data sets. Coverage varies by age class and over time. Ages 1 and 5+ receive less coverage in general when compared with the middle three age classes, presumably due to the catchability characteristics of the gear employed. Some surveys are missing in the most recent years due to a time lag in availability. Information on age classes 0 and 2 through 4 will therefore have the greatest influence on the overall fit of the model because those data are the most abundant.
The log-transformed mean number per tow of summer flounder at age derived from the survey data is the relative abundance measure used by the assessment model. Trends in abundance can be seen by viewing a single age class across years as tracked by the individual surveys (NEFSC, 1997), but these trends can also be viewed in summary by examining the averaged log(CPUE+1) across surveys (see Figure D-1a, Figure D-1b and Figure D-1c). The log transformation is used to achieve stability in the variances and to facilitate contrast across the widely ranging abundance values. Such a transformation is commonly applied to fishery survey indices prior to inclusion in assessments because of the log-normal variation often seen in these indices; consequently, a log scale is the most appropriate scale on which to view these data in an exploratory analysis.
When survey-by-survey data are viewed (see Figures 4-8 in Terceiro, 1999), the NEFSC winter survey shows a much greater range in data values than the other surveys and thus must contribute a significant signal to the fit even though the length of the time series is limited. The second point, which can be seen more easily in the figures averaged across surveys (see Figure D-1a, Figure D-1b and Figure D-1c) is that the relative abundance indices may have peaked with the 1995 cohort. The trends indicated by the two observations available for the 1998 cohort go in opposite directions, creating a conflict in the data that may lead potentially to equally likely but conflicting trends in model estimates. These are important features to track in interpreting the assessment results below.
Another data set that goes into the assessment is the total catch at age (Figure D-2) representing the combined catch from the commercial and recreational fisheries. From the data available, we have no way of knowing how commercial and angling effort has changed in recent years, but the upward trend in ages 3 through 5, the slightly downward trends in age classes 1 and 2, and the significant drop in catch of age-0 fish should, in conjunction with the survey indices, determine the nature of the fit by the assessment models. Interpretation of these data is complicated by the 1992 changes under Amendment 2 to the summer flounder fishery management plan, which included an annual landings quota, a minimum size limit at 33 cm, and a minimum mesh size of 140 mm. These management actions probably influence what is seen in the combined commercial and recreational catch data. Shifts from commercial to recreational harvest also should influence the catch at age as each fishery exhibits different size and age selectivity patterns. Finally, weight at age goes into the assessment. The summer flounder weight at age has not shown any directional trends over time (Figure D-3). Thus, the dynamics exhibited in model estimates should reflect mainly the catch and relative abundance indices.
A unified picture of the factors contributing to stock dynamics can now be developed using a system of equations linking the various components. This system of equations (a model) is derived from basic principles that are assumed to represent the system. These models add structure to the estimation process by characterizing relationships that exist among population variables (e.g., population size) and observations (e.g., catch and catch per unit effort). When using models for analysis, one should always keep in mind that (1) models are simplified representations of the system; (2) assumptions made in modeling impose a structure on the information that will influence assessment results; (3) no model works well with poor data (see NRC, 1998a); and (4) even if the data are high quality