Avian Conservation and Management

INTRODUCTION

A species’ breeding distribution is determined by the extent of ecosystems that contain resources associated with their survival, reproduction, and population viability (Patton 1992). The presence of certain species of wildlife (e.g., habitat specialists) can therefore act as indicators of the resources and attributes present in an area, which reflect more broadly, ecosystem function and integrity (Lambeck 1997, Canterbury et al. 2000, Betts et al. 2024, Brunk et al. 2025). For example, the Black-backed Woodpecker (Picoides arcticus) is used as an indicator of ecosystem responses to management actions in burned forests (U.S. Forest Service Pacific Southwest Region 2007). Similarly, the Three-toed Woodpecker (Picoides tridactylus) is used as an indicator of structural diversity and bird species richness in boreal forests (Roberge et al. 2008, Pakkala et al. 2018). The relationships between indicator species and ecosystem health, species diversity, and management outcomes highlights the utility of monitoring their populations to inform forest management.

The Blue Mountains of northeastern Oregon and southeastern Washington, USA is a highly complex and heavily managed landscape that has historically lacked a robust wildlife monitoring program. The region was once covered by a mosaic of ponderosa dry forest, mesic mixed conifer forest, riparian woodland, and various shrublands and grasslands (Altman and Bresson 2017, Hanberry et al. 2020). However, late successional ponderosa pine-dominated forests have substantially declined in the region. The decline is attributed to a combination of insect pathogens, the preferential harvest of old-growth ponderosa pine trees for human goods and services, and a long-standing effort to suppress fire in this historically fire-adapted landscape (Hessburg et al. 1994, Johnston et al. 2025). Many forested areas in the region have therefore transitioned to dense stands of mid-successional fir-dominated forests (Altman and Bresson 2017, Hanberry et al. 2020). More recently, there has been a concerted effort to restore these forests to their historical structure and composition for various objectives including improving ecosystem health, reducing wildfire risk, and providing a sustainable supply of timber.

White-headed Woodpeckers (Leuconotopicus albolarvatus) are a species of conservation concern (Kozma et al. 2025) that are considered an effective indicator species for mature ponderosa pine ecosystems with open understories, varying canopy cover, and snags (Latif et al. 2015, Altman and Bresson 2017). White-headed Woodpeckers commonly nest in 25–50 cm diameter-at-breast height sized snags in areas of varying fire severity with high amounts of edge habitat (i.e., > 75 m/ha; Latif et al. 2015, Lorenz et al. 2015). White-headed Woodpeckers mainly eat seeds from large-coned pines like the ponderosa pine, but will supplement their diet with insects, including wood boring beetles (Buprestidae) found in dead wood and recently burned stands (Kozma et al. 2025, Stillman et al. 2022). Despite our ecological understanding of White-headed Woodpeckers, there is a large knowledge gap related to the distribution of this indicator species in the northern Blue Mountains because of a lack of a targeted monitoring effort in the region. Quantifying the distribution of White-headed Woodpeckers in the Blue Mountains can aid our understanding of the forest ecosystems and help with the restoration of ponderosa pine ecosystems in the region.

In 2021, we evaluated the efficacy of passive acoustic monitoring (PAM) to detect White-headed Woodpecker in mixed ponderosa pine forests within two watersheds on the Wallowa-Whitman National Forest that were known to be occupied by the species. We found PAM was more effective and efficient than traditional survey techniques (i.e., playback surveys) at detecting White-headed Woodpeckers when making inferences across multiple sampling stations in remote forest environments (Gaylord et al. 2023). In this study, our objectives were to (1) describe the distribution of White-headed Woodpeckers on the Umatilla and Wallowa-Whitman National Forests, (2) identify ecologically and management relevant habitat variables associated with forest structure and composition related to White-headed Woodpecker occurrence, and (3) examine the relationship between White-headed Woodpecker occurrence and an independently created White-headed Woodpecker Habitat Suitability Index (i.e., Latif et al. 2015) currently used in land management decision making. Habitat suitability models can have limited accuracy when applied outside the region used to develop them (Latif et al. 2020). Our analysis therefore provides essential information on if the habitat suitability model is accurately identifying habitat that currently supports breeding White-headed Woodpeckers. Last, we use previously reported White-headed Woodpecker juvenile dispersal distances (Lorenz et al. 2024) to understand the extent to which juvenile dispersal may be limiting the colonization of suitable White-headed woodpecker habitat. To accomplish our objectives, we expanded the PAM effort to gather detection/non-detection data for the White-headed Woodpeckers across the Umatilla and Wallowa-Whitman National Forests. We then used Bayesian occupancy modeling to quantify occurrence probability while correcting for imperfect detection (Royle and Kéry 2007, MacKenzie et al. 2017). As a whole, our study promotes the conservation and understanding of a priority indicator species in a relatively data deficient area.

