The development of an improved scat survey method for koalas (Phascolarctos cinereus)

As an iconic species in Australia, koala (Phascolarctos cinereus) populations have been declining rapidly over recent decades (Smith and Smith 1990; Lunney et al. 2002; Preece 2007; Seabrook et al. 2011; Hanger et al. 2017), with their conservation status downgraded to ‘Vulnerable to Extinction’ by IUCN in 2016 (Woinarski and Burbidge 2016). The situation was much worsened by the catastrophic bushfire crisis in 2019 where allegedly thousands of koalas were killed and millions of hectares of their habitat burnt (Zhou 2019). Therefore, effective and practical survey methods are critical in the understanding of koala abundance and distribution and hence the evaluation of their conservation management. Various methods are available to investigate free-ranging koala populations in terms of their abundance and/or distribution. A count of these animals by strip/line transects is a typical direct survey technique (Gordon et al. 1990; Mitchell and Martin 1990; Melzer and Lamb 1994; Dique et al. 2001, 2003). Given the cryptic nature of koalas, however, searching for them in the wild is extremely time consuming and the results can be disappointing, with as low as 23% spotting rates for koalas (Hanger et al. 2017), especially in inland regions with sparse koala populations and low density (Melzer 1994; Melzer et al. 2000; Sullivan et al. 2004). To overcome these limitations, censuses based on indirect signs of koala activity, especially koala scats, are widely used.

Koala scats are quite distinctive from most other Australian mammal scats due to their oval shape often with a pointed end and slightly ridged surface, and they can persist in the field for several months under dry conditions (Cristescu et al. 2012). As a result, scat surveys have been a critical part of koala distribution studies (Rhodes et al. 2006; McAlpine et al. 2008), population density/abundance surveys (Sullivan et al. 2004), tree species preference studies (Phillips and Callaghan 2000; Phillips 2000) and dietary analysis (Ellis et al. 2002). This paper focuses on the scat survey techniques used for koala presence/absence surveys that give binomial data to determine koala presence in an area. For this purpose, there are two main methodologies that are broadly applied in the field: the Spot Assessment Technique (Phillips and Callaghan 2011) and the Koala Rapid Assessment Method (Woosnam-Merchez et al. 2012).

The Spot Assessment Technique (SAT) was initially established in the Koala Habitat Atlas project (Phillips and Callaghan 2000; Phillips et al. 2000) in 2000 and refined and published in 2011 (Phillips and Callaghan 2011). Each SAT survey site comprises 30 trees including a centre tree central to the nearest 29 trees. After a survey site is chosen, the centre tree is subjectively selected by the following criteria: (1) any tree with koala scats around, and/or (2) any tree with a koala spotted using it, and/or (3) any tree considered important for koalas, such as a food tree. For all 30 trees, a consistent search effort of two person-minutes per tree is conducted within a fixed 1-m radius around the tree base. Each search ends when either a koala scat is found or the designated time is up.

SAT has been widely used in large regional areas with unknown koala presence (Phillips and Callaghan 2000; Phillips et al. 2000). However, by allocating a fixed search time (two person-minutes) for all trees, SAT overlooks the variety of tree sizes and density, which influences the search area, the variety of ground vegetation and koala scat density, all of which can substantially influence the detectability of scats (Cristescu et al. 2012; Woosnam-Merchez et al. 2012). Other issues have also been raised about the potential bias when subjectively choosing the centre tree of each site (Dique et al. 2004). Woosnam-Merchez et al. (2012) argued that the subjective selection of the centre tree breaks the principle of randomness and so can lead the survey effort into areas with a higher chance of koala presence, making the results unreplicable by other surveyors. Furthermore, given that SAT only searches the area within 1-m radius around tree bases, it was criticised for its likelihood of missing koala scats (Lunney et al. 2000; Woosnam-Merchez et al. 2012), because their distribution around koala roosting trees tends to be patchy rather than a gradual radial pattern (Ellis et al. 1998).

