Optimizing the power of human performed audio surveys for monitoring the endangered Houston toad using automated recording devices

Introduction

Monitoring of endangered anuran populations is required to gain an understanding of population dynamics (Pechmann et al., 1991) and the effects of management actions (Walls et al., 2014). Researchers commonly conduct auditory surveys to determine presence or absence of anuran species (Bridges & Dorcas, 2000; Crouch & Paton, 2002; Schmidt, 2003; Pierce & Gutzweiller, 2004; Weir & Mossman, 2005; Jackson et al., 2006). Data from these surveys can also be used for estimates of the relative abundance of calling male anurans (Zimmerman, 1994) or for determining the cadence or phenology of chorusing behavior (Saenz et al., 2006). These surveys are also used to inform land use and development decisions, in addition to ecological research and endangered species management.

There are presently 14 native anurans, with the inclusion of Puerto Rico, classified by the United States Fish and Wildlife Service (USFWS , hereafter) as threatened and endangered (Anaxyrus californicus, Anaxyrus canorus, Anaxyrus baxterii, Anaxyrus houstonensis, Eleutherodactylus cooki, Eleutherodactylus jasper, Eleutherodactylus juanariveroi, Peltophryne lemur, Rana chiricahua, Rana draytonii, Rana muscosa, Rana pretiosa, Rana sierrae, and Rana sevosa). The USFWS has published guidelines for conducting detection/non-detection surveys for only four of these 14 (A. houstonensis, A. californicus, R. chiricahua, and R. draytonii; USFWS, 1999; USFWS, 2005; USFWS, 2006; USFWS, 2007). Two species (R. pretiosa and R. sierrae) have general overviews of currently applied survey methods, authored by the United States Forest Service, but these have not been established as approved policy guidelines. Federal guidelines for conducting surveys of an additional two species (A. canorus and R. muscosa) are reported to be pending approval. The remaining six species lack formal protocols to confirm species presence at a site. These guidelines are intended to ensure that independent researchers are performing standardized surveys designed to maximize the likelihood of detecting the species when present. This lack of survey guidelines, or the existence of poorly designed guidelines, can have serious negative consequences for populations of endangered species. Incorrectly concluding a population of an endangered species is absent from a site can lead to “take” by development of the site through the loss of breeding wetlands and/or upland habitat, as well as potential mortality caused by having individuals present and active in the development site, resulting in preventable losses to the species and expensive fines and protracted delays for the development projects.

Among the existing published guidelines, recommendations for surveying Houston Toads (A. houstonensis) are some of the most specific (USFWS, 2007). These guidelines dictate that at minimum, six 5-minute audio surveys, per year, are required at each listening post (i.e., potential breeding location); surveys must be conducted for three consecutive years; surveys should be spread out between February through April; temperatures must be at or above 57°F; surveys do not begin until about 30 min after sundown and cease if a drop in temperature occurs (presumably below 57°F, however, this is unclear); and wind speeds must not exceed 15 miles per hour (USFWS, 2007). These guidelines also include less quantitative recommendations intended to increase the likelihood that Houston Toads are chorusing on nights that are chosen to survey. These include nights in which humidity is greater than 70%; cloud cover is present or the moon is not full; and rainfall occurring or recent rainfall has occurred. The efficacy of these guidelines has been studied previously; as a result the updated policy now recommends increasing the number of surveys necessary within each season from six to twelve (Jackson et al., 2006; USFWS, 2007).

