Contents lists available at ScienceDirect

Biological Conservation

journal homepage: www.elsevier.com/locate/biocon

Using metapopulation models to assess species conservation–ecosystemrestoration trade-offs

Connor M. Wooda,⁎, Sheila A. Whitmorea, R.J. Gutiérreza, Sarah C. Sawyerb, John J. Keanec,M. Zachariah Peerya

a Department of Forest and Wildlife Ecology, University of Wisconsin-Madison, 1630 Linden Drive, Madison, WI, USAbU.S. Forest Service, Pacific Southwest Region, Vallejo, CA, USAcU.S. Forest Service, Pacific Southwest Research Station, Davis, CA, USA

A R T I C L E I N F O

Keywords:Metapopulation modelPatch occupancyRecolonizationSpotted owlStrix occidentalis

A B S T R A C T

Ecological restoration is needed to counter global-scale ecosystem degradation, but can conflict with endangeredspecies conservation when restoration impacts habitat quality. In such cases, prioritizing long-vacant patches forrestoration is an intuitively appealing strategy for minimizing the effects on endangered species. Metapopulationmodels grounded in empirical data potentially provide a rigorous framework for developing theoretical “patchvacancy thresholds” (i.e., duration of vacancy required before implementing restoration) and assessing theimplications of such criteria for restoration objectives. We develop such a model for spotted owls (Strix occi-dentalis), which embody the species–ecosystem dilemma given their preference for closed-canopy forests that arealso susceptible to severe fire and drought and hence the center of debates about forest restoration intended toreduce fire and drought risk. We leveraged a> 20-year territory occupancy dataset to parameterize a StochasticPatch Occupancy Model (SPOM) to assess relative risk to a metapopulation of owls in California under alter-native conservation guidelines, including a range of vacancy thresholds. Territories with greater amounts of owlhabitat were more likely to be recolonized and less likely to go extinct. Importantly, the probability of a vacantowl territory becoming recolonized declined as length of vacancy increased; territories vacant for 1 and 10 yearshad annual recolonization probabilities of 0.34 and 0.06, respectively. Based on our SPOM, projected territoryoccupancy rates declined as the vacancy threshold decreased and as habitat within territories was impacted byrestoration. However, more liberal territory vacancy thresholds were projected to increase the proportion ofterritories (and thus landscape) that could be restored and that restored conditions could be maintained withrepeated treatments. Reintroducing natural disturbance regimes, which eliminated the need for repeatedtreatments, was projected to reduce risk to owls, particularly with relaxed vacancy thresholds. We provide asimple, yet novel, metapopulation framework for quantifying how alternative conservation guidelines mightimpact owl occupancy and influence forest restoration guidelines. Similar analyses could facilitate restorationefforts in other systems by more explicitly quantifying tradeoffs between species–ecosystem objectives.

1. Introduction

Human activities have resulted in both biodiversity loss and eco-system degradation worldwide (Barnosky et al., 2011; Vitousek et al.,1997). Consequently, endangered species conservation and ecosystemrestoration have emerged as two important paradigms in natural re-source management (Groom et al., 2005). Considerable synergy existsbetween the two paradigms because habitat restoration has improvedthe status of many threatened or endangered species (Hogg et al., 2013;Lawson et al., 2014; Rannap et al., 2009; Rinkevich, 2005; Webb andShine, 2000). However, the trade-offs between species conservation and

ecosystem restoration (hereafter species–ecosystem) objectives are in-creasing in this era of global change (Fraser et al., 2017; Peery et al.,2017). Even when endangered species are expected to benefit fromrestored ecosystems, restoration practices can produce short-term coststo habitat quality that increases extinction risk before predicted benefitsof restoration accrue (Warchola et al., 2018). Therefore, uncertaintyexists between the potential short-term costs and long-term benefits ofrestoration for many species such as prairie and forest birds (Powell,2008; Wilson et al., 1995), grassland insects (Thomas et al., 1986;Warchola et al., 2018), forest reptiles and marsupials (Cunninghamet al., 2007), and freshwater fish (Lintermans, 2000). In such cases,

https://doi.org/10.1016/j.biocon.2018.05.001Received 18 October 2017; Received in revised form 27 April 2018; Accepted 4 May 2018

⁎ Corresponding author at: A223 Russell Labs, 1630 Linden Drive, Madison, WI 53706, USA.E-mail address: cwood9@wisc.edu (C.M. Wood).

Biological Conservation 224 (2018) 248–257

0006-3207/ © 2018 Elsevier Ltd. All rights reserved.

T

developing effective ecosystem restoration practices that minimizeshort-term impacts to endangered species is a key to meeting bothobjectives. Developing rigorous analytical frameworks that quantifyspecies–ecosystem tradeoffs in the currencies of population viabilityand restoration accomplishments under alternative management stra-tegies would facilitate such efforts (Fraser et al., 2017).

Metapopulation theory and models have provided conceptual andanalytical frameworks for conserving endangered species in fragmentedlandscapes for over two decades (Hanski, 1998; Levins, 1969;McCullough, 1996) and, we suggest, hold considerable promise forevaluating trade-offs between species and ecosystem conservation ob-jectives. While prioritizing vacant patches for restoration is an in-tuitively appealing approach for minimizing effects on endangeredspecies, metapopulation theory predicts that reductions in the quality ofunoccupied habitat patches resulting from restoration practices reducespopulation viability (Hanski, 1998). That is, prioritizing patches forrestoration based purely on occupancy status is challenging becausepatches in a metapopulation regularly switch between occupied andunoccupied states. Nevertheless, when targeting long-vacant patchesfor restoration carries relatively low risk, restoration may lead to me-tapopulation growth in the long-term (Lawson et al., 2014). A keyquestion is: “how many years of vacancy are necessary before restora-tion can occur without reducing metapopulation viability?” The ques-tion “how do such thresholds influence the ability to meet restorationobjectives?” is equally important given that requiring long vacancyperiods before implementing restoration actions limits the fraction of alandscape that can be restored. For a vacancy threshold to be in-formative, a negative relationship must exist between the duration ofpatch vacancy and the probability that a vacant patch will be re-colonized, otherwise the impacts of restoration to the overall metapo-pulation are expected to be independent of the time they are vacant.

