structure introduces implicit assumptions, and some models are built to be robust by “hard-wiring” the way some factors are handled. These assumptions often go unstated, but in this case conflicting signals in the data dramatically influenced model results. It is important to note that our analysis does not necessarily indicate that the NMFS results are incorrect or that any of our specific results are correct. However, the analysis does show that the assumptions are influencing the results and that their influence should be acknowledged explicitly and used to help understand the dynamics of fish populations.

In recent years, where the estimates are most critical, the model estimates of spawning stock biomass differ most dramatically from each other (Figure 1). All of our model biomass estimates are lower than the NMFS estimates in recent years. In particular, NMFS assessments show an increase in spawning stock biomass in recent years, whereas our results indicate that the spawning stock biomass may have stabilized or fallen during the same time period. The committee was able to narrow down the likely causes of these differences and believes they are related to the structure imposed on fishing mortality-at-age in the various models. This structure is implemented in the assumption of separable selectivity and full recruitment fishing mortality in the CAGEAN model and by the level set for the “shrinkage” control assumption in the VPA and ADAPT models. The separable selectivity assumption in the CAGEAN model forces selectivity-at-age to remain fixed over time. The shrinkage factor, used in the ADAPT and VPA models, controls how quickly fishing mortality (F) is allowed to change from year to year.

FIGURE 1 Summer flounder spawning stock biomass as estimated by NMFS, as compared with three independently conducted model applications. NOTE: NMFS = National Marine Fisheries Service, NRC = National Research Council, and VPA = virtual population analysis.

The CAGEAN model shows the most distinct drop in spawning stock biomass in recent years because it does not have the flexibility to deal with the decrease in younger fish seen in both the catch-per-unit-effort (CPUE) and the catch-at-age data. The behaviors of the VPA and ADAPT models appear to be related to the degree of variability in fishing mortality allowed through the shrinkage applied to fishing mortality-at-age. Although the controls imposed by these models take place through different mechanisms, both types of controls decrease the degree



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement