Avian relationships with wild ﬁ re at two dry forest locations with di ﬀ erent historical ﬁ re regimes

. Wild ﬁ re is a key factor in ﬂ uencing bird community composition in western North American forests. We need to understand species and community responses to wild ﬁ re and how responses vary regionally to e ﬀ ectively manage dry conifer forests for maintaining biodiversity. We compared avian relationships with wild ﬁ re burn severity between two dry forest locations of Arizona and Idaho. We predicted di ﬀ erent responses to wild ﬁ re between locations due to regional di ﬀ erences in historical ﬁ re regime. We conducted point count surveys for 3 yr following wild ﬁ re (Arizona: 1997–1999; Idaho: 2008–2010) and used multispecies hierarchical models to analyze relationships of bird occupancy with burn severity. Consistent with our prediction for mixed- severity ﬁ re regimes characterizing the Idaho location, we observed proportionately more positive species occupancy relationships and, consequently, a positive species richness relationship with burn severity in Idaho. We also observed the opposite pa (cid:308) ern in Arizona, which was congruent with our prediction for the low- severity ﬁ re regime characterizing that location. Cavity nesters and aerial insectivores occupied more severely burned sites following wild ﬁ re, corresponding with predicted increases in nesting substrate and foraging opportunities for these species. In contrast, canopy- nesting foliage gleaners and pine seed consumers exhibited negative relationships with burn severity. Our results were consistent with predictions based on species life histories and with pa (cid:308) erns from the literature, suggesting generality of observed relationships and locational di ﬀ erence in relationships with wild ﬁ re. We therefore suggest that optimal management strategies for maintaining avian diversity could di ﬀ er regionally. Speci ﬁ cally, intensive fuels management may be ecologically less appropriate for promoting biodiversity in areas such as the Idaho location where mixed- severity wild ﬁ res and dense forest stands were historically more common.


INTRODUCTION
and Kennedy 2012 ). Species and community responses to a wildfi re depend on its burn severity (fi re eff ects on vegetation; Agee 1993, Smith 2000 and time passed since fi re (fi re history; Nimmo et al. 2014 ). Stands that vary in fi re history will vary in the species they support (Fontaine et al. 2009 ). Consequently, landscapes containing a diversity of stands varying in fi re history are expected to support the greatest array of species (Clarke 2008 , Fontaine andKennedy 2012 ). Forest managers aiming to conserve and promote biodiversity must recognize the role of wildfi re in maintaining overall diversity by supporting fi re-adapted species and communities within forested landscapes. A nuanced understanding of species and community responses to wildfi re would allow ecologists to be er inform forest management decisions.
Such information would particularly inform ongoing research and debate regarding the ecological role of wildfi re and related management strategies in lower elevation dry conifer forests. Such forests have generally been characterized historically by frequent low-or mixed-severity fi res (Hessburg et al. 2005 ). In the Inland Northwest, these fi res maintained low and variable tree densities, light and patchy ground fuels, simplifi ed forest structure, and favored fi retolerant trees, such as ponderosa pine ( Pinus ponderosa ), and a low and patchy cover of associated fi re-tolerant shrubs (Hessburg et al. 2005(Hessburg et al. , 2007. Historically, in the Southwest, a drier climate has favored forest stands that were relatively homogenous, were lower in both tree and understory density, were heavily dominated by ponderosa pine, and experienced more frequent and lower severity fi res (Moir et al. 1997 , Bock andBlock 2005 ). In the central west, moister conditions have encouraged denser stands, more landscape-scale heterogeneity in stand structure, more mixed tree species composition, and less frequent and more mixed-severity wildfi re (Schoennagel et al. 2004. The range and scale of spatial and temporal variability characterizing dry conifer forests is currently under investigation and debate (Pierce et al. 2004, Freche e and Meyer 2009, Williams and Baker 2012, Odion et al. 2014 ). Nevertheless, researchers generally agree that lower density stands associated with higher frequency; lower severity fi res were more widespread in the Southwest (Agee 1993, Schoennagel et al. 2004, Nimmo et al. 2014.
Within the last ~100 yr, anthropogenic fi re suppression and concurrent logging, development, livestock grazing, and climate change are thought to have reduced fi re frequency and increased severity in lower elevation dry conifer forests (Dale et al. 2001, Brown et al. 2004, Schoennagel et al. 2004 ). Many forest managers implement fuels reduction treatments of low-severity prescribed fi re and timber harvest along with continued suppression of wildfi res as part of a large-scale eff ort aimed at restoring historical fi re regimes (Fulé et al. 2012 ). Some expect this approach to be the best possible strategy for maintaining natural processes in managed ponderosa pine-dominated forests (Miller andThode 2007 , Crimmins et al. 2013 ). This strategy may be more appropriate in the southwestern United States, where biological communities have experienced more frequent lower severity wildfi res (Schoennagel et al. 2004, Bock and Block 2005, Illán et al. 2014. In contrast, central Rocky Mountain forests may benefi t from use of more high-severity wildfi re to restore historical conditions and maximize biodiversity (Schoennagel et al. 2004. Central to this debate is our understanding of how organisms respond to wildfi re. If historical fi re regimes shape wildfi re responses, biological relationships with wildfi re should vary regionally by fi re regime, in which case forest management strategies aimed at promoting biodiversity should refl ect these diff erences. Birds off er useful opportunities for studying biological wildfi re relationships. Avian communities can be surveyed without specialized equipment (Sutherland et al. 2004 ) and are therefore used to evaluate management strategies aimed at biological conservation . Bird species with diff erent ecological and life history traits are aff ected diff erentially by wildfi re , Smucker et al. 2005, Kotliar et al. 2007, Fontaine and Kennedy 2012, Seavy and Alexander 2014. For example, wildfi re can create openings to enhance foraging and nesting opportunities for shrub-nesting and ground-foraging species, and create snags that provide important nesting and foraging resources for cavity-nesting species (Hu o 1995, Kotliar et al. 2002 ). In contrast, tree mortality a er wildfi re reduces resources for canopywww.esajournals.org LATIF ET AL.
nesting and foliage-gleaning species (Kotliar et al. 2007, Fontaine et al. 2009 ). Diff erences among species in relationships with wildfi re and consequent changes in community composition have been documented repeatedly (see reviews by Kalies et al. 2010 , Fontaine andKennedy 2012 ).
Despite numerous studies of avian wildfi re relationships, regional variation in these relationships is poorly understood. Individual studies typically examine population or community relationships at one location at a time (e.g., Smucker et al. 2005, Kotliar et al. 2007 ). Comparisons across studies reveal some consistencies and also diff erences between studies in diff erent locations. Potential explanations for diff erences among studies include diff erences in burn severity, severity measurements, population parameter measurements, study length, spatial scale, and fi re regime (Bock and Block 2005, Kalies et al. 2010, Fontaine and Kennedy 2012. Recent reviews emphasize burn severity and methodological factors for explaining diff erences among studies (Kalies et al. 2010 , Fontaine andKennedy 2012 ). Regional diff erences in climate, vegetation structure, and historical fi re regime, however, could also be important. Studies comparing pa erns among regions that control for methodological factors are needed to understand the role of such variation in shaping avian responses to wildfi re.
We compared avian relationships with wildfi re between two dry coniferous forest locations of the western United States; one in a southwestern forest and the other in the central Rocky Mountains. We surveyed forests following wildfi re at sites distributed along a burn severity gradient and compared avian occupancy relationships with burn severity between the two study locations. We used similar survey protocols at both locations and controlled for remaining diff erences in methodology by estimating and accounting for studyspecifi c detectability. Given diff erences in historical fi re regime and related vegetation structure, we predicted proportionately more bird species at the Idaho location would relate positively with burn severity, and consequently species richness to relate positively. Conversely, we expected the opposite at the Coconino location. We examined how species and community relationships with burn severity were similar and how they diff ered between the two locations to evaluate this predic-tion. Additionally, to understand the mechanistic basis for and generality of community-wide patterns, we considered the consistency of observed relationships with species life history traits and pa erns reported in the literature. Finally, we discuss the implications of our results for management of dry conifer forests.

