B. Censusing Fish at Aggregation Sites

While seemingly a straightforward task, counting aggregating fish and calculating the increased densities that represent an aggregation can be a challenge in both design and execution. Nonetheless, the challenge is well worth addressing. The information to be gained, especially if part of a long-term monitoring program, can be very valuable for gaining an understanding of changes in numbers over time as well as better understanding the nature and dynamics of the aggregation itself. Remember that in order to identify a site as a spawning aggregation site, surveys of the area outside of the spawning aggregation may also be necessary. In any event, the methodology discussed in this section can be adapted to both aggregated and non-aggregated fish.

There are three approaches to assessing numbers of fish in aggregations, underwater visual census (UVC), collection of fisheries dependent data and remote surveillance techniques. Each has its advantages and disadvantages and each must be interpreted with an understanding of associated limitations and biases. In this section we will cover UVC and remote surveillance. Monitoring via the fishery will be covered in Section VI.

Why is it such a challenge to meaningfully monitor aggregating fishes? In the first place the diving conditions are often difficult; deep water and limited bottom time, at dusk or even at night time if spawning is to be observed, strong currents and often the presence of hooks or other gears in the water. But these may be the easier problems to deal with. The single most difficult task is accurate assessment of the number/density of fish at an aggregation site. As we have come to learn more about aggregations of different fish species, or the same species over time, we have also come to know how variable aggregations can be in time and space. For example, the area of greatest fish density can vary within a given aggregation site from year to year as can its timing in a given month or in relation to moon phase (e.g. Epinephelus guttatus). There may be diurnal patterns in density or total numbers at specific aggregation sites (e.g. Plectropomus leopardus). Numbers, density and sex ratios can change substantially during the days leading up to spawning (e.g. E. polyphekadion) or at any one moment at different places within a given aggregation. The timing of aggregation formation of the same species can vary even within different aggregations located within 20 km of each other (e.g., P. areolatus). In other words, it is not possible to simply go out to an aggregation site, do a couple swims and expect the counts to be meaningful. Careful planning is essential, it must take into account the various factors that could influence the quality of your results, and ensure that your data are representative of the natural situation.

This section covers different approaches available for measuring fish numbers and assessing density, the biases involved and problems associated with the various approaches. It covers the considerations essential for developing a robust, standardized and repeatable sampling protocol and briefly touches on other aspects of underwater surveys such as the assessment of size and sex of fish underwater. We also touch on questions of accuracy (=the closeness of a measurement, or estimate, to the true value of the variable being measured, or parameter being estimated) and precision (= a measure of the degree of concordance among a number of measurements or estimates for the same population, reflected by the variability of the estimate). We outline details of the methods that have been used or could be adapted to count fish in aggregations and discuss their respective merits and demerits, discuss the question of validating fish counts, address sources of error and identify some solutions to problems raised. We also provide illustrative examples (boxed) of different approaches, whenever possible, that have actually been applied and published in the scientific literature. We finish with a brief summary of the key points to consider when designing and implementing a monitoring program based on UVC or remote surveillance.

A Note of Caution

Given the many problems of assessing spawning aggregations, we can not, unfortunately, provide an easy step-by-step guide because one size can not possibly “fit all”. Each field situation is unique and the available techniques need to be understood and adapted to each different circumstance, bearing in mind the considerable natural variability expressed even within single aggregations of single species. What we provide are guidelines, examples and options. To be valid, the sampling protocol must be properly designed and repeatable, not easy to achieve in the underwater environment, under any circumstances, let alone under the dynamic conditions often associated with and exhibited by spawning aggregations.

Estimating the Number of Fish by Underwater Visual Census (UVC)

If aggregations are small or fish are few, it might be possible to count all the fish. However, most cases are different. When there are too many fish to count, an estimate of the total number is achieved by assessing the number of fish in a small and known area of an aggregation and factoring up these numbers by an estimate of the total area over which the aggregation extends (see below). As a general rule we can say that most studies that have been successful in obtaining high confidence data on relative population numbers and density have either limited the area surveyed, if the overall aggregation is large, or had relatively low numbers of fish. However, in most cases, we must accept that such estimates have a high, and unknown, error value. There is no easy way to check the accuracy of such estimates, except by photographing or video recording the entire extent of an aggregation and subsequently counting all the fish present (which doesn't include those temporarily hiding or outside the area), or by having fishers capture nearly all the aggregation which can then be documented (posthumously). Neither approach is likely to be possible under most circumstances, nor, is the latter case desirable (and would certainly not be repeatable).

Methods of assessing the numbers of fish in an aggregation by UVC should be repeatable. Basic standards for underwater sampling (e.g. English et al., 1994) and diving safety apply, and workers should familiarize themselves with these.

Underwater Visual Census (UVC) Methods

There are many important decisions that must be made when designing UVC surveys; when, where, how and why should the surveys be done are the questions to be addressed in the following section. It is important, however, that, before planning any survey of an aggregation, a preliminary roving dive at the site be conducted to provide basic information on extent of aggregation, depth range involved, water conditions, and to assess the order of magnitude of fish present and their responses to divers. Without this important information, it will be very difficult to plan a safe and scientifically meaningful survey.

