Background

Managers use prescribed fire and mechanical fuel reduction treatments to reduce the risk of high-severity wildfire and to manage changes in forest integrity and ecological functions (Covington and Moore 1994; Covington et al. 1997; McIver et al. 2013). Following a century of fire suppression, land management agencies have increased the use of prescribed fire in recent decades to reduce fuel loads and restore wildlands to desired conditions in dry conifer forests of the western USA (Morgan et al. 1994; Ryan et al. 2013; Stephens et al. 2016). Such efforts follow legislative directives designed to reduce fuel loads on public lands (i.e., National Fire Plan [US Department of Agriculture (USDA) 2000], Healthy Forest Restoration Act [USDA 2003], Healthy Forest Initiative [White House 2004], Collaborative Forest Landscape Restoration Program [(Schultz et al. 2012)]). The use of prescribed fire, however, requires careful consideration of treatment effects on wildlife species and communities. In particular, the uncertainty of management outcomes given potentially novel ecological conditions and processes with the interaction of climate change and human disturbance elevate the need for quantifying ecological responses (Millar et al. 2007; Seastedt et al. 2008; McKelvey et al. 2021).

Understanding wildlife responses to prescribed fire is needed to assess the consequences of forest management practices intended to reduce fuels. Avifauna offer useful opportunities for assessing ecological integrity and biodiversity in relation to forest management actions (Sutherland et al. 2004). Forest managers pay particular interest to birds in light of policies under the National Forest Management Act (USDA 1976) for the management of biodiversity, protections afforded migratory species through the US Migratory Bird Treaty Act, and requirements under the 2012 Planning Rule and Directives to consider persistence of species of conservation concern (USDA 2012). Birds, in particular, serve as an appropriate taxon for investigating the effects of fire because they have varied and predictable relationships with vegetation structure.

Scientific understanding of avian responses to prescribed fire in Western North American forests includes persistent knowledge gaps (Russell et al. 2009; Bagne and Purcell 2011; Fontaine and Kennedy 2012; McIver et al. 2013). McIver et al. (2013) conducted continent-wide research evaluating forest treatments on avian communities of seasonally dry conifer forests, but small-sized experimental plots (~40 ha) limited their conclusions and inference. Consistent patterns have been reported in other prescribed fire studies, including reduced numbers of ground and shrub nesting birds (Wilson et al. 1995; Artman et al. 2005; Blake 2005), benefits to cavity-nesting bird populations (Blake 2005; Russell et al. 2009; Bagne and Purcell 2011), and increases in aerial insectivore and ground-foraging species (Artman et al. 2005; Blake 2005; Russell et al. 2009). Knowledge is lacking on response differences among different fire regimes. Additionally, sparse information exists from studies conducted at appropriate spatial scales for drawing inference about landbirds with varying home range sizes.

Avian responses to fire depend on burn severity (fire effects on vegetation; Agee 1993; Fontaine and Kennedy 2012) and time since fire (Saab and Powell 2005). Forested landscapes with a varied fire history are expected to support the greatest diversity of species (Clarke 2008; Fontaine et al. 2009; Fontaine and Kennedy 2012). Varied fire histories are more typical of forests with mixed-severity fire regimes occurring at intermediate frequencies, in contrast to frequent, low-severity regimes of the southwestern USA (Hood et al. 2021).

In the central and northern Rocky Mountains, mixed-severity fires burned principally in late summer and maintained heterogeneity in forest structure with mixed tree species composition, often favored by ponderosa pine (Pinus ponderosa) (Schoennagel et al. 2004; Hessburg et al. 2007) (Hood et al. 2021). In the southwestern USA, forests with lower tree densities were more widespread and associated with higher frequency, lower-severity fires that burned typically in spring. A drier climate in the Southwest favored forest patches that were relatively homogenous, lower in tree and shrub densities, and heavily dominated by ponderosa pine (Moir et al. 1997; Schoennagel et al. 2004; Nimmo et al. 2014). Considering this variation in ecological context, birds may respond differently to prescribed fire in locations with historically low- versus mixed-severity fire regimes, with potential implications for the role of prescribed fire in management strategies that include promoting and conserving biodiversity.

Here, we evaluated the influence of prescribed fire treatments on avian species occupancy and richness at locations representing both mixed- and low-severity fire regimes in dry mixed conifer forests across the interior western USA, known as the Birds and Burns Network. We used a before-after-control-impact (BACI) design with plot sizes averaging 300 ha for rigorous evaluation of treatment effects (Morrison et al. 2008; Popescu et al. 2012). Based on habitat changes expected within 5 years of low-severity prescribed fire treatments (Saab et al. 2006), we predicted changes in species’ occupancy rates concurring with life history traits (Table 1), including reductions of ground and shrub nesting birds (Wilson et al. 1995; Artman et al. 2005; Blake 2005), benefits to cavity-nesting bird populations (Blake 2005; Russell et al. 2009; Bagne and Purcell 2011), and advantages to species that forage in the air and on the ground (Artman et al. 2005; Blake 2005; Russell et al. 2009).

