METHODOLOGICAL INSIGHTS: Comparing the ... - Wiley Online Library

Journal of Applied Ecology 2004 41, 1185–1196

METHODOLOGICAL INSIGHTS

Blackwell Publishing, Ltd.

Comparing the precision and cost-effectiveness of faecal pellet group count methods D. CAMPBELL,* G. M. SWANSON* and J. SALES† *Strath Caulaidh Ltd, Algo Centre, Glenearn Road, Perth PH2 0NJ, UK; and †Biomathematics and Statistics Scotland, King’s Buildings, Edinburgh EH9 3JZ, UK

Summary 1. Faecal pellet group (FPG) count data are widely used to estimate animal abundance, with two alternative methods normally employed. Faecal accumulation rate (FAR) techniques measure the daily accumulation rate of pellet groups, while faecal standing crop (FSC) techniques measure overall density. To estimate abundance, both methods require estimates of the animal defaecation rate. FSC techniques also require an estimate of pellet group decomposition rate. In general, FAR techniques are considered less prone to bias while FSC methods are considered more precise and cost-effective. On balance, the majority of authors and practitioners prefer FSC methods, although little empirical evidence supports this decision. 2. FPG count data were obtained from 26 study areas to compare the precision of FAR and FSC count techniques when applied to wild deer populations in the UK uplands. The time needed to collect count data was quantified in 10 study areas. 3. The coefficients of variation (CV) of FSC pellet group count data ranged from 9% to 23% and were approximately 0·7–0·9 times those of equivalent FAR data. The precision of both methods was related to the density of pellet groups. On average, FSC count data took 80 min per plot to obtain, with FAR taking 1·6–1·9 times longer. 4. For the precision of FSC and FAR abundance estimates to be comparable in the range of conditions studied, decomposition rate trials would require a CV of 5–20%. While a number of studies report this to be possible, estimates of the time needed to obtain this level of precision generally exceed the net available time that results from the deployment of FSC rather than FSC pellet group counts. 5. Synthesis and applications. Using the levels of finance available to most deer managers in the UK uplands, deer abundance estimates obtained using FSC techniques on individual study sites up to 20 000 ha appear generally less cost-effective than FAR when compared in terms of their overall precision. As FAR methods are also thought to have less potential for bias when applied in the appropriate environmental conditions, they should be preferred over FSC when estimating deer abundance in concealing habitats. Key-words: deer management, dung counts, estimating abundance, optimal sample allocation, sampling costs, woodlands Journal of Applied Ecology (2004) 41, 1185–1196

Introduction Faecal pellet group (FPG) count techniques are widely used to estimate animal abundance in concealing habitats (Neff 1968; Barnes et al. 1995; Plumptre & Harris

© 2004 British Ecological Society

Correspondence: D. Campbell, Strath Caulaidh Ltd, Algo Centre, Glenearn Road, Perth PH2 0NJ, UK (e-mail [email protected]).

1995; Marques et al. 2001; Webbon, Baker & Harris 2004). In the UK uplands, the techniques are commonly used by deer managers to inform and support decisions relating to the level of annual culls (Bailey & Putman 1981; Mitchell et al. 1985; Ratcliffe 1987b; Mayle 1996; Marques et al. 2001). There are two categories of FPG count commonly used, faecal standing crop (FSC) and faecal accumulation rate (FAR) (Putman 1984; Staines & Ratcliffe 1987; Buckland 1992; Mayle 1996; Marques

1186 D. Campbell, G. M. Swanson & J. Sales

© 2004 British Ecological Society, Journal of Applied Ecology, 41, 1185–1196

et al. 2001) and issues relating to their relative accuracy, precision and cost-effectiveness are therefore critical. FSC techniques measure the density of pellet groups present and relate this to their decomposition rate or the mean time that they have been present (McClanahan 1985; Ratcliffe 1992; Laing et al. 2003). These data are generally obtained in a separate trial conducted prior to collection of FSC count data. FAR techniques, in contrast, measure the rate of pellet group accumulation between two points in of time (Neff 1968; Mitchell & McCowan 1979; Bailey & Putman 1981; Buckland 1992; Webbon, Baker & Harris 2004). Separate trials to measure the decomposition rate are not required as the techniques are applied when negligible decomposition occurs (Mayle 1996; Laing et al. 2003). For both techniques, animal abundance is calculated by adjusting faecal accumulation rates by the mean deer defaecation rate over the period of accumulation (Mitchell & McCowan 1979; Buckland 1992). There are differences in the relative accuracy, precision and cost-effectiveness of each technique, both in terms of FPG counts and conversion to abundance estimates (Buckland 1992). Pellet groups in the FSC system accumulate over a longer period of time than those in the FAR system (Ratcliffe 1987a; Buckland 1992; Mayle 1996). As a consequence, the FAR method may suffer from a higher number of zero counts and smaller sample size of pellet groups per unit effort than FSC, leading to poorer count precision in equivalent conditions (Ratcliffe 1987a; Buckland 1992). Although it may be possible to improve the precision of FAR counts by leaving plots out for longer, bias from intermediate decay could affect the data obtained (Buckland 1992; Mayle 1996). Line transect techniques (Burnham, Anderson & Laake 1985) have been used to sample FSC pellet groups in the UK with the potential to further improve precision over FAR (Marques et al. 2001). Overall, however, the FSC technique is potentially less accurate because of the additional scope for bias from estimating the decomposition rate (Mitchell & McCowan 1979; Buckland 1992; Nchanji & Plumptre 2001). Moreover, many published studies using FSC techniques fail to incorporate the sampling error associated with the measurement of decomposition rate into the overall estimate of precision (Plumptre 2000). While decomposition rate sampling error has been shown to contribute less to overall precision than the count data in African conditions (Barnes et al. 1995), the effects of any error are still additive (Barnes et al. 1995; Plumptre 2000; Barnes 2002). FPG count data require more effort to obtain using FAR methods than FSC because of the requirement for two visits to plots (McClanahan 1985; Mayle 1996; Nchanji & Plumptre 2001). However, plots used to measure the decomposition rate for FSC data must also be visited, and at more frequent intervals than FAR plots (Ratcliffe 1992; Plumptre & Harris 1995; Barnes et al. 1997; Laing et al. 2003). Although the costs

