Sampling period, size and duration influence measures of bat species richness from acoustic surveys



1.  Understanding animal ecology depends on an ability to accurately inventory species. However, there are few quantitative data available, which allow for an assessment of the effectiveness of acoustic sampling methods for determining bat species richness.

2.  We assessed inventory efficiency, defined as the percentage of species detected per survey effort, using data from 7 to 9 Anabat bat detectors deployed concurrently between June 2008 and August 2009 at fixed locations. We examined sampling period and time of night to calculate the minimum duration of sampling effort required to detect the greatest percentage of species.

3.  In all cases, multiple survey nights at multiple sampling locations were necessary to detect higher levels of species richness using acoustic detectors. Additionally, continuous sampling throughout the night was important for detecting more species, especially during summer, fall and spring months.

4.  Species accumulation curves indicated that relatively few nights were needed to detect ‘common’ species at various sampling locations (2–5 nights on average); however, longer sample periods (>45 nights) were necessary to detect ‘rare’ species at some sampling locations. Accumulation curves indicated that the number of detector locations positively influenced the number of species detected during surveys periods.

5.A priori knowledge of sampling effort is fundamental for designing biologically robust inventories. We make recommendations for improving the efficiency of acoustic surveys using analytical methods that are broadly applicable to a range of survey methods and taxa.


A key component of understanding, and effectively managing communities, is gaining a basic understanding of inherent variation in their specific composition across space and time (Morris 1990). A failure to understand or address normal spatiotemporal variation in community structure can make the most basic ecological data problematic (Boulinier et al. 1998). Knowledge about the presence or absence (or detection vs. non-detection) of various species within a community over time is necessary to elucidate spatial and temporal patterns of ecosystems (Carroll, Zielinski & Noss 1999), especially when using survey data collected at the same sites over extensive sample periods (MacKenzie et al. 2002). Understanding normal levels of variation in species richness is fundamental for conservation efforts because managers must decide whether changes in species richness across space and time warrant conservation efforts. There are many spatial scales at which species richness can be assessed, including the landscape scale where anthropogenic change (e.g. fragmentation or restoration) is often considered in study design (Krohne 1997; Turner 2005).

The development of passive sampling technologies to record the acoustic calls of free-flying bats and potentially permit species identification (O’Farrell & Gannon 1999; Ochoa, O’Farrell & Miller 2000; Gannon et al. 2004) has spurred many surveys for bats. Given that most insectivorous bats use echolocation to detect prey and orient, acoustic detectors provide a useful tool for measuring species richness within and across large geographical areas and habitat types. Prior to the advent of inexpensive acoustic detectors, our understanding of bat species richness came from data generated by captures (i.e. harp traps and mist nets) or visual observations (i.e. roost surveys).

Acoustic detectors are portable devices that passively record sound frequency parameters and call patterns (duration, intensity, etc.) such that individual echolocation events may be identified to the guild or species at a later time (O’Farrell & Gannon 1999; Gannon et al. 2004). Acoustic surveys typically detect more species than active capture devices (Kalko, Handley & Handley 1996; Murray et al. 1999; O’Farrell & Gannon 1999; Ochoa, O’Farrell & Miller 2000). When compared with nets and traps, acoustic detectors are: (i) a non-invasive means of assessing activity by detecting individuals over a greater area, (ii) not limited to surveying local resources such as water bodies or open flyways that may be necessary when using capture devices, (iii) able to sample continuously and (iv) operational regardless of weather and environmental conditions.

Data from the acoustic detection of bats have been used to document activity patterns (Hayes 1997; Tibbels & Kurta 2003), resource use (Vaughan, Jones & Harris 1997; Williams, O’Farrell & Riddle 2006; Zukal & Rehak 2006), species richness (Wickramasinghe et al. 2003) and species distribution (Jaberg & Guisan 2001). However, explicit evaluations of the survey effort necessary to acoustically detect different species are generally lacking (Hayes 1997; Milne et al. 2004). Greater a priori knowledge about spatial and temporal variation in species detection rates should improve the design of future studies.