In order to overcome the issue of subjective centre tree selection, and the possibility of missing scats, another method, the Koala Rapid Assessment Method (KRAM), was developed based on SAT in 2012 (Woosnam-Merchez et al. 2012) and applied in the field (Cristescu et al. 2015). Two key modifications distinguish KRAM from SAT. These are: (1) random selection of the centre tree, and (2) search effort is not limited within a fixed radius around each tree base but includes all the ground area covered by the tree canopy.

With these modifications KRAM minimises the biases from SAT of the subjective centre tree selection and missing scats outside the tree base area. However, similar to SAT, KRAM does not address the heterogeneity of scat detectability because it also requires predetermined consistent search effort for all trees. This bias was acknowledged by KRAM, and to overcome this limitation it suggests that search time should not be strictly restricted and be proportional to scat detectability affected by various substrates. As a standardised survey protocol, this is far too vague to be applied in the field with consistency. Little was known about the impact of substrates on scat detectability (Cristescu et al. 2012), and thus the predetermined, proportionated search time may vary widely between different surveyors. Similar to the subjective centre tree selection bias in SAT, this arbitrarily determined search time in KRAM is biased and would subsequently lead to an unreplicable result.

Furthermore, searching all the canopy area of 30 trees in one survey site requires considerable time, especially for large trees. For example, it can take, on average, 15 min to complete a thorough scat search in a 1 m by 5 m plot with complex litter (Cristescu et al. 2012), let alone a search area more than 100 m² under a large tree canopy, which is not uncommon in some koala habitats. Thus, it is quite impractical to achieve an accurate result with this method in the field, particularly when studying large areas of land. In other words, KRAM compromises the benefit and purpose of scat surveys – the searching of vast areas with affordable time effort.

The limitations of SAT and KRAM related to varying scat detectability have been recognised but limited quantitative analysis has been made to evaluate this. Cristescu et al. (2012) reported that the time needed to find the first prescattered scat within a plot was always within 2 min and not associated with ground vegetation complexity. This has plausibly justified the validity of SAT where a fixed 2 min of search time is allocated to each tree and only the first scat is needed. However, this result has never been tested in the natural habitat of koalas in the wild and, as mentioned in their paper, their results may be biased because no ‘settling time’ was given for scats to be naturally covered; and the number of prescattered scats was probably too high (2.7 m^–2) – both can incur significant bias on experimental results by making the first scat much easier to find.

Scat surveys, as an indirect population census technique, play a fundamental role in koala ecology research, but the limitations of existing methods demand that a new koala scat survey method with improved accuracy and feasibility be developed. We:

1. developed and field tested a new balanced koala scat survey method (BKSS) and give details of our theoretical justifications; and

2. quantitatively analysed the weakness of previous methods and determined the improvement of BKSS on koala scat survey accuracy.

To overcome the potential biases and inefficiency of SAT and KRAM, we developed a Balanced Koala Scat Survey method (BKSS). The methodology of BKSS is based on SAT while incorporating the strength of both SAT and KRAM. The BKSS field protocol is described in detail below:

1. GPS coordinates are generated by predetermined random selection, then the nearest tree is located and marked as the centre tree of the survey site. As in SAT or KRAM, a tree is defined as having ‘a live woody stem of any plant species (excepting palms, cycads, tree ferns and grass trees) which has a diameter at breast height (DBH) of 100 mm or greater’ (Phillips et al. 2000: p. 2). This definition applies to all the trees in this study.

2. Thoroughly search the 1-m radius around the centre tree base with no time limit. In order to standardise the search procedure throughout the entire survey, conduct the first-round quick search to look for any obvious koala scats without disturbing the ground surface and vegetation. If no obvious koala scats are found then conduct the second-round thorough search by closely searching and, when necessary, lifting up leaf litter and removing vegetation to the level where potential scats have no space to go any deeper, such as on the ground. The search ends when either: (a) a koala scat is found and confirmed, or (b) no scats are found after a thorough search.