The spotted owl (Strix occidentalis) is an ideal species for usingmetapopulation principles (particularly vacancy thresholds) to assesstradeoffs between conservation efforts oriented towards short- andmedium-term objectives for focal species and those designed for longer-term ecosystem restoration objectives. Spotted owls' territorial behaviorand strong site fidelity result in spatially structured populations whosedynamics are consistent with those of a metapopulation, where in-dividual breeding pairs that occupy the same territories over a longtime are analogous to a network of interacting local populations(Gutiérrez and Harrison, 1996). Moreover, populations of this specieshave been intensively monitored yielding multi-generational territoryoccupancy histories (Dugger et al., 2015; Franklin et al., 2000; Tempelet al., 2016, 2014). Most importantly, spotted owls reside at the epi-center of regional and national forest management debates that hingeon the tradeoffs between the retention of critical spotted owl habitatand forest restoration treatments. While spotted owl occupancy ratestend to be higher in territories containing greater amounts of theclosed-canopy forest (Tempel et al., 2016, 2014), reducing tree den-sities with fuels reduction treatments in such stands are consideredessential for reducing large, severe fires and drought-related tree mor-tality (Collins et al., 2010; Dow et al., 2016; Hagmann et al., 2017).Developing forest restoration practices that minimize the potentialshort-term costs of reducing habitat elements important to spotted owlswhile maximizing the flexibility to achieve forest resilience is im-perative for achieving both ecosystem restoration and species con-servation objectives (Peery et al., 2017). Moreover, developing criteriafor selecting spotted owl territories (i.e., habitat patches) for ecosystemrestoration without reducing viability of owl populations will facilitatethis process.

We developed a Stochastic Patch Occupancy Model (SPOM)(Caswell and Etter, 1993; Gyllenberg and Silvestrov, 1994) to explorethe potential impacts of treating unoccupied patches under alternative,hypothetical forest restoration strategies on the viability of a spottedowl “metapopulation.” We evaluated forest restoration strategies thatvaried according to vacancy thresholds prior to restoration and the

amount of owl habitat within territories treated. Our specific objectiveswere to: (1) leverage a 22-year territory occupancy dataset to assess theeffects of habitat conditions and duration of territory vacancy on ter-ritory extinction and colonization rates, (2) use these statistical re-lationships to parameterize and implement SPOMs under differentforest restoration strategies, and (3) explore trade-offs between pro-jected changes in territory occupancy as a function of treatment fre-quency and number of territories treated under the different restorationstrategies. We predicted that forest restoration involving shorter va-cancy thresholds for treatments would result in greater declines inprojected occupancy but that shorter vacancy thresholds would in-crease the frequency that treatments could be implemented within owlterritories. This represented a simple yet novel metapopulation frame-work for explicitly quantifying trade-offs between the implementationof ecosystem restoration and predicted impacts to population viabilitythat also would be applicable to development of species conservationguidelines in a range of ecosystems.

2. Methods

2.1. Territory occupancy surveys

We modeled occupancy histories of 64 spotted owl territories sur-veyed as part of a long-term demographic study area in the centralSierra Nevada, California (Fig. 1). We used annual territory occupancyhistories from 1993 to 2014 based on surveys that followed standar-dized protocols (Forsman, 1983; Tempel et al., 2014). Briefly, we usedvocal lures at established call stations and walking routes covering theentire area. We attempted to capture and band all located owls withunique color-coded combinations to identify owls at individual terri-tories. We considered territories occupied when at least one owl wasdetected during daylight or twilight hours within or near (< 400m) theknown core area, and reduced false positives as recommended byBerigan et al. (2018). We excluded nocturnal detections from uni-dentified individuals because they may not be territory holders.

2.2. Modeling patterns in territory extinction and colonization

We included territory occupancy histories in the analysis beginningin the year a territory was first determined to be occupied and that alsohad uninterrupted survey effort through 2014. These two criteria en-sured that any recolonization events would occur with a known numberof years preceding vacancy. We used the same definitions for extinctionand (re)colonization as MacKenzie et al. (2003). We did not use aformal multi-season occupancy modeling approach because detectionprobabilities for bimonthly survey periods have been estimated to be0.68 (Tempel et al., 2016), which, over the four-month season, yieldsan overall detection probability of 0.99. Therefore, we assumed perfectdetection and treated the annual survey results as true presence/ab-sence data. We used binomial logistic regression to model associationsbetween extinction and, separately, colonization and predictor vari-ables (described below), which is mathematically equivalent to theprocess MacKenzie et al. (2003) used to test the influence of covariateson those same probabilities. Our process of estimating extinction andrecolonization rates was therefore effectively equivalent to that used toobtain detection-corrected estimates of those rates.

We treated the number of years of consecutive preceding vacancyand the amount of owl habitat within 400-ha circular areas (i.e.,~territory size; Tempel et al., 2016) as predictor variables in the re-colonization logistic regression. Our habitat variable was the propor-tion of a territory comprised of forests having both ≥70% canopy coverand trees with a quadratic mean diameter-at-breast height(QMD)≥ 61 cm (hereafter referred to as “owl habitat”). We extractedhabitat data from gradient nearest-neighbor (GNN) structure maps,which use Landsat imagery informed by forest measurements taken atFederal Inventory and Analysis (FIA) plots to give 30m coverage of the

C.M. Wood et al. Biological Conservation 224 (2018) 248–257

249

area (https://lemma.forestry.oregonstate.edu).We considered four competing models for recolonization prob-

ability: (1) intercept only, (2) years of vacancy, (3) habitat, and (4)years of vacancy and habitat. We modeled extinction probability as afunction of habitat (which we log-transformed to improve model fit)and compared this model to an intercept-only structure. We evaluatedmodel support with Akaike information criterion with a correction forsmall sample size (AICc), and averaged parameters from competitivemodels (ΔAICc < 2) when appropriate (Burnham and Anderson,2010). We conducted these analyses using R (R Core DevelopmentTeam, 2014), and packages MuMIn (Bartoń, 2015), mapplots(Gerritsen, 2013), and popbio (Stubben et al., 2016).

2.3. Modeling the effects of alternative restoration scenarios

We developed forest restoration scenarios that varied by (1) numberof years of observed territory vacancy required before a hypotheticalrestoration treatment was implemented (i.e., vacancy threshold); (2)amount of owl habitat retained within territories following a treatment;and (3) the frequency of restoration treatments. We considered 11territory vacancy thresholds ranging from 0 to 10 years. Territorieseligible for treatment based on observed vacancy history but containingless habitat than the treatment level were not treated. We assumed thatmanagers had no prior knowledge of territory occupancy, so observedvacancy was calculated as the years of vacancy beginning in year one ofthe simulation.

