| model | label | msy_sel_argument | data_file | stocks | fisheries | fishery_names | indices | index_names | fishery_age_comp_years | fishery_length_comp_years | end_year | n_parameters | nll | max_gradient |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| FN | FN base | default | FN.dat | 1 | 1 | FarNorth | 2 | Peru_Artis, Peru_Ind | 0 | 54 | 2025 | 618 | 629.4314 | 1e-04 |
| FN_selblocks | FN selectivity blocks | default | FN.dat | 1 | 1 | FarNorth | 2 | Peru_Artis, Peru_Ind | 0 | 54 | 2025 | 177 | 549.7744 | 6e-05 |
SCW17/Paper-01a Far North model selectivity
Rendered 14 June 2026, 23:32 CEST
1 Purpose
This document is a focused diagnostic for the FN Peru-only test model. Both runs use the same single-stock configuration with FarNorth catch, FarNorth length composition data from 0.15.dat, and the two Peruvian CPUE indices (Peru_Artis and Peru_Ind). The base run uses the default jjm behavior, while the FN_selblocks run estimates the fitted model with FarNorth selectivity allowed to change in 1980, 1999, 2010, and 2020. The fitted-model comparison below focuses on the rerun FN and FN_selblocks outputs, with run configurations summarized in Table 1.
This diagnostic follows up on items 11 and 12 in the SCW16 benchmark meeting report, which identified the need to test variable Peru fishery selectivity for catch length-frequency data and to link Peru survey length frequencies to survey selectivity estimation.
2 Fitted Model Comparisons
The diagnostics in this section compare the model predictions from the rerun FN and FN_selblocks fits.
| Metric | FN base | FN selectivity blocks |
|---|---|---|
| Estimated parameters | 618.00000 | 177.00000 |
| Total NLL | 629.43142 | 549.77436 |
| Delta total NLL | 79.65705 | 0.00000 |
| Catch biomass NLL | 0.29848 | 0.06038 |
| Fishery age-composition NLL | 0.00000 | 0.00000 |
| Fishery length-frequency NLL | 471.70400 | 475.97700 |
| Index NLL | 14.64230 | 18.12590 |
| Survey age-composition NLL | 0.00000 | 0.00000 |
| Survey length-frequency NLL | 0.00000 | 0.00000 |
| Fishery selectivity penalty | 113.95400 | 54.46570 |
| Survey selectivity penalty | 0.00000 | 0.00000 |
| Recruitment penalty | 28.25130 | 0.99292 |
| F penalty | 0.15466 | 0.07689 |
| Index q prior penalty | 0.01086 | 0.06125 |
| Other prior penalty | 0.00000 | 0.00000 |
| Residual | 0.41512 | 0.01392 |
| Maximum gradient | 0.00010 | 0.00006 |
| Mean index SDNR | 0.78818 | 0.89311 |
| Peru Artis SDNR | 0.67459 | 0.86040 |
| Peru Ind SDNR | 0.90178 | 0.92581 |
The selectivity-block model has a lower total negative log-likelihood than the base run, but Table 2 shows that the reduction comes mainly from lower FarNorth fishery-selectivity and recruitment penalties. The direct length-frequency and index likelihood components are higher for FN_selblocks, and mean CPUE-index SDNR also increases. The result is therefore useful as a diagnostic of how selectivity flexibility changes the fitted dynamics, but it does not by itself indicate a clear improvement in the data fits used for advice.
2.1 Selectivity Time Series
The ridge plots compare the fitted FarNorth fishery selectivity-at-age trajectories from the base FN and FN_selblocks runs. The selectivity-block run uses change years in 1980, 1999, 2010, and 2020. Ages are plotted as actual model ages 1-12.
The base run estimates a smoother progression in FarNorth selectivity through time, whereas FN_selblocks concentrates most years around a narrower age-3 peak and then shifts the most recent block toward older ages (Figure 1). The change is important because the lower total NLL in Table 2 is associated mostly with the selectivity-penalty structure rather than with improved catch-at-length or CPUE likelihood components.
2.2 Catch Biomass
The catch biomass panel compares the observed FarNorth catch biomass from the active data file with the fitted catch biomass from the FN and FN_selblocks runs.
Both model configurations reproduce the annual FarNorth catch biomass almost identically in Figure 2. This visual result is consistent with the small catch-biomass likelihood contribution in Table 2, indicating that the selectivity-block configuration does not materially change the catch-biomass fit.
2.3 Index Fits
Points and vertical intervals show the observed Peruvian CPUE values and nominal 95% intervals from the active data file. Lines show the fitted values from the FN and FN_selblocks runs.
The CPUE fits are broadly similar for both runs, but FN_selblocks tends to be slightly higher than the base run in several early and mid-series years (Figure 3). The index likelihood and both index-specific SDNR values are higher for FN_selblocks in Table 2, so the added selectivity blocks do not improve the Peruvian CPUE fit.
2.4 Catch-At-Length Fits
The FarNorth catch-at-length panels show observed length proportions as bars or black points and fitted proportions from the FN and FN_selblocks runs.
The aggregate and annual length-frequency panels show modest differences between the two runs, especially around the lower-length modes in the 2002-2015 and 2016-2025 periods (Figure 4, Figure 5, and Figure 6). However, the fishery length-frequency NLL is slightly higher for FN_selblocks in Table 2, so the extra selectivity flexibility should not be interpreted as improving the length-composition fit overall.
3 Conclusion
The most flexible selectivity model provides only limited improvement in fitting the length-frequency data relative to the base FN configuration. Based on these diagnostics, we propose using a more static selectivity set that is allowed to change during apparent availability-at-size shifts over time, rather than relying on highly flexible selectivity blocks to absorb the length-frequency patterns.