3. Record scat search results and move to the next nearest tree to the centre tree. Repeat the search procedure (Step 2) until all the nearest 29 trees are searched within a survey site. Record scat search results for each tree and mark trees with flagging tape when finished searching.

When conducting BKSS, relevant field data should be recorded as per study purposes, e.g. tree species, height, DBH, GPS coordinate location and ground cover category. If there are trees with overlapping scat search area, i.e. the distance between tree trunks is less than 2 m, only one tree should be included in order to avoid potentially double-counted scats.

BKSS can be viewed as an improved variation of SAT or KRAM that integrates the advantages of those two methods while minimising their weaknesses. The key differences between SAT, KRAM and BKSS are shown in Table 1.

Table 1.  Key differences between SAT, KRAM and BKSS as koala scat survey methods

The three key features of BKSS are: (1) random selection of the centre tree, (2) tree base search, and (3) unlimited search time. These features provide increased scientific rigor when investigating koala populations and its efficiency in the field:

1. Random selection of the centre tree instead of manual selection in SAT. The manual centre tree selection in SAT has two scenarios that are: (A) selecting a tree with scats around it, or there is a koala in the tree; (B) selecting a tree of high importance to koalas (e.g. koala fodder or roost trees) if Scenario A is not applicable. In Scenario A, manual selection appears plausible if conducted consistently as it increases the search accuracy. However, this means that a preliminary search is required before selecting a centre tree, either for scats or koalas. Consequently, this means that one needs to search first before a decision can be made as to whether to search for scats. Moreover, this preliminary search alone can be considerably time consuming and compromises the efficiency of the survey in the field. Scenario B is considered more common in the field, especially with the low density of koala populations in many parts of Australia. Unfortunately, the concept of a tree with high importance to koalas, i.e. known fodder or roost trees, is too vague to be consistently applied in the field. Koalas are known to feed on eucalypts, Melaleuca and Lophostemon species as well as to use various Acacia trees and other native bushes in which to rest (Handasyde et al. 1990; Sullivan et al. 2003; Preece 2007). Significant observer bias could arise by surveyors’ respective opinions of ‘importance’, which makes the result unreplicable (Woosnam-Merchez et al. 2012). Therefore, random selection of the centre tree in BKSS is necessary to both minimise observer bias and increase time efficiency in practice.

2. Tree base search instead of canopy area search in KRAM. To avoid potentially missing scats when only searching around the tree base, KRAM involves searching the area on the ground covered by the whole tree canopy. This reduces the chance of false negative outcomes, but, as mentioned above, is biased due to consistent search effort and in practice is quite time consuming and hence compromises its practical value. On the other hand, since only a few scats are needed to confirm koala occupancy, the tree base search can be legitimate if there is a high chance of any scats falling within the tree base area. In their koala scat distribution study, Ellis et al. (1998) found that 71% of all scats were dropped by koalas within a 4-m² quadrat around the surveyed tree base; while Phillips and Callaghan (2011) found 47% of scats were found within a 1-m radius (~4 m² depending on tree size). Both of these results indicate a significant proportion of missed scats (>30%) outside the search range. However, in the in-depth research of free-ranging koalas by Sullivan et al. (2002), tree base search (0.3-m radius around the tree) was shown to be highly reliable to identify trees with koala scats, with less than 7% false negative rates (scats were found within the tree canopy area but not in the tree base area) in most landform communities (i.e. Residue, Plains and Flood Plains) and 13.5% in riverine communities. Indeed, this failure rate could be even lower when BKSS searches the area in a 1-m radius, instead of 0.3 m, around the tree base, which can yield an eight-times-larger search area on a tree of 20 cm diameter. Those conclusions about tree base searches from the above studies, i.e. relatively low proportions of total scat deposition in search area versus high success rate to indicate scat presence, do not necessarily conflict with each other. It suggests that even though the proportion of scats deposited outside the tree base search area can be high (>30%), it is likely (~90%) that at least one scat will be found within the tree base range when any scats are found underneath the whole tree canopy area. In other words, as long as a definite number of scats is not required, a tree base search (1-m radius) for koala scats to indicate koala presence has been shown to be valid and reliable.