Study system
We studied avian occupancy in relation to wildfi re at the Paye e National Forest (NF) in the central Rocky Mountains of western North America and the Coconino NF in southwestern North America. In both forests, fi re suppression began ~100 yr prior to this study, and both were managed for multiple uses, including timber harvest, mining, recreation, livestock grazing, wildlife habitat, and watershed management (Hollenbeck et al. 2013 ).

Data collection
We surveyed bird species using point counts at the Paye e and Coconino NFs. We visited point count stations 2-3 times each between 22 May and 3 July over multiple years (twice annually 2008-2010 at Paye e NF; thrice annually 1997-1999 at Coconino NF). We began point counts just a er the dawn chorus and completed them within 5 h. Observers recorded all birds detected during a 5-min count at the Paye e and an 8-min count at the Coconino and estimated distances to each detected individual. Only detections within 100 m of the point were included in this analysis. Our sampling design was a robust design (Pollock 1982 ) with years as the primary periods and visits within years as the secondary periods. At both locations, point count stations were in areas varying in wildfire burn severity and surveyed for 3 yr following wildfire (20081997-1999. We used the ΔNBR index generated from comparison of Landsat TM imagery recorded before (June 14, 1994 at Coconino;August 11, 2005 at Payette) vs. after (June 22, 1997 at Coconino;August 24, 2007 at Payette) wildfire events (MTBS 2012 ) to quantify burn severity at the point count stations. Raw ΔNBR values were compiled at a 30 × 30 m resolution. We used mean ΔNBR values for 100-m radius neighborhoods centered on point count stations as a covariate of occupancy probabilities during data analysis. In similar habitats, Miller and Thode ( 2007 ) found ΔNBR values measured at a resolution similar to those in this study to be closely related with field collected burn severity measurements.
At Paye e, two of fi ve study units (31 points) were treated with prescribed fi re in 2004 and 2006. Eff ects of prescribed fi re on vegetation are described elsewhere (Saab et al. 2006 ). The correlation between prescribed fi re treatments (1 = treated, 0 = untreated) and wildfi re burn severity was low (Pearson ' s r = −0.186, n = 101 points), so we had minimal concerns over prescribed fi re confounding relationships with burn severity.