Acoustic survey effort influences the likelihood that various species are detected during survey events (Gorresen et al. 2008; Weller 2008). By not accounting for spatiotemporal variation in bat activity, researchers may fail to detect certain species at different locations (Moreno & Halffter 2000; Broders 2003; Milne et al. 2005) because it is difficult to predict the time needed to acoustically detect inconspicuous species. Typically, multiple surveys and methods are required to detect all species in an area (Krebs 1989; Hayes 1997; Moreno & Halffter 2001), and yet wildlife managers may be constrained by limited resources. While ecological attributes of bat communities (i.e. diversity and richness) typically vary across spatiotemporal scales, basic elements of general study design (i.e. survey intensity, duration and detector orientation) are applicable across locations and communities (Duchamp et al. 2006). To design studies intending to assess species richness, it is necessary to be familiar with the spatiotemporal variation inherent in bat assemblages so that variation in community structure across space and through time can be assessed (Sherwin, Gannon & Haymond 2000; Gannon & Sherwin 2004). For example, questions regarding seasonal variation in community structure or non-random distribution of species across the landscape of interest likely require different sampling strategies.

While understanding inherent spatiotemporal variation in community structure has proven critical in understanding long-term ecological attributes of wildlife communities (Morris 1990), there has been little formal evaluation of this variation in species richness of bat communities using acoustic detectors. To date, there have been no studies that have investigated nightly patterns of species richness of bat communities using long-term (>1 year) continuous acoustic survey data concurrently collected at multiple permanent sampling locations. Our objective was to assess the sampling effort required to detect fixed levels of bat species acoustically. Our broad expectation was that increased sampling would yield more precise estimates of richness, and that there would be differences across nights and detector sites during different seasons. We further expected that given the extensive nature of our data set, richness estimates would plateau at each site. Our second objective was to then develop specific recommendations for the design of acoustic monitoring studies intended to assess species richness.

Materials and methods

Study Site

Our study was conducted at Ash Meadows National Wildlife Refuge in southern Nevada (36° 25′ N, 116° 19′ W), a 95-km2 region of the Mojave Desert managed by the U.S. Fish and Wildlife Service. The location receives roughly 70 mm annual rainfall; however, it is a major discharge point for two underground aquifer systems. The area is one of the few places in the Mojave Desert with reliable year-round surface water and supports the highest concentration of endemic species in the United States per unit land area (Stevens & Bailowitz 2008). The Refuge is located between limestone mountains to the east and west and is characterised by a variety of different habitats. Approximately half of the total land area in the study area is composed of alkali shrub/scrub and creosote shrub land. Alkali meadows, wet meadows, invasive weed patches and riparian woodlands comprise roughly 25% of the total land area. The remainder of the Refuge is a mixture of vegetation types including mesquite bosque, alkali playas, seasonally flooded woodlands, dunes, open water and mixed Mojave scrub habitats.

Locations of Anabat detectors were chosen using a randomised block design that considered habitat types and current and historical restoration efforts. Each detector was deployed in a different habitat and/or vegetative structure. However, while habitat differences likely contributed to the variability in species richness across sampling locations, we did not assess habitat use by bats. Each detector was located between 2 and 6 km from other detectors, and the location of each detector was maintained (permanently) throughout the study. Seven detectors were deployed in June of 2008, and an additional two detectors were added the following February and June of 2009.

Echolocation Recording

We used frequency-division systems (i.e. Anabat II detectors; Titley Electronics, Balina, NSW, Australia) because they are best suited to record echolocation calls passively at remote sampling locations for extended periods of time (Corben 2000; Limpens & McCracken 2004). We collected data between June 2008 and August 2009. Each detector system was powered by a 12-V battery connected to a 5-watt solar panel and mounted on a pole 2·5–3·0 m above the ground. Polyvinyl chloride (PVC) hoods protected microphones from the elements, and a 15 × 15 cm acrylic glass plate mounted below each microphone reflected sound upwards into the microphone. The frequency-division setting was set to eight to provide the greatest call resolution (Corben 2000), and sensitivity levels were set to about 6·5 on all detectors to minimise background noise (i.e. wind, insects) while enabling the detection of most echolocation calls. Detectors were programmed to switch on automatically 90 min before sunset, and off 90 min after sunrise each day. Detectors remained switched off during day-time hours to recharge batteries. Data were saved internally to compact flash cards and later transferred to a computer.

Detector malfunction was prevalent during some periods (specifically spring); therefore, we only used data for any given unit when >20 nights per season were recorded. We made no attempt to compare detection rates among detectors.