3. Unlimited search time instead of consistent search time in both SAT and KRAM. The primary issue of SAT and KRAM is the predetermined consistent search effort for all the trees regardless of tree size, ground vegetation and scat density. First, given that SAT requires a search of 1-m radius around each tree base, different tree sizes can yield large differences in search area. For example, a small tree with a 0.1-m DBH has only a 3.45-m² search area whereas a larger tree with a 1-m DBH, which is not unusual in the study area, has 6.28 m², almost double the area. In the wild tree canopy sizes are likely to vary widely in the search area of KRAM. Second, ground vegetation cover, such as bare ground with or without some or a lot of dense vegetation, has been shown to be a major factor that directly influences scat detectability – the denser the vegetation cover the less chance of finding (and the longer time required to find) all scats (Cristescu et al. 2012). Finally, scat density can affect the detectability of scats as well, especially for the first scat: it is always easier to find the first scat when scat density is high. Thus, the varying search area, ground cover and scat density potentially bring substantial errors to SAT and KRAM where there is a relatively fixed search time for each tree. In BKSS, however, these biases would be greatly reduced, if not eliminated, by conducting thorough searches for all trees without a time limit.

With these three key features, BKSS was tested as described below.

This study was conducted from December 2017 to January 2019 on Old Hidden Vale (OHV), a 4858-ha property owned by the Turner Family Foundation, located ~7 km south of Grandchester and 30 km west of Ipswich in south-east Queensland. It has over 30% cleared land mainly used for cattle grazing, with most of the remainder covered with regrowth open eucalypt forest. The property contains typical koala habitat features including tree species favoured by koalas such as Eucalyptus tereticornis, E. crebra, E. melanophloia and Corymbia citriodora (Ellis et al. 2002; Sullivan et al. 2003; Melzer et al. 2014), and topography including hills, slopes, flat plains and riverine areas. It is considered high-quality koala habitat although, while koalas are often seen, little is known about their distribution, abundance or density across the property.

Koala scat surveys using BKSS, as described above, were conducted at OHV by one experienced surveyor, with 25 survey sites allocated by stratified random selection based on dominant Regional Ecosystems. Time spent on each search was recorded when either the first scat was found – denoted by Scat Search Time (SST) – or a thorough search was completed when no scats were found – denoted by Thorough Search Time (TST). By doing this, search effort within the first 2 min can be regarded as SAT, and hence the difference of efficacy between SAT and BKSS can then be evaluated by comparing survey results which were achieved within and over 2 min.

In addition, tree size and ground vegetation complexity were also recorded in this study. The tree sizes were estimated by measuring the DBH. As the tree base area in OHV was always covered by either tree leaves, bark, grass or shrubs, the ground vegetation cover around the trees were classified into four categories (Fig. 1): Leaf (L) – over 50% of search area covered by leaf litter; Bark (B) – over 50% of search area covered by fallen tree bark chips or slabs; Grass (G) – over 50% of search area covered by low grass less than 20 cm high; and High Grass (HG) – over 50% of search area covered by grass or shrubs (e.g. Lantana spp.) 20 cm tall or higher. The survey result of each site was expressed by Koala Activity Level (KAL), which represents the percentage of trees found with koala scats among the total 30 trees in one survey site. KAL has been shown to be positively associated with koala abundance where low/high KAL indicated low/high koala abundance of local area (Phillips and Callaghan 2011). We determined two levels of KAL based on the survey results: Low – KAL ≤ 33% (no more than 10 trees with scats out of 30 trees), and High – KAL > 33% (more than 10 trees with scats out of 30 trees).

