Accounting for false-positive acoustic detections of bats using occupancy models



  1. Acoustic surveys have become a common survey method for bats and other vocal taxa. Previous work shows that bat echolocation may be misidentified, but common analytic methods, such as occupancy models, assume that misidentifications do not occur. Unless rare, such misidentifications could lead to incorrect inferences with significant management implications.
  2. We fit a false-positive occupancy model to data from paired bat detector and mist-net surveys to estimate probability of presence when survey data may include false positives. We compared estimated occupancy and detection rates to those obtained from a standard occupancy model. We also derived a formula to estimate the probability that bats were present at a site given its detection history. As an example, we analysed survey data for little brown bats Myotis lucifugus from 135 sites in Washington and Oregon, USA.
  3. We estimated that at an unoccupied site, acoustic surveys had a 14% chance per night of producing spurious M. lucifugus detections. Estimated detection rates were higher and occupancy rates were lower under the false-positive model, relative to a standard occupancy model. Un-modelled false positives also affected inferences about occupancy at individual sites. For example, probability of occupancy at individual sites with acoustic detections but no captures ranged from 2% to 100% under the false-positive occupancy model, but was always 100% under a standard occupancy model.
  4. Synthesis and applications. Our results suggest that false positives sufficient to affect inferences may be common in acoustic surveys for bats. We demonstrate an approach that can estimate occupancy, regardless of the false-positive rate, when acoustic surveys are paired with capture surveys. Applications of this approach include monitoring the spread of White-Nose Syndrome, estimating the impact of climate change and informing conservation listing decisions. We calculate a site-specific probability of occupancy, conditional on survey results, which could inform local permitting decisions, such as for wind energy projects. More generally, the magnitude of false positives suggests that false-positive occupancy models can improve accuracy in research and monitoring of bats and provide wildlife managers with more reliable information.


Erroneous observations are a fundamental problem in wildlife surveys because they can obscure ecological processes that surveys are intended to illuminate (Royle & Dorazio 2008). Although research has typically focused on correcting errors in which animals are undetected (Williams, Nichols & Conroy 2002), some survey methods are prone to misidentifying animals and therefore generating false positives in which species that are absent from a site are erroneously judged to be present. For example, aural surveys for birds (Simons et al. 2007; Alldredge et al. 2008) and anurans (McClintock et al. 2010a; Miller et al. 2012) can produce false-positive identifications. However, misidentification may be more common in acoustic surveys for bats. In contrast to territorial and mating calls of birds and anurans that have undergone selection to communicate the species of the caller, echolocation calls of bats have been selected primarily to enable foraging in darkness. Accordingly, bat species occupying similar foraging niches often produce similar calls (Siemers, Kalko & Schnitzler 2001; Jones & Holderied 2007). Furthermore, in contrast to the stereotypical calls of many bird and anuran species, bats exhibit considerable plasticity in their echolocation and adjust their calls to suit their navigational task (Kalko & Schnitzler 1993; Berger-Tal et al. 2008). This combination of similarity among species and variety within species helps bats to forage efficiently, but it also increases the likelihood of misidentifications and false positives during acoustic surveys. Even false-positive rates of 1–2% can potentially bias survey conclusions (McClintock et al. 2010b; Miller et al. 2011), so the potential for frequent false positives during surveys for bats could significantly bias survey results and inferences.

The potential for misidentification of bat calls means that considerable effort has been directed towards developing more sophisticated analytic tools for correct identification of individual bat calls to species. Recent approaches include various refinements of discriminant function analysis, neural networks, support vector machines and classification trees (Preatoni et al. 2005; Redgwell et al. 2009; Armitage & Ober 2010; Britzke et al. 2011; Walters et al. 2012). Despite the sophistication of these efforts, misidentification rates typically range from 5 to 30%, depending on the community of bat species, methods used and other variables (Preatoni et al. 2005; Redgwell et al. 2009; Armitage & Ober 2010; Britzke et al. 2011; Walters et al. 2012). Furthermore, it has been argued that for species with similar calls, misidentifications cannot be eliminated, regardless of the technology brought to bear on the problem (Barclay 1999).