METHODS

Study location

We conducted PAM surveys on the Umatilla and the Wallowa-Whitman National Forest in the northern Blue Mountains, USA. The northern Blue Mountains are made up of multiple different mountain ranges that include many canyons and valleys that range between 762 and 2700 meters above sea level in elevation. Low elevation habitats receive less than 25 cm of precipitation while higher elevation, mountainous areas can receive > 203 cm of precipitation (Altman and Bresson 2017). As part of the Collaborative Forest Landscape Restoration Program, the U.S. Forest Service is implementing forest treatments (i.e., commercial logging, non-commercial thinning, prescribed fire) to promote the restoration of the historical forest structure and composition in the northern Blue Mountains. Specific restoration goals include creating open forest structures with fire-resistant tree species, reducing the prevalence of true firs, and enhancing the overall heterogeneity of the forests to promote forest health.

Data collection

We used a 500-hectare hexagon tessellation sampling frame that spanned both National Forests to select the sampling locations in 2022 and 2023 (Fig. 1). In 2022, we used a stratified random sampling design to select 44 hexagons to sample by first grouping hexagons based on their proposed forest treatment types (i.e., commercial thinning, non-commercial thinning, prescribed burns, and no treatment). In 2023, we used the same sampling frame but selected hexagons using simple random sampling (i.e., we did not group by proposed treatments before selecting). Before selecting locations to sample, we omitted hexagons that contained relatively little forest-capable lands (< 25% in 2022 and < 50% in 2023; Davis et al. 2015), were deemed unsafe for fieldwork, and/or largely comprised wilderness areas or experimental forests. We established up to four stations per hexagon, resulting in 172 survey stations in 2022 and 170 survey stations in 2023. We placed stations at least 500 m from other stations, 200 m from the edge of the hexagon, and, to reduce noise interference and theft, 50 m from a path, road, or loud running water. To maximize consistency in the quality of recordings across stations, we mounted all autonomous recording units (ARUs) on trees with a diameter-at-breast height of < 30 cm at approximately 1.5 m above ground level, and we cleared branches within 30 cm of the ARU microphones. We used a combination of Wildlife Acoustics Song Meter 4 units, Song Meter 4 Mini units, and Song Meter 4 Mini Bat units equipped with acoustic microphone stub attachments to collect soundscape recordings. These ARUs all have external microphones and have very similar sensitivity and signal to noise ratios. The differences in recording units are unlikely to influence our ability to detect White-headed Woodpecker vocalizations in a meaningful way, particularly given our study objectives (Wildlife Acoustics [date unknown]). We collected soundscape recordings during the breeding season of the White-headed Woodpecker, from mid-May to early July in 2022 and early May through early September in 2023. We programmed ARUs to record continuously for two hours before and after sunrise, one hour before and three hours after sunset, and the first 10 minutes of every hour not included in the continuous recording period. For more details related to the sampling design see Duarte et al. (2024).

Audio processing and validation

Before processing audio recordings, we first narrowed the monitoring period to only include recordings collected between 1 May to 31 June. This filtering scheme focused our processing efforts on recordings collected during the White-headed Woodpecker nesting period, as the home range of the species changes outside of this time window (Lorenz et al. 2015). We then processed audio recordings with BirdNET v.2.4 (Kahl et al. 2021). We filtered White-headed Woodpecker predictions to those with confidence scores 0.5 to prioritize reviewing high confidence predictions (Pérez-Granados 2023) and reduce the effort associated with the manual review process. We grouped BirdNET predictions by station and sampling week and treated each seven-day period as a unique sampling occasion or survey period. We then randomly selected up to 50 predictions per station per survey period (i.e., 1 week of monitoring) for manual review. We randomly selected the 50 predictions to prioritize selecting temporally independent predictions. This resulted in 1889 predictions for manual review. We used the Simple Passive Acoustic Monitoring protocol, as described in Vernasco et al. (2025), to conduct manual review. This protocol facilitates efficient review by allowing reviewers to simultaneously listen to audio and inspect spectrograms of BirdNET predictions. We marked each reviewed prediction as a true- or false-positive prediction. We classified a BirdNET prediction as a true-positive prediction based on the presence of a White-headed Woodpecker specific call type that we could confidently identify (i.e., the three- or four-note “pee-kik-kik” call). We required three true-positive predictions for White-headed Woodpecker to be considered detected at a station within a survey period (i.e., week). We chose a threshold of three to minimize the likelihood of false positives (i.e., classifying an unoccupied site as occupied because of a transient or dispersing individual, for example) and managed the amount of effort associated with manual review process. Additionally, as shown in Table 1 of Gaylord et al. (2023), sites with greater than three White-headed Woodpecker detections were also found to have an average of 17 days with detections, indicating three detections reflects birds consistently using an area.

Estimating BirdNET precision and recall

We used two sources of data to estimate BirdNET’s precision and recall, metrics that are essential to understanding the reliability of an acoustic classifier. Precision is the number of true positive predictions divided by the total number of true positive and false positive BirdNET predictions. Recall is the number of true positive predictions divided by the total number of true positive and false negative predictions.