Fig. 1.  The four ground cover categories, from left to right: Leaf, Bark, Grass and High Grass.

The hypothesis tested for this experiment was that ground cover complexity, tree size and KAL significantly affected the detectability of koala scats, in their natural environment, given a fixed search time. That is, SAT and KRAM is less likely to determine koala presence by yielding false-negative results when searching large trees and when ground cover complexity is high or scat density is low in survey sites.

The impact of ground cover condition and KAL on SST, as well as the impact of ground cover condition on TST, was analysed via ANOVA, with a 5% significant level. In order to reduce the influence of any outlier data and then skewed residual distribution, when conducting the ANOVA test, both SST and TST were transformed by common logarithm. The impact of tree size (DBH) on SST and TST was demonstrated by linear regression models. In addition, since SAT requires two person-minutes as search effort, scat trees with more than 2 min SST were also calculated to demonstrate the failure rate of SAT.

When no scats are detected, BKSS requires the user to conduct thorough searches to the extent that there are no scats in the area (as per BKSS searching protocol described above) before moving to the next tree. In this study this ‘search thoroughness’ was tested by a blind test in an attempt to recover a known number of prescattered scats. This also helps to test the effectiveness of BKSS in the field, i.e. whether BKSS can detect all the trees with prescattered scats. The methodology is similar to that of the scat detectability experiment of Cristescu et al. (2012), where a known number of scats were prescattered in plots of various ground covers and then a thorough search was conducted right away until the surveyor was confident that no more scats were left in the area. In this study, a few modifications were made.

1. All the test scats were marked with small red dots that could only be identified when checked closely. This avoided the bias caused by any possible pre-existing scats defaecated by resident koalas.

2. Tests were conducted in the tree base area of real survey trees. In this way, tests were site-specific rather than those in random arbitrarily selected test plots with ‘similar’ substrates to those in the real survey. Plus, by using marked test scats, a great amount of time was saved by completing the tests and the scheduled scat survey simultaneously without compromising either result.

3. There was a settling period of at least one week before searches were conducted after the test scats were deposited. This gave time for test scats to settle into the substrate (by wind or other factors, e.g. growth of grass) or be covered by newly fallen leaves and bark. However, this did not simulate the situation of older scats that may have lasted months in the wild.

In total, 71 trees, referred to as test trees in this study, with different ground cover categories (18 L, 15 B, 23 G and 15 HG) were randomly chosen from survey trees. A random number of marked scats, referred to as test scats, ranging from 0 to 30, were deposited in the 1-m-radius range around the test tree base. The surveyor was required to find all the test scats deposited as well as any potential natural scats dropped by resident koalas; that meant a thorough search was conducted for each test tree irrespective of whether natural scats were found or not. Search time was recorded for test trees when the first test scat was found, the first natural scat was found, if any, as well as the thorough search, which meant that test trees had both SST and TST analysed. As only one surveyor conducted the scat search across all survey sites potential biases between different surveyors were avoided.

In total, 102 trees were found with scats, known as scat trees. SST of scat trees ranged widely from 5 to 420 s, with mean SST of 93 s (s.d. = 90, s.e. = 8.9), and median SST of 68.5 s. Of those, 26 ± 4% s.e. (n = 27) SST were longer than 2 min (Table 2).

Table 2.  Time (s) required to detect the first scat (SST) under trees with koala scats across High and Low Koala Activity Level (KAL) sites, and the percentage of trees requiring SST longer than 2 min

The two-factor ANOVA test for SST showed that KAL has a significant impact on SST (F_1,94 = 10.02, P = 0.002). The mean SST of scat trees (n = 63) in High KAL sites was 73 s (s.d. = 83, range = 5–420 s), with 14 ± 4% (n = 9) requiring more than 2 min to detect the first scat. On the other hand, the mean SST of scat trees (n = 39) in Low KAL sites was 126 s (s.d. = 94, range = 5–369 s), with 46 ± 8% (n = 18) requiring more than 2 min to detect the first scat (Table 2).