To represent different effects fuel reduction treatments might haveon owl occupancy, we simulated treatments in which owl habitat wasreduced to 4 ha (“high-impact”), 12 ha (“low-impact”), or not affected(“no impact”); the range of habitat in a territory was 2.2–91.9 ha(median=25.5 ha). In practice, fuels reduction treatments may or maynot occur in owl habitat as defined in this study. Rather, multiple fuel

reduction strategies (e.g., variable-diameter-retention and “thinningfrom below”) could be implemented in multiple forest types with dif-ferent effects on forest structure, owl habitat quality, and occupancyrates, though it was beyond the scope of this paper to explore the po-tential impacts of all possible management strategies. Instead, as de-scribed above, we treated “owl habitat” as a proxy for the overallquality of an owl territory because it is a strong predictor of occupancystatus in our study population (Jones et al., 2018). Changing the area ofthis forest type allowed us to explore a range of hypothetical treatmenteffects (from no- to high-impact) that also produced realistic changes interritory occupancy parameters. In our population projections, we as-sumed that owl habitat in untreated territories increased at a rate of0.32 ha per year, the approximate growth rate we observed in the GNNdata from 1993 to 2012. Research conducted adjacent to our study areasuggests that canopy cover and, to a lesser extent, basal area recoveredto pre-fuel reduction treatment levels after 30 years (Collins et al.,2011). We set the post-treatment growth rate at 1.5 times the untreatedgrowth rate (0.48 ha/year), such that a territory with a median amountof habitat undergoing a 56% reduction (27.4 to 12 ha) would recover in30 years, to account for competitive release of remaining trees fol-lowing treatment. Thirty years after treatment, the increase in owlhabitat reverted to the original rate.

We considered two possible scenarios for the timing of additionaltreatments following the implementation of the initial treatment. First,territories were eligible for treatments after 30 years had elapsed sincethe previous treatment (i.e., the effective lifespan of fuel treatments;North et al., 2012). This scenario represented a case where desireddisturbance regimes were not achieved and periodic treatments wereneeded to reduce the likelihood of large high-severity fires and hightree-mortality events. The second scenario only allowed for one treat-ment per territory over the course of the simulation. This scenario torepresented successful restoration, where management was no longer

Fig. 1. The Eldorado demographic study area, with territory symbol size proportional to the number of years occupied (1993–2014) and color scaled to the numberof times the territory was recolonized.

C.M. Wood et al. Biological Conservation 224 (2018) 248–257

250

needed to maintain desired forest conditions (i.e., natural disturbanceregimes returned). For simplicity, we only considered single-entriesinvolving reductions in owl habitat to 12 ha. Thus, we developed fourhypothetical scenarios for each of the 11 vacancy thresholds: (1) owlhabitat reduced to 4 ha and repeated treatments allowed (“high-im-pact”); (2) owl habitat reduced to 12 ha and repeated treatments al-lowed (“low-impact”); (3) owl habitat not reduced and repeatedtreatments allowed (“no-impact”); and (4) owl habitat reduced to 12 hain a single treatment (“true restoration”).

To project owl occupancy at the 64 territories forward in time underdifferent restoration strategies, we use the statistical relationships be-tween colonization/extinction and their supported covariates identifiedby the logistic regression models to parameterize a SPOM. Future oc-cupancy was projected forward 75 years by simulating annual coloni-zation and extinction events according to the following framework:territories became colonized (or went extinct) if a randomly generatednumber (Ys,t or Zs,t) drawn from a random uniform distributionbounded by 0 and 1 was smaller than the expected colonization (orextinction) probability based on the values of covariates in the modelfor each year in the simulation. For each individual territory, the si-mulation occurred as follows:

⎧

⎨

⎪⎪⎪⎪⎪

⎩

⎪⎪⎪⎪⎪

⎧⎨⎩

> = + ∑

< = + ∑

⎧⎨⎩

< = + ∑

> = + ∑

−

=

=

=

=

if ψ

0,

then ψ0, if Y ~U(0, 1) logit(γ ) β (β x )

1, if Y ~U(0, 1) logit(γ ) β (β x )

1,

then ψ0, if Z ~U(0, 1) logit(ε ) β (β x )

1, if Z ~U(0, 1) logit(ε ) β (β x )

s t

s ts t t i i i s t

s t t i i i s t

s ts t t j j j s t

s t t j j j s t

, 1

,, 0 1

n, ,

, 0 1n

, ,

,, 0 1

n, ,

, 0 1n

, ,

i

i

j

j

where ψt represented the simulated binary occupancy state for territorys in year t, β0i

and β0jrepresented the intercept for either colonization or

extinction, xi, t and xj, t (where i or j=1, 2, …, n) represented spatial ortemporal variables i (colonization) or j (extinction) for territory s inyear t, and βi and βj represented regression coefficients that scaledcolonization or extinction probabilities to xi, s, t and xj, s, t, respectively.This process was repeated recursively 75 years into the future for eachof the restoration scenarios (see below). We used the occupancy statusand true years of vacancy (for unoccupied territories) on our study areain 2014 to initialize occupancy projections with the SPOM. True va-cancy was the number of years of vacancy observed prior to 2014. Weused owl habitat values derived from 2012 GNN data as starting pointsfor the model, as these were the most recently available estimates,where the median amount of owl habitat in the 64 territories was27.4 ha (range: 2.2–91.9 ha).

We ran 1000 simulations of the SPOM at each treatment level(“high-impact”, “low-impact”, “true restoration”, and “no impact”) andeach vacancy threshold (0 through 10 years of vacancy before treat-ment) plus a no-treatment scenario (45 total scenarios). For each

scenario, we recorded (1) the mean proportion of occupied territories atyear 75, (2) the mean number of times territories were treated, and (3)the proportion of territories treated at least once. Thus, the first outputprovided a measure of the potential detrimental impacts of forestmanagement on owl occupancy, whereas the second and third outputsquantified the degree to which these management strategies might re-duce fuel loadings across the landscape and thus achieve forest re-storation objectives. The second output, the mean number of timesterritories were treated, provides a measure of the frequency at whichrestoration practices could be implemented and thus an indication ofhow consistently the landscape might remain in a desired state overtime. The third output, the proportion of territories treated, measuredthe spatial extent of restoration at a landscape scale, an importantmetric given that the cumulative area within spotted owl territoriesencompassed approximately half of our study landscape. We conductedthese analyses with Microsoft Excel and @Risk (Palisade Software,2016).

3. Results

3.1. Patterns in territory recolonization and extinction

The most supported logistic regression model of territory re-colonization (w=0.55) indicated that the probability of recolonizationdeclined with the number of years of preceding vacancy (Table 1;Fig. 2a). The probability of territory colonization was 0.34 after a singleyear of vacancy and declined to 0.06 after 10 years of vacancy. Con-siderable support existed for the second-ranked model (w=0.45;ΔAICc=0.43), which indicated that recolonization probability wasgreater in territories containing a relatively high proportion of owlhabitat (Table 1, Fig. 2b) and declined as a function of years of vacancy.These top two models outperformed the null and owl habitat-only re-colonization models substantially (ΔAICc=18.87 and 19.30, respec-tively). Although the owl habitat parameter was statistically unin-formative in the second-ranked model, it added an element ofecological realism and was supported by previous studies that havedetected a link between territory recolonization and owl habitat quality(Tempel et al., 2016, 2014). Therefore, we model averaged parameterestimates from the top two recolonization models for the purposes ofdeveloping the SPOM, which yielded:

= − + −logit(colonization) 0.54 1.5X 0.23X1 2

where X1 was ha of owl habitat in the territory and X2 is years of va-cancy.

