Power and the passion: assessing statistical power with simulations to optimise monitoring of a threatened lizard

Introduction

Ecological monitoring is a vital component of biodiversity conservation because an understanding of threatened species population trends allows for the rapid implementation of effective management protocols (Field et al. 2005; Moore et al. 2023). In particular, a key aspect of monitoring is the ability for a survey method to confidently detect changes in relative or absolute abundance in threatened populations (Reynolds et al. 2011; Geyle et al. 2019; Moore et al. 2023). If a critical decline is confidently detected, it facilitates proactive management, including the establishment of captive insurance populations (Andrew et al. 2018) or employment of translocation efforts where individuals are moved to sites where threats are absent or better managed (Legge et al. 2018; Magoulick et al. 2024). Without effective strategies for detecting population declines and rapid commencement of management strategies, populations of threatened taxa can quickly decline below their minimal viable population size, resulting in an increased likelihood of population-level extirpation.

Cryptic and rare species are difficult to survey, owing to their often inconspicuous behavioural habits (e.g. small size, fossorial or nocturnal lifestyle, etc.), which leads to low detectability without significant survey effort (Mazerolle et al. 2007; Owens et al. 2024). Many reptile taxa are under-represented in the scientific literature and are classified as Data Deficient by the International Union for Conservation of Nature (IUCN) because of their life-histories, morphology, and behaviour (Meiri and Chapple 2016; Chapple et al. 2019; Wotherspoon et al. 2024). Although comprehensive extinction-risk assessments have been available for frogs, birds, and mammals for well over a decade, the same level of assessment for reptiles was achieved only recently, finding 21% of species are threatened (Chapple et al. 2019; Cox et al. 2022).

Difficulties associated with detecting cryptic reptile taxa create a major problem for ecological management where an inability to measure population trends limits the ability for effective and timely action. Inadequate survey designs lead to reduced statistical power to detect important population variation. Ignoring statistical power as a measure of appropriateness and reliability of monitoring methods may result in ineffective management strategies, poor data collection, and failure to record true population trends (Legg and Nagy 2006). Ensuring that monitoring programs are able to achieve their objectives effectively and confidently with limited resources remains one of the largest challenges for conservation managers globally (Reynolds et al. 2011).

The pygmy bluetongue (Tiliqua adelaidensis) is a highly specialised species that primarily occupies lycosid or mygalomorph spider burrows in fragmented grassland of south-central South Australia (Milne and Bull 2000; Fellows et al. 2009). These mesic grasslands have experienced significant anthropogenic degradation through landscape modification for agriculture and development (Fordham et al. 2012; Delean et al. 2013). Consequently, owing to significant range and habitat contractions and inadequate ecological knowledge, T. adelaidensis was for decades considered extinct (Milne 1999; Fordham et al. 2012). Their rediscovery in 1992 led to increased survey and research efforts to improve our understanding of their conservation status, geographic distribution, ecological requirements, threats, and best management practices (Milne 1999; Souter et al. 2004; Fellows et al. 2009; Clayton and Bull 2016). Currently, they are federally listed as Endangered under the Environment Protection and Biodiversity Conservation Act 1999 (EPBC Act) and are threatened by continued habitat degradation and loss owing to anthropogenic development and climate change (Fordham et al. 2012).

Adult T. adelaidensis spend significant periods of time concealed within their burrow, typically departing only to find mates or capture passing prey (Milne et al. 2003; Bull et al. 2015). They aggressively defend their burrows against conspecifics and, once established within a burrow, typically have a home range extending less than a metre from the burrow entrance (Fenner and Bull 2011; Bull et al. 2015). Survey practices have traditionally exploited their reliance on burrows and largely sedentary habits and population densities are calculated through individual and repeat counts of occupied burrows. Previous surveys have estimated population densities using 20 × 20-m or 30 × 30-m monitoring quadrats. However, quadrats were often specifically located in areas of high lizard density because higher sample sizes were beneficial for research objectives (Milne and Bull 2000; Souter et al. 2004; Clayton and Bull 2016). Alternatively, government-led surveys were undertaken using single 100 × 100-m quadrats (Milne 1999; Nielsen 2017). However, relying on a single large quadrat can lead to biased results, because it may not accurately represent the variability in habitat and population distribution across the entire occupied area.