Table 1 Predicted species-specific responses to fire by life history traits based on published literature and reviewed in Birds of the World (Billerman et al. 2020)
Full size table

We also expected species richness and average species responses to vary regionally depending on the different historical fire regimes (Latif et al. 2016b). Because prescribed fire is intended to burn at low severity, we predicted occupancy changes to be more positive and of stronger magnitude at locations characterized by historically low-severity fire regimes.

Methods

Study system

We selected paired study units (treatments and controls) within areas identified by seven national forests that planned to conduct fuel reduction treatments for the Birds and Burns Network (Fig. 1). Each study unit was approximately 100–400 ha (Table 2) and dominated by ponderosa pine that had not burned by wildfire for nearly 100 years or prescribed fire for more than 10 years. Sixteen study units were located in 3 national forests characterized historically by mixed-severity fire regimes: 6 in the Payette National Forest (NF; Idaho), 6 in the Okanogan-Wenatchee NF (Washington), and 4 in the San Juan NF (Colorado). Ten study units were located in 4 national forests characterized historically by a low-severity fire regime: 2 each in the Gila (New Mexico) and Coconino NFs (Arizona) and 3 each in the Apache-Sitgreaves and Kaibab NFs (Arizona; 1 treatment unit “paired” with 2 smaller control units at each location; Fig. 1). Thus, we established 12 treatment-control pairs consisting of 26 study units (Table 2). USFS District fire personnel determined the location and boundary of each prescribed fire treatment unit.

Fig. 1
figure 1

Study areas of the Birds and Burns Network located on 7 national forests of the interior western USA. Triangles represent study areas with historically mixed-severity fire regimes, whereas circles represent those with low-severity fire regimes

Full size image
Table 2 Locations, areas, sampling distributions, and historical fire regime, for study units where avian community changes in relation to prescribed fire were studied on 7 national forests in the Interior West
Full size table

Ponderosa pine trees > 23 cm diameter-at-breast-height (dbh) dominated overstory vegetation on all units with both low-severity and mixed-severity historical fire regimes. In low-severity locations, the understory was relatively open with few shrubs and dominated by grasses, i.e., Arizona fescue (Festuca arizonica) and blue gramma (Bouteloua gracilis), and elevations ranged from 2072 to 2500 m. In mixed-severity locations, the understory vegetation was comprised of multiple shrub species, including snowberry (Symphoricarpos spp.), spirea (Spirea spp.), ninebark (Physocarpus malvaceus), and serviceberry (Amelanchiar alnifolia), with Bluebunch wheatgrass (Pseudoroegenaria spicatus) and Idaho fescue (Festuca idahoensis) as the common grass species and elevations ranged from 670 to 1980 m.

Study units

In 2001–2003, we established 320 point count stations in units characterized historically by mixed-severity regimes and 278 in low-severity regime units (hereafter mixed-severity and low-severity units, respectively), for a total of 598 point count stations (Table 2). We placed point count stations at least 250 m apart and 250 m from the edge of study unit boundaries and visited each station multiple times (1–4 visits per point) annually between 22 May and 3 July.

Bird surveys

We surveyed birds for 1–4 years and 1–3 years before and after prescribed fire treatments at mixed-severity and low-severity units, respectively. For the BACI design, we collected data in control units that were never burned but measured during the same timeframe, i.e., before and after prescribed fire applied to the treatment units. Control and treated points were surveyed in the same years. For comparability across fire regimes, we restricted our primary analysis to data from 2 years before to 2 years after prescribed fire treatments (Table 3). A supplemental analysis included all available data from mixed-severity units (Additional file 1: Appendix A). We began point counts just after the dawn chorus and completed them within 5 h. Only detections within 75 m of the point were included in this analysis. Our sampling design included a robust design (Pollock 1982) with years as primary periods and visits within years as the secondary samples.