Echolocation Analysis

Recordings were analysed to identify species using Analook software version 3·7f (Titley Electronics, Balina, NSW, Australia). We used a reference library of >8000 echolocation calls from various species captured and recorded in the western U.S. (many within 500 km of the study area) and made the assumption that the identification of calls in the library was accurate. Twenty-one of Nevada’s 23 species of bat are reported to occur in the southern part of the state (Bradley et al. 2006).

Fragmentary call sequences consisting of ≤2 discernable call notes from an individual (O’Farrell & Gannon 1999) were classified as ‘unknown’ because of high levels of inter- and intraspecific call variation. A ‘call’ was defined as one note in a sequence of call notes, and a call file as a sequence of call notes from one or more individuals during a single recorded sequence (Corben & O’Farrell 1999). Single call files could contain multiple call notes from multiple individuals. To account for multiple species recorded within single files, each call file was independently viewed to ensure that all detected species were recognised and identified.

Calls were identified qualitatively following the methods of O’Farrell, Miller & Gannon (1999) and Milne et al. (2002). All calls were identified by SLS to prevent interobserver bias. Calls that could only be identified to the level of ‘bat’ because of the fragmentary nature of the call were classified as ‘unknown.’ Identification accuracy was vetted using discriminant function analysis (Parsons & Jones 2000; Russo & Jones 2002; Fukia, Agetsuma & Hill 2004) and tests of interobserver agreement. Tests of interobserver agreement were conducted by sending batches of randomly selected files to a third party for identification. Pearson’s r was used to ascertain whether different analyses (discriminant function analysis, third part individuals and SLS) produced statistically similar results.

Subsamples of identified call files were randomly chosen (including ‘unknowns’) and checked for species identification accuracy and consistency across nights. Subsample sizes were chosen using power analysis, and resulting call-file batches were selected randomly. By selecting subsample sizes using the total number of files identified for each species, we assumed that species classifications were fairly accurate. For tests of interobserver agreement, information about species, date, and time information were extracted from each subsample call file and then removed so that call files contained no information about initial identifications. Neither observer had any prior knowledge about the numbers or kinds of species contained in the selected files, and analyses were performed independently.

Discriminant function analysis was conducted using systat 11 for Windows (Systat Software Inc., Chicago IL, USA). Nine parameters were extracted from reference calls automatically using Analook software: minimum frequency (Fmin), maximum frequency (Fmax), characteristic frequency (Fc), the frequency of the ‘knee’ or inflection point of each call (Fk), the duration in time of each call (Dur), the time from the start of the call to Fc (Tc), the time from the start of the call to Fk (Tk) and the slope of the call at the Fc (Sc) (for definitions, see: Corben & O’Farrell 1999; Corben 2000; Milne 2002). We chose these parameters because they are representative of call structures across species. Species were set as the grouping variable, and call parameters were used as the predictive variables. All groups were weighted evenly.

Our verification tests indicated a high level of agreement in the total number of species detected, total number of echolocation call files detected, and the numbers of species detected among and across nights for most species. Identification accuracy and consistency was >90% for most species. There were minor issues identifying between Eptesicus fuscus and Lasionycteris noctivigans and discerning Lasiurus blossevilli from Parastellus hesperus. Both issues are not surprising, however, as these species pairs have calls that are similar in frequency, structure and pattern. Despite issues parsing between these species, identification accuracy remained >75%.

Analysis of Variability in Species Richness

For simplicity, we used 3-month blocks of the annual 12-month calendar to define the four seasons in temperate North America: spring (March–May), summer (June–August), fall (September–November) and winter (December–February). We treat seasons independently with the exception of summers 2008 and 2009, where data were combined for some analyses. Because detector malfunction occurred during some periods (specifically spring), we only used data where ≥20 nights per season were recorded by individual detectors in tests that required longer sample periods (i.e. species accumulation curves). Some detectors came online later in the study, and malfunctions eliminated usable data from some detectors during some seasons. Therefore, our results do not represent comparisons between seasons. Species richness was assumed to be zero at the start of each season.

Analysis of Species Detection Patterns

To measure bat activity, we quantified the number of echolocation events recorded over specific time periods. We found it difficult to discern detection patterns using file size (e.g. bytes: Broders 2003) or counts of the number of files recorded (Kalcounis et al. 1999) because multiple species were frequently encountered within individual files at some sampling locations. Therefore, we used the activity index (AI; Miller 2001) to calculate the magnitude of species presence across individual time blocks. Modified to include sampling effort, AI is defined as:


The AI is calculated by summing the number (n) of time blocks for which a species was present (P*) and dividing by sampling effort (e: time period over which the data were collected).