The most supported logistic regression model for territory extinctionindicated that the probability of extinction decreased as the amount ofowl habitat increased (Table 1, Fig. 2c). The probability that occupiedterritories with a first- and third-quartile amount of habitat (14.7 and36.9 ha, respectively) went extinct was 0.13 and 0.10, respectively.While the null model was competitive (ΔAICc=1.74), the 85% con-fidence intervals (−0.62, −0.09) associated with the parameter esti-mate for owl habitat in the top model did not overlap zero (Arnold,2010). Consequently, we used the top-ranked logistic regression modelto model extinction dynamics in the SPOM, which was expressed as:

= − −logit(extinction) 3.03 0.35 log X1

where X1 was the ha of owl habitat in the territory.

3.2. Projected changes in territory occupancy

Projections of the SPOM indicated that, in the absence of forestrestoration treatments (“reference scenario”), spotted owl occupancyrates would decline from ψ=0.56 in 2014 to 0.40 (SD=0.06) over75 years. Hereafter, we refer to the ψ=0.40 at year 75 as the “re-ference level” to evaluate the impacts of different restoration scenarioson owl occupancy. When multiple treatments were allowed in a

Table 1Binomial models of territory recolonization and extinction. Vacancy representsthe number of consecutive years of territory vacancy preceding year t. Owlhabitat represents the proportion of a 400-ha territory containing forestscharacterized by both ≥70% canopy cover and trees≥ 61 cm quadratic meandiameter at breast height.

Terms AICc ΔAICc w K

Recolonization Vacancy 346.70 0.00 0.55 1Vacancy+ owl habitat 347.13 0.43 0.45 2Null 365.57 18.87 0.00 0Owl habitat 366.00 19.30 0.00 1

Extinction Log(owl habitat) 623.30 0.00 0.71 1Null 625.04 1.74 0.29 0

C.M. Wood et al. Biological Conservation 224 (2018) 248–257

251

territory over the projection period, ending occupancy did not declinebelow the reference level under the most conservative restoration sce-nario (no-impact treatment, vacancy threshold=10 years) (ψ=0.40,SD=0.06). Under this scenario, 58% of territories were expected to betreated and territories were expected to be treated an average of 1.1times over the projection period. Projected ending occupancy ratesdecreased as the vacancy threshold was relaxed (fewer years of vacancyrequired before treatment) under both the low- and high-impact sce-narios (Fig. 3). The expected number of treatments per territory(Fig. 3a) and the proportion of territories treated (Fig. 3b) both in-creased as the vacancy threshold was relaxed. In the most aggressiverestoration scenario (high-impact treatment, vacancythreshold=0 years), occupancy rates declined 0.18 below the re-ference level (ψ=0.22, SD=0.05). Under this scenario, every terri-tory was treated and territories were treated an average of three times(the maximum possible).

Expected declines in occupancy were relatively small when only onetreatment per territory was allowed and low-impact treatments wereimplemented (i.e., true restoration scenario; Fig. 4). When the vacancy

threshold was between 0 and 4 years, true restoration scenarios resultedin greater ending occupancy rates than multiple-entry scenarios; with athreshold of 5 or more years, there was essentially no difference inending occupancy presumably because the number of treatments in themultiple-entry scenario was generally low (Fig. 4). Under the true re-storation scenario, a 10-year vacancy threshold resulted in a 0.01 de-cline in occupancy below the reference level (ψ=0.39 SD=0.05),whereas a 0-year vacancy threshold resulted in a 0.06 decline in oc-cupancy (ψ=0.34 SD=0.06; Fig. 4). These scenarios resulted in 57%and 100% of territories being treated over the projection period(Fig. 4a), respectively (the mean number of treatments and the numberof treatments were identical in the true restoration scenarios becausethe number of treatments was capped at one). The proportion of ter-ritories treated was nonetheless almost identical between the true re-storation and multiple-entry scenarios when low-intensity treatmentswere implemented (Fig. 4b).

The mean number of treatments per territory and the proportion ofterritories treated responded differently to changes in vacancythreshold. The mean number of treatments per territory steadily

Fig. 2. Logistic regression showing the probability of recolonization as a function of the years of vacancy (a) and habitat (b), and extinction as a function of habitat(c). These were the top models of each process (ΔAICc < 2.00; Table 1). Owl habitat is the amount (ha) of a territory (400 ha) that had high canopy cover (≥70%)and large trees (QMD≥ 61 cm). The histograms along the top and bottom of each panel represent the number of territories that were recolonized (top, a and b),remained vacant (bottom, a and b), went extinct (top, c), or remained occupied (bottom, c). iVacancies > 10 years were combined for clarity.

C.M. Wood et al. Biological Conservation 224 (2018) 248–257

252

increased as the vacancy threshold was relaxed when multiple entrieswere allowed, reaching a maximum at a 0-year vacancy threshold(Fig. 3a). In contrast, all 95 to 100% of territories were expected to betreated with a vacancy threshold of ≤3 years, with impacts to popu-lation-level occupancy depending on how much habitat was modified atterritories (Figs. 3b, 4b); shorter vacancy thresholds resulted in furtherdeclines in owl occupancy without substantial increases in the pro-portion of territories that were treated.

4. Discussion

Spotted owl territories that were vacant for long periods of timewere less likely to be recolonized on an annual basis than territoriesthat were vacant for shorter time periods. The mechanism(s) behindthis pattern is uncertain, but long-vacant territories may not containadequate amounts of some habitat elements such as large trees (Jones

et al., 2018; North et al., 2017), or a population process such as an AlleeEffect where individuals might serve as cues to settling responses couldhinder recolonization (Seamans and Gutiérrez, 2006). Consequently,projections of our metapopulation model indicated that adopting astringent territory vacancy threshold before restoration was allowedhad a relatively small impact on projections of future occupancy ratesbut limited the frequency and spatial extent of treatments. Territorieswere often not vacant long enough to be treated when stringent va-cancy criteria were adopted. Such a strategy, then, is expected to con-strain forest restoration (Stephens and Moghaddas, 2005). Alter-natively, a liberal territory vacancy threshold allowed greaterfrequency and spatial extent of restoration treatments but was projectedto incur greater direct impacts to owl territory occupancy unless “no-impact” treatments were implemented. While these findings underscoreexisting conservation conflicts between forest managers and owl con-servationists (Collins et al., 2010; Dow et al., 2016; Tempel et al.,

Fig. 3. Simulated changes in occupancy (displayedrelative to no treatment) in the high-, low-, and no-impact scenarios as the vacancy threshold changes.Occupancy decreases and treatment frequency(average number of treatments per territory) increasesas the vacancy threshold decreases (a). Point size isproportional to treatment frequency; up to threetreatments were possible in the 75-year simulation.Occupancy decreases and the extent of treatmentacross the landscape (proportion of territories treatedat least once) increases as the vacancy threshold de-creases (b). Point fill is proportional to the extent of thelandscape (proportion of territories) treated.