Table 3 Treatment and sampling timing at 26 study units established for the primary analysis of avian community changes with prescribed fire at 7 national forests in the Interior West. In mixed-severity locations where fire treatments occurred in spring of the same year as post-treatment bird surveys, fire treatments were always applied prior to the breeding season, in contrast with bird surveys, which always occurred during the breeding season
Full size table

Burn severity measurements and analysis

Prescribed fire treatments were implemented during 2003–2010 (Table 3). Fires were designed to reduce existing surface and ladder fuels of relatively small diameter (< 15.4 cm dbh), and create small gaps in the upper tree canopy, while retaining large pine trees and snags (>23 cm dbh). We measured burn severity using a composite burn index (CBI) representing a gradient of unburned (min CBI = 0) to severely burned (max CBI = 3; Key and Benson 2006; Additional file 1: Appendix B). We assumed CBI = 0 for all survey points in untreated units. One treatment unit in Idaho was burned by wildfire before we could measure post-treatment vegetation. We used the Bayesian approach to missing value imputation for this unit whereby we treated missing values as a parameter to be estimated during model-fitting (Link and Barker 2010). We used a truncated Gaussian prior distribution for missing CBI values in Idaho (mean = 0.32, variance = 0.07, min = 0), approximating the CBI distribution for another unit in the San Juan NF (CO) that burned at a qualitatively similar severity (based on visual assessment). Saab et al. (2006) reported effects of prescribed fire on vegetation. To briefly summarize these effects, overall downed woody material declined by 35% in the Southwest (low-severity fire regime) and 46% in the Northwest (mixed-severity fire regime). Large-diameter trees (>23 cm) declined by 19% in the Southwest with no change in the Northwest. Large-diameter snags increased by 72% on the southwestern forests and 29% on the northwestern forests (Saab et al. 2006).

Occupancy models

We used avian point count data in a hierarchical multi-species occupancy model (Dorazio et al. 2006; Russell et al. 2009) to identify changes in occupancy rates in relation to prescribed fire treatments and burn severity. Occupancy models leverage repeat-survey data to estimate species detectability (p) conditional upon occupancy (species presence within a specified time period and spatial unit), allowing unbiased estimation of occupancy probabilities (ψ) given sufficient data and adherence to model assumptions (MacKenzie et al. 2002; MacKenzie et al. 2018). We assumed that the occupancy states of species could change among years, but not between visits within a year. We used multi-species occupancy models to estimate species-specific parameters as random variables governed by community-level parameters. The use of a common distribution among species improves the precision of species-specific parameter estimates, particularly for rare species, facilitating estimation of species richness (Dorazio et al. 2006; Russell et al. 2009). We excluded raptors, owls, and grouse because they were not readily detectable with our survey methods, and we only included species breeding in our study areas. For mobile animals such as birds, detectability (p) estimated with surveys repeated over a season includes information on both within-season movement and surveyor ability (i.e., availability and perceptibility; sensu Chandler and Andrew Royle 2013; Amundson et al. 2014). Occupancy probabilities thereby represent the probability of a surveyed point intersecting at least one home range for a given species (Latif et al. 2016a).

We used occupancy patterns estimated from our analysis to evaluate predictions for species based on their individual life histories (Table 1) and on previous research (Russell et al. 2009; Gaines et al. 2010; Bagne and Purcell 2011; Fontaine and Kennedy 2012; McIver et al. 2013). We expected these predictions to describe general patterns while taking into consideration that each individual species has a unique life history that may not fit perfectly within the broad categories for which we had a priori predictions. Thus, we considered both general predictions for life histories and literature on individual species when evaluating whether patterns were consistent with current knowledge.

For each study location, we compiled a 3-dimensional data matrix y, where element yjit was the sum of binary indicators for species detection (Sanderlin et al. 2014). Given a binary indicator xjikt = 1, we detected species i (i = 1,…,N) at point count station j (j = 1,…,J) during visit k (k = 1,…,K) in year t (t = 1,…,T; T = 4). Because we did not have covariates that differed for detection between visits, we analyzed the sum of all binary detections for species i over all visits at each point count station j in year t, where \({y}_{jit}=\sum_{k=1}^K{x}_{jikt}\) and yijt ϵ [0,1,…,K]. We modeled these data given probability of detection pi, and occupancy latent state zijt using a Bernoulli distribution with probability of success pi × zijt:

$$\left[{y}_{jit}|{p}_i,{z}_{jit}\right]\sim Bin\left(K,{p}_i\times {z}_{jit}\right)$$
(1)

where the latent variable zijt for occupancy given probability of occupancy ψijt was modeled as:

$$\left[{z}_{jit}|{\psi}_{jit}\right]\sim Bern\left({\psi}_{jit}\right)$$
(2)