AI allows for comparison of the relative time contribution of each species compared with other species detected. For example, a species was recorded ‘present’ during the time blocks; it was detected regardless of the number of call notes (≥2) or sequences present in the given file. As the AI is applicable regardless of the number of individuals, species or call notes in a recorded file, it reduces the influence of an individual repeatedly circling within the range of the detector microphone. By plotting the amount of time that different species were detected during ½ hour time blocks across detectors, we plotted species-specific patterns of activity by time of night. This allowed us to assess whether sampling all night influenced species richness measures. Sunset and sunrise data were gathered using Moonrise software version 3·5 (Sidell 2002; Moonrise, Grande Rapids, MN, USA).

We used the program presence (Proteus Research & Consulting Ltd., Dunedin, New Zealand) to test whether all-night sampling was effective for detecting previously undetected (i.e. unique) species during nightly sampling events. presence is frequently employed to produce occupancy models from data on detection vs. non-detection, where detection probabilities are <1 (MacKenzie et al. 2002). Detection probabilities were weighted according to temporal distributions of detections. For example, a species detected only on the first five and last five nights of a 100-night survey would have a lower detection probability than a species detected every 10th night because the latter detection rate is more consistent. Missing observations (i.e. detector malfunction) provided no data about detection vs. non-detection and were therefore not used in the analysis.

Estimates of species occupancy were calculated for ½ hour time blocks during each night beginning at sunset. Occupancy estimates were a function of the probability that an individual species was detected ‘present’ during each time block at each detector location, and the probability that each species had previously been detected at each site. Final estimates for each time block were averaged across species and detector locations for each season. The analysis assumed that species richness at the start of the sampling night was zero and that detections at each site were independent of other sites.

Analysis of Sampling Period and Size

We evaluated sampling efficacy and completeness across time and among detectors using species accumulation curves (Moreno & Halffter 2000). Species accumulation curves are a class of linear dependence models, which assume the number of new species detected will decrease with increasing sampling effort, and are useful in situations where an assessment of the amount of effort necessary to detect X% of species is the objective (Soberón & Llorente 1993). Curves were calculated using ecosim ver. 7.72 (Gotelli & Entsminger 2001). Numbers of nights (n nights) and number of detectors (n detectors) were used as measures of sampling effort, which made it possible to predict the likelihood that greater species would be detected through time based on sampling effort. Accumulation curves reached an asymptote when the probability of detecting a new species approaches zero.

Sample nights were repeatedly randomly reordered 1000 times by Ecosim to produce smoothed curves. Reaching 100% species richness within models is unlikely because of the chance that some species remain undetected during the survey period. Therefore, the proportion of species detected during simulated sampling periods is typically used to assess sampling completeness. We selected 80% and 90% of total species richness detected during each season as conservative levels of survey completeness to allow for comparison of species richness within and among locations (Moreno & Halffter 2000; Shiu & Lee 2003). Per cent species richness values were derived from Ecosim calculations. We then used the species accumulation curves to calculate the mean number of nights needed to attain these values following Lamas, Robbins & Harvey (1991) as reviewed by Soberón & Llorente (1993) and applied by Moreno & Halffter (2000).


We analysed patterns of species richness based on data collected by 7–9 detectors on 2645 total sample nights between June 2008 and August 2009. We recorded data during 437 detector nights (seven detectors) during the summer of 2008, 637 detector nights (seven detectors) during fall, 610 nights (eight detectors) during winter, 337 nights (9) detectors during spring and 624 nights (nine detectors) during the summer of 2009. We identified 12 different species from a potential 14 species that are presumed to occur at our study site (unpublished data; U.S. Fish and Wildlife Service survey records on-file at Ash Meadows National Wildlife Refuge) and recorded 353 371 individual call files (Table 1). For 18·6% of the echolocation events, we classified the species as ‘unknown’. A review of the unknowns by an independent observer and discriminant function analysis suggested that the majority (∼85%) were the most commonly detected species: Myotis californicus, Myotis yumanensis and Parastrellus hesperus. However, ‘unknowns’ were too indistinct to confidently identify, and we left them as undiagnosed for our analysis.

