29th Session of the IPHC Scientific Review Board SRB029, Day 1 Part 1

Okay, yeah, let's go ahead and get started. I think we'll probably start with introductions. We have some new folks. So I am Olaf Jensen, Cornell University survey chair. Pass it to you, Ian.

Morning, Ian Stewart. I work on the stock assessment and help Alan with the MSE.

I'm a research biologist here. I work on right now, discard mortality and whale depredation. I'm Lynn Webster, I'm a biometrician, I do analyses and survey design. Joseph Blanes, I'm the manager of the Biological and System Science Branch. I've been here for about 11 years.

I'm Colin Jones, I'm a research biologist here, life history, doing mostly reproductive work right now.

And Alan Hicks, uh, work MSE and help Ian with the assessment.

Joanna Mills Fleming from Dalhousie University. Don't know exactly what my title is, you guys, external something or other. Yeah, that's our new member. Fantastic. And Dave Wilson, Executive Director.

Um, Giuseppe, did you want to highlight any of your team that are online? Yeah, Andy Jasonovich, who is online, is a microbiologist, but also—. Yeah, he'll be participating. Thank you. Great.

And let's then go to SRP members online, Anna and Mike. Could you—. Anna, you go.

Anna, we're just doing introductions, if you're able to turn on your mic and introduce yourself. Oh, sorry, I forgot to turn on the mic. Yes, my name is Anna Kuparinen, and I'm professor of aquatic environmental sciences at the University of Jyväskylä. I do like both lake and marine ecosystem modeling, life history work, and reproductive dynamics, allee effects, and that kind of things. Great, thank you, Anna.

Mike, do you want to go? Certainly. Hello everyone, sorry can't be there in person this meeting. My name is Mike Wilberg at the University of Maryland Center for Environmental Science, and my main areas of focus are stock assessment, management strategy evaluation, and quantitative fisheries. So thanks.

Great, thank you, Mike. And Annemarie, could you just briefly introduce yourself?

Good morning, my name is Annemarie Hoang. I am a biologist working for Fisheries and Oceans Canada, and I'm based out of Vancouver.

And Joanna, Annemarie's counterpart in the US, is not speaking, so comments separately, I guess.

Okay, great. So I'll turn it back over. Okay, fantastic. Thanks very much, Chair. If we could just have the agenda up, please.

Okay, so the agenda is pretty much the same, um, flow as we have from the last meeting. And as part of it, we also have built-in collaborative sessions where we'll go offline and SRB members will just work with various teams, either as a whole or small teams, and work on any additional components that just need a little bit more detailed discussion across the table here at the SRB. So I'll just pause and pass to you, Chair, to see if there's any amendments to the agenda or whether The agenda is kind of arranged.

Noting that we have the two SRB members remotely, so we'll just have to see if we can get you, Anna and Mike. And Annmarie, you're very welcome as well to try and join those sessions, and we'll just communicate them as we reach them in the agenda. But I understand that this, with the time differences, particularly for you, And we'll see how we go with that. And sorry, Anna, when we— and Mike and Anna, when we get to that point, I know sometimes we have kind of a free-flowing discussion where multiple people are talking, so it might be hard. Please feel free to jump in and ask us to take turns and make it easier for the hybrid.

Yeah. And just a note on that, we've made both of you presenters, which means you can activate your mic and speak at any time. You don't necessarily have to wait. So if you do feel like we're neglecting you, just, just speak up. Okay, well, with the agenda adopted, we can jump into agenda item 3, which is IPHC process.

And so 3.1 is just a general reminder that this is the second meeting in the SRB series. The first meeting is really about guiding and the secretariat on various tasks. Associated with the 5-year program of integrated research and monitoring. And this meeting is where you develop recommendations that you would like to put forward to the Commission. And now the Commission will consider them at both the interim meeting on the 1st of December, which is an online Commission meeting, and then also at the annual meeting in late January, up to the 25th, commences 25th of January.

So as you are developing recommendations, which are informal, just for the Secretariat and others to continue work on versus— sorry, requests rather, which are the informal version, and then the recommendations are the formal versions put in front of the Commission. Just keep that in mind as you draft. Then we can jump on to agenda item 3.2, which is an update on the actions arising. And so that is paper 3, which is on the screen at the moment.