UVC – Repeatable Methods

Visual estimates are based on quantitative measures that include the whole aggregation or that can be interpolated to include the entire aggregation. Most of these approaches fall into the category of “transect methods” although fixed-point counts have also been used. Also crucial is some formal measure of the area of bottom being surveyed, which is where a mapping survey is useful (see below). At the least, some type of quantitative area measure of some portion of the substrate is important. For repeatable surveys to be possible, it is essential to have some method of surveying the same area during each survey. This could be based on distinct natural features on the bottom, used for reference, or permanent floats or markers attached to the bottom. If natural features are used for reference, it is important that these be carefully documented, principally through mapping of their positions, so that someone else can repeat a survey of the same area at a later date (see mapping Section IV.A).

The problem of how to sample fish numbers in an aggregation is tricky and each species and site presents its own set of challenges. At present the best surveys have yielded only an approximation of actual numbers for any aggregations numbering more than 30-50 fish. The worst case is where fish are dense, distributed from the bottom up into the water column some distance, moving constantly, are disturbed by human presence and are often hiding in the reef. In this sort of a case, we would be fortunate to obtain a value that is within two or three times the true number.

If actual numbers are considered to be impossible to count even while running transects, or if transects can not, for some reason, be completed, then it is valid to make an estimate of fish numbers present by using some form of index of abundance. For the SCRFA global database, we created an index to describe the peak (maximum) number of fish observed for a given species in an aggregation at one time. While only approximate, these categories nonetheless provide an indication of aggregation numbers that can be compared over time: 1-10 fish; 11-50 fish; 51- 100 fish; 101-500 fish; 501-1,000 fish; 1001- 5,000 fish; 5,000-10,000 fish; > 5,000 fish.

If the observer is at all serious about accuracy of counts, and sufficient funding is available, it is recommended to video record the transects for later analysis while visual counts are being made. Modern video cameras require no more attention than to point and push the record button, so having the camera operate while swimming along counting fish is not a major problem. Potentially a divers' buddy could do the video recording at the same time that counts are being made. Concurrent manual counts and video taping will be useful in eventually addressing the questions of the differences in data obtained from the same aggregation by both methods. Such assessment will eventually allow us to refine our methods for doing surveys of aggregations.

Transects and Underwater Visual Census by Moving Divers

Once the aggregation area or site has been identified, measured and mapped (see Section IV.A above), it is then possible to plan the aggregation surveys. Here we summarize the basic techniques and principles applicable to underwater visual census surveys; note that these were largely designed for non-aggregation situations and will need to be adapted to surveying aggregations as our experience increases. The available survey methods need to be carefully adapted to each situation and species but always with an eye for repeatability, scientific rigor and type of statistical analysis to be conducted (if any). Careful questions also need to be asked prior to starting such surveys regarding the specific objectives to be achieved. Most frequently, these will be questions of aggregation size (in terms of fish numbers or density) but there may also be interest in recording social structure, changes in fish numbers or diversity during the day, or over the course of the aggregation, etc.

Decisions need to be made regarding the size and number of sampling units (i.e. length and width of transect and total number of transects needed), i.e., how long should each be and how wide (the latter could be determined in part by visibility and general density of fish) and how many should there be. The effective survey width must be estimated either visually (this needs experience) or by using markers previously placed on the substrate (see below). Length of transect will be determined by the area of the aggregation or subsection of aggregation to be sampled, as well as other factors such as depth, current, etc. and it would obviously be best to survey as large a proportion of an aggregation as possible. In terms of the length of transects; in most cases these will be used to sample larger, more mobile reef fishes in aggregations. As a rule of thumb transects should be at least 50 m long although this recommendation is based on general survey work, not that on spawning aggregations (Samoilys and Carlos, 2000).

If there are relatively few fish and the aggregation area is small, then one transect that covers the entire aggregation area could be used. If the same transect path is followed every time and enables counts of all fish present then these counts would be comparable between different points in time. To get some measure of precision such counts could be replicated but this may not be feasible given the often short duration of maximum numbers of fish in any given aggregation. This may or may not be important but does need to be considered for long term monitoring programs.

If more than one replicate is needed because not all fish can easily be counted or the aggregation is large, then a decision is necessary on how many replicate transects must be run to provide representative counts from a given aggregation site; note that the following discussion is based on standard UVC surveys that need to be modified to an aggregation setting but that many of the underlying design principles are relevant. It is important to run sufficient replicates to account properly for variability between transects that might reflect different densities throughout the aggregation and produce a representative transect count (this applies to cases where the fish are not randomly distributed which may require a stratified sampling approach whereby sampling effort is stratified according to factors that may influence fish numbers over the aggregation such as differences in habitat or depth). To determine the necessary number of transects, ideally a

Figure 13. Example using cumulative transect counts to determine the minimum number of transects necessary to run on a given aggregation to obtain a representative fish count per transect. A similar plot of standard deviation (SD) can also be calculated from the column of transect counts to provide an indication of precision using a formula, (SD/sqrt n)/mean, where n is sample size, which could be used to compare with the precision of other similar data. (YS)

preliminary study should be carried out, which involves more transect runs than you think you will need, to produce a simple plot of cumulative mean densities (per transect) with increasing number of transects run. With the data, plot a graph (Fig. 13) of the cumulative mean counts of fish per transect (this means that, say, after 4 transects, you sum all of your counts for the four transects and divide the total by four). Plotting this information will yield a graph that will indicate when the means are beginning to stabilize. In this case, and with the variability in the data provided, 7 or 8 transects are needed for the cumulative mean transect fish counts to become stable. This means that, in this case, a minimum of 7 to 8 transects were needed to assess the fish number and density at the time of the survey (note that such replication is often not possible). The results provide a measure of average density for the aggregation at the time of sampling (remember that density can change, however, from morning to afternoon of the same day and between days – see below under when to survey) and some measure of variability (by calculating the overall mean and standard deviation from all transects run). In non-aggregation survey work, and in the absence of preliminary data, a ‘rule of thumb’ suggests that about 10 replicates are advisable (Samoilys, 1997a, Samoilys and Carlos, 2000). Multiple transects can be run by a single diver or involve multiple divers.

