Site descriptions and survey methods

This study was conducted across multiple wetland sites, mostly in south-eastern Australia. The sites were selected from a larger national monitoring program on Latham’s Snipe. This ongoing citizen science program commenced in 2014 with about twenty sites in south-west Victoria [19]. The program focuses on determining patterns in distribution and abundance across the species’ Australian range. New sites have been added to the program over time, usually informed by local knowledge. The sites vary greatly in size, extent, topography, elevation, permanence (e.g., permanent artificial, natural ephemeral), water salinity, vegetation type and structure, and landscape context.

Latham’s Snipe surveys were conducted at wetland sites, initially over five consecutive months (October 2014 – February 2015), then three times a season (in mid-September, mid-November and mid-January) from 2015 onwards. The timing of surveys targets key movement and stationary periods across the species’ southerly range. All sites were surveyed on the same day (between sunrise and sunset) to reduce the likelihood of double-counting birds moving between areas overnight. The survey method followed Naarding [20]; observers walked a line transect slowly (1 - 2 km/h) through the site, pausing periodically (for 5 - 30 seconds), and counting birds as they flush. The frequency of pausing depends on vegetation thickness, and may be every minute or so in more thickly vegetated habitats. This ‘walk-pause-walk’ method was adopted because snipe are more likely to flush when a potential predator (i.e., walker) stops, and is the only reliable way to detect snipe [21]. Indeed, frequent observations were made of birds flushing from behind a stationary observer ( < 5 m away).

Observers estimated the distance (m) between themselves and the location where each bird flushed. This “flush distance”, is analogous to flight initiation distance (FID), being the distance between an animal and approaching human when escape is initiated [7]. Observers also estimated vegetation heights (by eye) during their surveys.

The Latham’s Snipe Project is conducted under Federation University Animal Ethics Committee approvals 15-005, 18-004 and 23-005.

Site selection for FID analyses

Latham’s Snipe surveys were conducted at over 350 national survey sites, however FID was not recorded during all surveys. Sites with reliable FID records were selected based on the following: (a) experience of the observer (i.e., observers that had been involved in the survey program from early in its inception or were otherwise suitably experienced); (b) continuity of the data (representativeness across the survey program, to avoid survey data clustered in a single time period), (c) landscape context (both urban and non-urban sites were included where sufficient data were available; see below); (d) FID records reported as single values (i.e., to remove reported ranges such as when a group of birds flushed at, for example, “10 - 20 m”); and (e) the total number of FID records (set at a threshold of >  20).

Sites were classified into either urban or non-urban adjacent land uses. Classification was done by spatial intersection of point coordinates (representing site centroids) with the Urban Centres and Localities 2021 (polygon) spatial dataset [22] in QGIS. Sites were categorised as non-urban when they fell within the classification “Rural Balance” and urban for all other classifications.

Observers were assigned unique codes either as individual surveyors or as a group. Group codes were associated with the most experienced observer in a group. If a new observer joined an existing individual or group, the existing observer code was used. This coding process recognised that the most experienced observer would influence the estimation of an FID. Where observers included pairings or groupings of experienced people who previously had their own observer code, this pair or grouping of observers was assigned a new code.

Data analyses

Variation in raw FID values between wetland site and observer code was explored initially by plotting. Based on variation in plots, a negative binomial generalised linear mixed model was fitted with observer code and site as fixed effects, and land use as a random effect. Negative binomial models were calculated using the package glmmTMB [23] in R [24]. Model fit was assessed by measures of overdispersion and residual plots. On the basis of model results, raw FID values were resampled with replacement using 10,000 bootstrapped replicates, grouping values based on site and observer code. Resampling was used to generate confidence intervals, while grouping by the main sources of variance (i.e., site and observer; see Results).

FIDs, estimating ADs and therefore buffer widths

Two metrics were initially calculated (1) 95^th percentile of raw FID, and (2) adjusted 95^th percentile (i.e., 95^th_StDev =  mean +  1.645 ×  standard deviation) of raw FID (after [5]). These have been used in previous studies to determine buffer widths, usually for non-cryptic species.

A variety of approaches can be used to infer AD from FID in cryptic shorebirds, and while none are unambiguously superior and validation is lacking, they offer an alternative to the metrics above that we consider more suitable for cryptic species. Here, we use two inferential approaches for estimating AD in Latham’s Snipe. First, a small sample of both AD and FID records for Latham’s Snipe (n =  8) obtained from a different study (D.T. Blumstein, unpubl. data) were analysed with a simple linear regression, and the regression results used to predict AD from our bootstrapped FID estimates. Second, we used linear regression to understand the relationship between AD and FID across eight scolopacid shorebird species during the non-breeding period in Australia (D.T. Blumstein and MAW unpubl. data): Limosa lapponica (n =  95), Calidris ferruginea (n =  18), C. ruficollis (n =  180), C. acuminata (n =  79), Numenius madagascariensis (n =  49), N. phaeopus (n =  13), Tringa nebularia (n =  21) and T. brevipes (n =  47). We choose these species because they are confamilial with snipe; we are not interested in the evolution associations of FIDs, thus we do not use phylogenetic controls. Two response variables were modelled: 1) the raw 95^th percentile (FID) and 2) the adjusted 95^th percentile (FID). Based on the linear model coefficients for each of the two FID metrics against AD, we then used the Latham’s Snipe bootstrapped FIDs from the present study to predict ADs.

