INTRODUCTION
Underwater visual census (UVC) techniques, as with all field census methods, are selective in focus with respect to factors such as size, appearance and behaviour of target organisms. As a consequence, studies of species assemblages are inevitably biased because of differential visibility of organisms and a proportion of the population not being detectable by the particular method used at time of sampling. Surveys
involving direct observation by divers include additional bias associated with
behaviour, experience and subjective decision-making of the individual. Given that bias can never be completely eliminated, the challenge is to recognise its extent and
minimise its effect thereby reducing misinterpretation of results.
Although a few studies have used remotely-operated equipment such as nets, grabs, cameras or videos, the majority of recent ecological surveys of species associated with shallow subtidal reefs have involved diver based UVC. Surveys of plant and sessile invertebrate density typically involve the random or haphazard placement of quadrats (Dethier et al., 1993; Benedetti-Cecchi et al., 1996), while transects are generally used
to quantify densities of organisms distributed at larger spatial scales, particularly fishes and epibenthic invertebrates. For fishes, the main categories of UVC method used are:
(i) strip transects – where density is estimated by a diver swimming or being towed along a strip of known or estimated length and width,
(ii) time transects – where a diver records the number of animals sighted during a fixed time interval,
(iii) line transects – where a diver swims along a line and estimates the distance and direction of target organisms from that line,
(iv) point counts – where the density (area counts) or distance and direction
(distance counts) of organisms are estimated by a diver scanning through 360° from a fixed point or while descending to a fixed point, and
(v) rapid visual techniques – where a diver lists fish species sighted in rank order of initial encounter for each species (DeMartini and Roberts, 1982; Sanderson and Solonsky, 1986; Thresher and Gunn, 1986; Mapstone and Ayling, 1993;
Kulbicki, 1998).
For more detailed descriptions of these methods see Kingsford and Battershill (1998). All methods can involve in situ recording of data by diver or post hoc examination of
video or photo records. Choice of the optimal UVC method will vary with the particular situation depending on aims of study and logistic constraints. In addition this will depend on the characteristics of target species (density, patchiness, animal size, mobility, behavioural response to divers, crypsis), level of training of data gatherers, water clarity, depth, habitat type, current speed and wave exposure.
Amongst the plethora of UVC techniques, strip transect methods are most widely used by ecologists (Cappo and Brown, 1996), and have also been our preferred method for monitoring long-term changes in marine protected areas (MPAs) (Buxton and Smale 1989, Edgar and Barrett, 1997; Edgar et al., 1997; Edgar and Barrett, 1999). Yet despite
general application, biases associated with different UVC methods not been well studied. What is known is that different techniques applied to the same fish community can generate quite different results (Davis and Anderson, 1989; Connell et al., 1998;
Kulbicki and Sarramega, 1999; Willis et al., 2000). Also, within strip transect
protocols, factors such as transect dimension (Sale and Sharp, 1983) and observer speed (Lincoln Smith, 1988) can greatly affect counts.
In the present study we assessed the extent to which several poorly-examined biases associated with UVC methods can confound interpretation of abundance and size- frequency data. Sampling biases examined included accuracy of size and density
estimates made for different species, by different divers, and at different sites and times. Accuracy was assessed by comparing density estimates obtained using strip transect methods with values independently calculated using a capture-resight method based on underwater resights of fish with unique colour-coded tags.
When considering sampling error, it is important to recognise that biases that remain consistent with respect to absolute number for different sites, times and divers are unlikely to greatly affect interpretation of results in comparative studies. Moreover, random errors that add noise to data and affect precision of estimates should not greatly confound studies, other than reducing the power of analyses (and in extreme cases perhaps resulting in Type II statistical errors). However, bias that is inconsistent in time or place is likely to generate misleading interpretations.
One potential source of inconsistent bias was differing detectability of fishes in different habitats. This bias was of considerable concern in monitoring studies of effectiveness of MPAs (Edgar and Barrett, 1999). In particular, high rates of fish and lobster predation within MPAs were found to reduce sea urchin populations in some situations, which allowed urchin-engineered barren grounds to transform to kelp forests (Shears and Babcock, 2002; Shears and Babcock, 2003). The possibility therefore exists that even with no change in absolute fish density within MPAs relative to external reference sites, transect counts may decline in MPAs if the efficiency of divers in sighting fish declines amongst kelp forest that regenerates in former barren grounds. As part of the current study, we assessed this potential problem by clearing laminarian and fucoid kelps from large blocks of seabed to determine whether the detectability of common fishes in strip transects differs between cleared and vegetated habitat patches.
METHODS Sites studied
Fishes were censused using both underwater visual transect and fish trapping methods at three rocky reef sites separated by ≈50 km distance in eastern Tasmania (Fig. 5.1, Chapter 5). Transects and trapping were undertaken within four to five day sampling occasions at time intervals of 0, 1 week, 1 month, 3 months, 6 months and 1 year after initiation of study at each site. Surveys commenced on 21 Feb 2000 at Return Point (42.633°S 148.025°E), 3 Oct 2000 at Little Swanport (42.276°S 148.015°E) and 6 Feb 2001 at Lobster Point (42.964°S 147.667°E). Because of poor weather, sampling at Lobster Point was omitted on the 1 week sampling occasion. Underwater visibility ranged between 5 m and 9 m at Lobster Point, 5 m and 20 m at Return Point, and 6 m and 10 m at Swanport.
On each sampling occasion a 200 m subtidal transect line was laid parallel with the shore in 3-6 m water depth. The transect line was relocated within 1 m on each sampling occasion by reference to permanent markers on the seabed. Reef at all sites was bounded at its deeper edge in 5-7 m depth by sand and extended linearly along the coast in a 30-70 m wide swathe. The sandstone reef at Lobster Point was heavily dissected with numerous rocks and crevices, whereas the low dolerite reefs at Return Point and Little Swanport were less structurally heterogeneous but still possessed occasional boulders, shelves and crevices. Habitats at all sites were dominated by a variety of fucoid and laminarian algal species. Ecklonia radiata and Sargassum
verruculosum were predominant at Lobster Point, Cystophora subfarcinata, Cystophora moniliformis, Caulocystis cephalornithos, Sargassum decipiens and S. verruculosum at
Return Point, and Phyllospora comosa at Little Swanport. Taxonomic authorities for all
species discussed here are listed in (Edgar, 1997).
Underwater visual transect techniques
Visual censuses of (i) fishes, (ii) large macro-invertebrates and cryptic fishes, and (iii) plants were undertaken using protocols described previously for reef monitoring studies of marine protected areas (Edgar and Barrett, 1997; Edgar et al., 1997; Edgar and
Barrett, 1999). For estimates of non-cryptic fishes, a diver swam beside the transect line at a distance of 2.5 m, recording on a waterproof notepad the abundance and size structure of fishes in a 5 m wide swathe (i.e., from the transect line to a distance 2.5 m past the diver). Divers swam slowly, estimating the number of the different fish species sighted within each 10 m length of transect. The size of each fish was estimated and transcribed within 25-mm size-classes. The survey process was then repeated when the diver returned along the other side of the transect line, with data from two adjoining sides of the transect added together. Twenty 10 m x 10 m blocks were thus censused, with data from 10 m x 50 m blocks generally aggregated. Three different divers (NSB, AJM and GJE) undertook fish censuses during the study, with five replicate surveys generally conducted on each sampling occasion. The divers all possessed considerable experience (>100 hours) in underwater visual transect work.
Large macro-invertebrates (echinoderms other than ophiuroids, large molluscs, large crustaceans) and cryptic fishes were counted along transect lines by recording animals in the same 10 m transect intervals as used for the fish survey, but within 1 m of one side of the line. Three replicate surveys were generally undertaken on each sampling occasion, with four different divers used overall (NSB, AJM, CRS and GJE). Plant cover was quantified by placing a 0.25 m2 quadrat at 10 m intervals along the transect line and estimating the percentage area of reef substrata covered by different macroalgal species. Percentage cover was assessed by counting the number of times each species occurred directly under 50 positions on the quadrat at which
perpendicularly placed wires crossed each other. Quadrats were located at the midpoint of 10 m blocks used for fish and macro-invertebrate counts. Two replicate surveys, occasionally more, were undertaken on each sampling occasion. Only one diver was used (NSB), other than at Return Point where plants were surveyed on the first four sampling occasions by a second diver (GJE).