Decisions also have to be made regarding where to place transects within an aggregation. There may be little choice because of depth constraints or because of the shape of the aggregation that may, for example, follow a shelf edge contour or run along the walls of a reef channel or slope (Fig. 14). Certainly transects should be placed in areas that appear to be representative but this may be very difficult to judge. Whatever are the practical considerations in the laying of transects, their placement should be systematic. If there are clear differences in densities around the aggregation, the best design is random stratified sampling, but in this case strata need to be identified (e.g., core area [see below], depth range, shelf edge contour, etc) as already discussed.

Figure 14. Typical reef slope with hypothetical aggregation showing types of transects that could be run (right panel). There could be a single transect (A) run in the direction of the current, a deep transect run first in the direction of the current and returning in shallower water upcurrent (B), or a series of transects up and down the reef slope working along the slope with the current (C). All transects are run relative to marker buoys set up on the bottom and long transects (such as A and B) would be counted as segments of equal length (PLC)

One approach that has been used in large aggregations, or where the entire aggregation is not accessible to divers, is to select a core area (i.e. a specific stratum) that appears to contain the densest concentration of fish, and follow trends in core (maximum) density within an aggregation period as an indication of aggregation build-up (e.g. Rhodes and Sadovy, 2002). This provides a maximum density measure at the time of sampling and may include a large proportion of fish assembled in the aggregation. The location of the densest concentration of fish should be tracked over time in case this moves and the core area shifts. However, care must be taken in applying the data collected from core counts to the aggregation as a whole since core densities are less useful than average densities for estimating total aggregation numbers. This is because lower densities of fish occur in non-core areas and the core numbers can not be assumed to represent a typical, aggregation-wide, density. Importantly, it is also possible that it may not provide a good indication of changes in aggregation numbers over time. This is because of the possibility that, even if the total numbers of aggregating fish decline (say due to fishing), a core of consistent maximum density might still continue to form despite an overall decline in aggregation fish numbers. At present we do not yet understand enough about aggregation dynamics to understand the significance of such ‘core’ measurements and so these must be carefully reported and interpreted accordingly. Further research might elucidate the significance of such cores but their use for monitoring should only be considered to address certain specific questions. This example also highlights the need to clearly establish why a survey is being done since this could substantially influence the survey approach taken.

In general, considerable care is needed at each decision point when designing a monitoring program for aggregations, taking into account the incredible variability of aggregation density that can occur over time and location, even within a species. Every attempt must be made to design a method that is repeatable, representative and provides some idea of precision, though total counts may not have a measure of precision. If not, the data obtained may be of little value, and the money and time spent getting them squandered. Attempts should also be made to evaluate accuracy by independent means although in many cases this may not yet be possible.

Underwater Visual Census and Stationary Divers

In some situations, possibly because of strong currents or because fish might be wary of divers in close proximity, it may be preferable for the diver to hold a fixed position and remain stationary (e.g., Sancho et al., 2000). As one example, observers may remain tethered well above the substrate at a fixed location recording all fish within a prescribed area within view. Such tethering can save the diver energy and allow longer survey periods (since the diver stays in shallower water) than if swimming along the substrate. The influence of moving divers on estimates of fish density has been demonstrated in non-aggregation studies (Watson et al., 1995) and needs to be considered when planning aggregation surveys.

UVC can be conducted by what are known as ‘point counts’ which involve a stationary diver who slowly turns through 360 degrees counting every target fish out to a pre-established, and known, radius (Samoilys and Carlos, 2000). As for transect width, this radius will have to be adapted to local conditions of visibility and fish density and is unlikely to be more than a few meters for aggregation studies. As for transects, replicate counts should be made at the same site and also consideration given to conducting point counts at several places randomly or haphazardly selected within the general aggregation area. Replicates in the order of 10 should be considered if possible (Samoilys, 1997a; Samoilys and Carlos, 2000), in the absence of preliminary surveys. For point counts it is also necessary to determine how long a single 360o rotation should take; again, preliminary trials should ideally be conducted to ensure that rotation is slow enough to count all or most fish but not so slow that fish are double counted. An excellent reference for stationary visual census is Bohnsack and Bannerot (1986).

In summary, as for transects, the number of replicate point counts must be determined. With insufficient samples there is not enough power to distinguish between mean measures from different times and places. On the other hand, effort and funds are wasted if many more samples are taken than are needed for scientific rigor. Methods for determining adequate sample sizes are discussed by English et al. (1994) and Samoilys (1997a) and references cited therein and should be consulted. The good news is that in most cases, fancy models are not needed, just good planning and design. The bad news is that there is no ‘sure’ formula for aggregation surveys that must be adapted to different circumstances and constraints.

How to Measure Aggregation Areas

If there is interest in assessing overall aggregation numbers, then at some point, the extent of the aggregation site must be measured. Many times it is most convenient, particularly when an aggregation is of limited duration or fish disturbed by too much diver activity within the aggregation site, to mark the edges of the site in advance, or at the time, by some type of marker that can be found later. The extent of the aggregation can be determined later based on the location of the markers, and an underwater survey done with compass and tape. If an accurate map is already available for the bottom, the edges of the aggregation could be plotted on that relative to known locations indicated on the map.