Data analysis
We analyzed avian relationships with wildfi re using multispecies occupancy models (Gelman and Hill 2007, Royle and Dorazio 2008 ). Occupancy models leverage repeatsurvey data to estimate species detectability ( p ) conditional upon occupancy (species occurrence), allowing unbiased estimation of occupancy probabilities (Ψ) given suffi cient data (MacKenzie et al. 2002(MacKenzie et al. , 2006. We assumed that the occupancy states of species could change among years, but not between visits within a year. We used multispecies occupancy models to estimate species-specifi c parameters as random variables governed by communitylevel hyper-parameters. The use of a common distribution among species improves the Table 1 . Predicted species-specifi c responses to fi re by life history traits based on published literature , Kotliar et al. 2007, Fontaine et al. 2009 ).

Life history traits Expected response
Primary cavity-nesting N , beetle-foragers F (i.e., bark-drilling woodpecker species) Strongly positive due to increases in standing dead wood (snags and dead portions of live trees), in which these species excavate nest cavities and forage for bark-and wood-boring beetle larvae Secondary cavity-nesting N Positive following increased availability of nest holes excavated by primary cavity nesters (i.e., woodpeckers), assuming other life requisites are present Foliage and bark-gleaning insectivores F Negative due to desiccation of foliage used for nesting and foraging substrates, especially a er high-severity crown fi re Pine seed consumers F Negative due to loss of pine seed production, especially a er high-severity crown fi re Shrub-or ground-nesting N , near-ground insectivores F Positive following regrowth of understory vegetation, which is stimulated by opening of the canopy, particularly at Paye e location where shrub understory was prevalent Aerial insectivores F Positive due to decreases in canopy cover, which provides more space for foraging maneuvers Open-cup canopy nesting species N Negative due to desiccation of foliage used for nesting, especially a er high-severity crown fi re Note: Nesting and foraging traits are indicated by superscript "N" and "F", respectively.
precision of species-specifi c parameter estimates, particularly for rare species (Dorazio et al. 2006 ). We excluded raptors, owls, and grouse because they were not readily detectable with our survey methods. We included only species that bred in our study areas. For mobile animals such as birds, detectability ( p ) estimated with surveys repeated over a season includes information on within-season movement and the observation process (i.e., availability and perceptibility; Chandler and Royle 2013 , Amundson et al. 2014 ). Occupancy probabilities in this study therefore characterize the probability that a surveyed point overlaps at least one home range for a given species. Predictions for species and communities shaped our analysis. Primarily, because birds have historically experienced more high-severity fi re at the Paye e compared to the Coconino location, we expected birds would have greater affi nity with severely burned forests at Paye e. Consequently, we predicted proportionately more positive species relationships and a positive species richness relationship with burn severity at Paye e, whereas we predicted the opposite at Coconino. Additionally, to understand the mechanisms underlying community-level patterns, we evaluated predictions for species burn-severity relationships based on life history categories described in the literature (Table 1 ; , Kotliar et al. 2007, Fontaine et al. 2009 ). We expected these predictions to generally describe pa erns but recognized that individual species have unique life histories that do not fi t perfectly within broad categories. Thus, we considered both general predictions for life history categories and literature on individual species to evaluate the likely generality of observed relationships across locations and regions.
We used a three-dimensional data matrix y , where element y ĳ t was a sum of binary indicators for species detection, rather than the binary indicators of species detection typically used in single-or multispecies occupancy models. When binary indicator of species detection x ĳ t = 1, we detected species i ( i = 1, …, N ) at point count station j ( j = 1, …, J ) during primary sampling occasion year t ( t = 1, …, T ; T = 3). Because we did not have covariates that diff ered for detection between secondary sampling occasions, we analyzed the sum of all binary species i detections over all secondary sampling occasions ( s ) at each point count station, where y ijt = ∑ 3 s=1 x ijts and y ĳ t ε {0,1, …, K } ( K = 2 at Paye e and 3 at Coconino). We modeled the probability of Bernoulli latent variable z ĳ t for occupancy given probability of occupancy Ψ ĳ t as: (1) We analyzed relationships between avian occupancy and burn severity for comparisons between our two study locations (Paye e and Coconino NFs). For both locations, we modeled occupancy probability Ψ ĳ t for species i at point count station j in year t as a linear function of the covariate for burn severity (ΔNBR). We examined the medians and 95% Bayesian credible intervals (BCI) for this relationship ( β Δ NBR ) to evaluate statistical support for relationships with burn severity. For numerical reasons, we standardized ΔNBR values using the same scaling constants in both locations to ensure comparability of β Δ NBR estimates (justifi ed given the similar range of burn severities observed at both locations). We included additional covariates to account for major sources of variation present in each data set. For Coconino data, Ψ ĳ t was partially dependent on the probability of occupancy in the previous year, i.e., we assumed a Markovian process described the current occupancy state. Thus, where β 0, i is the intercept, φ i is a Markovian parameter for local probability of persistence from the previous year, z ĳ 0 ~ Bern( Ψ i 0 ) (see Russell et al. 2009 ), β Δ NBR ,i is the burn severity eff ect, and all three are species-specifi c normal random eff ects. The Paye e data set contained more inter-annual variability than Coconino data, so we assumed Ψ ĳ t was non-Markovian. Paye e sampling was insuffi cient ( n = 81 points) to support both year and Markovian eff ects (a model with both did not converge within a reasonable timeframe and posterior predictive GOF tests ( Appendix S1 ) found li le unexplained varia- tion by a model with only year eff ects). Thus for Paye e data, where β 0, it was a species-specifi c and a yearspecifi c normal random eff ect. We considered parameters describing burn severity relationships ( β Δ NBR, i , Eqs. 2 and 3 ) whose 95% BCIs did not overlap zero to be strongly supported and other relationships as either marginally supported (if posterior mass was mainly positive or negative) or unsupported. We compared relationships between locations for species that had strongly supported relationships at one or both locations. We also highlight species whose relationships were marginally supported at both locations but diff ered in direction. We mainly compared the direction (positive or negative) of speciesspecifi c relationships between locations, but we also evaluated overlap of 95% BCIs between parameter estimates to recognize diff erences in magnitude. Recognizing the potential for sampling error, we tempered interpretation of strongly supported parameters estimated with small sample sizes (species detected during ≤10 point × year occasions).
In addition to species-specifi c relationships, we report emergent changes in species richness with burn severity. We calculated species richness ( N jt ) at each point count station j and year Russell et al. ( 2009 ) and unlike others (Dorazio et al. 2006, Kéry et al. 2009 ), we did not model unobserved species, so community-level inferences were restricted to the subset of members observed at least once during our studies.
For all models, we modeled the probability of observing species i at point count station j during primary period t , y ĳ t , given K secondary periods, probability of detection p i , and occupancy latent state z ĳ t using a binomial distribution with K trials and probability of success p i × z ĳ t : (4) Ninety-fi ve percent BCIs for parameters relating burn severity with detection probability overlapped 0 for all species for both data sets, and adding this parameter did not improve mod-el fi t to the Coconino data set (deviance information criterion increased by 2529.9). Therefore, we only modeled detectability as a species-specifi c normal random eff ect b 0, i : where p i is the probability of detecting species i during a survey of a given point count station in a given year when the species was present. We modeled heterogeneity among species using a correlation term ( ρ ) between species intercepts of detection probability ( b 0, i ) with occupancy probability ( β 0, i ) Royle 2005 , Kéry et al. 2009 ). We assumed a multivariate logit-scale normal distribution, where only the non-zero off -diagonal elements of the variance-covariance matrix for occupancy and detection parameters were between b 0, i and β 0, i .
We partly accounted for diff erences in survey protocol between locations (i.e., diff erences in point count duration and number of surveys per year) by running separate analyses by location and allowing for diff erent numbers of secondary periods with our statistical models. Nevertheless, because we did not model unobserved species, longer point count duration and more repeat surveys increased chances of observing rare species, potentially elevating species richness estimates at Coconino relative to Paye e. These diff erences were of minor importance, however, given our interest in comparing burn severity relationships rather than overall occurrence rates or species richness.
We sampled posterior parameter distributions for all models using either WinBUGS v. 1.4.3 (Lunn et al. 2000 ) or JAGS v. 3.3.0 (Plummer 2003 ) programmed with associated packages (R2WinBUGS and R2jags; Sturtz et al. 2005 , Su and Yajima 2014 ) from R (R Core Team 2013 ). We used independent noninformative priors for all parameters (for priors, see Appendix S2 ; for model code, see Appendix S3 ). We ran 3-6 parallel MCMC chains (Coconino: three chains of length 40,000 it , burn-in 25,000 it , and thinning 10 it ; Payette: six chains of length 51,000 it , burn-in 1000 it , and thinning 100 it ) to sample posterior distributions until n effective ≥ 100 and R ≤ 1.1 for all parameters (Gelman and Hill 2007 ). We ex-  amined model goodness-of-fit using posterior predictive testing based on both omnibus and targeted descriptions of the data ( Appendices S1, S2 ; Gelman and Hill 2007 ).

RESULTS
Surveyors detected 81 species: 37 species were found at both locations, and 16 and 28 species 18.2 20.8 † ΔNBR = the change in remotely sensed normalized burn ratio from before to a er wildfi re.

Fig. 3 . Posterior parameter estimates (medians and 95% BCI s) describing species occupancy relationships
with wildfi re burn severity ( ̂Δ NBR ). We compared relationships for the Coconino NF , Arizona ( COAZ ), and the Paye e NF , Idaho ( PAID ). Panel (a) presents estimates for species observed at both locations. The remaining panels present species only observed at Paye e (b) or Coconino (c) locations. Statistically supported positive and negative relationships are colored orange and blue, respectively. We used this information to identify similarities and diff erences in avian relationships with wildfi re between study locations. See Table 2 for common and taxonomic species names. were unique to Paye e or Coconino locations, respectively (Table 2 ). Chipping Sparrow, Western Tanager, Yellow-rumped Warbler, Redbreasted Nuthatch, and Hammond ' s Flycatcher were detected most frequently (during the most point × year occasions) at the Paye e location (see Table 2 for taxonomic names). Western Bluebird, Western Tanager, Dark-eyed Junco, Yellow-rumped Warbler, and Western Woodpeewee were detected most frequently at the Coconino location. Occupancy and detection were correlated at both locations (Paye e ρ = 0.71; 95% BCI = 0.37-0.95; Coconino ρ = 0.46; 95% BCI = 0.01-0.74). Point count stations sampled areas ranging widely in burn severity (Table 3 ). Detection probability median posterior estimates varied among species (species-specifi c median posterior p = 0.01-0.49 [min-max] at Coconino sites and 0.02-0.68 at unburned Paye e sites; Appendix S4 ). Posterior predictive tests provided no evidence indicating lack of model fi t with respect to diff erences in apparent occupancy (the proportion of point × year survey occasions a species was detected) among fi re conditions. Furthermore, we found evidence for lack of fi t with respect to inter-annual variation in apparent occupancy and apparent turnover for only a few species, suggesting our models adequately described the data ( Appendices S1, S2 ).

Relationships with wildfi re burn severity
We found both similarities and diff erences in relationships with wildfi re burn severity between the Coconino and Paye e locations for the 37 species observed at both locations. Relationships for 21 species (including 13 observed at both locations and 8 observed at only one location) were statistically supported (BCIs did not overlap zero) at one or both locations (Fig. 3 ). We observed consistently positive relationships with burn severity for House Wren, Mountain Bluebird, Olive-sided Flycatcher, and Hairy Woodpecker, and we observed consistently negative relationships for Steller ' s Jay and Mountain Chickadee (Figs. 3  and 4 ).
Other species exhibited relationships that were less consistent between study locations (Figs. 3 and 5 ). Townsend ' s Solitaire exhibited a negative relationship at the Coconino location but a marginally supported positive relationship at the Paye e location. Chipping Sparrow exhibited contrasting relationships between locations (negative at Coconino, positive at Paye e), although both were marginally supported. Yellow-rumped Warblers, White-breasted Nuthatches, and Dark-eyed Juncos exhibited negative relationships at Coconino, whereas relationships at Paye e for these species were weak and unsupported. American Three-toed Woodpeckers and Western Wood-pewee ex- Fig. 5 . Predicted occupancy ( ̂p red ) relationships with wildfi re burn severity (Δ NBR ) for eight species that exhibited inconsistent relationships either in direction or magnitude between Coconino (squares with dashed lines) and Paye e (circles with solid lines) locations. Predicted occupancy probabilities are plo ed for low burn severity (Δ NBR = −94), moderate burn severity (Δ NBR = 268; mean for Paye e sites), and high burn severity (Δ NBR = 630). Predicted occupancy probabilities are presented to aid interpretation of the magnitude of eff ect sizes, i.e., how much occupancy changed with varying burn severity. See Table 2 for common and taxonomic species names. hibited positive relationships only at Coconino, although sample sizes were limited at both locations and Paye e, respectively. Cassin ' s Finches were positively related with burn severity only at Paye e, but sampling was limited at Coconino. Although consistent in direction, relationships for Hairy Woodpecker and Mountain Chickadee clearly diff ered in magnitude between locations (Fig. 3 ).
Of species that were unique to either location, only Lesser Goldfi nch at Coconino exhibited a statistically supported positive relationship (Fig. 3 ). One species only at Paye e (Townsend ' s Warbler) and six species only at Coconino (Plumbeous Vireo, Pinyon Jay, Violet-green Swallow, Brown-headed Cowbird, Pygmy Nuthatch, and Grace ' s Warbler) exhibited statistically supported negative relationships.
Overall, proportionately more species at the Paye e location exhibited statistically supported positive rather than negative relationships with burn severity (four vs. two species, respectively), whereas the opposite was true at the Coconino location (6 positive vs. 12 negative relationships). Species richness pa erns also diff ered between locations. Species richness tended to increase at the Paye e location but decrease at the Coconino location with increasing burn severity (Fig. 6 ).

