact forest. Although we recognize that breeding habitat configuration represents only one component of full life-cycle conservation for migratory species, it remains an essential and directly manageable element of recovery strategies for this threatened species through regional forest policy.

LITERATURE CITED

Alberta Biodiversity Monitoring Institute [ABMI]. 2024. Open data portal. https://abmi.ca/abmi-home/data-resources/data-portal-main/data-portal.html?rootPath=/&itemPerPage=6

Alberta Biodiversity Monitoring Institute [ABMI] and Boreal Avian Modelling [BAM] Project. 2023. Canada Warbler (Cardellina canadensis). https://beta.abmi.ca/species/cardellina-canadensis

Andrén, H. 1994. Effects of habitat fragmentation on birds and mammals in landscapes with different proportions of suitable habitat: a review. Oikos 71:355-366. https://doi.org/10.2307/3545823

Bale, S., K. F. Beazley, A. Westwood, and P. Bush. 2020. The benefits of using topographic features to predict climate-resilient habitat for migratory forest landbirds: an example for the Rusty Blackbird, Olive-sided Flycatcher, and Canada Warbler. The Condor: Ornithological Applications 122:1-19. https://doi.org/10.1093/condor/duz057

Ball, J., P. Sólymos, F. Schmiegelow, S. Haché, J. Schieck, and E. Bayne. 2016. Regional habitat needs of a nationally listed species, Canada Warbler (Cardellina canadensis), in Alberta, Canada. Avian Conservation and Ecology 11. https://doi.org/10.5751/ACE-00916-110210

Betts, M. G., L. Fahrig, A. S. Hadley, K. E. Halstead, J. Bowman, W. D. Robinson, J. A. Wiens, and D. B. Lindenmayer. 2014. A species‐centered approach for uncovering generalities in organism responses to habitat loss and fragmentation. Ecography 37:517-527. https://doi.org/10.1111/ecog.00740

Betts, M. G., A. S. Hadley, N. Rodenhouse, and J. J. Nocera. 2008. Social information trumps vegetation structure in breeding-site selection by a migrant songbird. Proceedings of the Royal Society B: Biological Sciences 275:2257-2263. https://doi.org/10.1098/rspb.2008.0217

Bourque, J., and A. Desrochers. 2006. Spatial aggregation of forest songbird territories and possible implications for area sensitivity. Avian Conservation and Ecology 1(2): 3. https://doi.org/10.5751/ACE-00043-010203

Burnham, K. P., and D. R. Anderson. 2002. Model selection and multimodel inference: a practical information-theoretic approach. Springer-Verlag, New York, New York, USA. https://doi.org/10.1007/b97636

Crosby, A. D., L. Leston, E. M. Bayne, P. Sólymos, C. L. Mahon, J. D. Toms, T. D. S. Docherty, and S. J. Song. 2023. Domains of scale in cumulative effects of energy sector development on boreal birds. Landscape Ecology 38:3173-3188. https://doi.org/10.1007/s10980-023-01779-8

Cupiche-Herrera, V. J., B. E. McLaren, and A. R. Westwood. 2024. Influence of boreal forest disturbance and conspecific attraction on the Canada Warbler (Cardellina canadensis) habitat choices during the breeding season. Avian Conservation and Ecology 19(2):26. https://doi.org/10.5751/ACE-02763-190226

Environment Canada. 2016. Recovery strategy for Canada Warbler (Cardellina canadensis) in Canada. Species at Risk Act Recovery Strategy Series. Ottawa, Ontario, Canada. https://www.sararegistry.gc.ca/virtual_sara/files/plans/rs_canada%20warbler_e_final.pdf

Environmental Systems Research Institute, Inc. (ESRI). 2024. ArcGIS Pro Version 3.4.2. ESRI, Redlands, California, USA. https://www.esri.com/en-us/arcgis/products/arcgis-pro/

Fahrig, L. 2013. Rethinking patch size and isolation effects: the habitat amount hypothesis. Journal of Biogeography 40:1649-1663. https://doi.org/10.1111/jbi.12130

Filicetti, A. T., M. Cody, and S. E. Nielsen. 2019. Caribou conservation: restoring trees on seismic lines in Alberta, Canada. Forests 10:185. https://doi.org/10.3390/f10020185

Fischer, J., and D. B. Lindenmayer. 2007. Landscape modification and habitat fragmentation: a synthesis. Global Ecology and Biogeography 16:265-280. https://doi.org/10.1111/j.1466-8238.2007.00287.x

Fisher, J. T., and A. C. Burton. 2018. Wildlife winners and losers in an oil sands landscape. Frontiers in Ecology and the Environment 16:323-328. https://doi.org/10.1002/fee.1807