Fortunately, recently developed models enable unbiased estimation of probability of presence even when misidentifications occur (Royle & Link 2006; Miller et al. 2011, 2013). Such models are attractive because they obviate the need for error-free species identification. Of particular interest is an extension of occupancy models that combines results of presence/absence surveys that generate false positives, such as acoustic surveys, with surveys that do not, such as capture surveys using mist nets (Miller et al. 2011). These false-positive occupancy models have obvious applications for bat research and monitoring, such as improved parameter estimates in habitat selection studies (e.g. Yates & Muzika 2006; Weller 2008), estimating species range or biodiversity indicators and for monitoring the effects of White-Nose Syndrome (WNS) and climate change on bat distributions (Rodhouse et al. 2012). Similarly, such a model could improve acoustic surveys conducted at the behest of wildlife management agencies (Pennsylvania Game Commission 2007; Ohio Department of Natural Resources 2009). For example, the U.S. Fish and Wildlife Service is amending survey guidelines for the federally endangered Indiana bat Myotis sodalis Miller and Allen 1928 to incorporate acoustic surveys (U.S. Fish & Wildlife Service 2014). These presence/probable-absence surveys have economic significance for development project proponents, as well as ecological significance for bat populations, so accurate determinations are important. Here we present (i) estimates of false-positive rates encountered in an acoustic survey for the common and acoustically non-descript little brown bat Myotis lucifugus LeConte 1831, (ii) estimates of the effect of false positives on other parameters of interest, such as proportion of area occupied; (iii) a formula to estimate the conditional probability of occupancy at a site when false positives occur; and (iv) recommendations for effectively incorporating this approach into a sampling design.

Materials and methods

Study Area

We conducted surveys for little brown bats during 2006–2009 in Oregon and Washington, USA, as part of the Bat Grid Inventory and Monitoring Project (Hayes, Ober & Sherwin 2009). The study region spanned 427 156 km2 of diverse habitat, including large areas of sagebrush steppe and coniferous forest and smaller areas of juniper, agriculture, grasslands and barrens.

Data Collection

The Bat Grid consists of both a sampling design based on a grid overlaying the study region and a standardized survey protocol designed to maintain data quality (Rodhouse et al. 2012). The sampling grid consists of 4500 sampling units measuring 10 × 10 km, a size selected to match the scale of summertime home ranges of bats in the region (Pierson 1998). The Bat Grid survey protocol (P.C. Ormsbee, unpublished manuscript) describes standardized methods for training, field equipment, survey methods and oversight, as well as data collection, submission and management. Sampling units were selected using a hierarchical sampling process designed to ensure adequate representation of different habitats and availability of public lands or private lands where long-term access was assured. Although selection of sampling units was not strictly probabilistic, we assume that the survey design was adequate to assess the level and effect of false positives in acoustic surveys.

Within selected 10 × 10 km sampling units (or sites; we use the terms interchangeably), observers conducted repeated surveys for bats at multiple survey locations. Specific survey locations were selected from available habitats to maximize the probability of detecting the community of species in the region (Verts & Carraway 1998). Survey locations considered suitable for mist-netting and acoustic recording included accessible water features, meadows and dry washes. Given the large home ranges of bats in the region, we considered the 10 × 10 km sampling unit to be the unit of inference, rather than the survey location.

Bats were captured by placing mist nets over streams and other features likely to funnel active bats into nets (Kunz, Hodgkison & Weise 2009). Captured individuals were identified morphometrically using a regional key (Verts & Carraway 1998), which was supplemented by DNA and call analysis to distinguish among similar Myotis species (Weller et al. 2007; Rodhouse et al. 2008). All surveys occurred between 1 June and 15 September and lasted ≥2 h.