C.M. Wood et al. Biological Conservation 224 (2018) 248–257

253

2014), our modeling framework provides explicit estimates of how al-ternative conservation guidelines (i.e., territory vacancy thresholds),given assumptions, might impact both future population status and theimplementation of forest restoration treatments over space and time.The quantification of species – ecosystem tradeoffs provides forestmanagers with a scientific basis for considering restoration strategiesthat would be most likely to meet both objectives or, alternatively, abasis for prioritizing objectives.

While the adoption of a liberal vacancy threshold was projected toincrease both the frequency and spatial extent of treatments, we ob-served a notable difference in the response of these two restorationmeasures. The projected frequency of treatments in spotted owl terri-tories steadily increased as the vacancy threshold was reduced from 10to 0 years (Fig. 3a), whereas the projected proportion of territoriestreated was effectively maximized (95% treated) with a vacancythreshold of 3 years (Fig. 3b). Yet, territory occupancy rates were

projected to continue declining with vacancy thresholds< 3 years,suggesting little gain in the spatial extent to which landscapes can berestored relative to stricter criteria. While the frequency at which ter-ritories can be treated is expected to increase below 3-year vacancythresholds, which potentially increases the ability to meet ecosystemrestoration objectives, such a strategy comes at a cost to projected oc-cupancy rates. Identifying points at which the benefits to one objective– restoration or species conservation – cease to accrue, while the coststo the other objective continue increasing (e.g., 3-year vacancythreshold) are important to bracketing the range of acceptable man-agement actions.

The effects of adopting a liberal vacancy threshold of ≤3 years onspotted owl occupancy rates could be reduced – potentially con-siderably so – by (i) developing and implementing restoration treat-ments with reduced impacts to owl habitat and (ii) the restoration ofnatural disturbance regimes after a single treatment. Notably, projected

Fig. 4. Simulated changes in occupancy (displayed re-lative to no treatment) in the true restoration and low-impact scenarios as the vacancy threshold changes.Treatments affected owl habitat equally in both sce-narios, but eligible territories were only treated once inthe true restoration scenarios, and up to three times inthe low-impact scenarios. Occupancy decreases andtreatment frequency (average number of treatments perterritory) increases as the vacancy threshold decreases;point size is proportional to treatment frequency (a).Occupancy decreases and the extent of treatment acrossthe landscape (proportion of territories treated at leastonce) increases as the vacancy threshold decreases (b).Point fill is proportional to the extent of the landscape(proportion of territories) subject to treatment.

C.M. Wood et al. Biological Conservation 224 (2018) 248–257

254

differences in the change in occupancy between no-, low-, and high-impact treatments were greatest when the vacancy threshold wasshortest. This finding highlights the importance of developing treat-ments that do not result in significant impacts to owl habitat, particu-larly the loss of large trees, and thus occupancy, if short vacancythresholds are adopted (Jones et al., 2018; North et al., 2017). Re-storation treatments in habitat dominated by medium and small trees,rather than the large-tree-dominated areas seemingly important toowls, are more likely to reflect a low- or no-impact scenario. Yet, todate, it has proven challenging to empirically ascertain the effects ofrestoration on spotted owls and their habitat for multiple reasons(Tempel et al., 2016, 2014). For example, treatments in the large-tree,closed canopy habitat are likely to decrease canopy cover while in-creasing average tree size, leading to mixed, rather than directional,change in the attributes that we used to define owl habitat. Clearly,additional empirical information on the effects of restoration treat-ments on spotted owls at the territory scale would provide opportu-nities to refine the modeling of potential species–ecosystem tradeoffs,as well as the design of restoration treatments that minimize negativeimpacts to this species. Moreover, the restoration of natural disturbanceregimes following initial treatments could eliminate the need for ad-ditional treatments that have potentially adverse effects on spotted owls(North et al., 2012). Our results support this hypothesis because the“true restoration” scenario we modeled was projected to lead to occu-pancy rates that were 0.062 and 0.044 greater than when multipletreatments were allowed using 0- and 1-year vacancy criteria, respec-tively. It also underscores the need to restore natural disturbance re-gimes under which owls evolved, rather than relying on repeated re-storation treatments (Gutiérrez et al., 2017).

As is the case with all population projection models, our metapo-pulation analysis involved several simplifying assumptions that couldimpact inferences regarding the effects of adopting a given vacancycriterion. First, our modeling approach did not allow us to quantifyeither: (i) the potential long-term benefits that ecosystem restorationcould provide to owls by reducing risk of severe fire; or (ii) the risk ofnot conducting restoration, namely that habitat loss resulting from se-vere fire could effectively remove territories from the metapopulation(Jones et al., 2016a,b). Either or both of these factors would incentivizeshorter vacancy thresholds, but incorporating them into our modelingwould have required linking specific restoration treatments to changesin fire behavior and vegetation change. Existing fire behavior modelscannot replicate megafires known to impact spotted owls or be simu-lated with enough replicates to link them to population modelingprojections (S. Stephens pers. comm.). As a consequence, inferences ofeffects on owls in our study were limited to comparing the potentialdirect impacts of different vacancy thresholds on owl territory occu-pancy among alternative forest restoration scenarios. Second, we as-sumed that factors responsible for relatively low annual recolonizationrates at long-vacant territories on our study area would remain un-changed following vacancies in our simulated model projections re-sulting from, for example, the implementation of treatments. Third,extended periods of vacancy at some territories in our study may havereflected the absence of potential recruits resulting from a decliningpopulation (Tempel et al., 2016; Tempel and Gutiérrez, 2013). If re-cruits are lacking, the recolonization of territories – and ultimatelypopulation-level occupancy – may be less sensitive to a given vacancythreshold than estimated. Indeed, a large-tree deficit resulting from theselective removal of such trees may have contributed to recent declinesin population-level occupancy on our study (Jones et al., 2018). Thissuggests that lost habitat elements at a landscape scale could contributeto long vacancy at some territories by driving population declines,while their absence at a local scale could make the recolonization ofparticular territories unlikely. These interrelated factors could con-tribute to a negative feedback loop where low-occupancy areas aremore likely to be treated and thus less likely to become occupied in thefuture. Fourth, we made several potentially important assumptions

about temporal changes in forest conditions including: (i) projectedrates of habitat regeneration reflected recently published rates; (ii) overthe 75-year projection those regeneration rates would remain constant;and (iii) the matrix outside the habitat patches had and would continueto have no value to the owls. However, the complex effects of climate-driven changes in temperature, precipitation, and fire on the quantity,quality, and distribution of owl habitat are uncertain. High-elevationterritories could become key refugia for spotted owls, and thus benefitfrom a more cautious management approach (Jones et al., 2016a,b).The value of the matrix could increase if woodrats, a key prey item,colonize higher elevations as the climate warms (Moritz et al., 2008).Finally, habitat conditions were the only exogenous condition in themodel that influenced site occupancy. Competition, particularly frombarred owls (S. varia), could lead to substantial declines in spotted owloccupancy independent of restoration activities (Dugger et al., 2015).Despite these caveats, our metapopulation approach provided a simpleanalytic framework with explicit assumptions that could facilitatemanagers' ability to quantify species–ecosystems tradeoffs associatedwith alternative conservation guidelines and restoration strategies (i.e.,Fig. 3) rather than relying on, for example, expert opinion.