To estimate precision, we selected the first three reviewed predictions from each week of each station (n = 870 predictions). These first three predictions represent three randomly selected predictions above a confidence score 0.5. We selected up to the first three predictions reviewed because the manual validation approach used to review predictions herein (i.e., stopping upon the third positive prediction) causes the dataset as a whole to contain a greater number of false positive predictions. We calculated precision by dividing the number of true positive predictions by the total number of predictions selected for estimating precision. To estimate recall, we used a selection of White-headed Woodpecker vocalizations gleaned from audio recordings collected as part of a separate PAM study from the same region (i.e., Gaylord et al. 2023). White-headed Woodpecker vocalizations were identified by manually scanning the recordings. Audio segments containing the vocalization were extracted by selecting the time the vocalization began to the time preceding when there was two seconds of audio with no White-headed Woodpecker vocalization. This two second buffer ensures that there is a White-headed Woodpecker call within every three second segment of audio, the duration of BirdNET’s detection window. After identifying the length of the audio clip, a half second buffer was added to the beginning and end of the audio segment. The duration of each audio segment was then trimmed to the nearest multiple of three to ensure only complete 3 second windows were considered. We determined the potential number of White-headed Woodpecker predictions based on the length of the audio segment (e.g., three seconds = 1 potential prediction, 6 seconds = 2 potential predictions, etc.). All audio segments were then processed with BirdNET using the default settings, as described above. We filtered the BirdNET predictions to those with a confidence score 0.5 to match the threshold we used in the review of prediction data used for our study. We then calculated recall by dividing the total number of White-headed Woodpecker predictions by the total number of 3 second audio segments containing White-headed Woodpecker vocalizations.

Data analysis

We fit a single species, single season Bayesian occupancy model (Royle and Kéry 2007, MacKenzie et al. 2017) to the detection/non-detection data to quantify the probability of White-headed Woodpecker occupancy at a station, while accounting for the possibility that the species may sometimes go undetected even when present at a station. We considered ecologically and management relevant covariates describing forest structure and composition in the occupancy model and sampling-related covariates in the detection model.

Our habitat variables include the amount of prominent tree species, measured in basal area (m²/ha), and characteristics of forest structure previously found to be ecologically relevant to White-headed Woodpecker occurrence (Table 1). We quantified habitat variables within a 200 m radius, circular buffer centered on each survey station to capture the habitat characteristics while also minimizing overlap between spatially adjacent stations (Table 1). We gathered data for each tree species from the latest data provided by the Landscape Ecology, Modeling, Mapping & Analysis (LEMMA) research group (Bell et al. 2024). We also gathered data on forest structure from the LANDFIRE database (LANDFIRE 2016). All variables were measured on a 30-m pixel resolution and average or sum values were calculated using a moving window analysis in the terra package (Hijmans 2023). Although ecologically relevant, we did not consider the fire history of the area because only 27 of our stations have burned at any severity in the last 10 years. Examining the effects of fire requires consideration of the time since fire because of the successional changes that occur and even fewer of our sites have burned within ecologically relevant time intervals (i.e., 1 year since n = 8, 2–5 years since n = 12, and 6–10 years since fire n = 7).

We considered detection covariates that are likely to influence our ability to detect White-headed Woodpeckers. We included the hours of audio recordings collected each survey period to control for differences in sampling amounts. We also considered the Julian day of the first day of the sampling week to capture potential seasonal changes in White-headed Woodpecker vocalizations. Because forest structure can influence how sound travels, we included the average canopy cover for each station calculated using a 200m radius, circular buffer centered on each survey station. Last, forest structure and species abundance can change with elevation and we therefore also included an effect of elevation on our detection covariate.

We conducted model selection using indicator variables (Kuo and Mallick 1998). Specifically, we multiplied each model coefficient by a latent binary variable, where a covariate is included in the model when the indicator variable equals one. Each indicator variable was assigned a Bernoulli prior, assigning equal prior probability (0.5) of including or excluding each covariate. In this process, each unique sequence of indicator variables (i.e., combinations of ones and zeroes across detection and occupancy covariates) represents a candidate model (Royle et al. 2014). We used the model structure with the highest posterior probability for inference (Table 2). To help with model fitting and to ensure covariates did not artificially drop out, we implemented this model selection approach using slab-and-spike priors (Mitchell and Beauchamp 1988). To do so, we first fit the global model without indicator variables using uninformative priors. We then used the resulting coefficient estimates and their standard deviations for the values of the spike priors. This process ensures the informed prior distributions are near the posterior distribution for each coefficient, as recommended by Dellaportas et al. (2002).