Table 1.  Summary of identified echolocation events by season
  Summer 08Fall 08Winter 08–09Spring 09Summer 09Total Sp. Detections
  1. Summer 2008 (n nights = 437, n detectors = 7), fall (n nights = 637, n detectors = 7), winter (n nights = 606, n detectors = 7), spring (n nights = 293, n detectors = 6) and summer 2009 (n nights = 624, n detectors = 9).

Antrozous pallidus 2743314425621312
Corynorinus townsendii 522195151751801087
Eptesicus fuscus 32 3412
Lasiurus blossevilli 18162 521269
Lasiurus cinereus 516 70394
Lasionycteris noctivigans 951 12375
Lasiurus xanthinus 4057704 15221887101
Myotis californicus 21625742189085452263261113
Myotis ciliolabrum 312 3 18
Myotis yumanensis 5033183871604263410180
Parastrellus hesperus 638651159315511839105868193320
Tadarida brasiliensis 101589516870422513006
Grand Total12009832934148035542163316353371
No. Species12125121112

Analysis of Species Detection Patterns

The mean ± SD number of species detected per night across each detector was highest during the summer of 2008 (3·6 ± 1·8) and 2009 (3·0 ± 1·6), and lowest in winter (0·3 ± 0·6; Fig. 1). The number of species detected per night across detectors during fall and spring averaged 2·6 ± 2·0 and 2·8 ± 2·1, respectively. Several species were detected at all detectors within and across seasons, but some species were only recorded at some detectors within and across seasons (Table 2). Additionally, the percentage of survey nights that various species were detected varied (Table 3).

Figure 1.

 Box-and-whisker plot showing mean (bold), median and 25th and 75th percentiles of the number of species detected per night from June 2008–August 2009. Vertical lines represent range from 10–90%. Data are cumulative averages of data from detectors that operated largely continuously between June 2008 and August 2009 (detectors 1–7, n nights = 2395).

Table 2.  Per cent of detector locations various species were detected at by season
 Summer 08 (%)Fall 08 (%)Winter 08–09 (%)Spring 09 (%)Summer 09 (%)
  1. Summer 2008 (n nights = 437, n detectors = 7), fall (n nights = 637, n detectors = 7), winter (n nights = 606, n detectors = 7), spring (n nights = 293, n detectors = 6) and summer 2009 (n nights = 624, n detectors = 9).

Antrozous pallidus 100·071·414·377·8100·0
Corynorinus townsendii 57·142·914·344·466·7
Eptesicus fuscus 28·614·30·011·122·2
Lasiurus blossevilli 57·142·90·011·144·4
Lasiurus cinereus 42·914·30·044·433·3
Lasionycteris noctivigans 28·614·30·022·233·3
Lasiurus xanthinus 42·928·60·033·333·3
Myotis californicus 100·0100·028·688·9100·0
Myotis ciliolabrum 14·314·30·011·10·0
Myotis yumanensis 100·0100·028·655·6100·0
Parastrellus hesperus 100·0100·071·488·9100·0
Tadarida brasiliensis 85·7100·014·377·8100·0
Table 3.  Per cent of survey nights that species were detected by season
 Summer 08 (%)Fall 08 (%)Winter 08–09 (%)Spring 09 (%)Summer 09 (%)
  1. Summer 2008 (n nights = 437, n detectors = 7), fall (n nights = 637, n detectors = 7), winter (n nights = 606, n detectors = 7), spring (n nights = 293, n detectors = 6) and summer 2009 (n nights = 624, n detectors = 9).

Antrozous pallidus 24·82·40·218·422·7
Corynorinus townsendii 17·88·31·212·511·4
Eptesicus fuscus 0·70·20·00·310·5
Lasiurus blossevilli 13·14·30·00·61·0
Lasiurus cinereus 0·91·30·04·20·3
Lasionycteris noctivigans 1·32·40·01·20·3
Lasiurus xanthinus 11·57·40·03·66·4
Myotis californicus 69·645·43·842·780·0
Myotis ciliolabrum 0·21·60·00·60·0
Myotis yumanensis 52·826·52·324·946·0
Parastrellus hesperus 84·156·77·241·396·5
Tadarida brasiliensis 21·722·00·230·625·4

