Evaluating differences in density estimation for central Iowa butterflies using two methodologies

Introduction

Ecologists have often struggled with the need to estimate the probability that an organism is detected during a survey, given that it is present. This concept is widely referred to as detection probability (Burnham & Anderson, 1984). Early studies tended to ignore it and assumed that all organisms were detected during surveys (Mackenzie, Nichols & Sutton, 2005). Later work shifted towards developing methods to directly estimate detection probability, which include distance sampling (Eberhardt, 1968; Gates, 1968; Buckland et al., 2001; Buckland et al., 2004), multiple-covariate distance sampling (Marques et al., 2007), mark-recapture (e.g., Haddad et al., 2008; Pellet et al., 2012), double observers (Nichols et al., 2000; Koneff et al., 2006), and a synergy of distance-sampling and double-observer methods (Kissling & Garton, 2006). If detection probability is estimated to be less than 1.0, that information is used to correct estimates of density and abundance to account for the fact that some fraction of the population is almost always missed during surveys (Buckland, Laake & Borchers, 2010). Still, despite increased awareness of imperfect detection, particularly in the field of conservation biology (Jarzyna & Jetz, 2016; Benoit, Jackson & Ridgway, 2018), estimation of detection probability is widely lacking (Kellner & Swihart, 2014; Kral et al., 2018).

The importance of addressing concerns about detection probability has clear and important implications for the conservation and monitoring of populations. Conservation plans often address the need to know the size of the population of interest (Yoccoz, Nichols & Boulinier, 2001; McGill, 2006; Farr et al., 2022), and such estimates repeated in time are useful for estimating the trend of the population (Buckland et al., 2001). The conditions for surveying a population can change over time for many reasons—differences in habitat, changes in observers, and a host of other factors. As such, trends that rely on changes in relative abundance may be biased because they assume that detection probability remains constant. Additionally, simulations have suggested that accurately identifying population trends becomes difficult when detectability <80% (Falaschi et al., 2022), such as for smaller, more cryptic species. Studies that directly estimate detection probability can greatly minimize this source of bias, resulting in more robust inferences about the population, and leading to more informed conservation actions.

Many types of surveys are used to estimate the size of a population, and one of the most common is line transects (Buckland et al., 2001). For butterfly surveys, the standard Pollard-Yates line transect has been the most extensively used means of surveying butterflies (e.g., Brown & Boyce, 2001; Collier et al., 2006; Nowicki et al., 2008). This method involves walking at a slow pace (ca. 10 m/min) along a predetermined line and counting only butterflies seen within a prescribed width, often 5.0 m (i.e., 2.5 m to either side of the observer; (Pollard & Yates, 1993)). There are several benefits to Pollard-Yates transects including the continuation of long-term studies (Van Swaay et al., 2008) and the ability to count many individuals efficiently (Isaac et al., 2011). Additionally, abundance estimates derived from Pollard-Yates surveys can be corrected to reduce the bias of double-counting individuals by averaging counts within sites and summing across sites (Pollard & Yates, 1993; Gross et al., 2007; Haddad et al., 2008). However, an implicit assumption of these surveys is that detection probability of butterflies is 1.0, or that detection probability is constant across the survey period. Violating this assumption means that estimates of density or abundance may not be comparable, and that any changes detected could result from true changes in the population or from changes in survey conditions (e.g., different observers). Data from Pollard-Yates transects are often converted to density estimates and used to make inferences about populations. The effectiveness of Pollard-Yates transects to correlate with population size has varied (Collier et al., 2008). The subsequent reliability of this method is likely dependent on a variety of factors including target species, and continuity of observers and environmental conditions. Fortunately, there are additional survey methodologies that can be used to directly estimate detection probability and yield “corrected” density estimates.

Distance sampling has been shown to be a beneficial sampling method for butterflies (Powell, Busby & Kindscher, 2007; Moranz, 2010; Isaac et al., 2011; Kral-O’Brien et al., 2020), given that the assumption of perfect detection can be tested by assigning a distance (or distance bin) to each individual (Burnham & Anderson, 1984). Under this sampling method, detection probability is assumed to decrease as distance to the transect line increases (Burnham & Anderson, 1984; Buckland et al., 2001; Buckland et al., 2004), which allows for butterflies to be sampled at greater distances than 2.5 m used in Pollard-Yates transects. Sampling at greater distances allows for more detections of rarer species and consequently, a more complete estimate of community composition. However, distance sampling is rarely used for butterflies, relative to other survey techniques (Brown & Boyce, 1998; Kral et al., 2018), and imperfect detection as a concept is widely unaccounted for in the majority of invertebrate papers (Kellner & Swihart, 2014). Not accounting for imperfect detection can result in underestimates of true abundance (Burnham & Anderson, 1984), which may have implications for conservation or management decisions when accurate population estimates are desired. Consequently, researchers have summarized the need to address sampling bias that stems from variation in butterfly detectability (Dennis et al., 2006; Kéry & Plattner, 2007; Haddad et al., 2008; Nowicki et al., 2008).

Species-specific variation in detection probability is likely heavily dependent on morphological characteristics (Iknayan et al., 2014). We found only one study that accounted for butterfly detection probabilities using species’ morphological traits (Kral-O’Brien et al., 2020). They used distance sampling to estimate the effective strip width for butterfly detectability as a function of species’ size and color, and found that larger, brighter species had higher detection probabilities. Almost all species with a wingspan under 40 mm had an effective strip width of 2.0−4.8 m (Kral-O’Brien et al., 2020), suggesting that some small species may be inadequately sampled via Pollard-Yates transects, particularly if they are also dull-colored. Conversely, studies that prioritize sampling larger, brighter species, such as monarchs (Danaus plexippus), may meet the assumption of perfect detection using Pollard-Yates transects.