Markers could be several types. Painted rocks have been used and different colors could represent different information, such as markers placed on different days (Shapiro et al. 1993). Small lead fishing weights could be similarly used, and short lines with floats could be added to weights to allow the markers to be more easily found and surveyed. If lines with floats are run to the surface and the area of the aggregation relatively large, the locations of the markers could be determined by GPS receiver from a small boat, and a rough area, within the limits of GPS determination, could be made (Rhodes and Sadovy, 2002). Markers can also be used to indicate the locations of specific fish, as was done by Shapiro et al. (1993), which are later analyzed for spacing and social structure. Similarly, the location of a single fish, such as a large identifiable (note that, for many species, slight natural differences in body markings make it possible to distinguish between different fish) male grouper, might be marked at regular intervals during any given time period (hours, days, weeks), by placing marked weights each time.

SOME CASE STUDIES

There are many ways in which a diver or divers can survey an aggregation site for numbers, density, distribution, etc. of aggregated fish. For example, multiple divers could follow parallel transects within marked-out areas covering a significant proportion of the aggregation, and pool their data (Shapiro et al; 1993). Alternatively, one or several divers could randomly survey transects across an aggregation site to produce replicate transects during single aggregation periods or multiple estimates at the site throughout the year (Samoilys and Squire, 1994; Samoilys, 1997b; Zabala et al. 1997a, b; Rhodes and Sadovy, 2002). In one case, multiple divers sampled different areas of an aggregation site and later validated their counts using video footage (Sala et al., 2001). Specific cases are given of methods used and how these have been adapted to local situations, but original papers must be consulted for full methodology of survey design.

Shapiro et al.(1993) surveyed a large grid of roped squares each 33m on a side that were, over a number of days, built up to seven square sections over a representative area of a red hind, Epinephelus guttatus, aggregation site. Squares were roped off into 8.25 m wide lanes and four divers were used to simultaneously swim across the entire grid in the four lanes. To determine the internal aggregation structure of fish, a colored rock was placed on the bottom at the location where a red hind was first seen and different colored rocks were used on different days. After the aggregation dispersed the position of each rock was determined by measuring the distance from the rock to the two nearest boundaries of the grid. For each day, the distance from each fish to its nearest neighbor was determined and the mean observed nearest neighbor distance was compared to the mean distance expected if all fish were randomly distributed within the grid. Internal structure of the aggregation was initially estimated by counting the number of individuals in 34 clusters (a cluster was defined when individuals were within 3 m of each other; in this study fish were significantly clustered in groups of 2-7 individuals in the aggregation). The sex ratio in 12 clusters was examined by spearing all 34 individuals within those clusters. Maximum fish density noted over several years of aggregation was 7.6 fish per 100 m2.

Samoilys (1997b) and Samoilys and Squire (1994) studied coral trout, Plectropomus leopardus, aggregations at Scott Reef and Elford Reef off Cairns, northern Great Barrier Reef. Fish were surveyed visually over a fixed route that was marked on a map of the aggregation site. Areas of aggregation were 1700 m2 for Scott Reef and 3200 m2 for Elford Reef. Counts of fish ranged up to 130 fish at Scott Reef and 60 at Elford Reef, equivalent to, at most, 44 fish per 1,000 m2 , a level at which fish can be relatively easily and accurately counted. Counts were conducted with a minimum visibility of 7 m. Regular monitoring was conducted year round at dusk around the time of new moon monthly during the spawning season (Aug to Dec) and bimonthly at other times. During the spawning season, surveys were done weekly or biweekly at various stages of the lunar cycle. During surveys, the observer counted and estimated size of fish seen and probably sex which were plotted on the map of the aggregation site. Censuses were standardized to 25 minutes. This took the observer around the aggregation site at a slow but steady pace. The observer could search a width of about 10 m, swimming about 1.5 m above the bottom. While the fish may move during the course of the census, the inaccuracy introduced was considered acceptable since the bias was present on all surveys. Sex was associated with specific color phases after confirmation of sex and color by spearing selected fish for dissection (Rimmer et al., 1994).

Zabala et al.(1997a and 1997b) used two different transects, one to estimate changes in density year round (200 m long by 10 m wide swum at mid-day) and the second to survey different depths and topographies where groupers occur (100 by 10 m swum a various times of day). These provided consistent, repeatable censuses for their work on the dusky grouper, Epinephelus marginatus in the Mediterranean. The first transect usually took about 40 min to swim while the second took about 20 minutes. Fish numbered up to about 40 per survey on transect 1 (20 fish per 1,000 m2 ) and up to 32 on transect 2 (32 fish per 1,000 m2 ). These numbers are easily and accurately counted, but the density overall is lower than that found in many tropical species, such as the Nassau or camouflage grouper for which this method is less likely to be applicable (see Fig. 6).

Rhodes and Sadovy (2002) did morning and afternoon visual censuses of an Epinephelus polyphekadion aggregation in Pohnpei, FSM. The fish were concentrated along a slope-reef wall interface. Because the main part of the aggregation was at 30-35 m depth (severely limiting time for the survey), censuses were done in the direction of the current (typically along the reef), swimming from one boundary of the aggregation to the other. The limits of the aggregation along the slope-wall interface were determined by putting surface marker buoys at those limits and later determining the positions of the buoys using GPS. An area of known size was used to count numbers of fish on each survey, providing data on trends in core (maximum) density for a given monthly aggregation. Maximum density recorded was about 5 fish per m2.