Ground cover had no impact on time taken to find the first scat. Correlation analysis showed no correlations between tree size (DBH) and SST (r = 0.10).

In total, 635 trees were found without any scats, known as non-scat trees. TST of non-scat trees ranged from 60 to 600 s, with mean TST of 235 s (s.d. = 91), and a median TST of 210 s.

Ground cover complexity had a significant effect on TST (F_3,631 = 37.51, P < 0.0001). Post hoc pairwise t-tests showed that High Grass cover took significantly longer time to search for koala scats than all other ground cover categories, and both Grass and Bark took longer than Leaf (P < 0.05).

There was a moderate positive relationship between tree size (DBH) and TST (r = 0.33, P < 0.0001). Linear regression analysis of DBH and logarithmic TST resulted in an adjusted R² = 0.1157. The fitted relationship was:

(F_1,633 = 83.94, P < 0.0001).

Time to find the first test scat (SST) of test trees (mean = 21 ± 3.9, s.d. = 33, median = 10, range = 2–200 s) was significantly influenced by ground cover (F_3,67 = 2.956, P = 0.0386), with Leaf requiring a significantly shorter SST than Grass. The SST of 97 ± 2% (n = 69) of all test trees (n = 71) was no longer than 2 min; only 3 ± 2% of test trees (n = 2) had SST longer than 2 min. Test tree SST was not correlated with tree size (r = 0.08).

Time spent to find all the test scats, i.e. thorough search time of test trees (mean = 319 ± 10.6, s.d. = 89, median = 310, range = 110–600 s), was influenced by ground cover (F_3,67 = 3.956, P = 0.0117), with High Grass and Bark taking significantly longer time than Grass and Leaf. Test tree TST was weakly correlated with tree size (r = 0.27, P = 0.0238).

Of the total 71 test trees in this study, the overall test scat recovery rate (number of test scats found per number of test scats scattered) was 65.5 ± 2%. Test scat recovery rate for L, B, G and HG was 71.4%, 61.5%, 68.3%, and 60.2% respectively (Table 3).

Table 3.  Test scat recovery rate of test trees across the four ground cover categories

For trees found with a scat, Koala Activity Level of the survey site had a significant impact on scat searching time (SST). Ground cover and tree size showed no significant association with SST.

It took significantly less time to find the first scat at trees from High KAL sites, than at those from Low KAL sites. This difference was likely due to the heterogeneity of koala scat density across survey sites of various KALs. KAL indicates the proportion of trees found with scats among all 30 trees in a survey site. Higher KAL was associated with higher koala abundance (Phillips and Callaghan 2011), and thus higher koala scat density. Higher scat density in a given search area decreased the difficulty of finding the first scat, i.e. it was always easier to find the first scat when there were many scats in the search area. Provided the scat search time stopped when the first scat was detected, higher KAL led to a shorter search time, as shown in this study.

Ground cover complexity acted as a significant factor to search time where no scats were found around trees, as well as to both SST and TST of the test trees. However, for trees where scats were found, search time was not influenced by ground cover. Scat density could have contributed to this result. When there was a relatively high number of koala scats in the search area, the easiness of finding the first scat may have outweighed the variation of searching difficulty due to the different ground cover complexity. As a result, it always took a similar time to find the first scat no matter how complex the ground cover was. However, if the search needed to be continued to find all the scats, i.e. a thorough search, the significance of ground cover effects on search effort was demonstrated by the trees without scats and test trees with prescattered scats in this study. This is supported by Cristescu et al. (2012) where complete search time and scat detection rate were affected by litter complexities but not the time to find the first scat.