We fit models using JAGS (Plummer 2003), in program R v4.2.0 (R Core Team 2022) using the jagsUI package (Kellner 2024). Before models were fit, we rescaled all continuous variables to have a mean of zero and a standard deviation of one. We inspected correlations between variables to ensure that all correlation coefficients were > -0.7 and < 0.7. We included a random effect of hexagon ID on occupancy probability to account for any potential spatial autocorrelation associated with the spatially clustered stations within our hexagonal sampling design. We used uninformative priors with a mean of zero and precision of 0.368 for all model intercepts and coefficients. We also used an uninformative prior for the variance of the random effect by specifying a uniform distribution that ranged from 0 to 20. We fit models using three independent chains, each consisting of 20,000 iterations with a burn-in and adaption phase of 10,000 each. All parameters converged (< 1.01; Brooks and Gelman 1998). We used the top-supported model identified using the model selection process described above (i.e., indicator variables) to estimate occupancy and detection probability. We describe model parameters using their mean values and 95% credible intervals (CIs) and report results using the odds ratios (i.e., the odds of detection or occupancy for every 1 standard deviation change in the covariate; Hosmer et al. 2013) of model coefficients found to be important to occurrence. Last, we assessed the fit of the global model using a posterior predictive check by calculating the Bayesian p-value (Gelman et al. 1996). Specifically, we simulated data using the estimated model, calculated the sum of the absolute value of residuals of our simulated data and our empirical data, and estimated the Bayesian p-value as the proportion of times the simulated data were more extreme (i.e., had a larger sum of the absolute value of residuals) than our empirical data (Kéry 2010).

Evaluating the habitat suitability index

To evaluate the habitat suitability index produced by Latif et al. (2015), we compared the habitat suitability index at sites where White-headed Woodpeckers were and were not detected. The habitat suitability index was created using habitat data at White-headed Woodpecker nest locations collected within the eastside of the Cascade Mountains of Oregon. The specific habitat variables included in the index are topographical characteristics (i.e., slope, aspect), local- and landscape-scale canopy cover, proportional area of ponderosa pine forest, and the density of habitat edges (for additional details see Latif et al. 2015). The habitat suitability metric represents an estimated relative probability of habitat use quantified using Maxent (Phillips et al. 2006). We compared the relationship between White-headed Woodpecker occurrence and the habitat suitability at two spatial scales using linear regression. The first spatial scale we evaluated was at the station level. We calculated the average habitat suitability index for each station using a 200m radius, circular buffer centered on the station location. We included a random effect of hexagon ID to account for spatially clustered stations and fit the model using the lme4 package (Bates et al. 2015). The second spatial scale was at the hexagon level. We calculated the average habitat suitability index for each 500 ha hexagon and fit the model in Program R (R Core Team 2022).

RESULTS

During the White-headed Woodpecker nesting period, we monitored 294 stations for an average of 244.8 hours (SD = 105.6 hours) and 3.46 weeks (SD = 1.24 weeks), with 56% of ARUs recording for 4 weeks (Fig. 1, Fig. A.1). Our manual review of audio recordings from Gaylord et al. (2023) generated 1039 three second audio segments containing White-headed Woodpecker vocalizations. Of those, 856 were three seconds long, 92 were six seconds long, 48 were nine seconds long, 18 were 12 seconds long, and 25 were 15 seconds long. There were therefore 1381 3-second audio segments containing White-headed Woodpecker vocalizations. We estimated recall using a BirdNET confidence score threshold of 0.5 to be 0.469. Using the first three detections per survey period for each station (n = 870 predictions), we estimated precision to be 0.35.

White-headed Woodpeckers were detected at 22 stations, resulting in a naïve occupancy of 7.5% of stations occupied. The total number of sites estimated as occupied after correcting for imperfect detection was 23 (95% credible intervals: 22, 27). The proportion of variance in the occupancy probability attributed to the random effect of hexagon ID was 0.55 [95% CIs: 0.29, 0.78]. The estimated detection probability was, on average, 0.71 (0.60, 0.80) per week of monitoring. The cumulative detection probability reached 0.99 after 4 weeks of monitoring, meaning we were able to detect White-headed Woodpecker, when present, without error using our PAM protocol after 4 weeks of monitoring. Of the habitat covariates considered in the global model (Table 1), the top-supported occupancy model was found to include lodgepole pine basal area and the top-supported detection model was the intercept only model (Table 2, Table 3). The odds of occupancy were 5.26 times lower for every 2.87 m²/ha increase in lodgepole pine basal area (Fig. 2). We estimated the Bayesian p-value of the global model to be 0.24, indicating adequate model fit. We provide a summary of the global model in Table A.2.

We found stations wherein White-headed Woodpecker were detected to have significantly higher habitat suitability indices (Fig. 3A, Table 4A). The adjusted intra-class correlation coefficient of the random effect of hexagon ID was 0.75 [95% confidence intervals: 0.61, 0.80), indicating most of the variation in habitat suitability values were among hexagons. Hexagons with White-headed Woodpecker also had higher suitability indices (Fig. 3B, Table 4B).