Acoustic surveys were conducted using Petterson D240X full-spectrum time-expansion bat detectors (Pettersson Elektronik, Uppsala, Sweden) placed along flight paths, either paired with net locations or at distinct locations. Detectors were elevated c. 1·5 m above the ground on a pole and aimed upwards at approximately a 45° angle. Detectors were turned on prior to sunset and retrieved at the end of the survey period. Bat call sequences were identified to species using a hierarchical decision engine trained on up to 72 time-frequency and time-amplitude parameters extracted from a library of >10 000 species-known recordings (Parsons & Szewczak 2009; Redgwell et al. 2009) implemented in sonobat version 3.1 (Szewczak 2010), followed by manual vetting and confirmation of species identifications using known-call characteristics (Szewczak et al. 2011). After filtering out noise and low-quality call sequences, sonobat separately classified sequences by evaluating mean parameter values of acceptable calls within a sequence and by evaluating classification agreement among individual calls within a sequence. Sonobat estimated posterior probabilities for identified call sequences using a discriminate function model trained on the library of known calls. To reduce misidentifications, we only accepted call sequence identifications that had a consensus between the two decision classifiers with an estimated probability of correct identification of ≥ 0·95. To further reduce errors, we only used sequences with call types exhibiting species-discriminating characteristics for establishing species presence at each site (Szewczak et al. 2011).

Data Analysis

We used a false-positive site occupancy model (Miller et al. 2011) to estimate false-positive rates encountered in acoustic surveys for little brown bats in Washington and Oregon. This model can be applied when two survey methods are used, an uncertain method that may generate false positives and a certain method that does not. The term ‘certain method’ only denotes certain identification of detected individuals; false negatives may still occur. Each survey method generates a binary detection history represented by a vector of 1s and 0s indicating detection or non-detection, respectively. We considered a species to be detected if at least one bat call was classified as that species (Yates & Muzika 2006; Weller 2008). The detection histories are then used to estimate ψ, probability of presence; r, probability of detection for the certain method; p, probability of detection for the uncertain method; and f, probability of false positives with the uncertain method. Intuitively, the difference in apparent rates of occupancy estimated from certain and uncertain detection histories informs estimates of false-positive rate, f (Miller et al. 2011). In contrast, the standard occupancy model assumes = 0 (MacKenzie et al. 2002).

In the current application, we assume that mist-net captures do not produce false positives, given our use of genetic and morphometric techniques to confirm the identity of similar species (Weller et al. 2007). In contrast, we expected that acoustic surveys might admit false positives, given that tests using known recordings typically detect some misidentified calls (e.g. Preatoni et al. 2005). The model also assumes that sampling units are closed during surveys, so that their occupancy status does not change during the course of surveys. We limited analysis to surveys conducted between June 1 and September 15, so that seasonal bat movements had minimal effects on the occupancy status of sampling units (Fenton & Barclay 1980; Rodhouse et al. 2012). In addition, we limited inferences to the 10 × 10 km sites because individual survey locations within sites were likely not closed between surveys.

We emphasize the difference between false-positive rates in occupancy models and acoustic misidentification rates commonly reported (e.g. Armitage & Ober 2010). Misidentifications occur on individual bat calls, while in our model, false positives occur at sites. For example, if species A and B were present and detected at a site, no false positive could occur for these two species, even if a few calls were misidentified. However, if species A was present and recorded and species B was absent from a site, then any misidentifications would produce a false positive. If we expected to encounter 10 calls of species A and a 10% misidentification rate at sites from which species B was absent, this would yield an expected false-positive rate for species B of 1-(1–0·1)10 = 0·65. Accordingly, the two rates are related but not equivalent because the false-positive rate is affected by the misidentification rate, probability of occupancy, community of species present in the study area and species-specific abundances of bats.