incurred in measuring decomposition rates are likely to be significant, they are rarely considered (Mitchell & McCowan 1979). To measure decomposition rates in the UK, some authors (Ratcliffe 1992; Mayle 1996) advocate a ‘steadystate’ approach that assumes that equal numbers of pellet groups enter and leave the FSC at a particular point in the year (McClanahan 1985). Only one cohort of groups then needs be followed to estimate the rate (Ratcliffe 1992; Mayle 1996). However, this can lead to bias (Marques et al. 2001; Laing et al. 2003), caused by a number of factors including variations in decomposition rates recorded within, and between, habitats and times of year (Mitchell & McCowan 1979; Putman 1984; Welch et al. 1990; Buckland 1992; Laing et al. 2003). To reduce bias, numerous cohorts of groups must be observed in each trial (Plumptre & Harris 1995; Barnes et al. 1997; Marques et al. 2001) and rates measured retrospectively (Laing et al. 2003). Given the volume of relevant literature, there is surprisingly little empirical evidence available to guide managers in the UK uplands regarding the relative accuracy, precision and cost-effectiveness of the two FPG count methods. In two of the few studies carried out (Ratcliffe 1987a; Smart, Ward & White 2004), the methods were applied to a single site where FAR count data were found to be less precise than FSC. Plumptre (2000) and Laing et al. (2003) suggested that the tradeoffs between FSC- and FAR-based techniques might be investigated more thoroughly. The aim of this study was to quantify the levels of precision obtained when using FSC and FAR methods, in equivalent conditions, across a range of sites in the uplands of Scotland. The levels of effort needed for gathering and processing the two types of data were also quantified. This facilitated an investigation of the likely levels of precision required in, and effort available for, trials to measure the pellet group decomposition rate for the FSC technique. The resultant information was used to gain an understanding of the conditions in which each method is likely to prove the most costeffective for a given overall precision of the resultant abundance estimate. It should be noted that the coefficient of variation (CV) was used as the measure of precision throughout.

Methods   :   FPG count data were collected from 26 predominantly upland sites in Scotland. In the main, the study areas were owned by the Forestry Commission although some sites incorporated private land. The local names of the study areas are therefore not disclosed. The 26 areas (ranging in size from 1483 to 16 162 ha) represented a large latitudinal (55–58°N) and altitudinal (0–940 m a.s.l.) range. All were dominated by Sitka spruce Picea sitchensis L. of varying age class, but with

1187 Faecal pellet group count methods

grassland, heathland and native woodland also present. Different combinations of roe deer Capreolus capreolus L., sika deer Cervus nippon Temminck and red deer Cervus elaphus L. were present.

  :   Most FPG count studies in the UK uplands recognize the benefits of stratification by forest habitat type (Welch et al. 1990; Mayle 1996; Marques et al. 2001) to improve precision (Table 1). Data from a forest database, containing information on each study area in terms of compartments of land and their attributes, were used for this purpose. Database entries described whether compartments were planted or unplanted, along with the species and age class of trees where present. Up to four strata were used (Table 1) where they occurred within each study area. The area and location of each compartment within each stratum, and the total area of each stratum, was obtained from the database. The information within the database related very closely to conditions on the ground and a priori stratification worked effectively, particularly when augmented by advice from local staff at the design stage. Compartments within each stratum were selected for sampling at random, with selection probability weighted by contribution to the overall area of the strata. Exact start positions within each chosen compartment were selected at random. If an estimate of the standard deviations of counts in each stratum was available from a previous study, sampling was weighted using a Neyman allocation (equations 1– 4). FPG count data were collected from all start positions on random bearings using plots that were 75 m in length, unless in pre-thicket when 50 m was used due to the difficulty of laying out a survey line. Plots were 1·5 m wide, unless the vegetation was difficult to search, i.e. consisted largely of heather Calluna vulgaris L., bracken Pteridium aquilinum L. or purple moor grass Molinia caerulea L. In this case, plots of width 1 m were used. The plot width was split equally on either side of the central line. Observers searched each side separately to maximize search success. It was assumed all groups on plots were detected using this method. Multiple plots could fall in one compartment but plots could not overlap. Plots were also restricted to their selected stratum, such that if the allocated bearing led into another type, the remaining portion of the plot was run from the start point at 180° to the bearing (or 90° then 270° if 180° was unsuitable). In reality, rerouting

of plots occurred in less than 10% of cases. This approach could create bias if it was apparent that the spatial distribution of pellet groups differed between the edges and ‘interior’ and that the scale of contiguous areas of sampled habitat was small relative to plot length. In fact, most of the blocks of a particular habitat type sampled were very large (> 50 ha) and few were of an elongated shape. All plots were sampled using a ‘combination’ system, where each was first visited to locate and mark existing pellet groups. In general each plot was then left in place for a period of 2– 4 months before to a second visit, after which the number of new pellet groups deposited, and original groups decomposed, was recorded. FAR data comprised the number of new groups. FSC data, obtained on the second visit, comprised all groups remaining from the first visit as well as all new groups on the second. The FSC data from the second visit were used in preference to the first as most studies in the UK are carried out in late spring (Table 2).