In this paper, we test and compare 30 × 30, 50 × 50, and 100 × 100-m quadrats in their capacity to detect T. adelaidensis population trends. We hypothesised that 100 × 100-m quadrats would require excessive survey effort, 30 × 30-m quadrats would under-represent site-level variability in occupancy, and 50 × 50-m quadrats would provide a suitable compromise with consideration for survey effort and statistical power. We used field data to generate simulations and assessed statistical power of different quadrat sizes and their practicality within a sampling timeframe. Whereas small-scale monitoring and research have occurred at this site for multiple decades, site-wide monitoring of T. adelaidensis has yet to occur. We developed this method to support long-term, site-wide monitoring strategies and future research on T. adelaidensis at Tiliqua Nature Reserve.

Materials and methods

Preliminary surveys and population simulations

To select a population density for simulation, we conducted preliminary on-ground surveys for T. adelaidensis at TNR. We surveyed two 100 × 100-m quadrats (each divided into four smaller 50 × 50-m quadrats), one of which was thought to be of higher-quality habitat on the basis of prior research projects, and thus to have a higher density, and one was presumed to be in lower-quality habitat. After quadrats were delineated, field personnel thoroughly searched them by walking sequential 1-m belts between pinned markers to ensure systematic and complete coverage. All spider burrows that were located and deemed suitable for T. adelaidensis occupancy (>13-mm diameter and >100-mm depth; Milne and Bull 2000) were marked with a temporary pin. After searching the entire quadrat, marked burrows were inspected using an optic fibrescope (Yateks M Series Videoscope) to confirm lizard occupancy.

To assess the utility of power analysis to optimise quadrat and sample sizes for assessment of T. adelaidensis population change, we simulated a 1-km² area with either a high-density population (100 individuals per hectare) and a low-density population (five individuals per hectare) in the R environment (R Core Team 2021) via the RStudio interface (ver. 2024.09.0+375; see https://posit.co/download/rstudiodesktop/; RStudio Team 2024) where points represent a simulated lizard individual. Points were placed randomly and represented the irregular distribution of the individuals we observed during preliminary surveys. We also created a mixed density simulation (where a random 25% of the 1-km² area was of high density (100/ha) and 75% was of low density (5/ha)), alongside the low- and high-density levels to represent the more realistic patchy distribution of lizards. We randomly placed non-overlapping quadrats of differing sizes (small; 30 × 30-m, medium; 50 × 50-m, and large; 100 × 100-m) within the 1-km² plot to count the number of T. adelaidensis individuals per quadrat.

The number of quadrats per treatment type was determined as the maximum number that could be feasibly and thoroughly searched by a team of two researchers within a 2-week period as previously described. As setting up multiple small quadrats inherently leads to disproportionately more setup time than does setting up a single larger quadrat, despite potentially surveying an approximately equivalent area, the time taken to complete the survey is not linear. Additionally, in the case of the survey methods used for T. adelaidensis, higher-density quadrats also take more time at the same quadrat size. Whereas the amount of time spent performing transects to locate burrows remains approximately the same to avoid adding biases, the time spent checking all of the burrows with the optiscope is often much longer in high-density quadrats because they typically contain a higher number of total burrows. These on-ground differences are important to take into consideration when designing a realistic in situ survey by using these a priori methods. Approximate time values were developed following the preliminary surveys and then were refined further following consultation with three independent advisors knowledgeable on the species (Supplementary Table S1). It is currently unknown how temporal trends in vegetation cover and survey effort affect detectability of T. adelaidensis; thus, we assumed that lizard detectability remained consistent between surveys.

We reduced the simulated population size by 10%, 25% and 50% to simulate declines and repeated the power analysis on the three quadrat sizes (Fig. 1). Because population dynamics of T. adelaidensis are poorly understood, these decline levels were chosen as proxies for slight, moderate, and extreme declines respectively, which may be observed in populations and might trigger different management interventions. This process was repeated 10 times per treatment, totalling 36 monitoring designs (i.e. three density levels, three decline levels and one baseline level, and three quadrat sizes), with 10 replicates each, equating to 360 total simulations.

Fig. 1.

A conceptual outline of the project design. Preliminary surveys informed the densities to simulate for the low-, mixed-, and high-density populations, where 10%, 25%, and 50% declines were applied. Quadrats of 30 × 30, 50 × 50, and 100 × 100-m were randomly placed to calculate power curves and number of days/quadrats required to reach a power of 0.8 (1−β) and a significance threshold (α) of 0.05.