Given the survey design, sampling units were likely closed within each year, but not between years (Rodhouse et al. 2012). Therefore, we selected just 1 year of survey data for each of 135 sample units. To maximize precision of estimates, we selected the year of each site with the most extensive surveys. If n is the number of mist-net surveys and a is the number of acoustic surveys, we defined the most extensive survey as the survey that maximized the value of n0·5*a0·5 + (n + a)/(+ 1). This formula strikes a balance between maximizing total surveys and minimizing reliance on one survey type. Using this criterion, 20 sample units had equally extensive surveys in multiple years. For those 20 sites, we randomly selected one of the equally extensive survey years for inclusion in the data set. We performed this random selection 101 times and kept the data set that produced the median estimate of false positives in our most general model. Each sample unit was included once; therefore, this data set was the largest data set that met the closure assumption. However, the estimated occupancy parameter does not correspond to any particular year. We believe this is an acceptable trade-off because bat demographics suggest stable populations across years (Barclay et al. 2004; Rodhouse et al. 2012) and because our primary goal was to demonstrate the potential for false positives and their impact on occupancy estimates, rather than to estimate occupancy at a particular time and location.

We considered several models for the distribution and detection of little brown bats in the survey region. We developed six models of detection probability. These included models in which detection probability was constant, varied by detection method (mist net or detector), year of survey, or an interaction of year and method. We also modelled detection probability as a function of the square root of survey duration because bat activity tends to be highest early in the night (Hayes 1997). Given the seasonal nature of bat activity (Hayes 1997), we also modelled detection as a quadratic function of ordinal date. We centred survey date and survey duration before analysis. To model false-positive probability, we used similar models, with either a constant false-positive probability or a probability that varied with year, survey duration or date. We eliminated detection method as a predictor because we assumed no false positives occurred during mist-netting. We also considered models in which probability of occupancy was constant or varied by year. These combinations yielded 48 total models. We fit all models in program R 2.15.2 (R Core Team 2012) using the “occuFP” function in the “unmarked” package (Fiske & Chandler 2011). We then used AIC rankings to select the most parsimonious model from the model set (Burnham & Anderson 2002). We compared parameter estimates of the most parsimonious false-positive occupancy model to those of an equivalent standard occupancy model (MacKenzie et al. 2002).

One common use of detection probability estimates is to calculate the probability that the focal species is present at a surveyed site, given that it was never detected, that is, the conditional probability of occupancy (MacKenzie et al. 2006; Wintle et al. 2012). For example, wildlife management agencies must assess the risk to endangered bats from proposed development projects, such as wind energy projects (U.S. Fish & Wildlife Service 2014). Assuming that false positives never occur is likely to affect calculations of the conditional probability of occupancy. Therefore, we compare conditional probability of occupancy calculated under the most parsimonious false-positive occupancy model with that calculated under the equivalent standard occupancy model.

Under the standard occupancy model, the conditional probability of occupancy at a site with no detections is the probability the species was present and undetected divided by the probability it was undetected. This denominator is more easily expressed as the probability the species was present and undetected plus the probability the species was absent. Recalling that r is probability of detection for mist nets, p is probability of detection for bat detectors, ψ is probability of presence, n is the number of replicate capture surveys, and a is the number of replicate acoustic surveys, then the species may be present and undetected with a probability ψ(1−r)n(1−p)a, or it may be absent with a probability of (1−ψ). Therefore, the conditional probability of occupancy is:

display math

Under the standard occupancy model, any acoustic detections indicate the species is present. However, when false positives are allowed, acoustic detections do not prove presence, and we want a conditional probability of occupancy for sites with acoustic detections, but no captures. This conditional probability is the probability the species was present and not detected with certainty divided by the probability it was not detected with certainty. Therefore, we revised the formula to allow uncertain detections, d, and replaced (1−p)a with pd(1−p)ad. Similarly, the denominator must express that an absent species may be falsely detected. Therefore, we replaced (1−ψ) with (1−ψ)fd(1−f)a-d. It follows that:

display math

Under either model, at sites where the species is detected with certainty, the conditional probability of occupancy is simply 1. We populated the above formulas with parameter estimates from our analysis to calculate the conditional probability of occupancy at each site, and we used bootstrap techniques to estimate 90% confidence intervals around estimates. We compared results obtained from the false-positive occupancy model to those of the standard occupancy model.