  :   Plot starting points were located in the field, marked with fluorescent tape and a central axis temporarily delineated using a thread distance measurer. Sections of bamboo cane (30 cm) were inserted at regular intervals to mark the line and to allow for exact relocation on the second visit. Each plot was then searched for pellet groups. A minimum of 18 pellets was considered to form a group when pellets showed no signs of breakdown and were fully formed. Where pellets showed signs of breakdown, a minimum of six whole pellets constituted a group. When a group was found, a small wooden marker was inserted in the centre, and its distance from the line, and position within the numbered segments of the plot, was recorded. Whole groups outside the plot but within 10 cm of the plot edge, or partial uncountable pellets and groups within the plot, were destroyed. The spatial distribution of a counted group, i.e. whether it was scattered, piled or strung in a line, was noted to ensure that it was not confused with another group in its environs on the return visit. On completion of the first sampling phase, canes were left in place and the thread removed. On the second visit to each plot, the thread line was re-instated. Each unmarked group was categorized as a new group, unless its location coincided with that of a previous group that had lost its marker due to disturbance. The disappearance of older, original groups (fewer than six whole pellets remaining), while infrequent, was also noted.

Table 1. The stratification used in each study area. Strata reflect probable differences in deer density brought about by differences in available food, shelter and cover


Stratum

Name

Description

1 2 3 4

Unplanted Restock Pre-thicket Closed canopy

Land with no commercial tree crop planted Commercial tree crop of 0 – 3 years old Commercial tree crop of 4 – 9 years old Commercial tree crop of > 9 years old

1188 Table 2. The details of the sampling schemes used. Visits 1 and 2 represent the first and second visit dates. The accumulation period is the days that FAR plots were left between visits. The number of plots sampled in each stratum (1, unplanted; 2, restock; 3, pre-thicket; 4, closed canopy) is given, along with D. Campbell, theM. areaSwanson (ha) of each. ‘–’ represents no area and therefore zero plots sampled; Y = yes; N = no G. & J. Sales Study area

Datevisit 1

Accumulation period (days)

Datevisit 2

Time data collected

Plots1

Plots2

Plots3

Plots4

Area1

Area2

Area3

Area4

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26

December 2000 November 2002 November 2002 February 2002 January 2001 January 2001 April 2002 January 2001 January 2002 February 2003 February 2003 February 2000 December 2002 March 1999 November 2000 November 2002 February 2001 January 2003 December 2002 January 2002 December 2001 November 2002 December 2001 January 2001 February 2003 December 2000

111 98 131 119 128 107 62 135 67 88 88 82 93 67 116 100 95 86 96 62 110 101 78 103 76 192

April 2001 March 2003 March 2003 June 2002 June 2001 May 2001 June 2002 May 2001 March 2003 May 2003 May 2003 May 2000 March 2003 May 1999 March 2001 March 2003 March 2003 April 2003 March 2003 March 2003 March 2002 March 2003 March 2001 May 2001 April 2003 June 2001

N Y N N N N N N Y Y Y N Y N N Y N N Y Y N Y N N Y N

70 57 61 44 32 11 11 16 7 21 16 23 28 7 41 13 19 40 35 3 10 v5 21 12 11 17

– – – – 16 9 8 11 4 3 6 – – 4 4 – 11 4 – – – 11 3 – 4 –

–

26 39 18 22 79 25 38 44 49 46 39 43 41 31 47 56 55 48 32 59 53 40 35 31 23 42

1790 4905 6656 2638 3264 491 1193 470 265 708 1193 1180 1965 362 2819 486 2151 1985 967 374 601 715 1668 496 191 695

– – – – 695 186 297 325 185 69 35 – – 77 234 – 435 412 – – – 174 87 – 58 –

–

2 183 4 647 4 788 3 530 11 071 3 225 3 462 3 195 2 216 2 176 4 909 2 324 3 092 1 975 4 461 1 029 4 949 5 306 1 823 3 222 2 876 2 225 2 230 2 884 1 083 5 056

  :    A study was carried out during the winter of 2002–2003 to quantify the effort involved in gathering FSC and FAR count data. Data were collected in 10 of the 26 areas included in the analysis. In nine out of 10 areas, data were collected for half the plots, whereas in the tenth data were collected for all the plots sampled (Table 2). FSC fieldwork time was taken to include all travel within each area, set up time on plots and measurement time on the first visit. FAR fieldwork time included all time needed for the second visit to plots, as well as that needed on the first. The administration times required for survey designs, data entry and analyses were estimated from staff records. FSC administration time included that for survey design, visit 1 data entry and all analysis time. FAR administration time included all FSC administration time as well as data entry time for visit 2 data.

  Pellet group density © 2004 British Ecological Society, Journal of Applied Ecology, 41, 1185–1196

To improve precision and efficiency in operational contracts, data are normally analysed using different plot sizes in each stratum. However, for this study plot size was standardized at 50 × 1 m (50 m2) to facilitate com-

4 – 9 13 11 11 1 9 – 10 7 – 17 9 – 11 9 – 7 3 11 12 17 12 7

117 – 271 1133 1186 655 264 191 – 545 250 – 380 721 – 1256 297 – 98 20 158 457 806 150 214

parison of standard deviations within and between strata and study areas. The numbers of FAR and FSC pellet groups found on the second visit were calculated for each plot. From this, the mean number of pellet groups, Xi, and the standard deviation of groups, si, was calculated for each stratum i in each study area. The survey design treated the overall area Ai of each stratum as being divided into equal-sized notional plots, Ni, and ni plots were then selected at random from the Ni plots. Therefore: estimated total number of pellet groups in a study area = ∑ NiXi eqn 1

estimated standard error (SE) =

∑N

2 i

2  ni  si 1 − N  n  i i

eqn 2 The CV was used as a measure of the precision of the total pellet group estimate within each study area: CV =

SE (sample statistic) × 100 value of statistic

eqn 3

The Neyman allocation was used in some study areas to determine the number of plots within each survey that should be allocated where σi is the population standard deviation in each stratum. The allocation thus minimized the variance of the term:


∑N

2 i

2  ni  σ i 1 − for a total, fixed sample size n = ∑ ni  N i  ni 

The allocation of plots to each stratum was therefore: ni =

Ni σi n ∑ Ni σi

eqn4

The Neyman allocation, which assumed that pilot data were obtained using equal-sized plots within each stratum, gives the optimum number of plots per stratum only when the population standard deviations for each stratum, σi, are known. In practice, estimates of the standard deviations were taken from previous studies. Estimates of the total number of FAR and FSC pellet groups, and the associated standard errors, in each study area were used to calculate a measure of the precision (CV) for each. A ratio of the CVFSC /CVFAR was then calculated as the measure of relative precision in each study area. Scatter diagrams of standard deviation, si, per stratum against mean pellet group count, Xi, were used to describe the distribution of groups within the FSC and FAR systems. For this comparison, stratum-specific FAR and FSC estimates of mean pellet group count, and standard deviation, in the 26 study areas were used. However, estimates whose mean count was based on five, or fewer, plots were excluded. This is because estimates of standard deviation based on very small sample sizes are inaccurate and including them would have obscured the relationship between the mean and standard deviation. Strata with zero counts were also excluded. Data collection time The mean sampling time per plot in each study area was calculated from the total time required, and the number of plots used, in the collection time study (Table 3). The mean time per plot was adjusted to allow for the fact that many plots in the collection time study were originally sampled using 75-m lengths, whereas shorter, standardized plots of 50 m were used for the analysis (Table 3). The time was reduced by assuming that 75-m plots would take 50% longer to survey than 50-m plots.

Prior to analysis of collection time data, the records for total travel times on visit 1 for study area 10 and study area 16 were reduced, by 1400 min and 750 min, respectively, because of the onset of snowy conditions during set up. These conditions resulted in a proportion of travel time within the areas being unrepresentative of the time the work would otherwise have taken. The adjustments were made from accurate diary records detailing the number of occasions on which surveyors had to abandon work. The mean travel time per plot was calculated from the adjusted total time required, and the number of plots used, in the collection time studies (Table 3). The mean administration time per plot was calculated from the total time required, and the number of plots used, in the pellet group studies (Table 3). Estimates of the total time needed for FSC and FAR methods were calculated on a per plot basis for each study area, by adding the relevant sampling, travelling and administration times. These data were converted from hours to days by assuming that a 7·25-h block of work represented one working day. Obtaining abundance estimates Abundance estimates from FPG methods are formed from estimates of a number of variables that are assumed to be independent. For the FSC method three estimates of precision are required to quantify the precision of the abundance estimate: these are the precisions of the defaecation rate per animal (production), the count of pellet groups in an area (count) and the estimated mean time to decay (decomposition). The overall precision of the abundance estimate may be approximated using equation 5. The FAR method does not require an estimate of the decomposition rate and so CVdecomposition is zero in this case: 2

2

2

CVoverall = CVproduction + CVcount + CVdecomposition eqn 5 The decomposition rate of pellet groups relating to the FSC at the point of the second visit, and the associated sampling error, was not measured in this study. Instead, the minimum CV that would have had to be obtained from a local decomposition trial to ‘break even’ with the overall precision of the FAR count was estimated.

Table 3. Details of a study quantifying the collection and processing times for FSC and FAR pellet group count data, obtained in winter 2002–2003


Study area

Area (ha)

No. of plots where pellet groups were sampled

No. of plots where time data were collected

Proportion of plots with 75-m length where pellet groups were sampled

2 9 10 11 13 16 19 20 22 25

9669 2856 2952 6680 5057 1515 2789 3649 3271 1482

100 69 70 71 69 69 67 69 67 50

100 35 35 35 52 35 35 35 35 25

0·89 0·94 0·93 0·73 0·87 0·94 0·90 0·90 0·97 0·46


This was compared with published studies to assess the likelihood of the FSC technique being more cost-effective overall. Equation 5 was used to obtain the minimum CV of the decomposition rate, CVdecomposition, needed for the overall CV, CV overall, to be equal for FAR- and FSC-based methods, assuming that equal numbers of plots were used in each study. The precision of the production component is likely to be similar for both methods and was ignored in the calculations. The difference in time between the two methods applied in equivalent conditions was then estimated to ascertain the likelihood of sufficient time being available to obtain estimates with the required level of precision. Collection time data were used to estimate the average time needed to apply FAR techniques to one plot. A range of times needed to apply FAR at various overall sample sizes was then generated. The ratios of difference in time per plot to apply FAR compared with FSC were then used to estimate the approximate net time available for an FSC count study of given sample size. Two designs of decomposition rate trial, one based on the methodology of Ratcliffe (1992) and the other using the methodology of Laing et al. (2003), were used to ascertain whether the net available time was sufficient. For the comparison, the time needed to sample an individual decomposition trial plot was assumed to be equivalent to the mean time needed to sample an FSC pellet group density plot in this study. The maximum number of possible visits to a number of trial plots based on the recommended sample sizes, was then calculated from the range of net times available.

Results   :   In 25 of the 26 study areas visited, the CV of the FSC pellet count estimate was lower than that of its FAR equivalent (Table 4). The degree of difference between methods was generally small, although variation was apparent (Table 4). In general, the longer the period of time for which plots were left to accumulate groups in the FAR method, the smaller the difference between methods in the CV of the overall count (Fig. 1).

Fig. 1. Scatter diagram of the ratio of differences in CV for FSC- and FAR-based estimates of pellet group density in each of 26 study areas, against accumulation period (days) for FAR data.

