Model exploration tasks

[brianlangseth-NOAA]

From model 3_1_5_update_tri_index here are some potential explorations and the rationale for them

• [x] Set survey selex to dome - To see if fit to combo survey length comp improves (going from 3_1_2 to 3_1_5 shifted combo survey left)

3_1_6 does this. Result is no change in dynamics, but combo survey selectivity shifts back to the right. This tells me that the fit to the combo survey length comps matters little. Q for the combo increases up to 0.56 and for the triennial 0.33, which I cant say is wrong but its creeping up

• [x] Tune sigmaR and update bias adj - To see if recdev pattern improves in early time period 4_0_1 looks at this, but the tuned R is already at the suggested value (0.5) and the bias adjustment, matches the main pattern, but doesn't when there is the long tail of negative early devs.

  • [x] Try fixing sigmaR at a value (4_0_2)
  • [x] Try turning off early devs (4_0_3) From 4_0_2 and 4_0_3, the lower early devs are to accommodate higher devs in recent years. Higher sigmaR leads to greater decline but large devs recently. Turning off is the opposite. My thought is selex blocks may inprove some of this so that changes in length/age aren't being adjusted with recdevs
  • [x] Update blocking Update blocking in 4_1_1 does change rec devs in recent years but does not resolve early devs being negative
  • [x] Turn off sum to 1 This does not fix it. Instead rdevs in recent years push higher to better fit age and length comps
  • [x] Set steepness to a higher value This does not fix it. Recovery is faster in recent years

• [x] Check what else changed when updating triennial survey index - to better understand why going from 3_1_2 to 3_1_5 changed things.

I looked at this and the only noticeable changes were in recdevs and survey q's. Originally thought that the change in combo survey selex leftward causes lower recdevs which is biggest reason for driving down population, however model 3_1_6 results in similar selex shape as 3_1_2 and lower recdevs persist. Not entirely sure why recdevs lower

• [x] Replot mixed triennial index overlayed on lognormal triennial index - to better understand why going from 3_1_2 to 3_1_5 changed things

• [x] Play with weighting to adjust fits to data from the best model - To explore what the most important data source is

Did this with 4_9_X across the board. Could do fleet by fleet but that is rather endless. Main point is that length data up the scale and increase the recovery, age data lower the scale and decrease the rate of recovery. Have a slide with image showing difference across explorations in model 4_9_5_upweightLen and 4_9_8_orTWLlen1. This doesnt really inform for 2015 model but does inform the patterns we are seeing. To better compare the effect of data weighting would need to compare the individual changes before and after.

• [x] Reduce lambda for rec length comps - To determine if the dome shaped pattern (and difficulting fitting it) matters

Done in 4_2_1 and result is larger recdevs in recent years, along with higher M and larger Linf. Fits are very similar. Seems to be some trade offs in rec versus non-rec

• [x] Go back to data workshop and see if the bimodal patterns in OR/WA rec can be explained by a fleet issue - To confirm fleet structure isn't dictated the bimodality. Confirmed. Biomodality is visible in both sectors.
• [x] See what weightings changed within bridging - To see where scale changed and to explain the new dynamics

*Added a spreadsheet of tuning weights in commit dbfe837. First and foremost, the CAAL were not downweighted. Second, all areas within a fleet are given the same weight. We do not do that though most weights average together and are close. Third, NTWL ages, which were in the 2015 model, were not downweighted. All together given the new samples, especially new age samples, alongwith the new use of trip for rec instead of number of fish, comparing these is really challenging. Overall, we have higher length comp var-adj than 2015 (not downweighting as much) and higher age comp var-adj other than surveys than 2015 (also not downweighting as much).

• [x] Adjust M to be value based with coastwide - To better isolate pattern of change between coastal and spatial

Done. Models 335 and 336 in briding runs

• [x] Update M alternatives - To see effect of choice of M on outcome and to better compare changes with 2015 base
  • [ ] 'profile' over a breakpoint ages, say 6-24 by 2

Done in models 4_3_X. M choice matters. Fits improve with more decline in old females, but M-constant model does not want to change things. Breakpoint at 20 improves recdev pattern.

I also played around with estimating female and male M together with the requirement that they are the same. Model was to estimate it high, and the pattern in early rec devs is resolved, but the trajectory is all weird. I also estimated steepness to see if there is correlation with female M (which I allowed to be estimated). There was no change in estimated female M and steepness went to 0.92.

[okenk]

Another option to possibly try:

• [x] Actually combine the triennial index into a single survey so we can float Q. UPDATE: Done in models 4_10_x. Had some difficulty with this setup. Include as a sensitivity in #63

[okenk]

If combo selex is driven left and the small fish are not there that would imply lower rec devs. (Maybe you already figured that out.)

[brianlangseth-NOAA]

Reason for low early recdevs is unresolved. Increasing old female M reduces pattern more than anything else I've tried, but it still is there.

Ive tried all I can think of.

UPDATE: Allowing sex dependent selectivity improves

[brianlangseth-NOAA]

Other things based on ongoing explorations include

• [ ] Estimated extraSD for indices due to fit mostly for triennial that is juuust on the cusp of reaching the edges of the values
• [x] Maybe not estimating CA rec male dependent selectivity given that the comp data is not sex dependent (only sex = 0) and seems to be nearing a boundary for parm 4 (when male 4 + exp(fem 4 offset) is near 9). UPDATE: Done in 5_0_0 and 5_0_1
• [x] Adjust what prerecruit index is tracking (recruits, spawning adults)

[brianlangseth-NOAA]

This unchecked tasks are described elsewhere. Closing this