Little brown bats were captured at 26 of 94 sites where mist-net surveys were conducted, yielding a naïve occupancy rate of 0·28 for the certain method. Combining both survey methods, little brown bats were detected at 85 of 135 sites for a naïve occupancy rate of 0·63. The best supported false-positive site occupancy model indicated that detection probability varied with survey method, while occupancy and false-positive probability were constant across years, survey duration and date. Six other models had substantial support (ΔAIC < 2), but they were all more complex versions of the top model, suggesting that additional predictors had relatively little value (Grueber et al. 2011). Therefore, we only report results from the top model. The equivalent standard occupancy model received less support (ΔAIC = 10·04), indicating the value of modelling false positives in this data set.

The estimated false-positive rate when using bat detectors was 14%, meaning that in a hypothetical 10 × 10 km sampling unit that was never occupied by little brown bats, our acoustic survey methods had a 14% chance of producing one or more spurious little brown bat detections during each night of surveys. Note, however, that absence cannot be directly observed, only inferred. In addition, estimates of other parameters of interest differed between the two modelling approaches (Table 1). In particular, estimated detection rates were higher in the false-positive model than in the standard occupancy model, although there was some overlap in confidence intervals. The higher detection rates and the possibility that some acoustic detections were erroneous also yielded a much lower estimated occupancy rate for the false-positive model, albeit with some overlap in confidence intervals.

Table 1. Parameter estimates under competing models that do or do not allow false positives in acoustic surveys for little brown bats Myotis lucifugus in Washington and Oregon, 2006–2009. Estimates with 95% confidence intervals in parentheses for occupancy (ψ), detection rate of mist nets (r), detection rate of bat detectors (p) and false-positive rate for detectors (f)
Modelψ r p f
  1. NA indicates not applicable.

Standard occupancy0·86 (0·74–0·93)0·19 (0·14–0·25)0·49 (0·43–0·54)NA
False-positive occupancy0·65 (0·47–0·79)0·25 (0·17–0·35)0·57 (0·49–0·64)0·14 (0·07–0·27)

Conditional probability of occupancy estimates can be divided into three groups: sites with mist-net captures, sites with acoustic detections only and sites with no detections or captures. At the 26 sites with mist-net captures, conditional probability of occupancy was 1 because the evidence (capture) was assumed to be error-free. At the 59 sites with acoustic detections only, conditional probability of occupancy was 1 under the standard occupancy model, but <1 under the false-positive occupancy model. Across our study area, estimated conditional probability of occupancy under the false-positive model ranged from c. 1·00 at a site with six acoustic detections after six acoustic surveys and three mist-net surveys to 0·02 at a site with one acoustic detection after eight acoustic surveys and four mist-net surveys (Fig. 1). At the 50 sites with no detections, conditional probability of occupancy was lower using the false-positive model because estimated detection rates were higher, making negative surveys more informative (Fig. 2).

Figure 1.

Conditional probability of occupancy and 90% confidence intervals under a false-positive occupancy model at 59 sites with no mist net detections but at least 1 acoustic detection of little brown bats Myotis lucifugus in Washington and Oregon, 2006–2009. The different detection histories among sites lead to the variation in estimates of conditional probability of occupancy and their associated confidence intervals. Under a standard occupancy model, occupancy would be 1·0 at all sites.

Figure 2.

Conditional probability of occupancy and 90% confidence intervals at 50 sites with no detections of little brown bats Myotis lucifugus in Washington and Oregon, 2006–2009. Solid circles represent estimates based on a false-positive occupancy model, while open circles represent estimates based on a standard occupancy model.


Identifying animals, particularly bats, from their calls is a challenge that requires a surveyor to ask, ‘is that you baby, or just a brilliant disguise?’ (Springsteen 1987). Given this difficulty, it is generally understood that errors may occur during acoustic surveys for bats and that such errors may affect study results (Britzke, Gillam & Murray 2013). To date, the preferred approach to handling such errors has been to try to lower them to a degree that they do not significantly affect study conclusions. Here, we have demonstrated an alternative approach that seeks to model false positives, rather than eliminate or ignore them (Royle & Link 2006; Miller et al. 2011, 2013). Using a false-positive occupancy model, we estimated false-positive rates in acoustic presence/absence surveys for bats and illustrated the effect of false positives on occupancy parameters. We also provided a simple method for estimating the conditional probability of occupancy – probability of occupancy at a site given its detection history – when false positives occur.