These two inferential approaches produced three estimates of AD. Estimate 1 was based on the observed relationship between AD and FID for Latham’s Snipe. Estimate 2a was based on the multi-species modelled 95^th percentile of AD against FID (excluding Latham’s Snipe), and predicted Latham’s Snipe AD from bootstrapped FID values, and Estimate 2b used the same approach but on adjusted 95^th percentile (bootstrapped FID values). These three estimates were derived separately for each wetland site and for all data combined. Bootstrapped FID values enabled confidence intervals of AD to be generated (5%, 80%, 90% and 95%). All reported distances are in metres.

Estimating AD

AD was highly correlated with FID in the reference study despite the small sample size (R² =  0.98, Adj. R² =  0.97, SE =  1.4, unstandardised slope 1.1 ±  0.1, mean ±  SE; S1). Using the reference study data, a median estimated AD of 44.6 m (5% CI =  16.4 m and 95% CI =  84.7 m) was calculated from the bootstrapped FID in the present study (Estimate 1 in Table 2).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Overall estimates of AD, coefficient of variation (COV) and range (maximum minus minimum) across all estimates of AD (and 5, 80, 90 and 95% confidence intervals) based on 10,000 bootstrapped replicates of raw FID. Estimates in bold indicate values deemed suitable for determining disturbance buffer widths, depending on the level of confidence of the decision-maker.

https://doi.org/10.1371/journal.pone.0317081.t002

There was a very strong relationship between FID and AD across Scolopacid species (see Fig 3). Fig 3 shows means, and includes the small sample of Latham’s Snipe which clearly falls close to the linear predictions of the other taxa (R² =  0.97, Adj R² =  0.96, SE =  6.7, slope 1.5 ±  0.1). Two linear regression models for all species excluding Latham’s Snipe were: AD versus 95^th percentile FID (R² =  0.98, Adj R² =  0.98, SE =  8.6, slope 1.4 ±  0.1), and AD versus adjusted 95^th percentile FID (R² =  0.96, Adj.R² =  0.95, SE =  11.8, slope 1.4 ±  0.1). These models were used to generate median AD estimates of 56.1 m (Estimate 2a) and 61.6 m (Estimate 2b), together with the coefficient of variation of those estimates (Table 2). Bootstrapped AD estimates plus the 5/ 95% confidence intervals for individual sites are given in Table 1, and those estimates at a larger range of confidence intervals are provided in S1 File. Estimates 2a and 2b were the largest of the three, ranging from 75.7 m (Est.2a) and 80.6 m (Est.2b) at a confidence of 80%, to 108.0 m (Est.2a) and 94.9 m (Est.2b) at a confidence of 95%. As the sample size underpinning Estimate 1 was very small, and its use would involve extrapolating well beyond the bounds of the data, we recommend using Estimates 2a and 2b as the basis for informing appropriate buffer widths to avoid disturbance of snipe.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Regression line of mean (

 ± SE) flight initiation distance in metres (FID) versus mean ( ± SE) alert distances in metres (AD) for nine shorebird species (unpubl. data). This illustrates the strength of these relationships, and the consistency between this relationship and the small sample of Latham’s Snipe available. Separate regressions to estimate AD from FID excluded the small sample of Latham’s Snipe, and used raw or adjusted 95^th percentiles.

https://doi.org/10.1371/journal.pone.0317081.g003

Use of buffer width recommendations

In this study, we use estimates of AD to provide a range of buffer width recommendations we consider suitable for reducing disturbance impacts on Latham’s Snipe. Our estimates of AD are derived from an extensive dataset of Latham’s Snipe FID collected from a range of sites across south-eastern Australia, suggesting at least some generalisation is reasonable. We also suggest that the estimates we present converge to a reasonable extent; the largest difference between our three estimates was 23 m (lowest COVs were only 12%). We recommend using the largest two of the three estimates (Estimates 2a and 2b), which were 75.7 m (Est.2a) and 80.6 m (Est.2b) at a confidence of 80%, and 108.0 m (Est.2a) and 94.9 m (Est.2b) at a confidence of 95%. This conservative approach reflects other recommendations about developing buffer zones (e.g., [7]) and produced buffer widths consistent with other studies. For example, a study on shorebirds using staging sites in North America determined buffer widths for nine species ranging from 61 m for smaller species like Least Sandpiper Calidris minutilla (approximately 30 g) to 186 m for larger species like Grey Plover Pluvialus squatarola (~160 - 275 g) [15].