DISCUSSION
Observed diff erences in avian occupancy relationships with burn severity are consistent with expected responses to historical fi re regimes and likely refl ect life history adaptations to environmental conditions generated by wildfi re. Because we used comparable methods and sampled a comparable range of burn severities, variation in observed pa erns can be more clearly a ributed to diff erences between locations. Forests of Idaho (i.e., central-northern Rocky Mountains) historically experienced a mixed-severity fi re regime (cf. Schoennagel et al. 2004, Block et al. 2012, Mellen-McLean et al. 2013, whereas those of Arizona (i.e., American Southwest), a low-severity regime (Schoennagel et al. 2004, Bock and Block 2005, Hu o and Belote 2013. Birds in the Paye e and Coconino forests likely evolved strategies specifi c to these locations for when, where, and how to forage and nest. If disturbance (wildfi re in this case) results in conditions comparable to what these populations experienced historically, they will likely benefi t more from fi re. If resulting conditions are not comparable, populations may be reduced or extirpated in aff ected areas. We predicted that proportionately fewer species would respond positively to severe fi re and, consequently, that species richness would decline with increasing burn severity in areas characterized by a low-severity fi re regime. In contrast, we predicted the opposite in areas with historically mixed-severity fi re regimes. Our results were consistent with these predictions.
Species relationships with burn severity and diff erences in relationships between locations were generally consistent with species life histories. Wildfi re benefi ts species with certain life history traits, and these species tend to associate with severe burns. In contrast, species with life histories that confer negative responses to wildfi re will generally tend to be less prevalent in the most severely burned habitats. The consistency of observed relationships with species life history traits suggests some generality of species Fig. 6 . Species richness posterior estimates ( N ; median and 95% BCI s) for point × year survey occasions plo ed against wildfi re burn severity (Δ NBR ). Best-fi t lines showing species richness trends are also depicted. We compared species richness trends for sites in the Coconino NF , Arizona (a) vs. the Paye e NF , Idaho (b). We used this information to compare community relationships with burn severity between study locations. relationships, which are foundational to community pa erns.

Consistency with life history and generality of observed patterns
By comparing only two locations, our inferences were limited by lack of replication. Such limitations are pervasive for studies of postfi re ecology given the unpredictability of wildfi re and infeasibility of experimental study. Nevertheless, species relationships with burn severity observed here were consistent with species life histories and relationships reported in the literature (Smucker et al. 2005, Kotliar et al. 2007, Fontaine and Kennedy 2012, Seavy and Alexander 2014, suggesting some generality of observed pa erns. Because species relationships are foundational to community pa erns, consideration of species relationships suggests underlying mechanisms and generality for community pa erns. Hairy and American Three-toed Woodpeckers, both bark-drilling species, tended to relate positively with burn severity, congruent with their reliance on resources enhanced by wildfi re: standing dead wood for nesting, and bark and wood-boring beetle larvae (e.g., Scolytidae and Cerambycidae , respectively) for food (Covert-Bratland et al. 2006 ). Although sample sizes were limited for American Three-toed Woodpecker, our results were consistent with those from a Colorado study (Kotliar et al. 2008 ). Positive relationships for bark-drilling species in this and other studies (Kotliar et al. 2008 ) suggest generality of the direction of wildfi re responses for this group, although the magnitude of responses may vary.
Consistent with our predictions and other studies (Kotliar et al. 2002, Smucker et al. 2005, Fontaine and Kennedy 2012, secondary cavity-nesting species (House Wren and Mountain Bluebird) and aerial insectivores (Mountain Bluebird and Olive-sided Flycatcher) exhibited positive relationships with burn severity, suggesting generality of these relationships. Western Wood-peewee exhibited a positive relationship at the Paye e location only, but sampling was limited at Coconino. Violetgreen Swallows forage above the canopy rather than in canopy openings and do not necessarily use woodpecker-excavated cavities for nesting (Brown et al. 2011 ), so wildfi re may aff ect this species diff erently than other secondary cavitynesting aerial insectivores.
Some species that nest in cavities but require live substrate for foraging exhibited negative relationships with burn severity (Mountain Chickadee, White-breasted Nuthatch, and Pygmy Nuthatch). Relationships for these species were more strongly and defi nitively negative at Coconino, however, and are also mixed across studies (Smucker et al. 2005, Kotliar et al. 2007, Fontaine and Kennedy 2012, Seavy and Alexander 2014. Mixed responses to wildfi re and other natural disturbance may refl ect a mix of lost foraging opportunities and gained nesting opportunities (Saab et al. 2007 , Norris andMartin 2010 ). Our results suggest these species may adjust their foraging in response to wildfi re more eff ectively at locations where they are more accustomed to high-severity fi re.
Open-cup canopy-nesting species that forage in live trees (Townsend ' s Warbler at Paye e; Yellowrumped Warbler, Grace ' s Warbler, and Plumbeous Vireo at Coconino) generally exhibited negative relationships with burn severity. These relationships were consistent with our predictions and other studies (Smucker et al. 2005, Kotliar et al. 2007, Fontaine and Kennedy 2012. Negative relationships with burn severity also followed predictions for other open-cup canopynesting species, Stellar ' s Jay and Pinyon Jay. Pine seed foraging by Pinyon Jays may also contribute to negative wildfi re responses. Pinyon Jays, however, have a diverse diet (Balda 2002 ) and positive or negative relationships with wildfi re are not widely observed (e.g., none reported by Kalies et al. 2010 , Fontaine andKennedy 2012 ). As generalist foragers, Steller ' s Jay may take advantage of food resources generated by fi re in some cases, possibly explaining positive relationships with relatively moderate burn severity levels reported elsewhere (Kotliar et al. 2007 ).
Finches as a group tended to exhibit positive relationships with burn severity at the locations where they were relatively common (Cassin ' s Finch at Paye e; House Finch and Lesser Goldfi nch at Coconino), a pa ern consistent with the literature , Smucker et al. 2005, Fontaine and Kennedy 2012, Seavy and Alexander 2014. Finches generally favor open-canopy forests and frequently forage on the ground or on www.esajournals.org LATIF ET AL. seeds of herbaceous vegetation whose growth is stimulated by fi re (Hahn 1996, Hu o et al. 2014, Wa and Willoughby 2014. Relationships with burn severity exhibited by species that nest and forage in the understory were inconsistent between locations and tended to be more negative at Coconino. A relatively dense understory layer at the Paye e location provided for relatively fast regrowth (cf. Saab et al. 2006 ), potentially providing more nesting and foraging opportunities for species such as Dark-eyed Junco, Chipping Sparrow, and Townsend ' s Solitaire. In contrast, the relative lack of a substantial shrub layer at the Coconino location (cf. Saab et al. 2006 ) may explain their more negative relationships with burn severity in the Arizona forest. Positive relationships for shrub-associated species with low-to-moderately burned habitats are reported in other forests with mixed-severity fi re regimes (Smucker et al. 2005 , Fontaine andKennedy 2012 ). Foraging resources for Townsend ' s Solitaire, however, are not necessarily tied to shrubs (Bowen 1997 ), so other mechanisms related to food may be more important for this species.