Many studies have revealed the advantages of automated audio recording systems in determining the influential exogenous environmental factors associated with vocalizing behavior when compared to manual call surveys (Bridges & Dorcas, 2000; Oseen & Wassersug, 2002; Hsu, Kam & Fellers, 2005; Acevedo & Villanueva-Rivera, 2006; Dorcas et al., 2009; Willacy, Mahony & Newell 2015; MacLaren, McCracken & Forstner, 2018b). In general, automated approaches have the capacity to easily collect significantly more survey data over a wider range of environmental conditions and accordingly provide greater statistical power in examining these relationships over manual surveys. Recently, the use of automated monitoring methods has been more formally recommended for Houston Toad detection/non-detection surveys (USFWS, 2020). Thus, the data provided by automated methods provides an excellent opportunity to investigate the true efficacy of manual survey as recommended by the USFWS (2007) guidelines. We are unaware of other studies that model the potential outcomes of a manual survey protocol using data acquired from an automated recording system, but this information is critical to better understand how well the recommended survey guidelines are detecting Houston Toads in particular, and other endangered anurans in general. The use of automated recording methods allows us to investigate the consequences of various important choices in monitoring program design in a systematic and unbiased manner. Thus purpose of this study is to evaluate the efficacy of the existing Houston Toad survey guidelines, (2) to investigate any opportunity to improve manual surveys by increasing the duration of each survey and the total number of surveys conducted each year, and (3) to provide updated recommendations for Houston Toad surveys.

Materials & Methods

Results

We collected detection/non-detection data on 433 dates across all four years (86, 123, 101, and 123 dates for years 2015-2018, respectively). Out of 92,652 5-minute intervals (intervals, hereafter) we detected Houston Toad vocalizations in 3,975 intervals (approximately 4%), occurring among 123 dates (18, 35, 37, and 33 dates for years 2015-2018, respectively). Environmental variables measured for the dates which include these intervals are given in Table 1. Only 122 (∼3%) of intervals containing Houston Toad vocalizations occurred outside of Feb-April. All results henceforth refer to sampling within this peak chorusing period only (72,359 total intervals, 328 total dates, 3,853 intervals with detections).

In the first survey scenario, when we selected surveys randomly, the mean detection probability was 0.063 (95% CIs [0.038–0.100]) for surveys 5 min in duration, and 0.121 (95% CIs [0.088–0.171]) for surveys 60 min in duration (Fig. 1, Table 2). These probabilities result in requiring a mean of 47 (range = 29–79) surveys to be 95% confident in determining absence of the species when conducting 5-minute-long surveys, and on average 24 (range = 16–33) when conducting 60-minute-long surveys (Fig. 1, Table 2).

Our second survey scenario, in which we removed all dates that did not reflect the environmental thresholds given by USFWS (2007), resulted in reducing the number of available survey dates to 118 out of 328. Under this scenario only 1,737 intervals containing detections are available to be sampled, leaving 1,895 intervals (49%) known to possess Houston Toad vocalizations unobservable to surveyors (3,853 intervals total). The mean detection probability was 0.08 (95% CIs [0.042–0.119]) for surveys 5 min in duration, and 0.142 (95% CIs 0.093 –0.186) for surveys 60 min in duration (Fig. 1, Table 2). These probabilities result in requiring a mean of 36 (range = 24–70) surveys to be 95% confident in determining absence of the species when conducting 5-minute-long surveys, and on average 20 (range = 15–31) when conducting 60-minute-long surveys (Fig. 1, Table 2).

In our final scenario, calculating the proportion of detections to non-detections over a range of environmental thresholds revealed that a unique combination of temperature (>16 ° C), precipitation (>0 mm/day), and change in barometric pressure (<−0.07 mmHg) provided the greatest advantage to surveyors. These thresholds allow 133 dates to be surveyable, comparable to USFWS (2007), yet provide 2,569 intervals containing detections. Under this scenario, the mean detection probability was 0.105 (95% CIs [0.066–0.146]) for surveys 5 min in duration, and 0.179 (95% CIs [0.133–0.229]) for surveys 60 min in duration (Fig. 1, Table 2). These probabilities result in requiring a mean of 28 (range = 20–49) surveys to be 95% confident in determining absence of the species when conducting 5-minute-long surveys, and a mean of 16 (range = 13–22) 60-minute-long surveys to be 95% confident in determining absence of the species (Fig. 1, Table 2).