DISCUSSION

We provide the first landscape-scale assessment of White-headed Woodpecker occupancy in the Umatilla and Wallowa-Whitman National Forests of the northern Blue Mountains, USA. Our study found that White-headed Woodpecker occupancy probability in the northern Blue Mountains region was low, while the probability of detecting the White-headed Woodpecker with PAM, given it is present, was relatively high. We also found that White-headed Woodpecker occupancy declined with lodgepole pine basal area. Further, stations and hexagons wherein White-headed Woodpeckers were detected had significantly higher habitat suitability indices. Our results suggest areas with relatively high habitat suitability (e.g., > 0.3 HSI at the station scale and > 0.2 HSI at the hexagon scale), but unknown White-headed Woodpecker occupancy likely offers more suitable breeding locations compared to areas below these HSI thresholds. More broadly, our results provide further evidence that the distribution of late-successional ponderosa pine forests capable of supporting wildlife dependent on such forests is limited in this region. Because wildlife is considered a useful indicator of ecosystem function and integrity (Chase et al. 2020, Matricardi et al. 2020, Betts et al. 2024), this study provides important insight into the health of the dry forests in the northern Blue Mountains.

Our results indicate a low occupancy probability for White-headed Woodpeckers in the northern Blue Mountains. Notably, the random effect of hexagon ID explained a large proportion of the variance in occupancy (i.e., 0.55), highlighting there are extensive differences among hexagons in occupancy (i.e., some hexagons have higher occupancy probability than others). In other words, overall occupancy is low across the landscape, but in a few spatially adjacent locations, occupancy probability is comparatively high. White-headed Woodpeckers therefore exhibit a sparse, but spatially clumped distribution in the northern Blue Mountains. The negative relationship between occupancy and lodgepole pine we observed could be related to contrasts between the preferred habitats of White-headed Woodpeckers and forests containing lodgepole pine. Specifically, White-headed Woodpeckers rely on dry forests composed of mature ponderosa pine and an open understory that is maintained by low severity fires (Kozma et al. 2025). Lodgepole pine, on the other hand, is more often associated with moist mixed conifer forests that tend to have more closed understories and burn less frequently with mixed fire severity. In the northern Blue Mountains, dry ponderosa-dominated forests occur at relatively lower elevations and the moist mixed conifer forests tend to be found at relatively higher elevations (Stine et al. 2014). Given the differences between the types of forest ponderosa pine and lodgepole pine are most commonly associated with, further study will be needed to understand the extent to which lodgepole pine directly affects White-headed Woodpecker occupancy or rather indicates broader habitat characteristics associated with unsuitable White-headed Woodpecker habitat. Importantly, the large proportion of the variance in occupancy explained by the random effect represents variation in occupancy not attributed to lodgepole pine basal area. Additional, unidentified site-level differences are therefore contributing to the observed patterns of White-headed Woodpecker occurrence.

PAM was previously shown to provide a similar cumulative detection probability estimate following 7 days of monitoring and after validating all detections produced by a custom classifier trained on recordings from the same study region (Gaylord et al. 2023). We used a classifier trained on publicly available White-headed Woodpecker recordings (i.e., BirdNET v2.4; Kahl et al. 2021) and validated only 50 detections per survey period, yet still obtained a cumulative detection probability > 0.9 after 2 weeks of monitoring. Both our results and the previous results therefore suggest PAM is a highly effective means for monitoring White-headed Woodpeckers. Based on these detection probability estimates, it seems PAM survey data collected and processed using our protocol, the same ARUs models, and BirdNET v2.4 can be used to directly inform forest management decisions without using statistical models that account for imperfect detection when > 4 weeks of audio recordings are collected at a site.

Mature ponderosa pine forests of the northern Blue Mountains are a priority restoration ecosystem because of their ability to support multiple species of conservation concern and for the regional decline of the ecosystem (Altman and Bresson 2017). The relatively low occupancy estimate identified in our current study may indicate that the amount or health of mature ponderosa pine forest in the northern Blue Mountains is insufficient or incapable of sustaining a broad distribution of White-headed Woodpeckers. This could be especially indicative of a lack of specific forest resources, such as snags, in areas that contain otherwise suitable habitat. Indeed, suitable habitat is seemingly unoccupied because there are areas without detections that have habitat suitability indices similar to stations with detections. It is also possible the Habitat Suitability Index does not completely identify suitable habitat because it was created using topographic variables, measurements of canopy cover, ponderosa pine prevalence, and forest edge density (Latif et al. 2015). White-headed Woodpeckers rely on snags and open understories for nesting and also select areas with variable canopy cover (Lorenz et al. 2015, 2024). Importantly, these forest characteristics are not represented by the Habitat Suitability Index developed by Latif et al. (2015). As additional occurrence data accumulates, updating the habitat suitability model can further improve the utility of the White-headed Woodpecker Habitat Suitability Index in our region (Duarte et al. 2025).

