Effect of sampling effort and species detectability on volunteer based anuran monitoring programs

A central question to the usefulness of surveys for estimating population trends is whether the surveys adequately represent the biological communities that are being monitored. There are two approaches to address this issue; the first is to statistically adjust site occupancy using species detection probabilities, and the second is to determine the minimum sampling effort required to adequately represent the communities. We focused on the latter approach, using data from two anuran monitoring programs, the Ontario Backyard Frog survey (1992–2001) and the Ontario Bait Frog survey (2001–2003). We determined the minimum sampling effort required to adequately represent the anuran community, and demonstrated the pitfalls of examining population trends when detection probabilities are less than one and vary either temporally or spatially. We found that approximately 12 and 24 randomly sampled nights were required to detect 80% and 90% of species richness, respectively. Detection probabilities varied among species, and were highest for spring peepers (Pseudacris crucifer) and green frogs (Rana clamitans), and lowest for wood frogs (Rana sylvatica) and northern leopard frogs (Rana pipiens). Variation in detection probabilities is likely associated with species specific differences in calling strategies and breeding periods. Anuran monitoring programs that use three stratified sampling periods require site occupancies to be adjusted using detection probabilities in order to adequately represent anuran communities. Failing to account for differences in detection probabilities may result in false significant differences in site occupancy rates. The issues raised here are not limited to monitoring anurans, but are suitable for surveys of many taxa.

Introduction

Reports of amphibian declines (e.g. Pechman et al., 1991; Wake, 1991) have persuaded both Canadian and American government agencies to implement programs to monitor amphibian populations. In recent years, geographically extensive anuran surveys have been used to determine if amphibian populations are declining (Bishop et al., 1997; Lepage et al., 1997). Many of these surveys are based primarily or solely upon volunteers to collect the data, and involve monitoring the calls of adult male anurans. Although much of the focus has centred on the use of these surveys for site occupancy, calling codes (an index of population size) and counts of calling males (a direct measure of population size) are generally also collected. Currently, many volunteer-based acoustic anuran surveys are active in Canada and the United States, most of which are funded, run by, or modeled after the North American Amphibian Monitoring Program (NAAMP), Environment Canada, and Frogwatch USA. Call surveys are also used in non-volunteer programs (e.g., Amphibian Research and Monitoring Initiative [ARMI]), as well as university funded projects.

Is the sampling effort of anuran call surveys sufficient to represent the anuran community? This question is central to the utility of acoustic surveys for monitoring population trends; unfortunately, instances of insufficient or inappropriate sampling effort have been well documented. Generally, monitoring programs are designed with the intent of sampling during peak calling times for anurans; nevertheless, success has been mixed. Bridges and Dorcas (2000) found that the typical timing of anuran surveys in South Carolina occurred when southern leopard frogs (Rana sphenocephala) had a small probability of calling. Crouch and Paton (2002) found that not all anuran species were detected using acoustic surveys, and that four sampling periods were necessary to monitor the seven species that were detected. Many acoustic survey programs have three sampling periods, not four, although the NAAMP protocol includes three seasonal, but also allows for an optional earlier sampling period to target wood frogs. Paszkowski et al. (2002) found that sampling effort of acoustic surveys was too low when they evaluated different survey techniques, as they sampled after the peak calling times for some species. Detection probabilities for many species can be low, which may underestimate site occupancy (Storfer, 2003). However, probability detection can be incorporated into models estimating temporal changes in anuran occupancy rates (MacKenzie et al., 2002). Although the problem of call detection is becoming well recognized (Bailey et al., 2002; Genet and Sargent, 2003; Weir et al., 2003), it is doubtful that monitoring programs have been designed with respect to this issue.

MacKenzie et al. (2002) described an approach to compare temporal changes in site occupancy when the probability of detection is as low as 30%. Currently, at least two amphibian monitoring programs are using this approach to examine differences in amphibian occupancy in light of suboptimal detection probabilities (Genet and Sargent, 2003; Weir et al., 2003). While we have no objections with statistically adjusting the data to account for detection probabilities of less than one, knowledge of detection rates does not aid in determining whether or not a species is present at any particular sites. Surveys can only accurately determine species presence or absence by assuring that the sampling effort is sufficiently high so that nondetection ceases to be a major issue.

We evaluated the sampling effort of two acoustic anuran monitoring programs: the Ontario Backyard Frog survey (1992–2001) by Environment Canada, and the Bait Frog (northern leopard frog; Rana pipiens) survey, 2001–2003 by the Ontario Ministry of Natural Resources (OMNR). Previous work has shown that volunteer observers generally correctly identify species by their call (Genet and Sargent, 2003; Shirose et al., 1997). Call intensity and count data are generally unsuitable for monitoring small areas or routes or for rarer species, because the combination of small sample size with large sampling error would reduce the precision of estimates of relative abundance (Shirose et al., In review), and so reduce power to unsuitable levels. Volunteers also had difficulty in estimating calling intensity (Genet and Sargent, 2003). Thus, we focused only on species richness and species detectability; specifically, we determined the sampling effort required for the Ontario Backyard Frog survey to adequately estimate species richness, and the relative detectability of different species. Lastly, we demonstrate using the Bait Frog survey of the OMNR how differences in detectability may give false significant differences in site occupancy among years.