Table 4. Estimates of overall pellet group density per ha2 (and SE) in each of the 26 study areas. Data were obtained by merging estimates from each of the strata sampled in each study area, using equation 2. The CV ratio is calculated as CV FSC/CVFAR


Study area

−2 Pellet groups ha FSC

SE

−2 Pellet groups ha FAR

SE

CV(%)FSC

CV(%)FAR

CVFAR /CVFSC

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26

771 755 390 255 480 565 989 134 354 549 799 764 202 565 791 855 727 214 528 660 1146 622 619 365 467 235

97 65 51 45 50 106 149 27 46 85 116 97 33 74 94 100 92 49 80 105 116 108 104 56 87 51

244 464 217 132 161 142 162 68 110 244 176 177 83 119 325 281 145 87 131 204 438 247 156 129 92 91

32 50 33 27 20 32 31 15 25 45 36 36 18 26 49 35 25 21 25 47 52 41 29 26 22 26

12·58 8·66 13·17 17·48 10·41 18·81 15·04 19·94 12·89 15·56 14·53 12·69 16·30 13·04 11·89 11·66 12·66 23·13 15·14 15·83 10·09 17·30 16·79 15·46 18·67 21·81

12·94 10·87 14·97 20·70 12·31 22·57 18·93 22·44 22·52 18·35 20·61 20·51 22·00 22·25 15·02 12·29 17·45 23·89 19·17 23·08 11·98 16·61 18·70 20·08 24·12 28·98

0·97 0·80 0·88 0·84 0·85 0·83 0·79 0·89 0·57 0·85 0·71 0·62 0·74 0·59 0·79 0·95 0·73 0·97 0·79 0·69 0·84 1·04 0·90 0·77 0·77 0·75


Fig. 2. (a,b) Scatter diagram of (a) untransformed and (b) log10-transformed standard deviations against mean pellet group counts. FAR data, circles, and FSC data, crosses, were obtained from each of four strata sampled within the 26 study areas, with each point relating to a stratum-specific estimate based on n > 5 plots.

However, the data for study area 26 did not follow this pattern. Pellet groups were spatially aggregated, with the standard deviations approximately equal to the means in both the FSC and FAR systems (Fig. 2a). The standard deviation relative to the mean in each stratum tended to be higher where pellet group density was lowest (Fig. 2a), and was approximately linear on a log–log scale (Fig. 2b). In all areas, more FSC groups were counted overall compared with FAR (Table 4 and Fig. 2a). Therefore, FSC count data were expected to have generally improved precision in equivalent conditions (Table 4), particularly where pellet group density within a study area was very low (Table 4), FAR plots were left out for a relatively short time (Fig. 1) and decomposition rates were slow.

  :   In all but one of the studies, the time needed to sample pellet groups on the first visit was longer than that needed on the second (Table 5). Mean travel time per

plot was generally higher than mean sampling time for both methods (Table 5). The time needed to administer a FAR study was 20–30% higher than that for FSC (Table 5). Collection of FSC data required approximately 80 min per plot, compared with approximately 140 min for FAR (Table 5). Each FAR count study needed more time to be carried out overall, taking 1·6–1·9 times longer than its FSC counterpart (Table 5).

  :    Given that CV obtained from the FAR counts ranged from 10% to 30% and for FSC from 9% to 23%, parallel decomposition rate trials would have needed CV in the range 5–15% to ensure that the overall precision of the FSC-based abundance estimates was equivalent to that of FAR (Fig. 3). We estimated that trials would need a CV of 4–7% for an overall target CV of 10%, 6–11% for an overall CV of 20%, and 9–14% for a target CV of 25% (Fig. 3). This assumed that the CVFSC /CVFAR ratio

Table 5. Data collected during the winter period of 2002–2003 in a study of the effort involved in processing pellet group data using FSC and FAR techniques. Mean travelling time (min) relates to the time spent travelling between plots within each study area. Administration time (min) relates to the time spent on design, data entry and analysis for each study area

Study area

Estimated mean sampling time per plot on first visit (min)

2 22 9 20 10 22 11 23 13 23 16 26 19 24 © 2004 British 20 19 Ecological Society, 22 23 Journal of Applied 25 26 Ecology, Mean 41, 23 1185–1196

Estimated mean sampling time per plot on second visit (min)

Mean travel time per plot on first visit (min)

Mean travel time per plot on second visit (min)

Mean FSC admin. time per plot (min)

Mean FAR admin. time per plot (min)

Estimated mean time per FSC plot (min)

Estimated mean time per FAR plot (min)

Estimated mean time per plot (FSC / FAR)

18 14 19 24 19 19 17 16 14 18 18

47 38 53 54 26 46 37 47 56 35 44

45 26 32 49 34 31 44 53 52 32 40

11 13 12 14 14 13 14 11 12 15 13

14 16 15 16 17 16 17 14 16 19 16

80 71 87 91 64 85 75 77 91 77 80

146 114 141 167 119 139 139 148 159 130 140

1·8 1·6 1·6 1·8 1·9 1·6 1·9 1·9 1·8 1·7 1·8


Fig. 3. The minimum level of precision (maximum CV) that would have been needed from a decomposition rate trial within a study area to equalize the overall precision of FAR- and FSC-based abundance estimates (assuming equal numbers of plots are used in both FAR and FSC sampling frameworks). Scenarios are based on the main range of CV ratios measured in the 26 study areas (0·7, 0·8 and 0·9).

Fig. 4. The maximum number of days available to carry out decomposition trials in a study area. The projection assumes that equal numbers of plots are used for both FSC and FAR surveys. The scenarios are based on a range of time ratios (FAR/FSC) of 1·7, 1·8 and 1·9 (Table 5). The scenarios also assume that FAR plots require an average of 2·35 h of sampling time per plot (Table 5) to derive the total time needed for application of both FSC and FAR methods in a study area.

is approximately 0·7 – 0·9 and ignores any contribution from CVproduction in both cases.

  :       


Assuming that 50 –100 plots were used for each count study, as is typical in UK conditions, the net time available to carry out decomposition rate trials would have been in the range 7–15 days (Fig. 4). For an assumed trial design of 24 plots (Ratcliffe 1992), requiring a total of 1·2–1·4 h plot−1 to sample and process, this would allow a maximum of one to three visits to each plot with the net time available (Fig. 5a). For the trial design employing 15 plots (Laing et al. 2003), also requiring a total of 1·2–1·4 h plot−1 to sample and process, a maximum of two to six visits to each plot would be possible in the net time available (Fig. 5b).