We found a range of 5–12 species recorded across individual detectors during summers 2008 and 2009, 4–12 species in fall, 0–5 in winter and 2–11 in spring. On average, four species were detected every 0·5 h per detector during summer, fall and spring between sunset and sunrise (Fig. 2). During winter, <1 species was detected every 0·5 h. Species were more or less active during different parts of the night (Figs 3–6). More species were detected earlier in the evening during fall and spring compared with summer. Some species exhibited crepuscular patterns of detection (e.g. P. hesperus), whereas some species exhibited more nocturnal patterns of detection (e.g. T. brasiliensis). The probability of detecting previously undetected species during the current night varied with time of night (Fig. 7), but was highest in the first 2 h after sunset in all seasons.

Figure 2.

 Mean species richness in relation to time after sunset during summer (solid circles), fall (open circles), winter (solid triangles) and spring (open triangles). Summers (2008: n nights = 437, n detectors = 7; 2009: n nights = 624, n detectors = 9), fall (n nights = 637, n detectors = 7), winter (n nights = 606, n detectors = 7) and spring (n nights = 293, n detectors = 6) months from June 2008 to August 2009.

Figure 3.

 Species-specific activity patterns relative to sunset during summer. Data are cumulative from all detectors during summer (2008: n nights = 437, n detectors = 7; 2009: n nights = 624, n detectors = 9). Figures 3–6 are arranged by activity levels, with the most active species in the top left and least active species in the bottom right. Values are multiplied by a factor of 100 to convert smaller values to whole numbers. We note, however, that for rarely detected species, these figures may be biased reflections of the actual activity patterns.

Figure 4.

 Species-specific activity patterns relative to sunset during fall. Data are cumulative from all fall detectors (n detectors = 7; n detector nights = 637).

Figure 5.

 Species-specific activity patterns relative to sunset during winter. Data are cumulative from all detectors in winter (n detectors = 8; n detector nights = 610).

Figure 6.

 Species-specific activity patterns relative to sunset during spring. Data are cumulative from all spring detectors (n detectors = 9; n detector nights = 337).

Figure 7.

 Probability of detecting a previously undetected species (within a single sampling night) by time of night. Values (filled circles) are cumulative from detectors that operated during summers (2008: n nights = 437, n detectors = 7; 2009: n nights = 624, n detectors = 9), fall (n nights = 637, n detectors = 7), winter (n nights = 606, n detectors = 7) and spring (n nights = 293, n detectors = 6).

Analysis of Sampling Period and Size

Species accumulation curves approached an asymptote at different rates within and among seasons (Fig. 8). The number of nights needed to detect 80 and 90% of the possible species at each detector varied with season (Fig. 9). It took a mean ± SD of 22·5 ± 17·9 nights to detect 80% of the known species during summer, 15·5 ± 7·4 nights in fall, 20·6 ± 9·9 in winter and 9·7 ± 5·3 in spring. To detect 90% of species at each detector in summer, 46·4 ± 29·2 sampling nights were needed, 30·0 ± 15·1 nights in fall, 30·8 ± 15·4 in winter and 21·2 ± 9·6 in spring.

Figure 8.

 Species accumulation curves by season. Symbols and lines represent different detector locations.

Figure 9.

 Mean (dot) number of sample nights for all detectors required to detect 80% and 90% of the bat species detected during each season. Vertical lines represent 95% confidence intervals. Values (filled circles) are cumulative from detectors that operated during summers (2008: n nights = 437, n detectors = 7; 2009: n nights = 624, n detectors = 9), fall (n nights = 637, n detectors = 7), winter (n nights = 606, n detectors = 7) and spring (n nights = 293, n detectors = 6).

The probability of detecting species increased with an increasing number of detectors (Fig. 10). The number of detectors necessary to detect 80% and 90% of species varied with season (Fig. 11). Detecting 80% required a mean ± SD of 3 ± 3·1 detectors in summer, 4 ± 5·8 in fall, 5 ± 1·6 in winter and 5 ± 3·8 in spring. To detect 90%, it took a mean of 6 ± 1·3 detectors in summer, 7·5 ± 3·5 in fall, 7 ± 1·9 in winter and 8 ± 1·8 in spring.

Figure 10.

 Accumulation curves with 95% confidence intervals showing the number of bat species detected during each season as a function of the number of detectors units sampling continuously. Smoothed curves were produced by randomly reordering data 1000 times. Data are cumulative from detectors that operated during summers (2008: n nights = 437, n detectors = 7; 2009: n nights = 624, n detectors = 9), fall (n nights = 637, n detectors = 7), winter (n nights = 606, n detectors = 7) and spring (n nights = 293, n detectors = 6).

Figure 11.

 Number of detectors calculated to detect 80% and 90% of the acoustically detected species richness within the study area across seasons. Bars represent variance. Data are from detectors that operated during summers (2008: n nights = 437, n detectors = 7; 2009: n nights = 624, n detectors = 9), fall (n nights = 637, n detectors = 7), winter (n nights = 606, n detectors = 7) and spring (n nights = 293, n detectors = 6).


The objective of many acoustic bat surveys is to determine which species occur in an area (Walsh, Barclay & McCracken 2004). Such surveys often strive to assess the relative importance of various habitat types based on the known distribution of species within the landscape (Sherwin, Gannon & Haymond 2000; Miller, Arnett & Lacki 2003). Despite the usefulness of acoustic detector data for monitoring bat communities, there is growing recognition that much of the data that result from acoustic surveys do not adequately account for variation in species activity across space and time (Hayes 1997, 2000; Gannon, Sherwin & Haymond 2003 & Miller, Arnett & Lacki 2003; O’Shea, Bogan & Ellison 2003; Milne et al. 2005; Fischer et al. 2009). An a priori understanding of the survey effort necessary will help to ensure statistically powerful sampling designs, clearer data interpretation and likely more successful management and conservation actions.

When Should Surveys be Conducted?

The duration of sampling within nights varies across studies (Gannon, Sherwin & Haymond 2003). For example, some authors sample continuously from dusk until dawn (e.g. Kuenzi & Morrison 1998; Williams, O’Farrell & Riddle 2006; Dzal et al. 2009), while others sample for only parts of the night (e.g. Wickramasinghe et al. 2003; Gehrt & Chelsvig 2004; Ford et al. 2005; Flaquer, Torre & Arrizabalaga 2007). A principal objective of all of these studies was to quantify species richness. Varying approaches to answering similar questions can lead to confusion about which method is best, differing interpretations from the data and can make it difficult to directly compare studies. We found that the probability of detecting a new species during a sample night generally declined 3–4 h after sunset after which most acoustically common species had been detected. However, bat detections did not appear to decrease linearly with time of night, and during summer and spring exhibited a bimodal peak where detection probability increased slightly in the 3–4 h before sunrise. This suggests that sampling all night is important to record less frequently detected species. In all but the winter, some acoustically rare species were only detected late at night or in the early morning. These species may not have been detected had surveys not lasted the entire night.

How Many Surveys were Needed to Detect X% of Possible Species?

We found that 2–5 survey nights were necessary to detect the 5–7 species (between 40% and 60% of the possible species richness) that were commonly recorded at most detectors during spring, summer and fall. Many more nights of sampling were required to yield increased richness estimates beyond this level. For example, in summer, it took on average >20 nights to detect 80% of possible species and >45 nights to detect 90% of possible species at some detectors. In spring, the number of nights needed to detect 80% and 90% of species was about half of that for summer. This may be due in part to seasonal variability in species richness among detector locations, and possibly lower detection probabilities of acoustically rare species in summer (i.e. migratory species like Lasiurus cinereus) vs. other seasons.

Longer survey periods will produce more precise survey results (Krebs 1989). Sampling should occur for as long as possible, especially given that the cost does not increase substantially when using passively operated detectors. However, extensive long-term data sets are more costly in terms of analysis.

How Many Sampling Locations were Necessary to Detect the Most Species?

Studies that have measured bat activity at multiple sites often report variation in species richness among detector locations (e.g. Milne et al. 2005; Williams, O’Farrell & Riddle 2006; Ciechanowski et al. 2007). Providing recommendations about the number of detectors needed to detect the most species for study areas other than our own is difficult because there are environmental, technological, financial and logistical factors that influence study objectives and the applicability of findings (Miller, Arnett & Lacki 2003). However, our general recommendations are broadly applicable to all geographical regions. We found that the probability of detecting more species increased with the number of detectors used. While there exists the statistical possibility of detecting all species using as few as one detector during the summer and fall seasons, and two detectors during the winter and spring seasons, we calculated there was a low probability of this actually occurring. Most importantly, a limited number of detectors make it difficult to assess spatial variation in species richness, which precludes extrapolation to broader spatial scales.