Discussion

This study demonstrates that the existing USFWS (2007) guidelines for conducting human performed surveys result in a likelihood of concluding absence of Houston Toads from a site when they may be truly present. Given the endangered status of the species, and the precedent set by Jackson et al. (2006), researchers should attempt to achieve 95% confidence in conclusions of absence. Currently, 36 (12 per year, for 3 years) 5-minute-long surveys (180 min total) are required to determine absence (USFWS, 2007). However, our simulation reveals that up to 79 surveys (395 total minutes) of this duration may be required to adequately determine Houston Toad occurrence at a single site. Failing to detect Houston Toads when they are truly present ultimately results in undetected populations, which in the event that the monitoring is being performed as part of a development project, may contribute to local extirpation events. This is especially true for populations outside of our study site, which do not receive any form of population supplementation, and may only support a few individuals. It is also important to add that our study, and the study conducted by Jackson et al. (2006), were carried out using data collected from locations where Houston Toads are likely at, or near, their highest abundance, and as such the data presented here could underestimate of the survey requirements necessary to detect other, smaller populations (Tanadini & Schmidt, 2011).

Our approach demonstrates that the false negative errors associated with manual surveys can be reduced in three main ways. First, manual surveys of longer duration can be performed. In each of our scenarios, the detection probability increases as survey’s duration lengthens. Second, the number of surveys performed can be increased. Third, surveyors can target occasions in which environmental conditions are most closely associated with chorusing behavior among Houston Toads. The trade-offs associated with each of these methods of decreasing false negative errors will depend on the timeframe and scope of the specific Houston Toad monitoring project. Our results indicate human observers must, at minimum, more than double the amount of time spent surveying in order to determine absence with 95% confidence. However, for human observers these changes may be too onerous to allow manual methods to be feasible for determining detection/non-detection from a site. An alternative approach would be the use of automated recording devices (ARDs), as they are demonstrably effective at the task of detecting anurans, especially when rare, due to the efficiency with which high cumulative detection probabilities are achieved (Hsu, Kam & Fellers, 2005; Acevedo & Villanueva-Rivera, 2006). Automated audio recording devices designed for monitoring wildlife are becoming smaller, more affordable, and are able to collect audio much more frequently than human observers (Saenz et al., 2006; Aide et al., 2013, Willacy, Mahony & Newell 2015).

The goals of this study, and that of the federal protocol we evaluate (USFWS, 2007; USFWS, 2020) are limited to determining occurrence of the Houston Toad at a single site (i.e., one human listening post or ARD location). It is our view that human observers should only be employed towards site specific detection of Houston Toads when automated methods are not suitable or potential recorder placement is not permitted. This is particularly relevant for the Houston Toad as the majority of Texas is privately owned lands and public roadways enable access across remaining habitat patches. Further, human observers might better be employed conducting survey methods that cannot adequately be conducted using remote, passive, methods. For example, in order to detect Houston Toads that do not chorus (i.e., subadults, females) human observers may be required to employ drift fence arrays (Brown, Swannack & Forstner, 2013) or sample aquatic habitats. Additionally, methods of determining anuran occurrence at larger scales (i.e., county or regional) often require visiting many sites in a single day, and have successfully been implemented using human observers (Gorman, 2009). Yantis & Price (1993) employed similar methods to determine the distribution of the Houston Toad within Texas.

One critical aspect in the application of our findings is the frequency with which survey dates containing suitable environmental conditions can be expected to occur. If we consider the four years utilized in this study (2015–2018), we find that conditions permissible under the currently accepted USFWS (2007) guidelines only occur on 27, 39, 39, and 31 dates within peak chorusing period each year, respectively. Our simulation indicates that up to 79 5-minute-long surveys should be conducted to minimize false negative errors, and accordingly surveyors would be required to choose an alternative approach (i.e., surveys of longer duration) in order to achieve confidence in their findings. If we consider the environmental conditions discussed in our third scenario, we find that for years 2015–2018, these conditions only occur on 40, 53, 5, and 50 dates within the peak chorusing period each year, respectively. While these results illustrate that in certain years this increases the abundance of surveyable days, during 2017 this method provides only five days of survey appropriate conditions, which reduces the practicality of this approach considerably. Additionally, it would be unwise for researchers to restrict their survey design to a limited set of environmental conditions when these events occur stochastically, and are not guaranteed to occur the necessary number of times to have confidence in conclusions of absence. A caveat within this scenario is that the range of conditions we identified during are derived from our complete dataset, and have not been applied to an independent group of data, and so our results from this scenario may be biased optimistically.

Given the environmental and temporal constraints we have identified, researchers have much to consider when scheduling acoustic surveys for the Houston Toad. The cost to researchers performing acoustic surveys using human observers is likely variable, and we will not speculate on the individual cost of each survey. Although ARDs are, in the long term, less expensive than human performed surveys, factors such as travel and wages can be minimized (Williams, O’Donnell & Armstrong, 2018). We believe our approach demonstrating detection probability as it increases with survey duration provides researchers under variable circumstances options to achieve confidence in absence determinations. For example, if travel costs are large (i.e., researchers must travel a great distance to reach their field sites) perhaps fewer surveys of greater duration may be favored, whereas researchers with many sites to visit may prefer shorter surveys, allowing more sites to be visited on each calendar date.

Conclusions

Given the findings of our simulations, we strongly recommend that human observers restrict their surveys to the peak of Houston toad activity that occurs during the 89 day period between February 1 and April 30. While we believe it is wise to use a priori knowledge of the environmental conditions in which chorusing generally takes place to improve the likelihood of detecting Houston Toads (MacLaren, McCracken & Forstner, 2018b), our study reveals that these events are rare, do not consistently elicit vocalization behavior, and may not allow for adequate effort to be put forth by human observers in any given year. For these reasons we feel our first scenario is most applicable, in which surveyors can choose to survey any date. Houston Toads are in decline throughout their native range (Forstner & Dixon, 2010) thus, we believe that, due to the serious consequences of potential false negative errors, the upper 95% confidence interval for randomly selected surveys be adopted as the minimum survey effort threshold (Table 2). Situations that trigger the need to conduct surveys following (USFWS, 2007; USFWS, 2020) are likely to occur in areas where local occurrence is not known (e.g., marginal habitats). Marginal populations are in the most need of stewardship, and maximum survey effort (i.e., the upper bounds of our confidence intervals, or beyond) is likely necessary to detect these less abundant population remnants.

We found that previously suggested environmental correlates to chorusing among Houston Toads offered improved detection probabilities over randomly selected surveys. However, we found that not all suggested weather criteria within USFWS (2007) were useful, specifically, moon illumination, humidity, and wind speed. This is either because these variables share no true relationship with chorusing within Houston Toads, as is the case for moon illumination, or because they do not serve as a hard boundary, as is the case for relative humidity. For example, relative humidity may range between 10% and 90% within a given single date in response to natural diel cycle. We identified definitive thresholds among temperature, precipitation, and shifts in barometric pressure that improve the probability of detection for Houston Toads beyond what USFWS (2007) currently suggests.

This study updates and expands upon the findings of Jackson et al. (2006). For perspective this previous study (Jackson et al., 2006) utilized twenty 5-minute surveys (100 min) within a given year at a single site, whereas within a single year one ARD provided us with approximately 60,000 min of audio from a single site. Using these vast and detailed data we found that detection probabilities, for surveys of any length, and under any sampling scenario, were lower than what has been previously estimated for this species (Jackson et al., 2006). By suggesting more accurate environmental thresholds under which surveys should be conducted, and evaluating surveys of varying duration, we have provided researchers and managers with an approach that should make the highest probability of detecting Houston Toads possible. Our approach to simulating survey effort allows researchers to choose the combination of survey duration and number of surveys they find most appropriate and maintain 95% confidence in determinations of absence. Like Jackson et al. (2006) our results suggest that the USFWS should modify the mandatory survey guidelines to require more surveys in each season than is currently specified. Moreover, for surveys that are designed to determine presence or absence towards regulatory decisions at a site, the conflict between the availability of suitable environmental conditions and the importance of conducting sufficient surveys based on environmental factors that increase the probability of survey success, ARDs should be strongly considered where possible. Finally, it is critical to differentiate absence determinations made from chorusing data from true absence of this endangered anuran from a potential disturbance site given the underlying nature of juvenile amphibian dispersal and adult use of upland habitats.

Supplemental Information