Population monitoring

Following the completion of the power analyses, we used QGIS (ver. 3.26.2; see http://qgis.osgeo.org; QGIS Development Team 2024) to randomly plot quadrats across TNR. We excluded an area of 28 ha that was considered to be unsuitable for T. adelaidensis (e.g. rocky hilltops, driving tracks, roadside verges). On the basis of the results of the power analyses, we surveyed 25 50 × 50-m quadrats in the remaining 53 ha of suitable habitat, which were marked with permanent iron posts to ensure consistency of area searched in subsequent years (Fig. 2). We thoroughly searched quadrats for suitably-sized spider burrows as described above. When detected, we classified each burrow by size (>16-mm diameter for adults; 10–16-mm diameter for juveniles and subadults), type (lycosid or mygalomorph), and inhabitant (lizard, spider, other invertebrate, or unoccupied) (Milne and Bull 2000).

Fig. 2.

A gridded layout of the 25 quadrats at the Tiliqua Reserve. Quadrats were randomly placed within the core pygmy bluetongue habitat. Each grid cell represents 50 × 50-m. Photo: Jack Bilby.

Results

Power analyses

It was possible only to detect the 25% declines in the high-density simulations within the 2-week survey period (Table 3). In this treatment, there was a significant difference among the quadrat sizes, with medium quadrats being significantly more effective at detecting 25% declines than were the large quadrats (Z = 2.769, P = 0.017). There were significant differences found in the mixed-density simulations, where both small and medium quadrats required fewer days to reach the β and significance thresholds.

Table 3.The mean number of quadrats and survey days (both rounded to the next whole number) required to achieve a power of 0.8 and p-value of 0.05 for the mean effect size.

Quadrat size (m)	Quadrat type	25% decline		50% decline
		Quadrats	Days	Quadrats	Days
30 × 30	Low density	3857 (CI 932–8645)	643 (CI 156–1441)	67 (CI 40–95)	12 (CI 7–16)
	Mixed density	203 (CI 51–355)	41 (CI 11–71)	33 (CI 28–37)	7 (CI 6–8)
	High density	31 (CI 11–51)	16 (CI 6–26)	5 (CI 5–6)	3 (CI 3–3)
50 × 50	Low density	97 (CI 67–127)	33 (CI 23–43)	38 (CI 13–62)	13 (CI 5–21)
	Mixed density	81 (CI 44–118)	33 (CI 18–48)	25 (CI 21–29)	11 (9– 12)
	High density	6 (CI 6–6)	6 (CI 6–7)	4 (CI 3–4)	4 (CI 4–4)
100 × 100	Low density	57 (CI 5–109)	58 (CI 5–110)	12 (CI 7–17)	12 (CI 7–17)
	Mixed density	103 (CI 63–142)	131 (CI 80–181)	19 (CI 14–24)	25 (CI 19–32)
	High density	4 (CI 3–4)	11 (CI 10–12)	3 (CI 3–3)	9 (CI 9–9)

95% confidence intervals are shown in parentheses.

Although the Kruskal–Wallis test found significant differences in the low-density simulation, no quadrat size was capable of detecting 25% declines within the 2-week survey period and no significant differences were found in the adjusted pairwise comparisons (Table 4, Fig. 3). All three quadrat sizes were able to detect 50% declines in every simulation within the 2-week survey period. However, there were significant differences between the number of days required to detect 50% declines for the mixed-density and high-density simulations, with large quadrats taking significantly more days than both small and medium plots (see Table S3 for details).

Table 4.Results from Kruskal–Wallis tests comparing the differences in the number of days required to reach a power of 0.8 among quadrat sizes across different treatments.

Density	25% decline	50% decline
Low density	H 2,29 = 7.443, P = 0.024	H2,29 = 0.111, P = 0.946
Mixed density	H 2,29 = 14.666, P = 0.001	H 2,29 = 22.408, P < 0.001
High density	H 2,29 = 7.760, P = 0.021	H 2,29 = 24.351, P < 0.001

Significant differences (P < 0.05) are shown in bold.

Fig. 3.

The number of survey days required to detect decline for all simulated scenarios: (a) low density – 25% decline, (b) low density – 50% decline, (c) mixed density – 25% decline, (d) mixed density – 50% decline, (e) high density – 25% decline, and (f) high density – 50% decline. Pairwise relationships are indicated by labels; n.s., P > 0.05; *P < 0.05; **P < 0.01; and ***P < 0.001.