Sala et al.(2001) conducted daily UVC on Belize Nassau grouper aggregations starting before spawning and before fishing activity began. Their census protocol had three divers and each diver surveyed different portions of the overall aggregation area. Groupers were found on coral ridges (spurs) and each ridge ‘population’ was counted. Shelf edge areas were counted using continuous 75 by 20 m transects within the actual spawning site covering approximately 1.5 ha. There were 1-3 large grouper schools within the aggregation; counts of groupers in the water column were carried out for each grouper school by each diver separately and later averaged across all divers. To evaluate estimates of grouper school size, schools were videotaped using digital video. Counts of each school were made on a monitor screen for comparison with UVC data.

Sancho et al.(2000) used a fixed area of 170 m2 within a larger spawning area to characterize fish abundance and quantify spawning activity. Observations were always done by one person to reduce inter-observer bias with the diver's recording position kept constant, lying on the bottom and 5 m away from the edge of the sampling area. Spawning observations were noted every minute during 30 min observation periods. The total number of spawning fish was estimated for 34 species at the beginning and end of each spawning observation period. However, only data from the 11 most active spawning species were reported.

Non-repeatable Survey Methods

Non-repeatable surveys can be undertaken, but their application to long-term study of spawning aggregations is limited. This is because they are either destructive (i.e., involve removing all fish) or because the survey design does not allow repeatability. A good example of this is the "timed swim" in which a diver visually counts the fish occurring along a swath of bottom, usually only a few m either side of the diver, determined by water clarity and bottom communities (Sluka, 2001). The diver swims for a set amount of time (generally 10-20 minutes, but in some cases up to an hour), rather than covering a certain and known distance, and counts all target fish(es). If the diver swims at 1 m min-1 (an average sort of speed while doing another task like counting one type of fish, and assuming no current), the diver would cover about 100 m in 10 minutes, with a swath width of 10 m, equaling 1,000 square meters. The direction of timed swims is usually "random" or more typically “haphazard” but in either case swims can easily cross habitat boundaries, with consequent differences in fish density. In the case of aggregations, timed swims may take the observer into or out of the area of an aggregation. If the diver swims to stay within the aggregation, then the direction is no longer random.

The true “timed swim” is somewhat similar to the manta tow method for estimating benthic habitats. It provides a semi-quantitative estimate that may be useful for pilot studies on aggregations and may be useful for preliminary work to find and initially survey aggregation sites. A timed swim can be improved somewhat by using a video camera to record what the swimmer is seeing, as the recording can provide a general impression of the abundance of fish and the conditions during the timed swim.

Remote Surveillance Techniques

Imaginative use of video techniques may hold promise for providing some answers to the questions of absolute abundance in the aggregation. For example, a defined area might be surveyed using normal visual census methods. Then an autonomous video camera system could be moored above the site (if the water visibility is suitable) and it could monitor the fish in the site over a given period of time. This has yet to be done, but, especially with recet advances in video technology, are like to be entirely feasible.

Some video techniques might provide useful information, particularly as the systems envisioned can function without a human present, eliminating most of the “human presence” factor. An autonomous video system mounted on a tripod that provides a view of an aggregation area could sample the aggregation at various times by recording for different periods during the day. Such a camera is limited in its field of view, but such a view would be consistent over time. Another possibility would be to “mount” a video system above an aggregation area, sort of an “eye in the sky” approach. The buoyant system would be moored on a three or four point mooring directly above the reef pointed straight down. The camera should be high enough to cover a broad area and low enough for the fish to be identifiable. It could be started to run continuously or connected to an Autonomous Underwater Video System (AUVS) timer. This and the mounted, unattended camera method have potential to address the “hidden fish” (see below) question, a big factor in visual surveys.

A useful method for perhaps covering an entire aggregation in a reproducible manner would be to mount a video system vertically on the front of a Diver Propulsion Vehicle (DPV), then from a height (depth) well above the aggregation, video record the entire area of the aggregation, making multiple passes if necessary over the site, until the entire area has been well covered. Tapes would be analyzed later and density data could be extracted also, if the area of the bottom (through a previous mapping survey?) was known. Such a method would depend on water visibility and behavior of the fish concerned.

When Should Monitoring be Done?

Having decided why, what and how to survey, decisions must also be made regarding when surveys should be conducted. If aggregations typically occur at the new moon, for example, then this would be an obvious period to concentrate monitoring activity but we have to first discover when an aggregation is most likely to occur. Moreover, we now know enough to say that timing of spawning can be quite variable, even for given species at particular spawning sites, both within and between years within species, such that monitoring should initially be done at different moon phases and in the typical non-spawning period to ensure that important information is not missed. We know, for example, that aggregation timing for a given species can vary within a country in a given year (e.g., Johannes and Lam, 1999; Johannes et al., 1999), occur in slightly different months in different years (Rhodes and Sadovy, 2002) or even at different moon phases (e.g. Sadovy et al, 1994a). Such examples warn us to be careful in determining timing of monitoring studies. Moreover, if aggregations form in different months each year at a particular lunar phase, do we sample them in all months (the decision will reflect why the survey is being done)? We do not yet know whether the same fish spawn each time or whether different individuals might be involved in different months of the same year. Whenever the timing of aggregation is not well-known, sampling of markets and discussion with fishermen may provide additional information useful for planning monitoring activities. Decisions also have to be made regarding what time of day to monitor an aggregation given that fish numbers can vary markedly at an aggregation site over the diel cycle (e.g., Samoilys, 1997b). Preliminary studies should involve regular and frequent surveys if a species is being studied for the first time.