Changes in density following algal clearance
In order to assess whether macroalgal cover affected densities of local organisms and visual estimates of fish density, divers cleared large patches of canopy-forming
macroalgae between the first and second sampling occasions. All large macroalgae were removed by pulling out plants by hand from either side of the transect in two 50 m long 12 m wide blocks. The cleared blocks were centred on the transect line and separated by 50 m distance; hence, one 50 m cleared and one 50 m control block alternated in each direction from the midpoint of the 200 m transect.
The effect of algal clearance on fish, invertebrate and macro-algal species richness and abundance was statistically examined using analysis of variance (ANOVA). In order to reduce heteroscedasticity, and because multiplicative changes were considered more important than additive changes, abundance data were Ln (x+1) transformed for analyses of significance involving F-tests, other than species richness data.
Relative changes over time in biotic relationships between cleared and reference areas were also examined graphically using non-metric multidimensional scaling (MDS). The data matrix was Ln (x+1) transformed for faunal abundance data but not for algal percent cover data, and then converted to a symmetric matrix of biotic similarity between pairs of sites using the Bray-Curtis similarity index. The stress statistic was calculated to assess whether the two dimensional MDS display provided an accurate depiction of relationships. Clarke (1993) suggests that a useful display is obtained with stress <0.2.
Comparison of visual and capture-resight density estimates
The absolute number of common fish species frequenting the transect area was estimated using capture-resight (CMR) techniques crossed between two sampling methods. Fishes were initially captured in fish traps baited with abalone viscera. They were then marked individually with colour-coded t-bar tags locked into the base of the dorsal fin (Edgar et al., in review). Tags were marked with three coloured plastic rings
(six colours: black, blue, green, red, white, yellow) that had been heat shrunk over the serial number, allowing underwater visual recognition of 216 colour combinations. Tagging was completed in less than 15 minutes onboard the capture boat, with fishes released within 2 m of the capture location following a 5 minute holding period to ensure recovery from the shock of tagging.
Traps were set for 1/2 hour intervals each 10 m along the 200 m transect line, near the centre of each block used during visual census. The complete line of traps was set and pulled 3-4 times on each sampling occasion, with the exceptions of 7 times during the initial sampling occasion at Swanport and 2 times during the second and fifth sampling occasions at Return Point.
In order to estimate absolute density, animals tagged during the first (0 week) and second (1 week) sampling occasions provided initial capture information. Recaptures were recorded as visual re-sightings on the third (4 week) sampling occasion (i.e.,
approximately three weeks after initial tagging). Data from visual censuses for animals less than the minimum size of animals captured in traps (150 mm length) were excluded
from analysis. Later sampling occasions were not utilised for density estimates because rates of tag shedding, fish mortality and emigration may have been unacceptably high. The total population size of common species was estimated using the NOREMARK statistical program to calculate Bowden’s Estimator (White, 1996), a variant of the Minta-Mangel Index (Minta and Mangel, 1989). Bowden’s Estimator, and associated 95% confidence intervals, were calculated using information for each common species on the variance of sighting frequencies of each fish marked with a decipherable tag plus number of untagged fish and tagged fish with undecipherable tags (White, 1996). Estimates calculated by NOREMARK relate to the fish population that frequents the 200 m x 10 m transect site; however, a proportion of that population will be roving outside the study area at any time when visual censuses are conducted. Densities (N) within the transect area were therefore determined after correction for animals displaced long-shore and cross-shore outside the transect area as:
N = B.(n200/n2000).(n10/nw)
where B is Bowden’s estimator, ni is the number of tagged fish sighted during searches
along distance i, and w is the mean width of the reef adjacent to the transect line from 1
m depth to the sand edge.
The long-shore correction factor n200/n2000 was assessed by swimming an extended
transect along the coast for 1 km each direction from the centre of the transect line, and recording the number of tagged animals sighted inside and outside the 200 m long transect area. Long-shore transects were undertaken on at least 8 occasions over the course of the study for each site.
The cross-shore correction factor n10/nw was assessed using data on distance moved
between underwater re-sightings within the transect area. The number of tagged fish resighted in the 10 m wide area located 5 m each side of initial sighting location (n10)
was related to total number of fish resighted within w/2 m distance for each species and
location, where w is mean width of reef habitat utilisable by fishes adjacent to the
transect strip. The mean width of reef habitat was calculated from rectified aerial photographs as the offshore distance from just below low water mark to the sand edge. Mean width of reef was estimated to be 30 m at Return Point, 40 m at Lobster Point and 50 m at Swanport.