Fig. 5. (a,b) The maximum number of visits that could be made to a set of plots to carry out a decomposition rate trial using sample sizes (a) assumed from Ratcliffe (1992) and (b) taken from Laing et al. (2003). Calculations are based on a time budget (h) per plot in the trial of 1·2, 1·3 and 1·4 (Table 5).

Discussion   :     The precision of each method was strongly influenced by the density of pellet groups present in study areas, as discussed by previous authors including Ratcliffe (1987a) and Buckland (1992). As expected, FSC count data were generally more precise than FAR in equivalent conditions. Across the wide range of sites included in this study, however, the relative differences were fairly small. Ratcliffe (1987a) and Smart et al. (2004) found that FAR precision was poorer than FSC by a margin that is equivalent to the lower end of the CV ratios found in this study. The range of relative differences in count precision was associated with the number of days that the FAR plots were left to accumulate pellet groups. In general, the longer the accumulation period the smaller the difference. As the study of Smart et al. (2004) left plots out for 60 days, this would explain why their FAR count data were less precise than their FSC. However, leaving FAR plots out too long risks biasing the results. This



may account for the lower than expected FAR precision in study area 26, which had an unusually long accumulation period of 192 days. Decomposition rates over winter, however, are generally considered to be very low in UK conditions (Mitchell & McCowan 1979; Welch et al. 1990) and appear not to have been important in the other study areas as the relationship between accumulation days and relative precision was approximately linear (Fig. 1). Slower decomposition rates, with groups carrying over from one year to the next (Mitchell & McCowan 1979; Welch et al. 1990), will result in a relatively large FSC and better precision being obtained than for equivalent FAR data. However, this ‘carry over’ may also bias parallel data on decomposition rates if trials are not run over a long enough period in the lead up to survey (Laing et al. 2003), since a short trial provides no reliable information on the duration of long-lived pellet groups. The use of the Neyman allocation in some study areas may also have increased FAR count precision at the expense of FSC as the weighting of sampling effort was carried out on the basis of previous over-winter accumulation rates and not FSC density. Laing et al. (2003) commented that there could be a threshold in deer density below which FSC methods may be more cost-effective than FAR because of the need to use more FAR plots to obtain equivalent count precision. Overall deer densities encountered in this study, calculated (Tables 2 and 4) using FAR data and an approximate defaecation rate of 20 pellet groups day−1, ranged from 2 to 23 km−2. However, there was no strong relationship between deer density and relative precision as measured by the CVFSC/CVFAR ratio while, as discussed, there was a clear relationship between return time and CV ratio. Smart et al. (2004) suggested that FAR count methods produced very poor results. We conclude that there is a consistent but small difference in precision in the conditions we studied. Our conclusions differ for a number of key reasons. First, Smart et al. (2004) did not consider the potential for variation between sites in relative precision. Also, the return period to their FAR plots was short (60 days). Given that the resultant CV of the FSC and FAR counts in their study was approximately 39% and 45%, respectively, but the deer density was high (16 –18 km−2), it would also appear that the allocation of samples to strata was suboptimal. As spatial aggregation of groups was clearly high in their study area, as in ours, the use of a random distribution of pellet groups in their simulation model is also potentially misleading. Because of the level of aggregation, their choice of quadrats over transects may also have reduced precision. Given the generally small differences in the relative precision of our count data, it is apparent that trials measuring decomposition rate would need to be relatively precise for FSC-based abundance estimates to obtain equivalent overall precision to those derived from FAR in the conditions studied. Even operating at

the lower end of the accumulation period range, where bias from intermediate decomposition was thought to pose least risk, the ratios of CVFSC /CVFAR varied from 0·6 to 0·8 and the precision of decomposition rate estimates would need to be high. Our estimates of maximum decomposition trial CV are based on the assumption that the sampling intensity is the same for FSC and FAR counts. In some situations it might be possible to reduce the overall CV for the FSC method by trading off the decrease in precision for CVcount against an increase in precision for CVdecomposition or vice versa. However, as these two quantities have been shown to be similar in magnitude, any gain in precision achieved by a reallocation of sampling effort will be small. Line transect methods could be used to improve the relative precision of FSC counts, having been suggested as improving precision over fixed width plots in many situations (Burnham, Anderson & Laake 1985; Marques et al. 2001). In theory, applying the technique to FPG counts should lead to an improvement in precision because of the wider area that may be searched for a given effort. Indeed, they have been applied to FSC techniques in Scotland (Marques et al. 2001). This could favour an FSC approach overall, as the method cannot be applied to FAR techniques that require that all pellet groups are detected on the first visit in order that those detected on the second can be treated as new. However, concern has been raised (Caughley & Sinclair 1994) over whether the improved precision, obtained by fitting models to allow for uncertain detectability, is worth the risk in some circumstances. Pellet groups are relatively difficult to locate in the ground conditions present in the UK uplands, even when located near the central line. We used widths of 50 cm in this trial while sampling in poor conditions, to ensure that equal search effort was allocated to counting pellet groups across the entire, albeit narrow, strip width. Operating in similar conditions that included heather and purple moor grass. Marques et al. (2001) used restricted plot widths of 2 m in their searches. In these circumstances, the allocation of part of the search effort to the periphery of the survey line (say the range 150–200 cm) to increase the number of observations on each plot could have resulted in surveyors missing groups closer to the central line (range 0–50 cm). Empirical evidence is therefore needed to determine whether the benefits to precision of applying line transect methodology to pellet group counting are not countered by ‘background detection errors’ or biases from searching too wide an area on plots. A range of additional biases must also be considered in any comparison. A potential bias of using the FAR method relates to vegetation disturbance. In this study the majority of vegetation present in the study areas was not permanently disturbed on the first visit, as it comprised conifer needles, mosses, short grasses and heath. When working in bracken and purple moor grass on the first visit to plots, surveyors were careful to


replace vegetation that had been searched beneath to ensure, as far as possible, that no differential growth occurred in the spring. However, many of the surveys were finished prior to the onset of plant growth. In other conditions, differential growth and changes to vegetation structure could attract animals onto sample plots and bias the count (Barnes et al. 1995). However, it is generally assumed that the scope for bias is higher when using FSC-based techniques. First, observers can experience greater difficulty in finding and classifying older, decaying pellet groups using FSC techniques (Neff 1968). Secondly, defaecation rates are normally estimated using pellet group count methods within known populations. Given that rates relevant to the FSC count include a proportion of the summer period, bias may be introduced because the defaecation trials are run during the time when groups are most often missed and decay more quickly (Mitchell & McCowan 1984). It is certainly necessary that decomposition rate trials are well designed to ensure that the resultant estimates are unbiased (Laing et al. 2003).