In this study, our objectives were to (1) evaluate butterfly density estimates derived from Pollard-Yates line transects and distance sampling, (2) estimate how detection probabilities for butterflies vary across sampling distances and wing length, and (3) offer advice on future butterfly sampling techniques to estimate population density.

Portions of this text were previously published as part of a thesis (Patterson, 2016).

Materials & Methods

Results

We recorded a sufficient number of detections for least skipper (n = 350), cabbage white (n = 117), clouded sulphur (n = 158), orange sulphur (n = 291), little yellow (n = 414), eastern tailed-blue (n = 247), viceroy (n = 88), and monarch (n = 301). Least skipper was the most abundant species recorded but this was primarily driven by high abundance at one site: Rock Creek Marsh WMA (Table 1). The two smallest species, least skipper and eastern tailed-blue, were detected in categories extending only out to the 5-10 m and 10–25 m bins respectively, whereas the other six species were recorded in all eight bins. However, in all instances, the two outermost bins accounted for a small proportion (<10%) of the detections, and the median detection distances occurred in the following bins for each species: least skipper (1−1.75 m), cabbage white (1.75−2.5 m), clouded sulphur (2.5−5.0 m), orange sulphur (2.5−5.0 m), little yellow (2.5−5.0 m), eastern tailed-blue (1−1.75 m), viceroy (2.5−5.0 m), and monarch (2.5−5.0 m). For every species, >80% of individuals were identified as flying, basking, or nectaring at time of detection, with resting, mineralizing, courting/mating, and ovipositing butterflies composing the remainder.

Butterfly detectability varied by distance and species’ morphology. At the <2.5 m truncation, which is the standard Pollard-Yates transect width, detectability among all species was >0.90, and for several species was estimated at 1.0. However, species-specific detectability began to decrease at distances >2.5 m (Table 2). Consequently, we focus most of our inferences at the 5.0 m truncation because it provides a more interesting comparison with Pollard-Yates densities. At the 5.0-m truncation, detection probabilities ranged from 0.53 for the least skipper to 0.79 for the much larger, more conspicuous monarch (Table 2). At this distance, the top model for all species included a hazard-rate key and simple polynomial adjustments (Table 2). Additionally, a post-hoc analysis also revealed a strong positive correlation (r = 0.91) between mean wing length and detection probability at the 5-m truncation, suggesting that detection probabilities were greater for larger species (Fig. 1). Detection probabilities at the 50-m truncation were low, ranging from 0.05 (95% CI [0.05–0.06]) for least skipper to 0.14 (95% CI [0.11–0.18]) for monarch.

The corresponding species-specific densities varied by sampling methodology. Densities derived from distance sampling were generally greater than those from the Pollard-Yates transects for multiple species at various truncations. Pollard-Yates densities for clouded sulphur and viceroy were below the 95% confidence intervals of distance sampling estimates at all truncations (Fig. 2). Least skipper, cabbage white, and eastern-tailed blue also had higher density estimates at either the 5-m or 50-m truncation compared to Pollard-Yates estimates. Generally, species occurring in the highest densities were least skipper, eastern tailed-blue, and little yellow according to distance sampling and Pollard-Yates. However, Pollard-Yates estimates suggest eastern tailed-blue and little yellow occur at similar densities (eastern tailed-blue: 23 no./ha, little yellow: 20 no./ha), while distance sampling predicted more than double the number of eastern tailed-blue compared to little yellow (eastern tailed-blue: 33 no./ha, little yellow: 14 no./ha) at the 5-m truncation (Fig. 2).

Discussion

Our study found that detection probabilities of eight common, widespread butterfly species in Iowa were at or near 1.0 in the standard Pollard-Yates transect but dropped considerably when the sampling area extended >2.5 m from the line transect. Detection probability was positively correlated with mean wing size and was greatest for the largest, most conspicuous species. Below we discuss our findings in the larger context of methods to estimate butterfly densities, and then comment on how the inclusion of distance sampling can help with conservation and management decisions for this taxon.

Conclusions

In our study, there was minimal variation in detectability at the width (5 m) employed by the traditional Pollard-Yates transect. Consequently, fixed-width transects assuming perfect detection may be adequate for estimating population densities if the target species are large and if transect width is ≤5 m. However, if the research objective is to adequately sample the butterfly community composition or to target smaller, less common species, then incorporating distance sampling with larger transect widths may increase detections and improved density estimates. Narrow transects might not adequately sample large habitat blocks, and avoidance behavior by faster species (e.g., viceroy) might lead to poor and inaccurate counts. By broadening transects and incorporating distance sampling, researchers are likely to detect rarer species and provide more robust estimates of population densities, which may have important implications for conservation actions that rely on accurate population and community estimates. Given the interspecific variation in detectability in our study and other studies (Moranz, 2010; Isaac et al., 2011), and the observer variation summarized by Kéry & Plattner (2007) and Isaac et al. (2011), we recommend incorporating distance sampling whenever possible, especially when transects are >5 m wide and when densities of smaller species (<6.0 mm median wing length) is of interest.

Line transects are straightforward in their implementation and can be used to repeatedly sample butterflies at multiple sites across a broad region in a short window of time. Distance sampling can be easily incorporated into the line-transect framework and analyzed using Program Distance (Buckland, 2006), which provides density estimates and associated measures of precision. As such, distance sampling represents an effective tool to survey butterflies and guide the management and conservation of butterfly populations.

Supplemental Information