Although false-positive occupancy models can accommodate false positives, we took steps to reduce misidentifications of bat calls prior to occupancy modelling so that our methods would be more consistent with general practices. These steps included filtering to remove lower quality calls, requiring a consensus between two different decision classifiers and using manual review to check species classifications. Even after these steps, we estimated a false-positive rate of 14% for little brown bats in our survey. False positives may also be common in acoustic surveys for other bat species and they have the potential to significantly impact survey results and inferences.

As expected, the false-positive rate for acoustic surveys of little brown bats in our study was higher than that reported for birds or anurans (Alldredge et al. 2008; McClintock et al. 2010a). However, false-positive rates will likely be sensitive to a number of factors, including the misidentification rate, amount of bat activity and probability of presence for the target species. A large literature has identified variables affecting misidentification rates, including the focal species, size and quality of the call library, acoustic software used, statistical techniques used, size and composition of the local bat community to be analysed, habitats surveyed, and behaviour of surveyed bats (Britzke, Gillam & Murray 2013). In addition, the overall level of bat activity will likely be positively related to false-positive rates because every bat call is an opportunity for a false positive. Therefore, future efforts to fit models that account for bat activity levels could improve estimation and inference. This could be made by incorporating bat activity as a covariate or extending the model to estimate detectability of individual calls (sensu Royle & Nichols 2003). In addition, false-positive rates will be affected by probability of presence because many unoccupied sites would offer more opportunities for false positives (McClintock et al. 2010b; Miller et al. 2011). In contrast, an omni-present species would never produce a false positive, even if misidentification rates were high.

Similarly, the potential for false positives to affect occupancy studies will depend on probability of presence and number of repeat surveys (Miller et al. 2011). For example, for a given false-positive rate, a low probability of presence can yield more false positives and inflate the probability of presence estimated in a standard occupancy model. In addition, more acoustic surveys will yield more false positives because every survey is an opportunity for a false positive (Miller et al. 2011). For example, with a 14% false-positive rate, the majority of unoccupied sites would yield spurious acoustic detections after five nights of surveys and 95% of unoccupied sites would yield spurious acoustic detections after 20 nights of surveys. With a standard occupancy model, such spurious detections could incorrectly suggest a high probability of presence.

False positives also affected estimates of the conditional probability of occupancy at individual sites. For example, among 59 sites with acoustic detections, but no captures, only 27% had a conditional probability of occupancy >90%, while 24% had a conditional probability of occupancy <50%. In contrast, a standard occupancy analysis would classify all of these as occupied with certainty. At sites with no detections at all, the two models also provide divergent estimates of the conditional probability of occupancy. In fact, under a false-positive occupancy model, a lightly surveyed site with no detections may have a higher conditional probability of occupancy than an extensively surveyed site with one or two acoustic detections.

Although false positives are an undesired aspect of acoustic surveys, our results help demonstrate the value of bat detectors. In particular, we found that for little brown bats, the detection rate was over twice as high for detectors, relative to mist-net surveys. Previous analysis with this data set also found that acoustic surveys contributed substantially to detectability (Rodhouse et al. 2012). The higher detection rate was also anticipated by several authors who reported greater species richness in acoustic surveys than mist-net surveys (Flaquer, Torre & Arrizabalaga 2007; MacSwiney, Clarke & Racey 2008; Robbins, Murray & McKenzie 2008). In contrast to our finding for little brown bats, a standard occupancy analysis of survey data for tropical bats found that most species were similarly detectable by acoustic or capture techniques (Meyer et al. 2011). The greater detection rates of bat detectors that we report, combined with labour costs that are often lower (Clement & Castleberry 2011) and lack of trap shyness (Robbins, Murray & McKenzie 2008), recommend them as a survey tool. However, frequent false positives, and their effects on parameter estimates, make it important to supplement detectors with a certain method (such as capture surveys) and to use models that account for misidentifications, such as false-positive occupancy models (Royle & Link 2006; Miller et al. 2011, 2013).