Discussion

Optimising survey designs offers substantial benefits to conservation management and monitoring of threatened species. In this study, we found that 50 × 50 and 100 × 100-m quadrats consistently detected 25% and 50% declines in simulated populations of the endangered T. adelaidensis. On the basis of the fewest survey days, we determined that 25 replicate 50 × 50-m quadrats allowed for the minimum effort required to achieve our designated levels of power and significance when surveying a 1-km² site, with the objective of detecting significant population declines of T. adelaidensis. We then set this number of quadrats as the goal for surveying the 53-ha TNR, so as to confidently sample beyond the minimum requirement. This experimental a priori analysis allowed us to confidently initiate an on-ground, site-wide monitoring project to estimate population size and subsequent temporal trends.

This project represents the first site-wide survey of T. adelaidensis, and the first informed population estimate for TNR. With TNR representing the only dedicated conservation reserve for this species, this study is valuable for practitioners and land managers working to conserve the species here and on other properties. Without ongoing and standardised surveys, understanding population dynamics of threatened species is substantially limited and conservation intervention may be delayed or ineffective (Lindenmayer et al. 2013; Robinson et al. 2018; Scheele et al. 2019). It is paramount to ensure that surveys are optimised so that resources are used efficiently, and that survey effort can obtain an accurate representation of a population and its characteristics (Marsh and Trenham 2008; Moore et al. 2023). The results of future annual surveys and corresponding vegetation assessments can be used to investigate temporal changes in detectability and develop a more optimised study design for population modelling. Whereas no population thresholds for conservation intervention currently exist for the pygmy bluetongue at TNR or other properties, the results of this project are the first step to allow for these to be developed. These a priori methods and baseline surveys will also set the framework for future annual monitoring at other properties and for other cryptic and threatened species, and for further work into environmental predictors of T. adelaidensis occupancy, more robust population estimations, and to identify critical population thresholds and declines that are detrimental to persistence of the population.

Although moderate and extreme declines were detectable with select methods, we were unable to confidently detect slight declines with any survey design or at any population density. We were also unable to confidently identify moderate declines in low-density populations. This has ramifications for the implementation of management, as expeditious and adaptive conservation action has greater chance of being successful (Martin et al. 2012; Walls 2018). Additionally, low-density populations are at greater risk of extirpation than are high-density populations because of reduced genetic and phenotypic diversity and increased vulnerability to stochastic events (Pimm et al. 1988; Griffen and Drake 2008). To gain a more holistic view of population trends, we suggest that conservation practitioners working on cryptic and threatened species use multiple complementary methods to identify declines, such as assessing changes in occupancy as well as quadrat density and broader population estimates (Bried and Pellet 2012; Einoder et al. 2018; Dennis et al. 2019).

In this study, volunteers were instrumental to achieving the survey goals in the restricted timeframe allocated to the survey and thus statistical power was considerably increased. With the help of volunteers, we were able to survey 25 quadrats in five days, rather than the eleven days that was predicted with just two observers. However, there is an inherent trade-off with inexperienced volunteers, as greater observer experience typically improves detection ability, and this is likely to be exacerbated in cryptic species (Garel et al. 2005; Fitzpatrick et al. 2009; Moyer-Horner et al. 2012; Albergoni et al. 2016). This was apparent during TNR surveys where false positives (e.g. flagging of burrows that were too small or flagging ant nests) occurred multiple times throughout the early survey activities. These issues were largely remedied by having experienced surveyors present and working in large groups, which enhances sampling accuracy (Schmeller et al. 2009; Albergoni et al. 2016). As such, with the addition of a brief training induction to discuss false positives and negatives (Newman et al. 2003), the benefit of substantially increasing coverage, and thus statistical power, outweighed any time taken to correct errors made during the survey period.

Effective conservation of threatened species relies on the ability of conservation practitioners to confidently detect declines and respond accordingly. This investigation has demonstrated the effectiveness of using 50 × 50-m quadrats for monitoring population trends in T. adelaidensis to achieve high statistical power within a feasible survey timeframe. The value of volunteers in the completion of on-ground monitoring cannot be overstated, although management of potential errors must be accounted for in any monitoring plan engaging inexperienced participants. Future monitoring should continue using and refining the approach described here to ensure accurate detection of population trends, enabling timely conservation actions. The processes applied in this investigation can also be applied to similar cryptic and threatened animal species, demonstrating the broad-scale application of such a priori methodology. This research offers significant insights into ecological monitoring methodologies, highlighting the importance of tailored survey designs for the effective conservation of cryptic and threatened species.

Supplementary material

Supplementary material is available online.