Sources of Errors

There are many possible sources of error in assessing fish numbers underwater, even when a monitoring protocol has been properly and carefully designed. These sources of error include, but are not confined to, substrate complexity, fish behavior, between-diver differences, aggregation types (some types are easier to assess than others), and double-counting (i.e., counting the same fish twice).

The humphead wrasse, Cheilinus undulatus, is one example of fish behavior affecting counts. Work in Palau using both snorkelers and SCUBA divers swimming along the reef would tend to drive all the fish present ahead of them, making it impossible to obtain anything approaching a reliable count of numbers. In this case, it made no difference what method was used to survey the fish, which simply could not be counted accurately by a human swimming in the water. For this species a stationary video camera was placed and the assembling fish were clearly visible in the absence of humans (see remote surveillance below) (P. Colin, unpubl. data).

The number of holes and other shelter available for fish can have an important effect on counts of fish found at a given site. Since most aggregations occur where there is significant coral cover, often with an abundance of hiding places, fish that remain within crevices, under ledges and otherwise hidden would produce an underestimate of the actual numbers. For example, a behavioral study on the southern Great Barrier Reef showed that individuals Plectropomus leopardus, a relatively non-cryptic diurnally active grouper, spend up to 35% of their time hidden under cover of coral during non-spawning times (Samoilys, 1987). Sadovy et al. (1994b) examined an aggregation area of Mycteroperca tigris off Vieques, Puerto Rico, in which the bottom consisted of layers of elevated coral heads that provided a huge number of hiding places and made any meaningful counts of fish abundance impossible. The number of fish hiding may also depend on the time of day, particularly for predators like groupers, and the disturbance effect of human presence. Without extensive study, it is not possible to say exactly how many fish are hiding at any given moment, although as already discussed, remote surveillance might be of value here.

A detailed discussion and attempt to deal with errors of hidden fish, or undercounting and double-counted by divers, on a species-specific basis was made by Johannes et al. (1999). Their solution of applying an exact percentage "correction factor" for hiding and double counts of each species was based on qualitative observations of fish behavior. To apply precise "correction factors" to counts of estimated numbers of fish (which may be inaccurate by a considerable percentage), however, confuses the issue of aggregation size. A more satisfying approach is not to attempt to apply correction factors, but simply state that a portion of the fish were missed or double-counted. The same study compared snorkel and SCUBA estimates (although not from the same time period) and tried to determine a sample design that minimized the amount of field time (and hence expense) necessary for the study objectives. The analyses were only variously successful but are interesting in their attempt to deal with some of these important issues.

Errors can also be introduced by what divers themselves are doing. For example, errors attributable to different divers being involved in monitoring (inter-diver error) are probably typically underestimated and rarely evaluated but can be important and could certainly be readily addressed by appropriate survey design. Many fish species appear to be affected by the noise/bubbles made by SCUBA gear. We may find that the introduction of rebreathing dive equipment may eliminate some of this diver error. See elsewhere (remote methods) regarding getting rid of divers altogether!

Finally, some aggregation types are simply more difficult to survey than are others. The hardest are those that involve very large numbers of fish packed densely and sometimes in 3- dimensions, coming up into the water column, or species that spread out over a large area and are very wary of divers (e.g. Lutjanus analis). Smith (1972) estimated at one site that the number of Nassau grouper was anywhere between 30,000 and 100,000, a massive range in estimated numbers. Species of acanthurids or lutjanids can likewise be very difficult to count while aggregated – there is a need for some bright biologist to develop survey methods applicable to such species. One possible approach might be release a known number of tagged fish into an aggregation and film or photograph the aggregation. If random mixing of tagged animals is assumed then an estimate of total aggregation size could be made by working out the proportion of tagged to untagged fish. There are likely to be some useful applicable methods from the bird literature that might be adapted for counting aggregating fish.

Validating Counts and Data

In the end there are few ways to validate the accuracy of any fish counts in aggregations even for properly designed surveys and hence interpretation of data must be made accordingly and carefully. When a permanent record of fish present is made, like a video survey of an aggregation, there is some ability to refine numbers by repeated analysis of tapes compared to numbers that might have been obtained in real time by counting fish (Sala et al, 2001). Even the best video survey and UVC, however, have to deal with the problem of fish that hide during the survey or taping and those that move can easily be double-counted.

Despite all these efforts to reduce errors, there are some species that are simply not going to permit a human being to accurately count them. For these fish, we have to accept that counts done by divers are highly inaccurate (but may reflect relative numbers over time), or find alternate methods to census them. For difficult species, video surveying with a remote system may be a solution, or fishery dependent methods may be more revealing (Section VI.A).

UVC AGGREGATION SURVEY DECISION TABLE
FOR REPEATABLE SURVEYS

  • Map the aggregation area
  • Qualitatively assess numbers of fish and any core group/dense schools
  • THEN, design survey and answer the following questions about the survey

Why?
What is the purpose of the survey? Must it be repeatable?

When?
When in the year, month and day should the surveys be conducted? Are non-aggregation surveys also needed for reference?

Where?
Survey aggregation site only? Survey both aggregation and non-aggregation sites? Survey entire site or only part of the site? Which parts to survey, a core area or randomly placed transects?