We analyzed changes in species occupancy patterns using a model that fully leverages our BACI sampling design for examining treatment effects (Popescu et al. 2012). Although our study design entailed surveying units with a priori assignments of treatment versus control, treatments did not realize homogenous impacts on vegetation structure and composition. We therefore measured shifts in occupancy from before to after treatment along a continuous burn severity gradient represented by CBI to evaluate treatment effects (CBI = 0 for unburned control units). We modeled occupancy (ψijt) as a function of burn severity measured after treatment (CBIj), treatment period (PERjt = 0 or 1 for before or after site j was treated, respectively), and the interaction between severity and period (CBIj × PERjt). Thus,

$$\mathrm{logit}\left({\psi}_{jit}\right)={\beta}_{0, il}+{\beta}_{PER, ir}\times {PER}_{jt}+{\beta}_{CBI, ir}\times {CBI}_j+{\beta}_{PER\times CBI, ir}\times {PER}_{jt}\times {CBI}_j$$
(3)

where β0,il is the intercept and βir parameters described additive or interactive effects of covariates PERjt and CBIj on occupancy of species i at site j in year t. All estimated parameters were species-specific normal random effects, β0,il was estimated separately by location (l = 1,…,L; L = 7), and covariate effects (βir) were estimated separately by fire regime (r = 1,…,R; R = 2). For numerical purposes, CBIj values were centered at the mean for point count stations in treated units (0.76) prior to all analyses. Unlike others (Russell et al. 2009), we did not model persistence as a Markovian process in our primary model to avoid stretching the limits of our data at low-severity locations. We did include Markovian species persistence, however, in a supplemental analysis of data from mixed-severity locations (described further below and in Additional file 1: Appendix A).

We primarily inferred species-specific prescribed fire effects from the extent to which occupancy shifted towards or away from severely burned (or unburned) points following treatment (hereafter treatment effect refers to βPER × CBI, ir in Eq. 3). We considered evidence for prescribed fire effects to be definitive for species with statistically supported treatment effects (90% BCI excluded zero). We also examined support for differences in treatment effects between fire regimes by deriving the 90% BCI for

$${\beta}_{diff}={\beta}_{mixed}-{\beta}_{low}$$
(4)

where βmixed and βlow represent estimated treatment effects (i.e., βPER × CBI, ir in Eq. 3) in mixed- and low-severity regimes, respectively.

Our sampling design afforded inferences that were stronger than purely observational studies but not equivalent to a fully controlled experiment (see Popescu et al. 2012 and literature referenced therein). We controlled for potentially confounding factors by randomly designating members of paired units as treatment versus control in most cases, replicating sampling (i.e., sampling multiple units and locations in each fire regime), and explicitly separating treatment effects (βPER × CBI, ir) from potentially confounding sources of variation (βPER, ir, βCBI, ir; Eq. 3) during analysis (Popescu et al. 2012). Nevertheless, burn severity within treated units was likely influenced by factors for which we did not explicitly control (e.g., vegetation structure, moisture levels, and topography). Furthermore, some species whose life histories typically confer effects of prescribed fire may exhibit subdued treatment effects if occupancy already favors desirable sites prior to treatment. Finally, our timeframe of sampling (2 years pre-, 2 years post-treatment) potentially limited the scope and strength of inference, especially because we expected some species to exhibit delayed effects of treatment (e.g., Taillie et al. 2018).

Considering these limitations, we supplemented our evaluation of BACI treatment effects (βPER × CBI, ir; hereafter treatment effects) by also evaluating post-treatment CBI-occupancy relationships (hereafter burn severity relationships = βCBI, ir + βPER × CBI, ir). We drew the strongest inference from treatment effects, and we also drew weaker but substantive inference from burn severity relationships that were consistent with our predictions based on species life histories. We followed up our primary analysis with two supplemental analyses. For one, we included data from additional years at mixed-severity locations and a Markovian persistence parameter to better account for variability among years (hereafter “extended sampling model”; Additional file 1: Appendix A). For the other, we analyzed data from each fire regime separately and estimated separate treatment effects for each post-treatment year (hereafter “yearly effect model”; Additional file 1: Appendix A). We examined species with statistically supported treatment effects or CBI relationships from our primary analysis and/or statistically supported treatment effects in supplemental analyses. We evaluated the strength of evidence for prescribed fire effects based on the consistency of patterns estimated across analyses and with biologically based predictions (Table 1).

We modeled detectability separately by location (fixed effect) and as a species-specific normal random effect b0,i:

$$\mathrm{logit}\left({p}_{il}\right)={b}_{0, il}$$
(5)

where pil is the probability of detecting species i at location l during a survey of a given point count station in a given year when the species was present. We modeled heterogeneity in detectability among species and assumed detectability did not change with treatment condition (preliminary models with treatment effects on detection converged poorly and were therefore abandoned). We modeled heterogeneity among species using a correlation term (ρ) between species intercepts of detection probability (b0,i) with occupancy probability (β0,i) (Dorazio et al. 2006; Kéry et al. 2009).

In addition to species-specific relationships, we calculated and plotted emergent changes between species richness with treatment condition. We estimated species richness (Njt) at each point count station j and year t: \({N}_{jt}=\sum_{i=1}^{\max (i)}{z}_{ijt}\). Community-level inferences were restricted to the subset of members observed at least once during our studies (cf. Russell et al. 2009; Latif et al. 2016b).

We sampled posterior parameter distributions for all models using JAGS v. 3.3.0 (Plummer 2003) programmed from R (Team 2013) (Su and Yajima 2014). We used independent non-informative priors for all parameters (for priors, see Additional file 1: Appendix C; for model code, see Additional file 1: Appendix D). We ran 6 parallel MCMC chains of length 100,000 it, burn-in 10,000 it, and thinning 10 it to sample posterior distributions. We verified that neffective ≥ 100 and \(\hat{R}\) ≤ 1.1 for all parameters (Gelman and Hill 2007). We examined model goodness-of-fit (GOF) using posterior predictive testing (Gelman and Hill 2007). Specifically, we calculated a Bayesian p-value representing the proportion of simulated datasets drawn from model posterior predictive distributions with deviance higher than deviance for observed datasets from each location, whereby p < 0.05 or p > 0.95 constitutes evidence for lack of fit.

Results

Ninety-five species were detected across all point count stations and years (Additional file 1: Appendix E). Forty-seven species were detected in both fire regimes, 19 unique to low-severity locations, and 29 at only mixed-severity locations. The five most commonly detected species at the three mixed-severity locations were western tanager, yellow-rumped warbler, chipping sparrow, red-breasted nuthatch, and Mountain Chickadee. The five most commonly detected species at low-severity locations were dark-eyed junco, pygmy nuthatch, Mountain Chickadee, western bluebird, and Grace’s warbler. Burn severity measured at points within treated units was variable (mean [SD] CBI = 0.92 [0.40], n = 274 points) and broadly overlapped among locations (Fig. 2). Detection probability estimates varied among species (median posterior p ranged 0.008–0.637; Additional file 1: Appendix F) and were highly correlated with occupancy (median estimate [90% BCLs] for ρ = 0.80 [0.75–0.85]). We found no evidence for lack of model fit (location-specific GOF p values from the primary model ranged 0.34–0.42).

Fig. 2
figure 2

Composite burn index (CBI) frequency distributions by national forest study location: Okanogan-Wenatchee in Washington (OKWA), Payette NF in Idaho (PAID), San Juan NF in Colorado (SJCO), Apache-Sitgreaves NF in Arizona (ASAZ), Coconino NF in Arizona (COAZ), Gila NF in New Mexico (GINM), and Kaibab NF in Arizona (KAAZ). Sample sizes (n) represent the number of point count stations where birds were surveyed. Vertical solid lines denote mean values and vertical dashed lines denote 1 SD above and below the mean

Full size image

Species-level prescribed fire effects and burn severity relationships

We identified 33 species for which we found statistically supported treatment effects and/or burn severity (CBI) relationships (Figs. 3, 4, and 5). Treatment effects were supported for 4 species in our primary analysis (American robin, western bluebird, hairy woodpecker, and ruby-crowned kinglet) and 8 additional species in supplementary analyses (black-backed woodpecker, American three-toed woodpecker, house wren, dusky flycatcher, western wood-pewee, gray flycatcher, red-faced warbler, and lark sparrow). Supported treatment effects for these species were primarily positive and observed at locations characterized by mixed-severity regimes (9 species). We found one negative treatment effect in mixed-severity locations (ruby-crowned kinglet) and two in low-severity locations (red-faced warbler and lark sparrow). We also found 36 statistically supported CBI relationships for 29 species (primary analysis), including relationships for 8 of 12 species listed above with supported treatment effects (Figs. 4 and 5). Supported CBI relationships included 11 positive and 12 negative at mixed-severity regime locations and 9 positive and 4 negative at low-severity regime locations.

Fig. 3
figure 3

Parameter estimates (posterior median) and 90% BCIs describing treatment effects (\({\hat{\beta}}_{CBI\times PER}\) from Eq. 3). Estimates are for locations with historically mixed-severity (circles with solid lines) and low-severity (squares with dashed lines) fire regimes. The 47 species observed in both fire regimes (left), 29 species observed only in mixed-severity regime locations (upper right), and 19 species observed only in low-severity regime locations (lower right) are shown. Treatment effects describe the extent to which occupancy shifted towards or away from burned sites with treatment application

Full size image
Fig. 4
figure 4

Statistically supported occupancy parameter estimates (posterior median) and 90% BCIs describing treatment effects (\({\hat{\beta}}_{CBI\times PER}\)) and post-treatment CBI relationships (\({\hat{\beta}}_{CBI}+{\hat{\beta}}_{CBI\times PER}\)) for 25 species observed at locations with historically mixed-severity fire regimes. Estimates from a primary model (A, B) are compared with those from supplemental models that included data from additional years and a Markovian persistence effect (C) or separated effects by post-treatment year (D, E). Treatment effects describe the extent to which occupancy shifted towards or away from burned sites with treatment application, whereas CBI (composite burn index) relationships quantify the post-treatment correlation only