Searching time was correlated with tree size when no scats were found. This result makes sense because DBH determined the size of the search area and a bigger area took longer to search. However, in this study there was nearly no correlation (r = 0.10) between DBH and time to find the first scat. Similar to KAL, the reason for this could be due to koala scat density. Koalas prefer medium to big trees (Hindell and Lee 1987, 1988; Matthews et al. 2007) and thus a relatively higher scat density around big trees was expected. Small trees may have fewer scats deposited because they have smaller canopies than large/medium trees for koalas to use as fodder and shelter. This could lead to: (1) less reason for koalas to visit – the fodder leaves are easily consumed so that koalas will not revisit until new leaves grow, and are less preferred by koalas seeking thick browse to rest in; and (2) shorter duration of koala visiting – koalas have been found using small young trees often to feed in for a short time (15 min) before leaving (Hanger et al. 2017). Hence, given the lower likelihood and time of a koala visiting, a lower scat density around small trees is likely. As a result, in relation to SST, the lower scat density in smaller search areas, and the higher scat density in bigger search areas could have cancelled each other out, so SST was not affected by tree size.

From the results of this study and the findings of Cristescu et al. (2012) that searching time for the first scat was not affected by ground cover and tree size, consistent search effort used in SAT and KRAM seems reasonably valid because only the first scat is needed to determine koala presence (Phillips and Callaghan 2011). However, the significant impact of koala activity level on the time to detect the first scat has revealed the weakness of SAT and KRAM from which false negative errors rise when fixed search time is used for every single tree regardless of varying koala activities (Cristescu et al. 2012; Woosnam-Merchez et al. 2012).

BKSS has successfully overcome this weakness by using unlimited search time for each tree, and providing more accurate survey results. The direct comparison between SAT and BKSS in this study, demonstrates that 74% of scat trees require no more than 2 min to find the first scat. This means that, when applying the recommended 2-min scat search time, SAT yields 26% false negatives. In other words, SAT is only 74% accurate compared with BKSS. More importantly, since time to find the first scat is associated with koala scat density, this 26% failure rate of SAT can vary accordingly. As shown in this study, in survey sites with High KAL where the first scat was easier to detect, the failure rate of SAT dropped to 14%. This means that the bias of SAT can be lower when conducted in areas with relatively high koala abundance, e.g. the koala coast in south-east Queensland (Gordon et al. 1990; Dique et al. 2004). In contrast, in survey sites with low KAL where it was harder to find the first scat, the failure rate of SAT was as high as 46%. This means the bias of SAT can be higher when conducted in areas with relatively low koala abundance, e.g. inland Queensland (Gordon et al. 1990; Melzer 1994; Dique et al. 2004; Sullivan et al. 2004; Adams-Hosking 2017).

The results from trees with test scats in this study generally matched the results from the normal survey trees. The thorough search time (TST) of the test trees, which aimed to detect all the prescattered test scats, was associated with ground cover and tree size. Although ground cover had a significant effect on the SST of test trees, this was probably because of the low sample size of each ground cover category of test trees (~15 trees each).

In test trees, the success rate of identifying trees with test scats by BKSS was 100%, i.e. BKSS successfully identified at least one test scat in all 71 test trees. However, among the 71 test trees, two required longer than 2 min to find the first scat. That meant that the success rate of SAT to identify test trees was 97% compared with 100% for BKSS. Such a high success rate for first scat detection was probably due to the high numbers of test scats (Cristescu et al. 2012), which in this study was supported by the fact that, among the 18 test trees that appeared to have natural scats as well, at only one tree (6%) from the high KAL site was the first test scat found after a natural scat.