Uncertainty in model projections resulting from the aforementionedassumptions could be reduced within an adaptive management contextby monitoring how the adoption of a particular vacancy criterion af-fects both the frequency and spatial extent of treatments and spottedowl site occupancy. If restoration targets are not achieved, the modelcould be refined using additional information on spotted owl re-colonization patterns and treatment impacts on site occupancy rates.Characterizing how restoration treatments affect spotted owl site oc-cupancy has proven remarkably challenging in part because a largesample size of territories needed to detect modest effect sizes (Popescuet al., 2012); however, a recently-developed bioregional-scale site oc-cupancy monitoring program (based on passive acoustic surveys) in theSierra Nevada may provide a novel and potentially statistically pow-erful means for quantifying the effects of treatments on this species(Authors' unpublished data). Moreover, as our understanding of thisecosystem grows, relaxing some of our modeling assumptions and in-corporating greater ecological realism will become increasingly pos-sible. In particular, emerging research provides opportunities forlinking treatments to fire occurrence, severity, and size at bioregionalscale (Keyser et al., 2017). Predictions of the spatial distribution ofsevere fire could then be linked to projected changes in spotted owl siteoccupancy using well-defined statistical relationships describing thenegative impacts of severe fire on owls (Jones et al., 2016a,b). Such anapproach would have the advantage of considering both the potentialdirect adverse effects of restoration treatments on spotted owls and thepotential benefits that treatments confer to owls by reducing severe firerisk. Regardless of the focal species, spatially extensive monitoringprograms can provide the foundation for (i) developing more nuancedmanagement criteria based on site conditions or classes, (ii) introducinga spatial dimension into subsequent models (e.g., Yackulic et al., 2012),and (iii) a more responsive adaptive management approach (e.g.,Martin et al., 2009).

4.1. Conclusions

The increase of large, severe fires and drought-related forest mor-tality events in dry forest ecosystems has had a range of undesirableecological and social consequences. While restoration practices thatreduce tree densities are generally believed necessary to restore theresilience of these ecosystems (Stephens et al., 2016), managers are alsoconfronted with the need to conserve sensitive species that depend onsome vegetation conditions believed to promote high fire risk. Ouranalysis indicates that prioritizing vacant spotted owl sites for forestrestoration treatments based on the duration of vacancy, with appro-priate assumptions and caveats about implementation of treatments(e.g., removing large trees reduces owl habitat quality), could help

C.M. Wood et al. Biological Conservation 224 (2018) 248–257

255

reconcile these two seemingly incompatible objectives and facilitateprogress in a long-standing conservation conflict. The restoration ofnatural disturbance regimes, for example, via managed and prescribedfire following initial treatments that eliminate the need for additionaltreatments, could be a key to meeting ecosystem restoration objectiveswithout increasing risk to sensitive species such as owls — particularlyif relatively short vacancy thresholds are adopted. A greater empiricalunderstanding of the duration and nature of treatment impacts on focalspecies would improve the defensibility of adopting a particular va-cancy threshold (Mutz et al., 2017).

The need to quantify ecosystem–species tradeoffs will intensify asecosystems become increasingly degraded and more species becomeimperiled (Cunningham et al., 2007; Lintermans, 2000; Powell, 2008;Thomas et al., 1986; Wilson et al., 1995). Moreover, with ecosystemsand their constituent species facing complex and interacting threats,management actions may conflict and be difficult to prioritize (Fraseret al., 2017), particularly when restoration induces short-term costs tospecies of conservation concern (Warchola et al., 2018). We expandedthe application of metapopulation biology to address the tension be-tween ecosystem and species conservation objectives by using patchoccupancy histories to develop empirical basis for prioritizing habitatpatches for restoration efforts. In doing so, we demonstrated that simplemetapopulation models can be employed to make tradeoffs betweenthese two objectives explicit and quantifiable. Thus, our approach canbe a valuable way to balance apparently competing objectives, parti-cularly when they potentially lead to dramatically different outcomes.However, we caution that modeling efforts such as ours should beviewed as hypotheses rather than definitive inferences or re-commendations because implementing widespread restoration activ-ities without proper evaluation (i.e., monitoring) can lead to a shiftingbaseline of deteriorating habitat conditions (Soga et al., 2018). There-fore, the value of this approach lies in its capacity to quantify thepossible outcomes of different management actions, make assumptionsexplicit, and provide testable predictions in the face of high un-certainty.

Authors' contributions

Authors RJG and MZP managed the long-term study and seques-tered funding for it. Authors MZP and JJK conceived the ideas andapproach; author SCS helped refine the approach; authors SAW, RJG,MZP, and CMW contributed to field data collection; author SAWmanaged data collection; and author CMW led the analyses and writingof the manuscript. All authors contributed to the analyses and writing.

Acknowledgements

We thank Doug Tempel, William Berigan, dozens of field techni-cians, and several anonymous reviewers. This work was supported bythe USDA Forest Service Region 5, USFS Pacific Southwest ResearchStation, US Fish and Wildlife Service, California Department of Fish andWildlife, and the University of Wisconsin-Madison.

References

Arnold, T.W., 2010. Uninformative parameters and model selection using Akaike'sInformation Criterion. J. Wildl. Manag. 74, 1175–1178.

Barnosky, A.D., Matzke, N., Tomiya, S., Wogan, G.O.U., Swartz, B., Quental, T.B.,Marshall, C., McGuire, J.L., Lindsey, E.L., Maguire, K.C., Mersey, B., Ferrer, E.A.,2011. Has the Earth's sixth mass extinction already arrived? Nature 471, 51–57.http://dx.doi.org/10.1038/nature09678.

Bartoń, K., 2015. Multi-model Inference. CRAN.Berigan, W.J., Jones, G.M., Whitmore, S.A., Gutiérrez, R.J., Peery, M.Z., 2018. Wide-

ranging Individuals Lead to False Positive Detections and Upwardly-biasedOccupancy Estimates. (in revision).

Burnham, K.P., Anderson, D.R., 2010. Model Selection and Multimodel Inference: APractical Information-Theoretic Approach. Springer, New York, NY.

Caswell, H., Etter, R.J., 1993. Ecological interactions in patchy environments: from patch-occupancy models to cellular automata. In: Patch Dynamics. Springer, pp. 93–109.