How?
First of all a preliminary survey of the site is needed to determine, depth, extent of aggregation, order of magnitude of fish numbers, effects of divers on fish and fish hiding behavior, etc. This information is critical for properly designing subsequent surveys.

(i) Swimming
Transect; width, length, replicates (one or more), stratified sampling?
(ii) Stationary
Point count: speed, radius, replicates (one or more)

Other General Considerations for Survey Decisions
  • Consider potential sources of error (inter-diver variability, effects of diver on fish, double-counting, hidden fish)
  • Determine statistical approach to analyze/compare data
  • When fish numbers are high, consider using an ‘abundance index’
  • Carefully document methodology so that others can repeat surveys and try to include a map of the site
  • Attempt independent assessment of fish numbers to assess accuracy, i.e., validation (e.g. underwater video)

Collecting Size and Sex Data from Aggregations

It may be possible to obtain size and sex (and thereby sex ratio) data from UVC surveys but considerable care must be taken in so doing and in clearly understanding the possible, and potentially considerable, errors that can be introduced by injudicious interpretation of results. For example, size data may be used to plot size frequency distributions of fish within an aggregation. While we recognize that it is reasonable to obtain some relatively crude information from estimates of fish size from aggregations to provide a general idea of size distributions subject to the caveats below, using such estimated information for performing comparisons between different aggregations and for the same aggregation over time requires caution. Data on sex of fish collected by UVC depends on earlier work linking particular color phases or behaviors with certain sexes, and for determining sex ratios there has to be some degree of confidence that it is equally easy to survey both males and females, as well as distinguish the sexes.

Size - care is needed in interpreting both the fish size data obtained by visual estimation underwater and the significance of the fish size data. In the first case, it is important to note that errors in fish length are involved each time a measurement is made. These can be several cm for a fish, or more, and may be different for different size classes. The use of calibration rulers can improve the accuracy of in-water length measurements and without such sticks divers may obtain a mean accuracy of 86% (St John et al., 1990). Nonetheless some error is likely to be involved. In the second case, we need to be cautious as to how the size data are interpreted, especially given the measurement error that might be involved (see Section IV.C for further discussion).

As long as the limits to visually determined size data are properly understood, meaningful comparisons of fish sizes over time or across locations are possible under the right circumstances. For example, Samoilys (1997b) did such a comparison between two reefs for size of P. leopardus. However, she had long experience with the sites and fish, and was dealing with only one species. In such a case, this comparative use of data can be applied with some confidence, but this study serves to point out the need for long and careful work to be able to apply size estimate data in any but the most rudimentary way. The problem with size estimate data is that, except in the case where a significant sample of fish can be measured from captured individuals, there is no true way to verify the accuracy or precision of such data. Training of observers, through wooden models of fish or other methods, is useful. However, considerable care is needed in training and retraining workers in size estimations using fish models if they are to apply their skills under field conditions and to living fish (Samoilys, 1997a). Such data can provide an indication of sizes (Fig. 15), but must not be confused with the level of accuracy obtained from specimens captured by fishermen. Another method involves a diving mask equipped with a calibrated bar mounted inside that allows fish lengths to be estimated to within 10% of their true value under water (Swenson et al., 1988).

Alternative techniques hold promise for providing accurate length data from aggregations, without having to resort to estimations of length by divers. Relatively straightforward is the use of a "laser scale" in which two or more laser pointers are enclosed in underwater housings at set distance(s) apart. The laser pointer effectively penetrates clear water

Figure 15. Comparison of estimated length by well-trained divers and actual length for models of fish assessed underwater. Source: Samoilys and Carlos (1992) Development of an underwater visual census method for assessing shallow water reef fish stock in the south west Pacific. Unpublished Final Report of Project PN8545, April 1992, Australian Centre for International Agricultural Research; used by permission.

for many meters and if two or more pointers are set up parallel to each other, either on a moving bar system (calipers) or at a fixed distance separating them, they could be used to measure the length of a fish underwater without coming close to it. The fish you wish to measure is video taped with the laser scale pointed on its side and the length is later determined by using stop frame video (Fig. 16). The main drawback of this system is that the laser scale needs to be pointed essentially at a right angle to the anterior-posterior axis of the fish for an accurate length (Fig. 17). Any deviation beyond about 10o from perpendicular to the fish axis results in measurements that are not accurate. The laser scale system illustrated in Figure 16 can have the lasers separated by either 5 or 10 cm easily and the interval selected depends on the size of fish you wish to measure. Visual estimates can be quite accurate but accuracy depends strongly on continued training and diver experience.

Harvey and Shortis (1997) described a stereo-video system for estimating fish length and later Harvey et al. (2000a, 2000b and 2002) compared this system to diver visual estimates of fish length. In general, they found the stereo video system to be more accurate than diver estimates of fish length. The ability of novice and experienced divers in a swimming pool to estimate fish length, based on plastic models, was similar for models of 10-50 cm in length (Harvey et al. 2000a). Small models (around 10 cm) were accurately estimated for size by divers, but at lengths

Figures 16 and 17. Fig 16 (Left) Laser scale system on top of underwater video camera housing. The four individual laser pointers are positioned 10 cm apart and project their red laser dots outward. Fig. 17 (Right) The laser dots (arrowed), 10 cm apart, on the sides of a surgeonfish, Naso literatus. The dots are not particularly visible in this black and white photo but more easily seen when in color. The total length of this fish was measured to be approximately 28 cm (PLC).