Management Implications

Bats are facing novel threats from the expanding wind-power industry (Kunz et al. 2007) and the spread of WNS (Frick et al. 2010), along with more common threats to wildlife, such as habitat loss and climate change. As such, there is widespread need for effective survey techniques at scales ranging from continental (Battersby 2010; Loeb et al. 2012) to local (Pennsylvania Game Commission 2007; USFWS 2014). For example, even formerly common species, such as the little brown bat, have declined precipitously due to WNS (Frick et al. 2010) and our approach could be used to improve efforts to monitor the spread of this disease. Similarly, a false-positive approach could improve estimates of the impact of climate change, habitat conversion or other stressors. By accounting for false positives, our approach could improve the information available for conservation listing decisions, range maps or biodiversity indicators developed from acoustic surveys. In addition, we calculate a site-specific probability of occupancy, conditional on survey results that could inform local permitting decisions, such as for development projects. The approach we demonstrate is attractive for these and other conservation uses because of its ability to leverage the confident identifications of capture surveys and the cost-effectiveness of acoustic surveys to estimate occupancy, even when misidentifications and false positives occur.

Implementing a false-positive occupancy model requires several design considerations. First, at least some study sites sampled with an uncertain method, such as acoustic surveys, must also be sampled with a certain method, such as a capture survey. This allows for the estimation of occupancy and false-positive rates even when it is not practical to perform capture surveys at all locations. However, sites selected for capture surveys should be representative of all study sites, or else relevant differences among sites (e.g. habitat and elevation) should be accounted for with site-specific covariates. Secondly, estimating false-positive rates will require a larger sample size to achieve the same estimator precision as a standard occupancy model, although we leave the question of the most appropriate sample size for future research. Thirdly, we recommend documenting covariates that may affect detection, occupancy or false-positive rates, such as habitat characteristics, weather data and overall levels of bat activity. Finally, we note that capture surveys for morphologically similar species may violate the assumption that captured bats are not misidentified, unless genetic analysis or some other certain technique is used (e.g. Weller et al. 2007).

If false positives in acoustic surveys are as common as our results suggest, then there are important implications for bat surveys. First, using our approach, it is not necessary to entirely eliminate misidentification errors to accurately estimate occupancy by bats. In fact, excessive efforts to eliminate misidentifications may generate additional false negatives and reduce estimator precision. Secondly, acoustic surveys may overstate the probabilities of detection and presence for bat species. Thirdly, acoustic detections alone may not be sufficient to determine that a site is occupied by a target species. This has implications for important management decisions, such as permitting of wind-power or other proposed development projects when no bats are physically captured during surveys at a site. If the research or management goal is focused on species presence at the site level, we recommend calculating a conditional probability of occupancy based on suitable estimates of the key occupancy and detection parameters, as we have demonstrated. Finally, it must be noted that the false-positive occupancy model does not identify bats by their echolocation, and therefore, known-call models implemented in SonoBat and similar software will continue to play a central role in acoustic surveys for bats.


We thank Wayne Thogmartin and 2 anonymous referees for their constructive comments on the initial manuscript. Funding for the Bat Grid was provided by the U.S. Forest Service, Bureau of Land Management and Department of Defense Legacy Program. The US Fish and Wildlife Service Region 3 Division of Endangered Species funded the data analysis and preparation of the manuscript. The National Park Service Upper Columbia Basin Network provided additional funding and support. We thank the many contributors to the Bat Grid, particularly Aimee Hart for her devoted efforts in the field and with the data base. Without their extraordinary effort, this research would not have been possible. Lew Cousineau provided invaluable data management support. Jeremy Hobson provided GIS support. J.M.S. developed and operates SonoBat. Any use of trade, product or firm names is for descriptive purposes only and does not imply endorsement by the U.S. Government.