Collins, B.M., Stephens, S.L., Moghaddas, J.J., Battles, J., 2010. Challenges and ap-proaches in planning fuel treatments across fire-excluded forested landscapes. J. For.108, 24–31.

Collins, B.M., Stephens, S.L., Roller, G.B., Battles, J.J., 2011. Simulating fire and forestdynamics for a landscape fuel treatment project in the Sierra Nevada. For. Sci. 57,77–88.

R Core Development Team, 2014. R: A Language and Environment for StatisticalComputing. R Foundation for Statistical Computing, Vienna, Austria.

Cunningham, R.B., Lindenmayer, D.B., Crane, M., Michael, D., MacGregor, C., 2007.Reptile and arboreal marsupial response to replanted vegetation in agriculturallandscapes. Ecol. Appl. 17, 609–619. http://dx.doi.org/10.1890/05-1892.

Yackulic, Charles B., Reid, Janice, Davis, Raymond, Hines, James E., Nichols, James D.,Forsman, Eric, 2012. Neighborhood and habitat effects on vital rates: expansion ofthe Barred Owl in the Oregon Coast Ranges. Ecology 93, 1953–1966. http://dx.doi.org/10.1890/11-1709.1.

Dow, C.B., Collins, B.M., Stephens, S.L., 2016. Incorporating resource protection con-straints in an analysis of landscape fuel-treatment effectiveness in the northern SierraNevada, CA, USA. Environ. Manag. 57, 516–530. http://dx.doi.org/10.1007/s00267-015-0632-8.

Dugger, K.M., Forsman, E.D., Franklin, A.B., Davis, R.J., White, G.C., Schwarz, C.J.,Burnham, K.P., Nichols, J.D., Hines, J.E., Yackulic, C.B., Doherty, P.F., Bailey, L.,Clark, D.A., Ackers, S.H., Andrews, L.S., Augustine, B., Biswell, B.L., Blakesley, J.,Carlson, P.C., Clement, M.J., Diller, L.V., Glenn, E.M., Green, A., Gremel, S.A., Herter,D.R., Higley, J.M., Hobson, J., Horn, R.B., Huyvaert, K.P., McCafferty, C., McDonald,T., McDonnell, K., Olson, G.S., Reid, J.A., Rockweit, J., Ruiz, V., Saenz, J., Sovern,S.G., 2015. The effects of habitat, climate, and barred owls on long-term demographyof northern spotted owls. Condor 118, 57–116. http://dx.doi.org/10.1650/CONDOR-15-24.1.

Forsman, E.D., 1983. Methods and Materials, for Locating and Studying Spotted Owls.(General Technical Report No. PNW-162). USDA Forest Service Pacific NorthwestForest and Range Experiment Station.

Franklin, A.B., Anderson, D.R., Gutiérrez, R.J., Burnham, K.P., 2000. Climate, habitatquality, and fitness in northern spotted owl populations in northwestern California.Ecol. Monogr. 70, 539–590.

Fraser, H., Rumpff, L., Yen, J.D.L., Robinson, D., Wintle, B.A., 2017. Integrated models tosupport multiobjective ecological restoration decisions. Conserv. Biol. http://dx.doi.org/10.1111/cobi.12939.

Gerritsen, H., 2013. mapplot. CRAN.Groom, M.J., Meffe, G.K., Carroll, C.R., 2005. Principles of Conservation Biology, Third

edition. Sinauer Associates, Inc., Sunderland, Mass.Gutiérrez, R.J., Harrison, S., 1996. Applying metapopulation theory to spotted owl

management: a history and critique. In: McCullough, D.R. (Ed.), Metapopulations andWildlife Conservation. Island Press, Washington D.C., pp. 167–185.

Gutiérrez, R.J., Tempel, D.J., Peery, M.Z., 2017. The biology of the California spotted owl(General Technical Report No. PNW-GTR). In: The California Spotted Owl: CurrentState of Knowledge, a Scientific Assessment Led by Pacific Southwest ResearchStation. USDA, Forest Service, Pacific Southwest Research Station, Albany, CA.

Gyllenberg, M., Silvestrov, D.S., 1994. Quasi-stationary distributions of a stochastic me-tapopulation model. J. Math. Biol. 33, 35–70.

Hagmann, R.K., Johnson, D.L., Johnson, K.N., 2017. Historical and current forest con-ditions in the range of the Northern Spotted Owl in south central Oregon, USA. For.Ecol. Manag. 389, 374–385. http://dx.doi.org/10.1016/j.foreco.2016.12.029.

Hanski, I., 1998. Metapopulation dynamics. Nature 396, 41–49.Hogg, R., Coghlan Jr, S.M., Zydlewski, J., 2013. Anadromous sea lampreys recolonize a

Maine coastal river tributary after dam removal. Trans. Am. Fish. Soc. 142,1381–1394. http://dx.doi.org/10.1080/00028487.2013.811103.

Jones, Gavin M., Gutiérrez, R.J., Tempel, D.J., Whitmore, S.A., Berigan, W.J., Peery, M.Z.,2016a. Megafires: an emerging threat to old-forest species. Front. Ecol. Environ. 14,300–306. http://dx.doi.org/10.1002/fee.1298.

Jones, Gavin M., Gutiérrez, R.J., Tempel, D.J., Zuckerberg, B., Peery, M.Z., 2016b. Usingdynamic occupancy models to inform climate change adaptation strategies forCalifornia spotted owls. J. Appl. Ecol. 53, 895–905.

Jones, G.M., Keane, J.J., Gutiérrez, R.J., Peery, M.Z., 2018. Declining old-forest species asa legacy of large trees lost. Divers. Distrib. 24, 341–351. http://dx.doi.org/10.1111/ddi.12682.

Keyser, A., Westerling, A.L., Jones, G., Peery, M.Z., 2017. Simulating high spatial re-solution high severity burned area in Sierra Nevada forests for California spotted owlhabitat climate change risk assessment and management. In: AGU Fall MeetingAbstracts.

Lawson, C.R., Bennie, J.J., Thomas, C.D., Hodgson, J.A., Wilson, R.J., 2014. Activemanagement of protected areas enhances metapopulation expansion under climatechange. Conserv. Lett. 7, 111–118. http://dx.doi.org/10.1111/conl.12036.

Levins, R., 1969. Some demographic and genetic consequences of environmental het-erogeneity for biological control. Bull. Entomol. Soc. Am. 15, 237–240. http://dx.doi.org/10.1093/besa/15.3.237.

Lintermans, M., 2000. Recolonization by the mountain galaxias Galaxias olidus of amontane stream after the eradication of rainbow trout Oncorhynchus mykiss. Mar.Freshw. Res. 51, 799–804.

MacKenzie, D.I., Nichols, J.D., Hines, J.E., Knutson, M.G., Franklin, A.B., 2003.Estimating site occupancy, colonization, and local extinction when a species is de-tected imperfectly. Ecology 84, 2200–2207. http://dx.doi.org/10.1890/02-3090.