We found a range of 6–12 and a mean ± SD of 8·9 ± 2·3 species recorded at each detector location over a relatively long (>12 months) period of survey effort. In instances where the number of detectors for use is not limited, our data suggest that sampling ≥6 locations for approximately 30 nights each should detect most acoustically detectable species in a relatively small landscape similar to ours. Additionally, because we found that the detection of acoustically rare species varied both temporally and spatially, we recommend that researchers first assess local variation via concurrent acoustic sampling and then modify the sampling design to best reflect local community dynamics.

Conclusions and Recommendations for Future Research

Many of the species we detected are broadly distributed throughout the south-western United States and northern Mexico. Despite their broad distributions, many of these species are considered to be at risk in all, or portions of their range. We focused on acoustic bat detectors as their use has increased exponentially (Hayes 2000; Fenton 2003; Gannon, Sherwin & Haymond 2003; Miller, Arnett & Lacki 2003). Understanding patterns of activity by various species may help future survey efforts, especially for rare or threatened species. However, given the diversity of bats, thorough surveys will require multiple sampling techniques. For example, we only detected 12 species out of a projected 14 (Myotis thysanodes and Eumops perotis californicus have been documented on Refuge lands previously), and only 12 species out of a potential 21 species known to occur in southern Nevada (Bradley et al. 2006). While our levels of species richness may reflect our sampling methodology, for example the sensitivity of Anabat detectors (Larson & Hayes 2000), they clearly show that even long-term acoustic data sets can be inconclusive for some species.

Our results support our original predictions that greater sampling effort would yield more accurate species richness estimates. On average, 20–30 sampling nights were needed to detect 80–90% of the species that occurred at each sampling location, and different species were detected with different frequencies across locations. Additionally, species richness varied at different sampling locations as did the rate at which species were detected. If only one detector is available, our analyses suggest that moving the detector between locations for 2–5 nights at each location would effectively detect the acoustically common species at our study site. This method is likely the best option where an inventory needs to be done quickly, and a general species list from across the landscape is the main objective. However, when a comparison of community structure, species richness, habitat associations and activity patterns among species is the focus of the study, longer periods of sampling effort with larger numbers of detectors will be required. Overall, we recommend sampling at ≥6 different locations for approximately 30 nights per location to detect less common species (Table 4), but acknowledge that our results may have differed slightly had we access to more detectors or sampled in a different geographic location.

Table 4.  Number of hours after sunset, nights and locations found necessary to detect 80% and 90% of detected species richness at permanent acoustic sampling locations
SeasonMinimum Sample Period Found
Survey period n nights n locations
  1. Values are the mean values from simulations that randomly drew subsamples from data pools within and across detectors. Summers 2008 and 2009 (n nights = 1061, n detectors = 9), fall (n nights = 637, n detectors = 7), winter (n nights = 606, n detectors = 7) and spring (n nights = 293, n detectors = 6).

WinterSunset + 5 h after213157

Our results lead to the following three cautionary suggestions about sampling design for future acoustic monitoring studies: (i) sample continuously throughout the night (dusk–dawn); (ii) sampling for repeated numbers of nights at the same locations increases chances of detecting acoustically rare species and developing more accurate estimates of species richness and (iii) sampling at multiple locations may not (statistically) be necessary to detect the acoustically common species, but will increase the rate at which species are detected and are necessary for robust comparisons of species richness between sampling locations.


This study would not have been possible without the help, guidance and support of many individuals. We thank C. Baldino & S. McKelvey and the staff at Ash Meadows National Wildlife Refuge for making this study possible. S.L.S. would like to thank K. Taylor for valuable help with data management, M.J. O’Farrell for help with species identification, R. Fisher for statistical assistance and C. Somers and R. Poulin for valuable comments that helped guide the direction of this analysis. This research was funded through a research grant to CNU from the U.S. Fish & Wildlife Service, facilitated through Biowest, Inc. R.E.S. wishes to thank D. Olsen of Biowest for his support and assistance with administration of the grant throughout the project. M.B. Fenton and three anonymous reviewers provided comments that improved an earlier draft of the manuscript.