Fiske, I., and R. Chandler. 2011. Unmarked: an R package for fitting hierarchical models of wildlife occurrence and abundance. Journal of Statistical Software 43(10):1-23. https://doi.org/10.18637/jss.v043.i10

Fletcher, R. J. 2006. Emergent properties of conspecific attraction in fragmented landscapes. American Naturalist 168:207-219. https://doi.org/10.1086/505764

Fletcher, R. J., R. K. Didham, C. Banks-Leite, J. Barlow, R. M. Ewers, J. Rosindell, R. D. Holt, A. Gonzalez, R. Pardini, E. I. Damschen, F. P. L. Melo, L. Ries, J. A. Prevedello, T. Tscharntke, W. F. Laurance, T. Lovejoy, and N. M. Haddad. 2018. Is habitat fragmentation good for biodiversity? Biological Conservation 226:9-15. https://doi.org/10.1016/j.biocon.2018.07.022

Flockhart, D. T. T. 2007. Migration timing of Canada Warblers near the northern edge of their breeding range. The Wilson Journal of Ornithology 119:712-716. https://doi.org/10.1676/06-111.1

Gregoire, J. M., R. W. Hedley, and E. M. Bayne. 2022. Canada Warbler response to vegetation structure on regenerating seismic lines. Avian Conservation and Ecology 17(2):26. https://doi.org/10.5751/ACE-02262-170226

Grinde, A. R., and G. J. Niemi. 2016. Influence of landscape, habitat, and species co-occurrence on occupancy dynamics of Canada Warblers. The Condor 118:513-531. https://doi.org/10.1650/CONDOR-15-168.1

Haddad, N. M., L. A. Brudvig, J. Clobert, K. F. Davies, A. Gonzalez, R. D. Holt, T. E. Lovejoy, J. O. Sexton, M. P. Austin, C. D. Collins, W. M. Cook, E. I. Damschen, R. M. Ewers, B. L. Foster, C. N. Jenkins, A. J. King, W. F. Laurance, D. J. Levey, C. R. Margules, B. A. Melbourne, A. O. Nicholls, J. L. Orrock, D.-X. Song, and J. R. Townshend. 2015. Habitat fragmentation and its lasting impact on Earth’s ecosystems. Science Advances 1:e1500052. https://doi.org/10.1126/sciadv.1500052

Hallworth, M., A. Ueland, E. Anderson, J. D. Lambert, and L. Reitsma. 2008. Habitat selection and site fidelity of Canada Warblers (Wilsonia canadensis) in central New Hampshire. The Auk 125:880-888. https://doi.org/10.1525/auk.2008.07115

Hart, T., A. Crosby, L. Leston, F. Schmiegelow, and E. Bayne. 2025. Edge effects reduce persistence but not colonization in a declining Setophaga virens (Black-throated Green Warbler) population in Alberta’s boreal forest. Ornithological Applications:duaf068. https://doi.org/10.1093/ornithapp/duaf068

Hesselbarth, M., M. Sciaini, K. With, K. Wiegand, and J. Nowosad. 2019. landscapemetrics: an open-source R tool to calculate landscape metrics. Ecography 42:1648-1657. https://doi.org/10.1111/ecog.04617

Hosmer, D. W., S. Lemeshow, and R. X. Sturdivant. 2013. Applied logistic regression. Wiley, Hoboken, New Jersey, USA. https://doi.org/10.1002/9781118548387

Hunt, A. R., E. M. Bayne, and S. Haché. 2017. Forestry and conspecifics influence Canada Warbler (Cardellina canadensis) habitat use and reproductive activity in boreal Alberta, Canada. The Condor 119:832-847. https://doi.org/10.1650/CONDOR-17-35.1

Kalukapuge, T., L. F. Leston, J. A. Martínez-Lanfranco, and E. Bayne. 2024. Response of boreal songbird communities to the width of linear features created by the energy sector in Alberta, Canada. Avian Conservation and Ecology 19. https://doi.org/10.5751/ACE-02730-190214

Kéry, M., and M. Schaub. 2012. Bayesian population analysis using WinBUGS. Academic Press, Boston, Massachusetts, USA. https://doi.org/10.1016/C2010-0-68368-4

Leston, L., E. Bayne, J. D. Toms, C. L. Mahon, A. Crosby, P. Sólymos, J. Ball, S. J. Song, F. K. A. Schmiegelow, D. Stralberg, and T. D. S. Docherty. 2023. Comparing alternative methods of modelling cumulative effects of oil and gas footprint on boreal bird abundance. Landscape Ecology 38:147-168. https://doi.org/10.1007/s10980-022-01531-8

MacKenzie, D., and L. Bailey. 2004. Assessing the fit of site occupancy models. Journal of Agricultural Biological and Environmental Statistics 9:300-318. https://doi.org/10.1198/108571104X3361