Locational and regional differences
The Coconino wildfi re was less severe and burned less area than the Paye e wildfi re, consistent with their respective historical fi re regimes. Species can evolve plasticity in their responses to disturbances that vary in size and severity, as well as diff erent life histories for diff erent disturbance regimes (Lytle 2001 , Lytle andPoff 2004 ). Such processes could explain pa erns observed in this study.
Heterogeneity in severity, size, and confi guration of burned patches likely diff ered in ways that could infl uence species and community responses (e.g., Cullinane-Anthony et al. 2014, Berry et al. 2015. For example, species experiencing a mix of costs and benefi ts following wildfi re (e.g., secondary cavity-nesting foliage insectivores) may benefi t greatest from fi res that leave mosaics of burned and unburned forest patches capable of fulfi lling diff erent habitat requirements (see also Wightman et al. 2010 ).
Not all species occurred at both locations, and of those that did, initial population size o en diff ered between locations. Regional diff erences in historical fi re regime and vegetation structure could fi lter for diff erent sets of species in diff erent locations. Additionally, responses to environmental disturbance by rare species (e.g., Cassin ' s Finch at Coconino) are diffi cult to detect due to sampling limitations. Conversely, wildfi re may aff ect abundance without aff ecting occupancy for common species (e.g., Yellow-rumped Warbler at Paye e).
Forest structure and composition may underlie some observed diff erences between Paye e and Coconino locations. The Coconino overstory contained more ponderosa pine and the shrub layer was extremely sparse. In contrast, the Paye e forest, although dominated by ponderosa pine, also included other conifer species in the overstory and more shrubs in the understory (cf. Saab et al. 2006 ). Occurrence and demographics of bark and wood-boring beetle species vary with host tree species and with time since wildfi re (Smith 2000 , Raff a et al. 2008 ), potentially infl uencing responses to wildfi re by bark-drilling woodpeckers (e.g., Black-backed, Hairy, and American Three-toed Woodpeckers; Smith 2000 ). As discussed above, diff erences in structure and composition of the shrub layer likely contributed to diff erences in occupancy relationships observed for understory nesting or foraging species.
The landscape context may have infl uenced observed pa erns. Wildfi res in the surrounding landscape provide alternative breeding locations for fi re associates and thus potentially aff ect their responses at any one location. The number and severity of wildfi res surrounding Paye e vs. Coconino locations during sampling therefore represent unexamined factors that could have contributed to observed pa erns. www.esajournals.org

LATIF ET AL.
We cannot completely separate the infl uence of regional diff erences in historical fi re regime from other location-specifi c characteristics. Many location-specifi c characteristics, however, are inter-related with fi re regime. Regional variation in vegetation structure parallels variation in fi re regimes due to both direct eff ects of fi re on vegetation and shared climatic drivers. Fire size, severity, confi guration of burned patches, and the number of fi res surrounding a location will likely follow the climatic and vegetation drivers that determine fi re regime. Prefi re species composition may in part be shaped by the historical wildfi re pa erns specifi c to a location. Thus, environmental or biological conditions that consistently diff er between regions with diff erent fi re regimes may suggest mechanisms for regional diff erences in wildfi re responses. Consistent with our results, a literature review restricted to southwestern studies reported more negative than positive avian relationships with highseverity wildfi re (Kalies et al. 2010 ) but another more geographically extensive review did not (Fontaine and Kennedy 2012 ).