  :  


FAR studies took 1·6 – 1·9 times longer to carry out than FSC. This variation related to differences in the level of vehicular access within and between study areas and in the number of plots sampled in each. The average distance between plots in each study area ranged from 470 to 1070 m. Our findings strongly contradict those of Smart et al. (2004), who calculated that FAR field data took three times longer than FSC data to collect. Their estimate of the time taken to collect FAR data appears to be surprisingly high compared with our experience. Crucially, Smart et al. (2004) failed to include administration time in their comparison. They also chose to remove pellets from the plots rather than mark them, which we have consistently found to be more time consuming. The fact that our FAR data took 1·6–1·9 times the time to collect compared with FSC data is easily explained. First, FAR plots took less time to locate on the second visit, as surveyors were by then familiar with the study areas. Secondly, FSC plots were harder to search, as many of the pellet groups were hidden under vegetation that collapsed in the first frosts. Almost all FAR pellet groups lay on top of this layer, making resurvey quicker due to improved visibility and a reduction in the time needed to define a group as countable. Also, a large proportion of the time required to design surveys and analyse data was common to both methods. Data entry from the second visit to plots was the only additional factor involved in processing FAR information. Critically, understanding the approximate relationship between overall FAR and FSC sampling time per plot facilitated the calculation of the difference in time

between methods derived from surveys of typical sample size. FAR surveys of 50–100 plots produced net time budgets of 7–15 days in our study, depending on the relative speed at which the techniques were deployed. However, the levels of precision obtained in some study areas would prevent anything other than large changes in density from being detected with the standard number of plots. If the number of sample plots increased from, say, 100 to 200, then the net available time for decomposition trials would increase to 25– 30 days. This would increase the likelihood of decomposition rate trials providing data with the appropriate level of precision in less than the total time available. However, most managers in Scotland accept poorer precision for low-density deer populations because monitoring proportional changes is less important under these conditions, given that a large proportionate error represents a small absolute error. Although some studies require very high precision to detect small changes in the abundance of rare species, the use of FPG techniques in these circumstances may prove inappropriate because precision declines so markedly with declining pellet group density. More work is needed to determine where this threshold lies for each set of local environmental conditions.

  :   Given the levels of count precision obtained, the question then arises as to how likely it would be to achieve levels of precision in decomposition rate within the target values of overall precision required. Marques et al. (2001) obtained CV of 9–15%, although the levels of effort required to collect these data were not published. Laing et al. (2003) obtained similar results during their study of roe and red deer abundance in Abernethy forest, although effort was similarly unquantified.

  :    The accuracy, precision and cost-effectiveness of decomposition rate trials will vary depending on the sampling framework used. Buckland (1992) highlighted the importance of using independently sampled locations to measure the decomposition rate to avoid potential bias. However, no specific mention of the spatial sampling design is made in published work (Ratcliffe 1992; Mayle 1996; Marques et al. 2001). In estimating cost in this comparative study, it was therefore assumed that randomly sampled plots would be used. Although various techniques have been described to estimate the decomposition rate itself, ‘prospective’ methods are unlikely to yield reliable estimates of pellet group decomposition rate, or associated sampling error, in UK upland conditions (Laing et al. 2003). The decomposition rates, and levels of precision, obtained


by Marques et al. (2001) and Smart et al. (2004) are therefore potentially unreliable. Trials used in the analysis were therefore assumed to be ‘retrospective’ and based on naturally defaecated pellet groups sampled on regularly observed, randomly placed plots (Laing et al. 2003). The minimum time required for sampling a decomposition trial plot would therefore be similar to that needed to sample pellet groups during normal counts, given the large proportion of travel time and difficulty of finding trial pellet groups in summer vegetation. The assumed time of 1·2–1·4 h to visit each decomposition trial plot therefore appears realistic. Employing a 100-plot survey design for the count, a maximum of 15 days of effort would be available for a trial and 1·3 h would be needed to visit each plot. This translates to three to four visits to 24 plots, or five to six visits to 15 plots. This compares poorly with the number of visits recommended by Ratcliffe (1992) and Laing et al. (2003). If the official decomposition trial visits were combined with the visit to the plots for the FSC pellet group assessment itself, each trial could be based on one more visit to the plots than indicated above. The number of visits may be sufficient to obtain some form of estimate, but appears unlikely to improve on the levels of precision quoted in published studies. Various authors (Barnes et al. 1997; Nchanji & Plumptre 2001) have collected decomposition rate data and searched for covariates to predict rates with varying success. To date, this type of work has not been published for UK conditions using a reliable method for measuring the rate. If decomposition rates were proven to vary little between study areas and years, FSC-based techniques could indeed prove very cost-effective (Buckland 1992; Laing et al. 2003). However, the scarcity of published evidence relating to UK conditions suggests that a large amount of prior research effort would be needed to gather the data and validate models across a range of environmental conditions and years.