Full size image
Fig. 5
figure 5

Statistically supported occupancy parameter estimates (posterior median) and 90% BCIs describing treatment effects (\({\hat{\beta}}_{CBI\times PER}\)) and CBI relationships (\({\hat{\beta}}_{CBI}+{\hat{\beta}}_{CBI\times PER}\)) for 17 species observed at locations with historically low-severity fire regimes. Estimates from our main model (A, B) are compared with those from a supplemental model that separated effects by post-treatment year (C, D). Treatment effects describe the extent to which occupancy shifted towards or away from burned sites with treatment application, whereas CBI (composite burn index) relationships quantify the post-treatment correlation only

Full size image

Evidence for treatment effects varied with time since treatment and fire regime. We found the most evidence for positive treatment effects in the mixed-severity fire regime (e.g., for American robin, western bluebird, and hairy woodpecker; Figs. 3 and 4). For some species, treatment effects and CBI relationships were not unequivocally supported in every analysis (i.e., 90% BCIs sometimes included zero) but were nevertheless consistent in direction (e.g., black-backed woodpecker and ruby-crowned kinglet; Fig. 4).

Positive treatment effects were also statistically supported for several species in the mixed-severity regime when data from additional years were considered (black-backed woodpecker, American three-toed woodpecker, brown creeper, western wood-pewee, house wren, dusky flycatcher, and gray flycatcher; Fig. 4). The yearly effect model showed treatment effects primarily arose in the second year following treatment (Fig. 4D, E). For dusky and gray flycatchers, occupancy changes became apparent only in the second year after treatment, suggesting lagged treatment effects.

For 21 species exhibiting 24 CBI relationships, we never found statistically supported treatment effects (Figs. 4 and 5). For some of these species, estimated treatment effects were nevertheless consistent in direction with CBI relationships and with predictions for aspects of their life histories (e.g., pine siskin, pygmy nuthatch, orange-crowned warbler; Table 1). Several species exhibited positive CBI relationships in the low-severity fire regime, but these relationships were not clearly reflected as a treatment effect (e.g., hairy woodpecker and western bluebird; Figs. 5 and 6). Although consistent with species life histories, some supported CBI relationships followed very low-magnitude treatment effects, supporting relatively weak inference (e.g., Nashville warbler, yellow-rumped warbler, warbling vireo, and Townsend’s warbler in the mixed-severity regime; white-breasted nuthatch, Northern flicker, and western wood-pewee in the low-severity regime; Mountain Chickadee in both regimes). Other species exhibited CBI relationships that were not accompanied by notable treatment effects (e.g., black-throated gray warbler, western tanager, green-tailed towhee) or were not necessarily consistent with their life histories (e.g., ash-throated flycatcher and Townsend’s solitaire), suggesting they were possibly spurious.

Fig. 6
figure 6

Predicted occupancy with burn severity (CBI) for example species showing treatment responses statistically supported in historically mixed-severity regimes but not supported in low-severity regimes. Relationships with CBI were estimated before (gray) and after (black) treatment in mixed-severity regimes (left) and low-severity regimes (right), and treatment responses are inferred from the change in slope between the two. Intercept terms for calculating model predictions were averaged (mean) across locations within each regime. Full species names are listed in Additional file 1: Appendix E

Full size image

Community-level patterns

Community-wide patterns and differences between regimes were also apparent but limited. Treatment effects were generally stronger in magnitude (i.e., deviated further from zero) in the mixed-severity regime, where effects were more positive than negative (Figs. 3, 4, and 5). Despite the apparent difference between regimes, we found no statistically supported difference in treatment effect between regimes for any one species (BCIs for βDiff all overlapped zero; Eq. 4). In burn severity relationships, however, we found differences between regimes for two species (gray flycatcher and spotted towhee). Although treatment effects were more positive in the mixed-severity regime (see above), treatment did not have a notable effect on species richness (Fig. 7). Instead, species richness varied much more among locations within and between fire regimes than with burn severity or treatment application. Treatment effects were generally stronger in year 2 compared to year 1 following treatment in both fire regimes (Figs. 4 and 5).