MacKenzie, D., J. Nichols, J. Hines, M. Knutson, and A. Franklin. 2003. Estimating site occupancy, colonization, and local extinction when a species is detected imperfectly. Ecology 84:2200-2207. https://doi.org/10.1890/02-3090

MacKenzie, D. I., J. D. Nichols, J. A. Royle, K. H. Pollock, L. Bailey, and J. E. Hines. 2017. Occupancy estimation and modeling: inferring patterns and dynamics of species occurrence. Academic Press, Boston, Massachusetts, USA. https://doi.org/10.1016/C2012-0-01164-7

MacKenzie, D. I., J. D. Nichols, N. Sutton, K. Kawanishi, and L. L. Bailey. 2005. Improving inferences in population studies of rare species that are detected imperfectly. Ecology 86:1101-1113. https://doi.org/10.1890/04-1060

Mahon, C. L., G. L. Holloway, E. M. Bayne, and J. D. Toms. 2019. Additive and interactive cumulative effects on boreal landbirds: winners and losers in a multi-stressor landscape. Ecological Applications 29:e01895. https://doi.org/10.1002/eap.1895

Matsuoka, S. M., E. M. Bayne, P. Sólymos, P. C. Fontaine, S. G. Cumming, F. K. A. Schmiegelow, and S. J. Song. 2012. Using binomial distance-sampling models to estimate the effective detection radius of point-count surveys across boreal Canada. The Auk 129:268-282. https://doi.org/10.1525/auk.2012.11190

Mazerolle, M. J. 2025. AICcmodavg: model selection and multimodel inference based on (Q)AIC(c). R package version 2.3.4. https://doi.org/10.32614/CRAN.package.AICcmodavg

Nocera, J. J., G. J. Forbes, and L. A. Giraldeau. 2006. Inadvertent social information in breeding site selection of natal dispersing birds. Proceedings of the Royal Society B 273(1584):349-355. https://doi.org/10.1098/rspb.2005.3318

Pebesma, E., and R. Bivand. 2023. Spatial data science: with applications in R. Chapman and Hall, Boca Raton, Florida, UK. https://doi.org/10.1201/9780429459016

Pickell, P. D., D. W. Andison, N. C. Coops, S. E. Gergel, and P. L. Marshall. 2015. The spatial patterns of anthropogenic disturbance in the western Canadian boreal forest following oil and gas development. Canadian Journal of Forest Research 45:732-743. https://doi.org/10.1139/cjfr-2014-0546

R Core Team. 2024. R: a language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. https://doi.org/10.32614/R.manuals

Reitsma, L. R., M. T. Hallworth, M. McMahon, and C. J. Conway. 2010. Canada Warbler (Cardellina canadensis), version 2.0. Birds of the World. https://doi.org/10.2173/bow.canwar.02

Sisk, T., and J. Battin. 2002. Habitat edges and avian ecology: geographic patterns and insights for western landscapes. Studies in Avian Biology 25:30-48.

Smith, A. C., M.-A. R. Hudson, V. I. Aponte, W. B. English, and C. M. Francis. 2024. North American breeding bird survey—Canadian trends website, data version 2023. Environment and Climate Change Canada, Gatineau, Quebec, Canada.

Stern, E. R., F. Riva, and S. E. Nielsen. 2018. Effects of narrow linear disturbances on light and wind patterns in fragmented boreal forests in northeastern Alberta. Forests 9:486. https://doi.org/10.3390/f9080486

Stralberg, D., S. M. Matsuoka, A. Hamann, E. M. Bayne, P. Sólymos, F. K. A. Schmiegelow, X. Wang, S. G. Cumming, and S. J. Song. 2015. Projecting boreal bird responses to climate change: the signal exceeds the noise. Ecological Applications 25:52-69. https://doi.org/10.1890/13-2289.1

Westwood, A., C. Staicer, P. Sólymos, S. Haché, T. Fontaine, E. Bayne, and D. Mazerolle. 2019. Estimating the conservation value of protected areas in Maritime Canada for two species at risk: the Olive-sided Flycatcher (Contopus cooperi) and Canada Warbler (Cardellina canadensis). Avian Conservation and Ecology 14(1):16. https://doi.org/10.5751/ACE-01359-140116

Whitaker, D. M., and I. C. Warkentin. 2010. Spatial ecology of migratory passerines on temperate and boreal forest breeding grounds. The Auk 127:471-484. https://doi.org/10.1525/auk.2010.127.3.471

Yackulic, C. B., J. D. Nichols, J. Reid, and R. Der. 2015. To predict the niche, model colonization and extinction. Ecology 96:16-23. https://doi.org/10.1890/14-1361.1