Comparison of methodologies
Despite widespread need for MR monitoring, appreciation that different monitoring methodologies have different accuracies and costs, and recognition that the best surveys will consider species-specific approaches to monitoring because there is no one-size-fits-all option, there have been relatively few tests or comparisons of survey methodologies. This is even more surprising, given the advance in survey technology during the last 30 years, from traditional diver censusing or potting or angling to remote camera operation (Cole et al., 2001, 2003; Davidson, 2001; Harvey et al., 2001a, b, 2002a, b; Léopold et al., 2009; Pelletier et al., 2011). From a management perspective, there is a clear need to know the accuracy, reproducibility and costs of each survey method so that appropriate decisions can be made about their use.
In this study, UVC underestimated abundances and size of P. colias in that fewer fish were recorded and variability of estimates among sites and seasons was high when compared to the two other methods. UVC has less ability to detect P. colias than the baited methods, having recorded the highest levels of zero counts of all three survey methodologies. Zero counts occurred, despite fish being observed within the area, either behind the diver or outside the transect. Fish behaviour towards divers was generally neutral, consistent with observations of diver-neutral behaviour of P. colias at KMR (Pande & Gardner, 2012), but different from studies in other NZ MRs where P. colias have been reported to be diver positive (Cole, 1994; Willis et al., 2000; Davidson, 2001). Large variation in UVC estimates of fish abundance and LT resulted in low statistical power and the inability to detect differences between these two indices even at the reserve and non-reserve level. Both baited methodologies, EA and BUV, had comparable ability to detect fish. No zero counts were made at any of the sites throughout the study using the baited methods, although individual replicates of the EA and BUV did record zero counts which equated to only 10 and 11%, respectively, of the 64 replicates undertaken. Differences did occur between these two methodologies where the highest zero counts were observed, but the overall trend was consistent with both methods recording a greater number of zero counts within non-reserve sites. The increased ability to detect P. colias also came with lower sampling effort compared to UVC (64 replicates v. 144 transects). The increased ability to detect fish using the baited methodologies resulted in smaller s.e. at each level tested and increased statistical power to detect change. Baited methodologies also proved superior in detecting fish on surveys where visibility was reduced (e.g. the July survey). Estimates obtained using the two baited methods were less variable between surveys than those obtained by UVC. This greater ability to detect individuals on the baited methodologies is attributed to the fact that the fish were attracted to a centralized area (the bait) reducing the area needed to record fish within (under camera or on a hook) as opposed to the greater area needed by the UVC (within a 125 m2 transect). Illustrating its utility, other studies using such baited methodologies have been undertaken in inhospitable environments including EA being used in surf zones (Bennett & Attwood, 1991, 1993) where high turbidity influences visibility, and BUV being used at depths beyond safe diving limits (Priede et al., 1994; Priede & Merrett, 1996).
The three methodologies appear to sample different portions of the P. colias population. Small fish (<90 mm) were not observed by any of the three methodologies, but the EA surveys recorded fish of 150–470 mm, the BUV recorded fish between 160 and 488 mm and the UVC recorded a few smaller sizes (100 mm, 1·6 % of the 248 individuals) up to a maximum LT of 400 mm. Thus, the UVC may underestimate mean LT of fish compared to the other methodologies. Larger LT estimates recorded by the BUV and EA may result because baited methodologies are (inadvertently) biased towards sampling larger fish (e.g. the bait may provide a stimulus for fish to move towards it and larger fish may be able to outcompete smaller fish for access to the bait). For BUV, this bias was most apparent for size data because this method returned the largest LT estimates: values were 105% larger than the EA estimates and 126% larger than the UVC estimates. Present observations indicate that bias in the BUV estimates may be related to the dominant behaviour of larger fish that actively guard the bait and prevent smaller fish from visiting the bait [Willis et al. (2000) for P. colias in northern NZ]. Dominant behaviour around the baited pot was not consistent across all replicates because such dominant behaviour was absent in areas of high abundance, possibly reflecting density-dependent territorial behaviour in P. colias (Cole et al., 2001; Davidson, 2001). Lesser bias in LT estimates from the EA surveys may occur because captured fish were retained on board the boat for the 30 min replicate, thus precluding recaptures and also allowing other, possibly smaller, fish to access the bait. This is supported by the extended LT–frequency distribution of fish for the EA, although size of P. colias and time of capture (i.e. sequence of capture) are not positively correlated (Davidson, 2001).
Any assessment of different survey methodologies needs to take into consideration the scale of sampling, both its spatial extent and its duration. In this study, three different survey methodologies were compared using the operating conditions for each that are now effectively standardized amongst NZ researchers. It is noted that there are substantial differences among the three standardized methods in terms of search time and spatial extent per replicate such that each methodology has its own likelihood of encountering a fish and each methodology will therefore also have its own variability. In all cases, however, the methodologies employed may be modified to increase or decrease transect time or area (for UVC), time of deployment (for BUV) and duration of soak time (for EA), and in all cases the number of replicates may be changed to decrease sampling variance. Ultimately, a recommendation for the choice of one survey method over another must consider the method's variability based on its sampling duration and its spatial extent. The present test of the three survey methods takes such factors into consideration indirectly in the sense that the comparison is based on the operating conditions that are most likely to be employed by any researcher.