Fig. 7
figure 7

Species richness estimates and 90% BCIs for surveyed points along burn severity (CBI) gradients estimated before (left column) and after (right column) prescribed fire treatments. Locations appearing in the top row historically experienced mixed-severity fire regimes (Okanogan-Wenatchee [OKWA], Payette [PAID], and San Juan [SJCO] National Forests), whereas locations in the bottom panels experienced low-severity regimes (Apache-Sitgreaves [ASAZ], Coconino [COAZ], Gila [GINM], and Kaibab [KAAZ] National Forests). Best-fit lines show trends in posterior median estimates. The change in slope of trend lines from left to right indicates treatment effect on estimated species richness at surveyed point count stations

Full size image

Discussion

Changes in avian occupancy related to prescribed fire treatments and relationships with burn severity generally supported our predictions. Accordingly, our findings followed the conclusions of previous prescribed fire studies (e.g., Hurteau et al. 2008; Dickson et al. 2009; Russell et al. 2009; Bagne and Purcell 2011; Fontaine and Kennedy 2012; White et al. 2016) and supported our predictions that prescribed fires benefited cavity-nesting, bark and ground insectivores. We found increases in occupancy related to post-fire treatments or burn severity for many cavity-nesting birds, including bark insectivores (American three-toed, hairy, black-backed woodpeckers, white-breasted nuthatch, brown creeper, and pygmy nuthatch) and ground insectivores (American robin, western bluebird, house wren). Increases in available snags created soon after fire likely increased nesting substrate and food resources (bark beetle larvae [Scolytidae]) for cavity-nesting, bark insectivores (Saab and Powell 2005; Saab et al. 2006; Russell et al. 2009). Notably, we recorded positive changes in occupancy related to low-severity prescribed fire for species known to favor higher severity wildfires (e.g., black-backed woodpecker, Saab et al. 2007), suggesting that fire management can be an effective tool to create habitat for some fire specialist species at least for a few years.

Further reductions in post-fire ground cover may have increased foraging opportunities across both fire regimes for open-ground feeding species (American robin, western bluebird, house wren), including Northern flicker (Bagne and Purcell 2011; White et al. 2016). Although we expected occupancy increases by several aerial foragers in relation to fire (Bagne and Purcell 2011), we detected few positive trends (western wood-pewee [both regimes], dusky and gray flycatchers [mixed fire regime]), suggesting that effects of low-severity fire treatments are variable for aerial insectivores. Habitat changes resulting from higher severity burns are likely more beneficial to aerial insectivores (Kotliar et al. 2002; Smucker et al. 2005; Russell et al. 2006; Kotliar et al. 2007; Latif et al. 2016b). Compared to low-severity fire, moderate- to high-severity burns potentially create more openings in the forest canopy, allowing efficiency in flycatching for insects, and release more soil nutrients that allow for shrub growth and increased abundance of associated arthropods (cf. Certini 2005), followed by increases of insect prey availability for aerial insectivores.

Negative relationships with burn severity followed our predictions for canopy foliage gleaners (mixed severity: Cassin’s and warbling vireos; ruby-crowned kinglet, Nashville, yellow-rumped and Townsend’s warblers, Mountain Chickadee; low severity: red-faced warbler and Mountain Chickadee) and were consistent with other prescribed fire studies (Bagne and Purcell 2011; Fontaine and Kennedy 2012; White et al. 2016). These species forage in live trees, contributing to their negative relationships with fires of various severities that can damage or kill portions of live trees. Live tree densities were reduced by 37% (Additional file 1: Appendix B Table 3) to 45% across our study locations (Saab et al. 2006), likely promoting the negative relationships between foliage insectivores and burn severity.

Unexpectedly, prescribed fire treatments provoked stronger responses, including lagged effects, at mixed-severity locations. Sampling effort could have played a role in the observed differences but we restricted the primary analysis to 1–2 years before and after prescribed fire in both regimes, thus standardizing the number of years of data collection. Differences in timing of burns (spring at mixed-severity locations vs. primarily fall at low-severity locations) could also contribute, although the timing of prescribed burns intentionally and consistently avoided historical wildfire seasons across both fire regimes. Spring burning at mixed-severity locations could have interfered with breeding the first year, although we did not find strong evidence for immediate interference of fire on breeding bird behavior. Rather, we found more lagged responses in the years subsequent to burning applications within both regimes.

Perhaps bird populations occurring in historically low-severity locations had fewer occupancy changes because fire is typically more frequent and regular. By comparison, fire events are relatively rare historically in mixed-severity locations, potentially eliciting more responses to an infrequent opportunity, even by species that are strongly associated with recently burned forests by wildfire (e.g., black-backed and American three-toed woodpeckers). This pattern suggests that fire management activities intended to reduce fuels and lower the risk of high-severity wildfire can be effective in creating habitat for some fire specialists at least in the short term. Historical conditions are especially meaningful when they encompass evolutionary relationships such as the role fire regimes play in structuring bird communities and species distributions (cf. Hutto et al. 2008).

We found no definitive evidence for either short-term prescribed fire treatment effects or burn severity relationships for the majority of bird species (61 of 95 species). For many species that were rarely detected, lack of evidence likely reflects low statistical power. Additionally, a lack of rapid responses to habitat changes after prescribed fire may be related to time lags created by site tenacity of breeding birds (Wiens and Rotenberry 1985), as indicated by our data for lagged positive responses by dusky and gray flycatchers at mixed-severity locations and lagged negative responses by red-faced warbler and lark sparrow at low-severity locations. Longer-term data may be necessary to quantify the timeframe of negative and positive impacts of prescribed fire on foliage gleaners and bark insectivores, respectively.