Until reliable decomposition rate data are collected across a broad range of conditions and years, managers of individual sites in UK upland conditions are best served by using FAR techniques. This will ensure that poor estimates of the decomposition rate do not compromise the quality of the local decisions they need to make to manage wildlife rationally and objectively. Wider information on levels of precision obtainable in, and effort required for, decomposition rate trials in UK conditions is required to justify fully our conclusions. Understanding the trade-offs between line transect and fixed-plot approaches, when applied specifically to FPG counting in the UK uplands, as well as those relating to the proportion of resources allocated to FSC counts against decomposition rate within a particular study area, would further aid decision making. Attempts to quantify the probable magnitude of relative biases, across a range of field conditions, are also required if

managers are to make a fully informed choice between the techniques.

Acknowledgements We thank Forestry Commission Scotland (FCS) for funding this study and the staff of Strath Caulaidh Ltd, past and present, for their commitment to collecting field data in consistently difficult circumstances. A number of FCS wildlife rangers also helped with data collection and we are equally indebted to them. We also thank Helen Armstrong, Iain McKendrick, David Elston, Steve Buckland and two anonymous referees for comments on an earlier draft, as well as Colin Legg for helpful discussions throughout the course of the study.

References Bailey, R.E., Putman, R. & J. (1981) Estimation of fallow deer (Dama dama) populations from faecal accumulation. Journal of Applied Ecology, 18, 697 – 702. Barnes, R.F.W. (2002) The problem of precision and trend detection posed by small elephant populations in West Africa. African Journal of Ecology, 40, 179 – 185. Barnes, R.F.W., Asamoah-Boateng, B., Naada Majam, J. & Agyei-Ohemeng, J. (1997) Rainfall and the population dynamics of elephant dung-piles in the forests of southern Ghana. African Journal of Ecology, 35, 39 – 52. Barnes, R.F.W., Blom, A., Alers, M.P.T. & Barnes, K.L. (1995) An estimate of the numbers of forest elephants in Gabon. Journal of Tropical Ecology, 11, 27 – 37. Buckland, S.T. (1992) Review of Deer Count Methodology. Unpublished report to the Scottish Office. Agriculture and Fisheries Department, Edinburgh, UK. Burnham, K.P., Anderson, D.R. & Laake, J.L. (1985) Efficiency and bias in strip and line transect sampling. Journal of Wildlife Management, 49, 1012 – 1018. Caughley, G. & Sinclair, R.E. (1994) Wildlife Ecology and Management. Blackewll Scientific Publications, London, UK. Laing, S.E., Buckland, S.T., Burns, R.W., Lambie, D. & Amphlett, A. (2003) Dung and nest surveys: estimating decay rates. Journal of Applied Ecology, 40, 1102–1111. McClanahan, T.R. (1985) Quick population survey method using faecal droppings and a steady state assumption. African Journal of Ecology, 24, 37 – 39. Marques, F.F.C., Buckland, S.T., Goffin, D., Dixon, C.E., Borchers, D.L., Mayle, B.A. & Peace, A.J. (2001) Estimating deer abundance from line transect surveys of dung: sika deer in southern Scotland. Journal of Applied Ecology, 38, 349 – 363. Mayle, B.A. (1996) Progress in predictive management of deer populations in British woodlands. Forest Ecology and Management, 88, 187 – 198. Mitchell, B. & McCowan, D. (1979) Estimating and Comparing Population Densities of Red Deer Cervus elaphus L. in Concealing Habitats. Institute of Terrestrial Ecology, Cambridge, UK. Mitchell, B. & McCowan, D. (1984) The Defecation Frequencies of Red Deer in Different Habitats. Institute of Terrestrial Ecology, Cambridge, UK. Mitchell, B., Rowe, J.J., Ratcliffe, P.R. & Hinge, M. (1985) Defecation frequency in roe deer (Capreolus capreolus) in relation to the accumulation rates of faecal deposits. Journal of Zoology, London, 207, 1 – 7. Nchanji, A.C. & Plumptre, A.J. (2001) Seasonality in elephant dung decay and implications for censusing and



population monitoring in south-western Cameroon. African Journal of Ecology, 39, 24 – 32. Neff, D.J. (1968) The pellet-group count technique for big game trend, census, and distribution: a review. Journal of Wildlife Management, 32, 597 – 614. Plumptre, A.J. (2000) Monitoring mammal populations with line transect techniques in African forests. Journal of Applied Ecology, 37, 356 – 368. Plumptre, A.J. & Harris, S. (1995) Estimating the biomass of large mammalian herbivores in a tropical montane forest: a method of faecal counting that avoids assuming a ‘steady state’ system. Journal of Applied Ecology, 32, 111–120. Putman, R. & J. (1984) Facts from faeces. Mammal Review, 14, 79 – 97. Ratcliffe, P.R. (1987a) The management of red deer in the commercial forests of Scotland related to population dynamics and habitat changes. PhD Thesis. University of London, London, UK. Ratcliffe, P.R. (1987b) The Management of Red Deer in Upland Forests. HMSO, London, UK.

Ratcliffe, P.R. (1992) Roe Deer Biology and Management. HMSO, London, UK. Smart, J.C.R., Ward, A.I. & White, P.C.L. (2004) Monitoring woodland deer populations in the UK: an imprecise science. Mammal Review, 34, 99– 114. Staines, B.W. & Ratcliffe, P.R. (1987) Estimating the abundance of red deer (Cervus elaphus L.) and roe deer (Capreolus capreolus L.) and their current status in Great Britain. Symposia of the Zoological Society of London, 58, 131– 152. Webbon, C., Baker, P.J. & Harris, S. (2004) Faecal density counts for monitoring changes in red fox numbers in rural Britain. Journal of Applied Ecology, 41, 768 – 779. Welch, D., Staines, B.W., Catt, D.C. & Scott, D. (1990) Habitat usage by red (Cervus elaphus) and roe (Capreolus capreolus) deer in a Scottish Sitka spruce plantation. Journal of Zoology, London, 221, 453 – 476. Received 7 February 2004; final copy received 20 August 2004