Considerations for future research
Occupancy only refl ects one aspect of population ecology. We did not measure abundance (or territory density), which wildfi re can aff ect independently of occupancy. Detection probability can co-vary with abundance for various reasons Nichols 2003 , Warren 2011 ). We implicitly accounted for this relationship by allowing occupancy and detection to co-vary among species. Analyzing abundance explicitly would be benefi cial, however, because as a continuous response metric, it can provide additional statistical power for revealing relatively complex relationships with burn severity (e.g., Smucker et al. 2005, Kotliar et al. 2007 ). In addition to occupancy and abundance, measures of fi tness (reproductive success, survivorship) are needed to fully understand wildfi re eff ects on species.
Detection probabilities were low for many species at both locations ( Appendix S4 ). Low detectability induces biased estimation of occupancy probabilities and covariate relationships and is particularly problematic for rare species (MacKenzie et al. 2002, Moreno and Lele 2010, McKann et al. 2013, Sanderlin et al. 2014. Sanderlin et al. ( 2014 ) suggest increased accuracy for rare spe-cies could be achieved by sampling more area. More simulation studies examining estimator properties for community occupancy models are needed to guide interpretation of occupancy estimates for rare species.
Studies measuring population shi s from before to a er natural disturbance are needed to control for confounding environmental variability among surveyed sites (Wiens andParker 1995 , Popescu et al. 2012 ). We lacked data necessary to analyze such shi s at the Coconino location. Other studies, however, have found broad consistency between burn severity relationships and population shi s from before to a er wildfi re (Smucker et al. 2005, Kotliar et al. 2007, Seavy and Alexander 2014. Examining changes in occupancy or abundance with time since disturbance is also necessary for a full understanding of species ecological relationships with wildfi re Powell 2005 , Si ers et al. 2014 ). Population changes with time since wildfi re over a longer timeframe (e.g., 12 yr; Saab et al. 2007 ) would likely have diff ered between locations, but we lacked data for verifi cation. Additionally, this study only considered breeding birds. Analysis of fi re relationships during the nonbreeding season could reveal additional insights (e.g., Covert-Bratland et al. 2006, Brown et al. 2015. Depending upon the pace and consistency of natural selection along with biological constraints, current populations may be adapted to current conditions, conditions experienced previously, or conditions experienced elsewhere by past immigrants (Grant andGrant 2002 , Lytle andPoff 2004 ). Given uncertainties about how wildfi re severity and associated structural conditions vary (Pierce et al. 2004, Freche e and Meyer 2009, Williams and Baker 2012, Odion et al. 2014, selective pressures infl uencing avian wildfi re responses are not well understood. The relevant spatial scale of variability for understanding natural selection likely varies with species ecology (e.g., home range size, gene fl ow). In addition to average severity, variability in burn severity at sampling points may be relevant for species with variable resource needs (e.g., foliage-or barkgleaning insectivores that nest in cavities; see also Wightman et al. 2010 ). Studying species and communities over a range of locations varying in current and historical conditions could complement our study for understanding these issues.

Management implications
Large-scale forest restoration eff orts currently implemented in western North American dry conifer forests aim to restore natural processes disrupted by anthropogenic activities over the last 100 years (Fulé et al. 2012, Franklin et al. 2014. Fuels reduction typically represents a major component of these eff orts. A fundamental assumption of intensive fuels reduction is that recent largescale severe wildfi res are uncharacteristic of historical conditions and therefore harm biodiversity in dry conifer forests, particularly in the Southwest (Lowther 2000, Arno and Fiedler 2005, Miller and Thode 2007. Our results support the growing consensus among avian ecologists that species relationships with wildfi re vary not only with life history traits but also regionally in ways that refl ect differences in historical fi re regime, vegetation, and climate (e.g., Powell 2005 , Hu o et al. 2008 ). Our results suggest that a greater proportion of bird species in southwestern forests respond negatively than positively to high-severity wildfi re (see also Kalies et al. 2010 ), and these relationships likely diff er from dry conifer forests in the central and northern Rocky Mountains. Thus, restoration targets should account for regional diff erences in historical conditions and consequent diff erences in the ecological roles of low-, mixed-, and high-severity wildfi re. In reality, fi redependent species occurred at both southwestern and central Rocky Mountain study locations. Even in the Southwest, where wildfi res are generally expected to burn at lower severities, some species related positively with severity (Kalies et al. 2010 ). Thus, in contrast with conventional thought, some mixedseverity fi res are likely needed to support the full suite of forest birds throughout dry conifer forests of western North America.
Fire regimes and vegetation structure are fundamentally driven by climate, which is rapidly changing (Agee 1993, Schoennagel et al. 2004. Consequently, recovery of historical fi re regimes is likely impossible through forest management alone (Pierce et al. 2004 ). Nevertheless, historical conditions provide an important reference for designing forest restoration targets (Franklin et al. 2014, Drapeau et al. 2016 ). Further work examining regional variation in avian fi re associations and integrating this information with data on historical conditions could elucidate the optimal distribution of severities for promoting biodiversity within particular landscapes (e.g., Kelly et al. 2015 ). Such work would benefi t from further examination of which life history traits confer fi xed vs. variable responses to wildfi re across regions and associated fi re regimes.