Designing EM reviewing rates for WCPFC fisheries

Electronic Monitoring/Observation of tuna fisheries in the western and Central Pacific Ocean is currently undergoing various trials throughout the region to ascertain how it can best complement existing at-sea observer programs. The types of electronic monitoring being considered typically includes both camera and metered observation systems (e.g. location and gear usage sensors). Camera systems typically include some review of the footage collected by an analyst once the vessel has returned to port. While reviewing the footage the analyst collects the relevant data in a manner equivalent to being on board the vessel at the time of fishing. As footage can be viewed at speeds faster than real-time there are potential cost savings associated with less time required to view the entire footage of a fishing trip. Noting that an intent of electronic monitoring is to potentially ensure all vessels and all trips have video footage than can be monitored, a key question is what proportion of trips by each vessel and what proportion of a trip duration needs to be viewed in order to obtain the desired precision in the fisheries data fields for scientific purposes. Only viewing the required amount of footage by the analyst to achieve the desired precision present opportunities for considerable cost savings for implementing electronic monitoring programs. SPC-OFP in collaboration with Patrick Cordue (Innovative Solution Limited) initiated analyses and simulations to estimate the precision for each observer data field associated with differing region, fleet and trip coverage rates for WCPFC longline and purse-seine fisheries. A copy of this report is attached to this Information Paper. In this study, a general sampling scheme involving the random sampling of vessels, trips, and sets was investigated. The primary data were the observer and logbook data for the years 2016-2019 inclusive. Analytical equations were developed for the variance and coefficient of variation (CV) of an unbiased estimator of total catch for a given species, area, and timeframe. It was found, when estimating total catch for a given species, that the simple approach of sampling a proportion of sets from every trip was almost always superior to any alternative approach. Even complex stratifications using information that would not generally be available at the time of stratification did not outperform the simple approach. This is because between-vessel and betweentrip variation in catch can be large. When all trips are selected the between-trip and between-vessel variation in catch does not affect the variance of the estimator. The main focus of the study was catch rather than nominal CPUE as the precision of estimators sampling sets was almost the same for both. This is because the CV for total hooks is small compared to the CV for total catch (when all trips are sampled). Intuitively this is clear as for a given trip there is unlikely to be much variation in the hooks deployed per set but the variation in catch per set (for a given species) can be very high. The preliminary analyses in this report and the development of operating models to scale observer data up to the level of logbooks laid the foundation for the study (though ultimately the operating models were not important). From the preliminary analyses of longline fisheries, it was clear that for target species over a large area and an annual timeframe that sampling 10% of the sets on each trip would provide good precision for total catch (e.g., CV less than 10%). This is because of the large number of sets that occur each year in broadscale areas and because target species are caught on a majority of the sets. Indeed, for some target species, a sampling proportion of 5% was adequate.
For bycatch species, an estimator’s CV is not always a good indicator of appropriate precision. For example, when there is a catch of 20 animals for a species it is not necessary to have a CV of 10% or less (e.g., a 95% CI of about 16-24). It is generally of little consequence whether there were 10, 20, or even 80 animals caught. Depending on the tolerance for precision, for infrequently caught species (5-20% of sets) a sampling proportion of 10-20% will generally be adequate. For the range of species looked at for longline, over shorter timeframes and smaller areas, a sampling proportion of 20% was also generally adequate (e.g., to estimate catch at a regional level). For purse seine catches a sampling proportion of 20% was also generally adequate. For rarely caught species (less than 5% of the sets) the main consideration may be whether any of the species were caught or not. In this case, it was more important to consider the number of sets involved and the proportion of sets on which the species is generally caught (i.e. defining the threshold for the probability of incorrectly reporting a zero catch). For very rarely caught species it may be that to avoid falsely reporting a zero catch that most of the sets need to be sampled. Hopefully, in these cases the number of sets that need to be sampled can be reduced by restricting the area to the known habitat of the species. It is recommended that SC18 note the framework used to simulate the “footage review rates” of potential electronic monitoring in the WCPFC fisheries for scientific data collection and the outcomes of the simulations undertaken: • To maximise precision with minimize review rates for estimating total catch for a given species, the capacity to sample a proportion of sets from every trip was almost always superior to the alternative of a stratified sample of trips. • Assuming a sufficient proportion of the fleet has EM, randomly selecting at least 10% of sets from each trip for analysis would be sufficient for the estimation of target species catch (with CVs <10%) at the WCPFC spatial scale. • That increased sampling proportions will likely be required when good precision is required at a sub-regional level (as a rule of thumb 20% of sets sampled at random from each trip may be more appropriate). • By design, a higher proportion of sets may need to be sampled to meet specific criteria (e.g., to determine if a species of special interest has been caught or not).