Build hydradata input object for Georges Bank data for mskeyruns.

[gavinfay]

[sgaichas]

@gavinfay input files have been added with 967751c and the model now runs but produces nans.
I've fixed several issues with the input data but it probably needs fresh eyes now if you have some time.

[sgaichas]

here are the current input survey and fishery time series. before we even look at lengths, I wonder if the really high catches first thing combined with some relatively low surveys are an issue?

[sgaichas]

@gavinfay I can work on this today. The growthprob_phi calculation is producing nans so I'll start there.

[sgaichas]

I changed to growth_type 3 instead of 4, which sets growthprob_phi to 0 if a length bin exceeds the vonB_Linf entered. No more nans but due to the mismatch between estimated Linf (the mean max length, not the max length) and observed max length, many species do not make it to the highest size bin (probability of leaving bin 4 or even 3 in a 5 bin model = 0).

We may want to consider adjustments so that there is a nonzero probability of growing into the largest length bins given this constraint. A hack would be increasing vonB_Linf to the max observed length but this will change growth from the observed fit if K isn't altered too. On the other hand, a curve through 5 bins isn't much of a curve...

[gavinfay]

Thanks @sgaichas. Yes, those early high catches from the foreign fleet are problematic. The surplus production over that period is incompatible with the survey time series. In part this is due to stock dynamics being driven by exploitation pattern not just on Georges Bank for some of our stocks (e.g. the pelagics, also silver hake and spiny dogs). In previous fits to MS-PROD, we got around this (also known as sticking one's fingers in one ears and saying 'lalalala') by removing the initial few years from the dataset. It might make sense to do that here, otherwise I suspect we will have a series of weird recruitment patterns estimated.... (or we allow for additional mortality but hope to have something simple to start with...).
Re growth, ah I knew our bandaid would rear its head again. One option could be to estimate the Linf and k, because we have information in the size comp data, but probably troublesome as you say to estimate a growth curve with only 5 size bins.

[sgaichas]

Well @gavinfay I am definitely missing something. I've fixed all the data errors I can find, but I cannot get a run where anything survives past the first timestep, even with catch set to 0 and the isprey matrix all 0.

[sgaichas]

OK, new input files starting in 1978 and with 0 replacing -999 in comps data (c001e53) and associated R list object (17d29c2) are up. This may make life easier:

[sgaichas]

@gavinfay on the growth issue, I can confirm that the simulated dataset also produces 0 growthprob_phi values for lengthbins < max for many species.
It is especially pronounced in the 10 bin model, with 0 probability of growing into the
next bin for as many as the 5 largest bins for a couple of species!

[gavinfay]

@sgaichas Recruits are not being populated because t % Nstepsyr is always 0 when there is only 1 time step per year, thus is always the last time step of the year and no first time step. Seems like there is a lot of places where this condition is used, so should/could we force the GB data to have more than 1 time step per year?

[sgaichas]

Aha! Thanks for finding that.
An option for forcing multiple steps per year with the data might be reverting to the original model bins, which were smaller for small sizes with one very large bin at the end. Faster growth through that small bin made multiple steps per year. The disadvantage of this approach is there will rarely be length data in the smallest bins, but maybe that is ok as long as there is some length data in something other than the last big bin.
This will leave that bug in place though. Could we add Nstepsyr == 1 || in front of the t % Nstepsyr == 1 && yrct <= Nyrs or would that break? I haven't thought it all the way through obviously...
It would be or the whole previous conditional

[gavinfay]

Hmm. I think adding the OR statement sounds like a decent fix for now. Will try and see what happens.

BTW: Stock Synthesis now has a predator module!

[gavinfay]

The OR statement seemed to do the trick.
Noticing that the InpN sample sizes for the comp data are huge. These look like they are Nfish for the fishery size comps, not sure about the surveys (looks like it could be Ntow in some case?). Can the fishery comps be the Number of samples instead?
Setting all to 100 for time being.

[sgaichas]

Great!

Yes, those comps inpN probably are Nfish for all; I will make an mskeyrun issue for those and see how quickly we can fix them. Setting to 100 sounds like a reasonable temporary fix.

The other data correction not currently included is to apply a discard mortality rate (< the implied 100%). Catches for dogfish and skates are currently overestimated because I haven't corrected for this. I will add an issue to mskeyrun for that as well. If those two species are getting killed off too quickly scaling the catch down a bit would be reasonable.

[gavinfay]

Aha. Will keep that in mind.
The values for the survey indices are very large also, are these estimates of total biomass? I would have expected them to be stratified mean kg per tow/hectare, but if they are meant to be total biomass estimates that helps think about catchabilities.