The absence of White-headed Woodpecker does not necessarily indicate that it is habitat that is limiting its distribution. Other factors, such as juvenile dispersal, could limit the distribution of these birds across the National Forests. Lorenz et al. (2024) quantified biased-corrected estimates of White-headed Woodpecker juvenile dispersal and found 90% of woodpeckers disperse at least 7 km while 5% dispersed 70 km. Figure 4 shows the minimum dispersal distances of 90%, 50%, and 5% of juvenile woodpeckers, based on juvenile dispersal distances reported by Lorenz et al. (2024), centered on the sites classified as occupied. This figure suggests that White-headed Woodpeckers are not dispersal limited if individuals are capable of dispersing the maximum distance previously observed. Within our study area, however, the dispersal buffers contained many non-forested areas and the extent to which such habitats limit dispersal is not known. If non-forested habitats are impervious to juvenile dispersal and many individuals do not disperse further than the minimum distance 50% of juveniles were observed to disperse (i.e., 24 km), then the ability of dispersing juveniles to colonize existing suitable habitat may be limited. Further resolving the habitat differences between occupied and unoccupied stations, particularly those containing suitable habitat, and understanding patterns of juvenile dispersal and survival represent promising future directions for understanding White-headed Woodpecker distributions in the northern Blue Mountains.

Here, we reveal that the White-headed Woodpecker distribution across the National Forests of the northern Blue Mountains is limited and spatially clumped. We found that sites where White-headed Woodpeckers were detected had significantly higher habitat suitability. With increased knowledge on the species’ distribution and the validation of a management tool currently used to identify suitable habitat, land managers can now more confidently identify areas that currently support this species of conservation concern during the breeding season and restore and monitor unoccupied areas that contain suitable habitat, as indicated by the habitat suitability model. The negative relationship between lodgepole pine and White-headed Woodpecker occupancy suggests restorations efforts may benefit from considering lodgepole pine management in suitable habitat. Forests, and the dry forests of the Pacific Northwest in particular, are facing a myriad of threats including increased frequency of drought, non-native insects and pathogens, and shifting wildfire regimes (Millar and Stephenson 2015, Johnston et al. 2025). Using PAM to identify and monitor the distribution of indicator species across managed landscapes, particularly those that are otherwise data deficient, represents a promising approach for understanding and quantifying changes in forest health and prioritizing areas for habitat restoration.

LITERATURE CITED

Altman, B., and B. Bresson. 2017. Conservation of landbirds and associated habitats and ecosystems in the northern Rocky Mountains of Oregon and Washington. Oregon-Washington Partners in Flight (www.orwapif.org) and American Bird Conservancy and U.S. Forest Service/Bureau of Land Management, Washington, D.C., USA.

Bates, D., M. Mächler, B. Bolker, and S. Walker. 2015. Fitting linear mixed-effects models using lme4. Journal of Statistical Software 67(1):1-48. https://doi.org/10.18637/jss.v067.i01

Bell, D. M., M. J. Gregory, and Z. Yang. 2024. Hindcasting and updating Landsat-based forest structure mapping across years to support forest management and planning. Forest Ecology and Management 572:122239. https://doi.org/10.1016/j.foreco.2024.122239

Betts, M. G., Z. Yang, A. S. Hadley, J. Hightower, F. Hua, D. Lindenmayer, E. Seo, and S. P. Healey. 2024. Quantifying forest degradation requires a long-term, landscape-scale approach. Nature Ecology & Evolution 8(6):1054-1057. https://doi.org/10.1038/s41559-024-02409-5

Brooks, S. P., and A. Gelman. 1998. General methods for monitoring convergence of iterative simulations. Journal of Computational and Graphical Statistics 7(4):434-455. https://doi.org/10.1080/10618600.1998.10474787

Brunk, K. M., H. A. Kramer, M. Z. Peery, S. Kahl, and C. M. Wood. 2025. Assessing spatial variability and efficacy of surrogate species at an ecosystem scale. Conservation Biology 39:e70058. https://doi.org/10.1111/cobi.70058

Canterbury, G. E., T. E. Martin, D. R. Petit, L. J. Petit, and D. F. Bradford. 2000. Bird communities and habitat as ecological indicators of forest condition in regional monitoring. Conservation Biology 14(2):544-558. https://doi.org/10.1046/j.1523-1739.2000.98235.x

Chase, J. M., S. A. Blowes, T. M. Knight, K. Gerstner, and F. May. 2020. Ecosystem decay exacerbates biodiversity loss with habitat loss. Nature 584(7820):238-243. https://doi.org/10.1038/s41586-020-2531-2

Davis, R. J., J. L. Ohmann, R. E. Kennedy, W. B. Cohen, M. J. Gregory, Z. Yang, H. M. Roberts, A. N. Gray, and T. A. Spies. 2015. Northwest Forest Plan–the first 20 years (1994–2013): status and trends of late-successional and old-growth forests. U.S. Forest Service, Pacific Northwest Research Station, Portland, Oregon, USA.