The resampling approach used here allowed the generation of AD estimates that account for variation in raw FIDs attributable to site and observer effects, and the production of confidence intervals. These confidence intervals allow decision-makers to decide what risk they wish to take when making buffer width decisions to address potential impacts of human proximity associated with land development on snipe. A buffer width based on the 95% confidence interval will reduce the risk that snipe are negatively affected by disturbance, but will require larger areas to be protected as habitat. Therefore, buffer designations represent a trade-off and managers must decide what level of risk they consider acceptable when implementing these widths.

Under the Environment Protection and Biodiversity Conservation (EPBC) Act, a precautionary principle to land development is advised in instances where there is uncertainty about the potential for significant impacts on matters of national environmental significance such as snipe. Consistent with this principle, we use estimates of AD to provide a more conservative measure than FID. This is common practice [9] and allows for the possibility that the disturbance effects on Latham’s Snipe occur at greater distances than FID [7]. From a management perspective, however, widths recommended here should be considered a minimum when designing buffers to mitigate disturbance [5]. If buffers are altered in some way (e.g., intersected with walkways, creating “buffer creep”), then widths may need to be increased, and human activity may require additional restrictions [25,26]. Larger vegetated buffers are also more effective for other management purposes such as pollution and nutrient retention, and provision of habitat for other species [27,28].

Uplisting of Latham’s Snipe to Vulnerable under the EPBC Act was underpinned by improved knowledge about contemporary threats to the species, the most notable being habitat loss caused by land development, which brings with it disturbance [6]. The application of appropriate setbacks will be a mitigation tool for reducing significant impacts caused by land development. At present, setbacks for species like Latham’s Snipe are determined arbitrarily, if at all. Here we present quantitative data to inform the mitigation of significant impacts that may result from proposed land use developments referred under the EPBC Act. This approach could equally be applied to other cryptic species in the EPBC Act Significant Impact Guidelines [14], and information exists for other species [12] that could support the development of a database for recommended buffer widths.

Caveats, potential improvements and opportunities for refinement

There are a number of caveats with the approaches used here. First, FIDs were estimated by eye, and there was variation in FID estimates among observers. Ideally, FID should be measured using laser range-finders to improve the accuracy (given that pacing is impractical for snipe; [3]). Many birds including shorebirds may exhibit non-flying escape responses such as walking or running which are adjusted in relation to the nature of the approach threat [29]. Non-flight responses were not detectable in the thick vegetation used by Latham’s Snipe, and may have occurred at distances which exceeded our measurement of FID. The nature of snipe responses to an approaching threat would ideally be investigated, perhaps using cameras on focal birds [8] or radio-tagged individuals [30].

Site-based context may influence response distances of snipe. The FID-AD calculations assume that detectability of snipe was uniform within and across sites. This is unlikely, as flushing distances varied over sites, which in themselves varied in vegetation structure and height (Table 1). We speculate that this variation between sites reflects major structural differences in vegetation at each the affects snipe alert responses. For example, sites that experienced high levels of spring rainfall often had exceptional vegetation growth, resulting in tall vegetation with thick cover – where this occurred, flushing distances appeared to be shorter (B. Hansen, pers. obs). Conversely, in sites that have shorter vegetation like Fox and Pub Lakes (estuarine sites with generally low saltmarsh: Table 1; Supplementary material S1 File), snipe may flush at longer distances, presumably due to their perception of cover and visibility of oncoming threats. However, this pattern does not appear to be consistent and determining the relationship between vegetation characteristics and flushing distance would require collection of systematic detailed vegetation data at the exact locations where birds flushed. Applying this approach at a large scale to multiple volunteer observers was not possible. Few volunteers had the skill, time, equipment or inclination to collect laser range-finder FID and structured vegetation data when conducting surveys (B. Hansen, pers. obs.). Further study could investigate detection probability across a range of habitat types and survey conditions to determine relationships between snipe flushing observations and presence, and whether this is affected by environmental factors [31]. Nevertheless, we chose a conservative approach through recommending use of the longest AD estimates (at 80 - 95% confidence intervals), which have been generated across multiple sites of varying size, vegetation height, and structure, to allow for the possibility of uncertainty caused by imperfect detection.

Any contexts which alter response distances may alter suitable buffer recommendations of snipe. We evoked snipe flushes by walking through habitat, but many other bird species may exhibit longer FIDs to other stimuli, such as dogs [32,33]. Pre-migratory restlessness and other temporal events may also influence FID [34], but our estimates deliberately span the non-breeding period (so exclude pre-migration periods). We also note that in general, buffer designations are required for implementation year round.