Variation in density estimates between divers, sites and time
The relative importance of three factors – diver, site and time – in affecting visual density estimates for common species was assessed by distinguishing different variance components using ANOVA. The ANOVA model used included random spatial factors ‘site’ (3 levels) and ‘block’ (2 levels nested within site), temporal factors ‘month’ (5 levels, representing sampling occasions from 1 mo to 12 mo) and ‘day’ (2 levels, representing different days within a sampling occasion, and therefore nested within month), and ‘diver’ (2 levels).
The data set used was that obtained for the algal clearance experiment, but with the elimination of information for the 50 m manipulated (cleared) sections of transect. Thus, the ‘block’ factor included aggregated data for 50 m sections of undisturbed (control) habitat to the east and west of the midpoint of the study area.
Data for the 1 week sampling occasion were also excluded from analysis because this sampling occasion was not censused at Lobster Point. Some data obtained by non- principal divers were also excluded. For both fish and macro-invertebrate transects, one diver (AJM) was used on all sampling occasions at all sites and a second diver (NSB fish, CRS macro-invertebrates) used at two sites and partially used at the third site. For treatment combinations lacking information from the second diver, data from a third diver (GJE) were used to maintain a symmetric model design. Variability between data collected by the second and third divers should not greatly affect outcomes, given that such variability will largely express itself as an increased ‘diver x site’ interaction term. Plant data were largely collected by a single diver, hence no analysis of this data set was made.
Variance components were calculated using non-transformed data. The statistical significance of the major treatment effects was also assessed using F-tests. In contrast to calculation of variance components, data for variates other than species richness were Ln(x+1) transformed before inclusion in the ANOVAs used to calculate F-values.
Effect of macroalgal cover on density estimates
The hypothesis that fish density is visually underestimated by divers in the presence of dense macroalgal cover was assessed by comparing observed changes in fish density in experimentally-cleared blocks with changes in control blocks, both for trap and visual estimate data. Trap data provided an index of fish density that is unaffected by the visual efficiency of divers.
Several specific predictions were tested using the ratio of fish density in cleared blocks to density in control blocks (the clearance ratio), as follows:
1. If fish density and visual efficiency are both unaffected by canopy clearance then clearance ratios for both visual and trap data should not change following habitat manipulation, and should approximately equal 1 (assuming no major pre-existing differences between control and manipulated blocks).
2. If fish density is unaffected by algal clearance but visual efficiency of divers increases then the clearance ratio for visual data should rise following habitat manipulation whereas the clearance ratio for trap data should remain constant. 3. If fish migrate into cleared areas following manipulation then clearance ratios
for both visual and trap data should rise.
4. If fish emigrate from cleared areas following manipulation then clearance ratios for both visual and trap data should decline.
Sampling errors associated with size estimates by divers
Errors made by divers when estimating fish length have been determined by comparing the estimated length of tagged fish sighted underwater with direct length measurements of the same fish after trapping. The three divers used to estimate fish length were experienced but lacked specific size-calibration training with objects of known size underwater. In analyses, length estimates for the same fish have been considered independent, providing that re-sightings were made by different divers or on different days.
Two components of visual error have been considered: (i) systematic errors in accuracy associated with divers consistently under- or over-estimating the length of fishes (bias),
and (ii) the variability in diver estimates of fish of a particular size (precision). Size
error was calculated as the difference between observed and measured size of animals. Precision has been determined as the standard deviation of bias measurements.
RESULTS
Changes in density following algal clearance
Changes in algal assemblages over the 12-mo period following plant clearance are shown as MDS plots for the three sites (Fig. 4.1). Algal assemblages at all sites were extremely diverse, with a total of 74 plant species possessing >5% cover in all quadrats recorded. Different macro-algal species dominated at different sites, hence variation between sites was much greater than that associated with experimental treatments. Within sites, algal assemblages changed little through the year for control blocks but exhibited a major change immediately after algal clearance in manipulated blocks, returning to near control values after 12 mo.
Fish assemblages generally showed comparable biotic patterns between sites, times and treatments as for plant assemblages, but with some differences in magnitude. Sites possessed lower biotic distinctiveness for fishes than for plants. Seasonal variation for fishes was greater than that observed for plants, and a higher level of overlap was evident between treatment and control blocks, particularly for the Return Point site (see Fig. 4.1).
1 3 Stress: 0.14