Martin, J., Runge, M.C., Nichols, J.D., Lubow, B.C., Kendall, W.L., 2009. Structured de-cision making as a conceptual framework to identify thresholds for conservation andmanagement. Ecol. Appl. 19, 1079–1090.

McCullough, D.R., 1996. Metapopulations and Wildlife Conservation. Island Press,Washington D.C.

C.M. Wood et al. Biological Conservation 224 (2018) 248–257

256

Moritz, C., Patton, J.L., Conroy, C.J., Parra, J.L., White, G.C., Beissinger, S.R., 2008.Impact of a century of climate change on small-mammal communities in YosemiteNational Park, USA. Science 322, 261–264. http://dx.doi.org/10.1126/science.1162547.

Mutz, J., Underwood, N., Inouye, B.D., 2017. Time since disturbance affects colonizationdynamics in a metapopulation. J. Anim. Ecol. http://dx.doi.org/10.1111/1365-2656.12689.

North, M., Collins, B.M., Stephens, S., 2012. Using fire to increase the scale, benefits, andfuture maintenance of fuels treatments. J. For. 110, 392–401.

North, M.P., Kane, J.T., Kane, V.R., Asner, G.P., Berigan, W., Churchill, D.J., Conway, S.,Gutiérrez, R.J., Jeronimo, S., Keane, J., Koltunov, A., Mark, T., Moskal, M., Munton,T., Peery, M.Z., Ramirez, C., Sollmann, R., White, A., Whitmore, S., 2017. Cover oftall trees best predicts California spotted owl habitat. For. Ecol. Manag. 405,166–178. http://dx.doi.org/10.1016/j.foreco.2017.09.019.

Peery, M.Z., Gutiérrez, R.J., Manley, P.N., Stine, P.A., North, M., 2017. Synthesis andinterpretation of California spotted owl research within the context of public forestmanagement (General Technical Report No. PNW-GTR). In: The California SpottedOwl: Current State of Knowledge, a Scientific Assessment Led by Pacific SouthwestResearch Station. USDA, Forest Service, Pacific Southwest Research Station,Albany, CA.

Popescu, V.D., De Valpine, P., Tempel, D., Peery, M.Z., 2012. Estimating populationimpacts via dynamic occupancy analysis of before–after control–impact studies. Ecol.Appl. 22, 1389–1404.

Powell, A.F.L.A., 2008. Responses of breeding birds in tallgrass prairie to fire and cattlegrazing. J. Field Ornithol. 79, 41–52. http://dx.doi.org/10.1111/j.1557-9263.2008.00144.x.

Rannap, R., Lõhmus, A., Briggs, L., 2009. Restoring ponds for amphibians: a success story.Hydrobiologia 634, 87–95. http://dx.doi.org/10.1007/s10750-009-9884-8.

Rinkevich, B., 2005. Conservation of coral reefs through active restoration measures:recent approaches and last decade progress. Environ. Sci. Technol. 39, 4333–4342.http://dx.doi.org/10.1021/es0482583.

Seamans, M.E., Gutiérrez, R.J., 2006. Spatial dispersion of spotted owl sites and the roleof conspecific attraction on settlement patterns. Ethol. Ecol. Evol. 18, 99–111. http://dx.doi.org/10.1080/08927014.2006.9522716.

Palisade Software, 2016. @RISK. Palisade Software, Ithaca, NY.Soga, M., Gaston, K.J., Kubo, T., 2018. Cross-generational decline in childhood experi-

ences of neighborhood flowering plants in Japan. Landsc. Urban Plan. 174, 55–62.

http://dx.doi.org/10.1016/j.landurbplan.2018.02.009.Stephens, S.L., Moghaddas, J.J., 2005. Experimental fuel treatment impacts on forest

structure, potential fire behavior, and predicted tree mortality in a California mixedconifer forest. For. Ecol. Manag. 215, 21–36. http://dx.doi.org/10.1016/j.foreco.2005.03.070.

Stephens, S.L., Miller, J.D., Collins, B.M., North, M.P., Keane, J.J., Roberts, S.L., 2016.Wildfire impacts on California spotted owl nesting habitat in the Sierra Nevada.Ecosphere 7, e01478. http://dx.doi.org/10.1002/ecs2.1478.

Stubben, C., Milligan, B., Nantel, P., 2016. popbio. CRAN.Tempel, D.J., Gutiérrez, R.J., 2013. Relation between occupancy and abundance for a

territorial species, the California spotted owl. Conserv. Biol. 27, 1087–1095. http://dx.doi.org/10.1111/cobi.12074.

Tempel, D.J., Gutiérrez, R.J., Whitmore, S.A., Reetz, M.J., Stoelting, R.E., Berigan, W.J.,Seamans, M.E., Peery, M.Z., 2014. Effects of forest management on California spottedowls: implications for reducing wildfire risk in fire-prone forests. Ecol. Appl. 24,2089–2106. http://dx.doi.org/10.1890/13-2192.1.

Tempel, D.J., Keane, J.J., Gutiérrez, R.J., Wolfe, J.D., Jones, G.M., Koltunov, A., Ramirez,C.M., Berigan, W.J., Gallagher, C.V., Munton, T.E., Shaklee, P.A., Whitmore, S.A.,Peery, M.Z., 2016. Meta-analysis of California spotted owl (Strix occidentalis occi-dentalis) territory occupancy in the Sierra Nevada: habitat associations and theirimplications for forest management. Condor 118, 747–765. http://dx.doi.org/10.1650/CONDOR-16-66.1.

Thomas, J.A., Thomas, C.D., Simcox, D.J., Clarke, R.T., 1986. Ecology and decliningstatus of the silver-spotted skipper butterfly (Hesperia comma) in Britain. J. Appl.Ecol. 23, 365–380. http://dx.doi.org/10.2307/2404023.

Vitousek, P.M., Mooney, H.A., Lubchenco, J., Melillo, J.M., 1997. Human domination ofEarth's ecosystems. Science 277, 494–499. http://dx.doi.org/10.1126/science.277.5325.494.

Warchola, N., Crone, E.E., Schultz, C.B., 2018. Balancing ecological costs and benefits offire for population viability of disturbance-dependent butterflies. J. Appl. Ecol. 55,800–809. http://dx.doi.org/10.1111/1365-2664.12983.

Webb, J.K., Shine, R., 2000. Paving the way for habitat restoration: can artificial rocksrestore degraded habitats of endangered reptiles? Biol. Conserv. 92, 93–99. http://dx.doi.org/10.1016/S0006-3207(99)00056-7.

Wilson, C.W., Masters, R.E., Bukenhofer, G.A., 1995. Breeding bird response to pine-grassland community restoration for red-cockaded woodpeckers. J. Wildl. Manag. 59,56–67. http://dx.doi.org/10.2307/3809116.

C.M. Wood et al. Biological Conservation 224 (2018) 248–257

257