above about 25 cm there was considerable variation (often times 10-30% of model length) from reality in visual estimates. The means of the overall estimates of many divers did arrive at reasonably close estimates (mean error of 2.1 cm for experienced and 2.3 cm for novice divers) of model length, but individual estimates often had substantial errors. Since usually only a single observer is making size estimates in the field, the errors in such single-person estimates may often be substantial. Two or more independent estimators would be better, but may not be practical in the field. In a field comparison of the methods, Harvey et al. (2002) used three experienced divers to estimate fish lengths and found, under optimal conditions, high accuracy, but low precision to these measurements, compared to the stereo video system. They state that the estimates of experienced diver scientists have much lower statistical power than stereo-video measurements when trying to detect changes in mean length of a group of fish. If there are low numbers of fish being estimated or when the program intends to detect changes in mean lengths of 30% or less, the power is much less. These results show why all interpretations of visual length estimates must be done carefully and, despite the wishes of the observer, the human eye underwater is not a particularly accurate tool for measuring fish length from a distance.

Sex - in cases where sexes can be distinguished externally by color differences, or by distinctive behaviors, some idea of sex ratios can be obtained. This is assuming that sexes can be reliably distinguished and that both sexes are equally visible. In general for groupers, sweetlips, and some other species, males become darker, often the black blotches, or particular color patterns are exhibited when courting and spawning (e.g. Fig. 18, Figs. 37 and 38). However, color changes can be rapid and some species may not constantly display courtship or other color patterns associated with aggregations during the time of the aggregation. While some species may not exhibit external color differences, females may become so swollen with eggs, there is little trouble identifying them (Fig. 19). Great care must be taken when associating color forms with sex, however, and with evaluating sex ratios from visual censuses. In one study, the presence of two color forms was assumed to represent male and female when in fact both sexes displayed both colors (Colin, 1992). In another study, female behavior appeared to differ from that of males with females tending to stay more hidden and males more active and visible at the aggregation site (Sadovy et al., 1994b). If these impressions are correct, then visual estimation of sex ratio would have been biased towards the more obvious males.

Figure 18. Grouper male, probably Epinephelus merra in uncharacteristically blotchy coloration from a probable spawning aggregation at Enewetak Atoll, date and time unknown (PLC).

Figure 19. A female red hind, Epinephelus guttatus, swollen with eggs. A female like this is probably no more than a few hours from spawning, as the eggs are almost certainly hydrating with the mass of the gonad expanding as a result. (C. Arneson)

Concluding Comments

Carefully designed surveys are critical for determining numbers of aggregating fish, density, sex ratios, and to make conservation and management decisions based on such information. It is risky to draw conclusions for parameters for which you don't have data, and important to be careful how you interpret what data you do have. It is easy to be misled by the results emerging from a particular sampling design, as some of our examples have illustrated. Conclusions drawn from the results of surveys must recognize the limits and constraints of the survey and be made within the bounds of the sampling protocol applied. Most importantly, when designing any survey, ask yourself where, when and how the survey should be done and why it is being conducted in the first place. Is the expected outcome worth the time and money involved?

Ideally, it would be nice to count all fish on an aggregation site, but more typically, subsampling is done. This can be by UVC, as we have just discussed, or by surveying the fisher catches (Section V.I). It is important to remember that any methods of subsampling will be biased and that we need to understand clearly the limits of our sample data. In monitoring aggregations of 10s to 100s of fish, total counts are probably possible, while for 1000s of individuals, subsamples will be needed. For numbers of fish, size and sex, there are few ways of verifying data, and little check on the accuracy of data. We should, whenever possible, use data collection techniques that provide a permanent record. It is also critical to undertake surveys that are repeatable, particularly by other observers. This "repeatability by others" factor should be considered from the start in planning any aggregation survey and is the only way meaningful long-term data are going to be obtained on aggregation sizes. No matter what methods are used the data have to be comparable between years and sites at some level clearly evident to readers or workers. The level of comparability must be understood to ensure proper data interpretation.

Finally, two other general points need to be considered, the cost and value of monitoring. As we have come to learn more about aggregations and particularly more about how variable they can be for given species or in given locations, we have had to continually refine methodology to address potential sources of error. This means that care is sometimes needed in evaluating older (and sometimes not so old) literature that may not have had to consider a wider range of confounding factors or have involved sufficient replicates. It has also become apparent that monitoring of aggregations may not always readily provide the necessary information for management or conservation. This is the case for species that are very difficult to assess for absolute number at aggregation sites. There is certainly a need to improve our ability to count aggregated fish that occur in large numbers and to be circumspect of studies that claim to do so with any degree of accuracy (for example one attempt to use acoustic methods to measure, without ground-truthing the methodology, numbers of aggregating groupers suggested fish numbers in orders of magnitude higher than was indicated by direct observation, and in another case acoustic assessments using echo integration were found to be very sensitive to the total number of fish present with gas-filled swimbladders). While this section is about monitoring, we also need to consider why the monitoring is being done and whether there are other approaches to, say, population assessment (see Section V.I), that can or should be done outside of aggregations. We can not answer these questions here but wish to make it clear that they need to be considered. Also to be considered is the cost of monitoring. This can be very expensive. Johannes et al. (1999) calculated that in Palau, the cost of monitoring aggregations probably exceeded the market value of the fish involved. This may simply not be justifiable in the long- run, an important consideration when establishing monitoring programs and determining the best way to use limited funds and human resources.