Logistical and time considerations play an important part in determining the methodology to be employed (a summary of advantages and disadvantages of the three methodologies is provided in Table III). Based on the present survey protocols, the EA and UVC methodologies were the hardest to organize, because the numbers of participants required in these surveys were greater than needed for the BUV. The EA required six people [four anglers, the principal investigator (P.I.) and the boat skipper] over 2 days. The UVC required seven people (six divers and the boat skipper), but only for 1 day. In contrast, the BUV required only three individuals per survey (the P.I., boat skipper and one assistant on the boat to deploy the camera), but for 2 days. It is appropriate to consider labour costs (time) because this reflects both the legal reality (e.g. health and safety regulations) to use properly trained individuals and also people who are available during an appropriate weather window, rather than having to rely on volunteers who may not be trained to a professional level and who are only available at weekends. Once the logistical organization was complete, the UVC surveys were the most straightforward of all the methodologies. Each buddy-pair undertook one dive at each of the four study sites and dives consisted of 20 min of bottom time: the complete survey took 1 day. In contrast, the EA and the BUV required a greater amount of field time: two full days were required. This was because the survey design included four 30 min replicates undertaken at each of the four study sites, with additional time needed to move among the sites and to set-up the equipment. The BUV was the only methodology that required large amounts of post-survey analysis. The other two methodologies required small amounts of data entry, but the BUV required an additional 3 weeks of processing to download tapes, determine the BCmax index and measure fish on the video. The processing time and field work made the BUV the most labour-intensive and time-consuming methodology. Such a large amount of laboratory analysis is an important consideration if limited time is available to undertake future research.
Table III. Summary of advantages and disadvantages of three survey methodogies to estimate Parapercis colias abundance and total length (LT)
Previous comparisons of fish survey methods, in particular for P. colias, have reported that the relative density of fish inside and outside the Cape Rodney to Okakari Point (Leigh) MR in northern NZ was similar for all the three methods tested here (Willis et al., 2000). In addition, it was reported that EA provided estimates of larger mean size compared to UVC and BUV, probably as a result of hook selectivity against smaller fish (Willis et al., 2000). In another study at Leigh, Willis & Babcock (2000) reported that P. colias were recorded at greater density inside the MR by UVC than by BUV. These two studies, plus the results presented here for P. colias at KMR in central NZ, illustrate the variability in estimates of fish size and abundance that is often reported for one species at different sites and at different times. In other studies, the biases and benefits of survey methods have been compared, but with contrasting outcomes. For example, Colton & Swearer (2010) tested UVC against BUV to survey fish assemblages in south-east Australia, whilst Langlois et al. (2010) compared stereo-UVC and stereo-BUV methods for surveying of fish assemblages in Western Australia. Both studies noted differences in the ability of the techniques to detect differences in fish diversity (e.g. all species, herbivores, cryptic species, mobile predators and territorial species) and fish richness (species and family level diversity). Ultimately, Colton & Swearer (2010) noted that if the need is to monitor fish diversity then multiple survey methods need to be applied, but if only one method can be used (e.g. because of cost or logistical constraints), then UVC is the method of choice. Their findings contrast with those of Langlois et al. (2010) who ultimately recommended the use of stereo-BUV for the surveying of fish assemblages, based on both greater precision and increased cost-effectiveness. The range of results and recommendations arising from these studies highlights the difficulties faced by managers and researchers alike as they try to find cost-effective and precise methods to survey fishes. Increasingly, tests of survey methodologies are highlighting the need to employ two or more methods when assessment of fish assemblages is being conducted (Murphy & Jenkins, 2010). Whilst increasing the accuracy and decreasing the bias of size and abundance estimates for fish of different guilds, this approach is inevitably more time consuming and more costly. The ongoing development of new technologies and the reduction of their costs will greatly assist with future fish monitoring which needs to be more precise and less intrusive (Murphy & Jenkins, 2010).
Final recommendations for future monitoring methods have to take into consideration not only efficiency and ease of use, cost, and power to detect change, but also other factors such as availability of historical data that may also be important. In the present context, the only available data for KMR are UVC counts (Battershill et al., 1993; Pande & Gardner, 2012), so although it has been demonstrated that the UVC methodology is arguably the poorest performing census method, it is recommended that it can be maintained for a short period of subsequent use to permit historical comparisons to be made with the existing data. Because of its ability to detect individuals, power to detect differences and low s.e. estimates, BUV is an excellent methodology for the monitoring of P. colias. In conjunction with UVC surveys, it is suggested that BUV surveys be conducted at sites inside and outside KMR as the most efficient way of censusing P. colias. While BUV may replace UVC for the reasons outlined above, a period of monitoring using both methods will provide some degree of comparability, and perhaps allow researchers to better interpret the historical UVC data in the context of present-day BUV data.
We thank J. Allen, R. Williamson, M. Fraser, G. Martin, M. Forsyth, M. Russell, B. Dudley, J. Brightwell, R. Hendry, B. Hudson, D. Herlihy, P. King, L. Lachowicz, J. Long, C. McDermott, R. McParland, A. Martin, K. Steger, J. Sweeney, R. Wake, B. Wear, C. Wightman, J. Williams and S. Williams for field and boat work, D. Wrightson (Kapiti Ranger) for assistance in the field and members of the Kapiti Marine Reserve Committee for their help, and S. Pledger (VUW) for statistical advice. Funding was provided by the New Zealand Department of Conservation to J.P.A.G.