Our findings that species richness was affected little by prescribed fire treatments concur with previous literature (George and Zack 2008; Hurteau et al. 2008; Russell et al. 2009). Post-fire bird communities may contain the same number of species as the pre-fire community, but nevertheless contain different species, including those not prevalent outside of recently disturbed forests, such as black-backed and American three-toed woodpeckers. Assessing both individual species responses to management practices and the overall contribution of a species to biodiversity on a larger regional scale (such as a forested area containing burned and unburned portions) is important for addressing specific management goals. Additionally, treatments on a study unit may affect shifts in species distributions only observable with a BACI study design that clarifies species responses. For example, consistent with their life history, house wren shifted their distribution toward burned/treated units, although this shift was not strong enough to completely negate or reverse their greater prevalence at unburned compared to burned units prior to treatment.

Our study design was unprecedented by the combination of large spatial scale, replication, multiple years, assessment of burn severity, and experimental plot sizes (179–402 ha). By designing our study to estimate changes in avian species occupancy and species richness at appropriate spatial scales, our study supports inference more relevant to landbirds than previous continent-wide research (e.g., McIver et al. 2013).

We evaluated occupancy changes for individual species and for trends in species grouped by life history traits. Although limitations apply to evaluating species grouped by traits (Fontaine and Kennedy 2012), we found evidence of changes in occupancy for many species that matched our life history trait predictions (e.g., patterns of positive changes for cavity-nesting, bark insectivores and negative changes for open-nesting, foliage insectivores). Most occupancy changes occurred at mixed fire regime locations. Some species exhibited changes with treatment overall, but lagged effects were more pronounced 2 years post-treatment. This was particularly evident in the mixed-severity fire regime locations, where the understory vegetation had greater potential to influence changes in occupancy. Evaluating post-treatment occupancy relationships with burn severity (i.e., disregarding pre-treatment distributions) revealed additional species that at least maintained distributions relative to treatment that were consistent with their life histories.

Management implications

Our results revealed primarily short-term benefits and limited negative effects of prescribed fire practices to the avifauna of seasonally dry forests across the interior western USA. Our data suggest that the longer-term potential benefits of prescribed fire for ecosystem resilience likely outweigh any potential near-term costs to avian diversity.

Unprecedented, extreme fire behavior resulting in rapid and extensive tree mortality is expected to be more common under changing climate conditions (Fettig et al. 2013), raising concerns by ecologists worldwide (Pickrell and Pennisi 2020). Prescribed fire and other fuel reduction treatments potentially reduce the risk of future severe wildfires, decrease tree mortality, and increase forest resilience to climate change (Stephens et al. 2018). Prescribed fire treatments are also potentially useful for creating near-term habitats for fire specialists that are more frequently found after wildfires. Fire suppression in the long term does not benefit avian species or biodiversity overall (Bagne and Purcell 2011). For example, broadscale contiguous tree mortality can result in homogeneity produced by fire suppression, reducing the fine-scale heterogeneity of forest conditions that contribute to resilience and biodiversity (Stephens et al. 2018). Prescribed fire and forest thinning could enhance adaptation to climate-induced stress if resources are focused on creating spatially and temporally variable patterns in seasonally dry forests that are aligned with local fire patterns (cf. North et al. 2009), accordingly supporting local avian communities.

Dry forested landscapes of the interior western USA support a diverse avifauna, including species of concern that rely on recent disturbance (e.g., black-backed woodpecker), old/mature forest specialists (e.g., red-faced warbler), and species that require multiple seral stages (e.g., white-headed woodpecker; Latif et al. 2015). Our results indicate that fire management practices promoting a mosaic of habitat conditions will best support the full suite of avian species native to seasonally dry conifer forests of western North America (Saab et al. 2005; Veech and Crist 2007; Fontaine et al. 2009; Fontaine and Kennedy 2012).

Conclusions

We implemented a regional interior western US study to estimate small landbird responses to prescribed fire treatments at spatial scales relevant to their ecology. We examined differences in treatment effects applied within historically mixed- vs. low-severity fire regimes. Bird populations in historically low-severity locations were relatively unresponsive to prescribed fire possibly because fire there is typically more frequent, regular, and expected by the avifauna. By comparison, fire events were relatively infrequent historically in mixed-severity locations, potentially eliciting more responses to an occasional opportunity, even by species that are strongly associated with recently burned forests by wildfire. Fire treatments intended to reduce fuels and lower the risk of high-severity wildfire potentially can be effective in creating habitat for some fire specialists over the short term.