The impact of bias in length frequency data on an age structured fisheries stock assessment model
Statistical age-structured models are widely used in fisheries stock assessment. These models have been become increasingly complex over recent decades, allowing them to incorporate a larger variety of fisheries data. These typically include information regarding annual fishery yields, indices of abundance and catch composition data, which reflect the distribution of ages in the harvested population each year. In some fisheries, age composition can be determined annually through the examination of annuli on hard parts, such as otoliths or scales. These methods are, however, costly, time consuming and require a relatively high level of expertise on the part of data collectors. Alternatively, length frequency distributions within the annual catch are relatively simple and inexpensive to acquire, and can be employed to extrapolate age structure given that some information regarding age length relationships in the population is known. This type of data is therefore critical for many age-structured fisheries models.
Length frequency data are compiled from length measurements of a sub-sample of the commercial catch. Even when they derive from a relatively large sample size, however, these data depend on a number of biological, economic and logistical factors. In some fisheries, for example, larger, more valuable fish may be separated from the overall catch and sold quickly, before port samplers have chance to gather sub-samples (Burns et al. 1983). This can reduce the relative frequency of large individuals in length frequency data. Alternatively, fish may become stratified in holding bins or storage containers according to size, due to their slippery texture and body shape (Hilborn and Walters 1992). With smaller, shorter individuals falling to the bottom where they are less likely to be picked up and measured, length frequency data may contain a disproportionately high frequency of large fish.
This study used simulations to examine the impact of these two types of bias in length frequency data on a statistical age-structured model. The model, which was similar to those used in stock assessments for black sea bass (Centropristis striata) and gag (Mycteroperca microlepis) in the southeastern United States, produced erroneous population estimates when given biased data. Length frequency data that contained too many small fish caused stock status estimates to became overly pessimistic, indicating that populations were more heavily depleted than was actually the case. This type of bias supported overly conservative management measures, which posed an unnecessary cost to fishermen. Conversely, when the data included too many large fish, estimates of stock status were overly optimistic, and supported management actions that did not effectively protect the stock from overfishing. These results indicate that the quantity of length frequency data alone does not protect against bias when using complex age-structured models. The likelihood and magnitude of bias in these must also be examined in order to determine whether results are likely to be biased. For a given fishery, it is therefore critical that potential sources of bias in length frequency data be thoroughly inspected, and that the modeling approach used to assess the stock be appropriate based on the availability and accuracy of the data.