Dellaportas, P., J. J. Forster, and I. Ntzoufras. 2002. On Bayesian model and variable selection using MCMC. Statistics and Computing 12(1):27-36. https://doi.org/10.1023/A:1013164120801

Duarte, A., R. S. Spaan, J. T. Peterson, C. A. Pearl, and M. J. Adams. 2025. Bayesian networks facilitate updating of species distribution and habitat suitability models. Ecological Modelling 501:110982. https://doi.org/10.1016/j.ecolmodel.2024.110982

Duarte, A., B. J. Vernasco, M. J. Weldy, R. S. Spaan, and J. Ratliff. 2024. Northern Blue Mountains wildlife monitoring, 2022–2023. U.S. Forest Service, Pacific Northwest Research Station, Olympia, Washington, USA.

Gaylord, M. A., A. Duarte, B. C. McComb, and J. Ratliff. 2023. Passive acoustic recorders increase White-headed Woodpecker detectability in the Blue Mountains. Journal of Field Ornithology 94(4):1. https://doi.org/10.5751/JFO-00330-940401

Gelman, A., X. L. Meng, and H. Stern. 1996. Posterior predictive assessment of model fitness via realized discrepancies. Statistica Sinica 6(4):733-807.

Hanberry, B. B., D. C. Justice, and D. C. Powell. 2020. Discovering Douglas-fir woodlands in the historical forests of Umatilla National Forest, eastern Oregon and Washington. Forests 11(10):1122. https://doi.org/10.3390/f11101122

Hessburg, P. F., R. G. Mitchell, and G. M. Filip. 1994. Historical and current roles of insects and pathogens in eastern Oregon and Washington forested landscapes. General Technical Report PNW-GTR-327. U.S. Forest Service, Pacific Northwest Research Station, Portland, Oregon, USA. https://doi.org/10.2737/PNW-GTR-327

Hijmans, R. J. 2023. terra: Spatial data analysis. https://doi.org/10.32614/CRAN.package.terra

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

Johnston, J. D., P. N. Stephens, M. R. Schmidt, S. M. A. Jeronimo, A. E. Brackett, A. E. Wuenschel, D. E. M. Ulrich, R. H. Paine, L. F. Duloisy, S. M. Greenler, A. J. Larson, and D. J. Churchill. 2025. Significant mortality of old trees across a dry forest landscape, Oregon, USA. Forest Ecology and Management 593(2):122907. https://doi.org/10.1016/j.foreco.2025.122907

Kahl, S., C. M. Wood, M. Eibl, and H. Klinck. 2021. BirdNET: a deep learning solution for avian diversity monitoring. Ecological Informatics 61:101236. https://doi.org/10.1016/j.ecoinf.2021.101236

Kellner, K. 2024. jagsUI: A wrapper around “rjags” to streamline “JAGS” analyses. https://doi.org/10.32614/CRAN.package.jagsUI

Kéry, M. 2010. Introduction to WinBUGS for ecologists: a Bayesian approach to regression, ANOVA, mixed models, and related analyses. Academic, Amsterdam, The Netherlands. https://doi.org/10.1016/C2009-0-30639-X

Kozma, J. M., T. J. Lorenz, M. G. Raphael, K. L. Garrett, R. D. Dixon, G. M. Kirwan, and S. M. Billerman. 2025. White-headed Woodpecker (Leuconotopicus albolarvatus), version 2.2. In P. G. Rodewald, B. K. Keeney, and M. G. Smith, editors. Birds of the world. Cornell Lab of Ornithology, Ithaca, New York, USA. https://doi.org/10.2173/bow.whhwoo.02.2

Kuo, L., and B. Mallick. 1998. Variable selection for regression models. Sankhyâ: The Indian Journal of Statistics, Series B 60(1):65-81.

Lambeck, R. J. 1997. Focal species: a multi-species umbrella for nature conservation. Conservation Biology 11(4):849-856. https://doi.org/10.1046/j.1523-1739.1997.96319.x

LANDFIRE. 2016. Fuel Disturbance. LANDFIRE 2. U.S. Department of the Interior, Geological Survey, and U.S. Department of Agriculture, Washington, D.C., USA.

Latif, Q. S., V. A. Saab, J. G. Dudley, A. Markus, and K. Mellen-McLean. 2020. Development and evaluation of habitat suitability models for nesting White-headed Woodpecker (Dryobates albolarvatus) in burned forest. PLoS ONE 15(5):e0233043. https://doi.org/10.1371/journal.pone.0233043

Latif, Q. S., V. A. Saab, K. Mellen-Mclean, and J. G. Dudley. 2015. Evaluating habitat suitability models for nesting White-headed Woodpeckers in unburned forest. Journal of Wildlife Management 79(2):263-273. https://doi.org/10.1002/jwmg.842

Lorenz, T. J., K. T. Vierling, J. M. Kozma, J. E. Millard, and M. G. Raphael. 2015. Space use by White-headed Woodpeckers and selection for recent forest disturbances. Journal of Wildlife Management 79(8):1286-1297. https://doi.org/10.1002/jwmg.957

Lorenz, T. J., A. N. Stillman, J. M. Kozma, and P. C. Fischer. 2024. Bias-corrected natal dispersal estimates fill information gaps for White-headed Woodpecker conservation. Avian Conservation and Ecology 19(2):4. https://doi.org/10.5751/ACE-02684-190204

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. Elsevier, Amsterdam, The Netherlands.

Matricardi, E. A. T., D. L. Skole, O. B. Costa, M. A. Pedlowski, J. H. Samek, and E. P. Miguel. 2020. Long-term forest degradation surpasses deforestation in the Brazilian Amazon. Science 369(6509):1378-1382. https://doi.org/10.1126/science.abb3021

Millar, C. I., and N. L. Stephenson. 2015. Temperate forest health in an era of emerging megadisturbance. Science 349(6250):823-826. https://doi.org/10.1126/science.aaa9933

Mitchell, T. J., and J. J. Beauchamp. 1988. Bayesian variable selection in linear regression. Journal of the American Statistical Association 83(404):1023-1032. https://doi.org/10.1080/01621459.1988.10478694

Pakkala, T., J. Tiainen, M. Piha, and J. Kouki. 2018. Three-toed Woodpecker cavities in trees: a keystone structural feature in forests shows decadal persistence but only short-term benefit for secondary cavity-breeders. Forest Ecology and Management 413:70-75. https://doi.org/10.1016/j.foreco.2018.01.043

Patton, D. R. 1992. Wildlife habitat relationships in forested ecosystems. Timber Press, Portland, Oregon, USA.

Pérez-Granados, C. 2023. BirdNET: applications, performance, pitfalls and future opportunities. Ibis 165(3):1068-1075. https://doi.org/10.1111/ibi.13193

Phillips, S. J., R. P. Anderson, and R. E. Schapire. 2006. Maximum entropy modeling of species geographic distributions. Ecological Modelling 190(3-4):231-259. https://doi.org/10.1016/j.ecolmodel.2005.03.026

Plummer, M. 2003. JAGS: a program for analysis of Bayesian graphical models using Gibbs sampling. Pages 1-10 in K. Hornik, F. Leisch, and A. Zeileis, editors. Proceedings of the 3rd International Workshop on Distributed Statistical Computing. March 20-22, Technische Universität Wien, Vienna, Austria.

R Core Team. 2022. R: A language and environment for statistical computing. R Core Team, Vienna, Austria. https://doi.org/10.32614/R.manuals

Roberge, J.-M., P. Angelstam, and M.-A. Villard. 2008. Specialised woodpeckers and naturalness in hemiboreal forests - deriving quantitative targets for conservation planning. Biological Conservation 141(4):997-1012. https://doi.org/10.1016/j.biocon.2008.01.010

Royle, J. A., R. B. Chandler, R. Sollmann, and B. Gardner. 2014. Spatial capture-recapture. Academic, Amsterdam, The Netherlands. https://doi.org/10.1016/C2012-0-01222-7

Royle, J. A., and M. Kéry. 2007. A Bayesian state-space formulation of dynamic occupancy models. Ecology 88(7):1813-1823. https://doi.org/10.1890/06-0669.1

Stillman, A. N., M. V Caiafa, T. J. Lorenz, M. A. Jusino, and M. W. Tingley. 2022. DNA metabarcoding reveals broad woodpecker diets in fire-maintained forests. Ornithology 139(2):ukac009. https://doi.org/10.1093/ornithology/ukac009

Stine, P., P. Hessburg, T. Spies, M. Kramer, C. J. Fettig, A. Hansen, J. Lehmkuhl, K. O’Hara, K. Polivka, P. Singleton, S. Charnley, A. Merschel, and R. White. 2014. The ecology and management of moist mixed-conifer forests in eastern Oregon and Washington: a synthesis of the relevant biophysical science and implications for future land management. General Technical Report PNW-GTR-897. U.S. Forest Service, Pacific Northwest Research Station, Portland, Oregon, USA. https://doi.org/10.2737/PNW-GTR-897

U.S. Forest Service Pacific Southwest Region. 2007. Sierra Nevada Forests Management Indicator Species Amendment: Record of Decision. U.S. Forest Service Pacific Southwest Region, Vallejo, California, USA.

Vernasco, B. J., A. Duarte, M. J. Weldy, S. R. Finch, and J. Ratliff. 2025. Workflow and training materials for validating species detections from passive acoustic monitoring: a case study from the northern Blue Mountains. General Technical Report PNW-GTR-1039. U.S. Forest Service, Pacific Northwest Research Station, Portland, Oregon, USA. https://doi.org/10.2737/pnw-gtr-1039

Wildlife Acoustics [date unknown]. Song meter mini quick start. Wildlife Acoustics, Maynard, Massachusetts, USA. https://www.wildlifeacoustics.com/uploads/user-guides/Song-Meter-Mini-EN-01032024.pdf