Test scat recovery rate also indicated the impact of ground cover on scat survey results. The overall test scat recovery rate, 66%, was similar to those of each ground cover category. This recovery rate was not only due to imperfect search but was also compounded by scat decay (Rhodes et al. 2011), which may cause up to 20% scat loss within a week (Cristescu et al. 2012). Nevertheless, since there was only one surveyor doing the survey, observer bias was able to be controlled. Most importantly, High Grass had the lowest test scat recovery rate (60%) while Leaf had the highest (71%). This indicates the varying degree of scat search difficulty across different ground cover complexities, and it matches the thorough search time (TST) analysis where High Grass/Leaf, as the most complex/simplest ground cover category, had significantly longer/shorter thorough search times than other ground covers. Similar conclusions were reported in the research of Cristescu et al. (2012) as well where test scat recovery rate decreased with an increase in ground vegetation complexity.

There were several limitations to BKSS and this study:

1. While providing a higher survey accuracy, BKSS could take a longer time, and subsequently have more labour costs, in the field than SAT, especially in low-koala-activity sites where thorough searches are needed for most trees. Compared with the fixed search time of 120 s in SAT, the thorough search time in BKSS averaged 235 s (s.d. = 91), which was less than 2 min extra time for each tree. This extra effort almost doubled the accuracy of survey results in the field where koala activities are low. In summary, the thorough search can make BKSS slightly more time consuming than SAT in the field (but still far less time consuming than KRAM); however, the significant improvement in accuracy by BKSS outweighs the slight compromise in efficiency.

2. By only conducting the tree base search, BKSS could incur false negative results due to the possibility that none of the scats dropped within the search area. This possibility is as low as less than 10%, demonstrated by the comprehensive research of Sullivan et al. (2002). Extra search effort for scats underneath the tree canopy is recommended to overcome this issue. Since a thorough search on the entire canopy-covered area for scats is impractical, alternative solutions, such as a quick scan over the surface layer or a thorough search on a random-chosen small plot (Lunney et al. 2000), should be considered. Future studies comparing the accuracies of BKSS and KRAM are recommended.

3. Although the association between KAL and koala abundance was suggested by Phillips and Callaghan (2011), KAL may not necessarily indicate the koala scat density level. For example, in one survey site, large amounts of scat may be deposited by koalas under a few favoured trees and hence incur a low KAL, whereas a few scats may be deposited under many trees and thus incur a high KAL. Future research is recommended to investigate the actual relationship between KAL and local scat density.

4. In this study the first 2 mins of BKSS was regarded as a SAT survey and the results of the two methods compared. This simulation may not be equivalent to a 2-min SAT survey, in the same search area, as the survey patterns may differ. This difference was, however, minimised by the standardised survey protocol of BKSS, i.e. a quick search for any obvious koala scats of the entire search area before a comprehensive search.

The Spot Assessment Technique and Koala Rapid Assessment Method as the most widely used koala scat survey techniques in the field, have been shown to be biased due to their fixed search effort for all trees surveyed for koala scats. Koala Activity Level, which represents the proportion of trees found with koala scats in a survey site, was shown to significantly influence search time for detecting the first scat, and this resulted in false negative results for SAT when search time went over 2 min. The failure rate of SAT was negatively associated with KAL: when KAL was low, SAT failure rate went up to 46%. Although both ground cover and tree size have impacts on thorough search time, time to detect the first scat was not affected. BKSS overcomes this bias by conducting an unrestricted thorough search around the tree base for every tree. In regional areas where koala density is low, BKSS has the potential to be especially powerful in determining koala presence when compared with SAT, which has a high failure rate in low-koala-activity areas.

Koala conservation management strategies are often made upon scat survey results, so a reliable and efficient koala scat survey method has great importance in koala conservation. Compared with SAT and KRAM, BKSS has been shown to provide results of higher accuracy than SAT, as well as being potentially more cost-effective by using less search time than KRAM. The application of BKSS in the field is expected to greatly enhance the reliability and feasibility of koala scat surveys in determining koala presence especially in low-koala-density areas, e.g. inland Queensland, and thus improve the efficacy of koala conservation management.

The authors declare no conflicts of interest.