Joanna, just in terms of process, this is something we together for each meeting. We just take the recommendations, requests for action from the previous meeting and put them into Appendix A in this case. Each one includes a specific action, deliverable, probably a timeline, and who's actually going to undertake it. So, in terms of Appendix A, I don't intend on going through these in detail, primarily because each of the presentations you're going to receive and discussion sessions over the coming days is going to address what the current status is. So for example, recommendation 1 on management strategy evaluation, Alan's going to walk you through that specific progress that he's made since the last SRB.

So the SRB made 2 specific recommendations which were also discussed at the recent Commission work meeting, and the Commission was very appreciative of, of the work that's being undertaken. Uh, and again, Bashar is going to give a little bit of an update later on recommendation 02 and Alan on recommendation 01. There were, uh, a number of, uh, additional requests. Uh, and again, as I said, biology and ecology will be updated by Giuseppe, his team during their presentations. Um, Ian will update on recommendation 04 as part of his presentation as well, and Alan deal with the rest.

So it's just a simple noting at this point, Chair, the progress that is incorporated within this paper, and that you're going to receive updates throughout the meeting. Joëlle, sorry, what's the difference between the recommendation and request? Good point, we should cover that because as a new SRB member, and also Mike there as well, so the requests are in an informal directive or request for action to be undertaken by a given party, usually in this case for the SRB as the Secretariat. And it's something that we try and potentially recalibrate some of the analysis, reexamine the way we're doing things, and then we'll do our best to try and fulfill that before the next SRB meeting, or if there's a different deadline. The recommendations are more formal where the SRB is providing very specific guidance for the Commission's awareness that you feel should be elevated to that Commission decision-maker level.

And I think if we look at those two up front, for example, maybe I'll pass to you, Olaf, to sort of articulate why the SRB at the last meeting felt that it was sufficient enough to elevate these for Commission awareness into session. Yeah, so we often use recommend when there's a management-oriented aspect of it, so that they're aware of it and have the ability to— their commissioners are aware of it, but the ability to step in and say, no, that's not really what we want, or like, yes, we want this, and take it off the request. So, and so some of the, the recommendations here are kind of higher-level issues that we'd like to Let's see, that's requests. It's more sensitivity analysis. Can you run it this way?

And sometimes requests, if they end up having something comes out of it that's mentioned relevant, we have a technical— the other verb that we use besides noted is All allows us to go back to a previous request to, like, bring it back for—. Okay.

Well, that closes that agenda item, and we can move on to 3.3. And this is simply paper 4, and it's simply an update on the actions arising from the Commission meeting. So this is the same paper that you would have seen at the previous SRB meeting. Senu for you, Joanna, it's the same principle. These are the actions arising from the previous Commission meeting that specifically relate to the SRB.

Most of the actions arising are related to, as you can see here, just some general noting and thanking and considering. If there is a redirection for the SRB, it will also be provided here. As the Chair was just saying, if it's a case of, "No, we don't want you to modify XYZ in a particular way," they'll provide that in writing. But otherwise, we'll just cut and paste the specific report text there as reference so that if you do have any questions as we progress through the meeting, you can quickly refer back to what the latest Commission advice, decisions, or general actions are. There was nothing specific coming out of that last meeting.

Yeah, I'll just add that this is really valuable because it's one of the primary ways that we learn about the Commission's priorities. For example, for the MSE not wanting to see evaluation of the distribution of the TC, that's really valuable. This is the place where we learn about those preferences.

With that, if we can just pop back to the agenda.

So, agenda item 3.4 is just an opportunity for our science advisors to provide updates on, on any recent requests directly from their commissioners. Anna? Sorry, Annemarie, if you can jump online, and yeah, that'd be great. Yeah, thanks. Thanks, Dave.

Olaf, I'll send you a quick summary of these points during break, but I have 4 broad categories that I've heard things from the commissioners on recently. We had a work meeting last week, so we're chatting with them. The first one would be around depleted, recovered, overfished, rebuilt concepts. The second one would be around catchability. The third one around whale depredation.

And the fourth one around aging methods. So I'll pop into the first one. Let me know if you want me to pause after each one or just keep going through. Sorry, Annmarie, if you could just bring your mic a little bit closer. Oh, sorry.

Does this work better? Yeah, it does. Thank you. Okay, great. So the first one around the concepts of depleted, recovered, overfished, rebuilt, and I'm trying to make sure that I'm pairing them together because I know MSAP gets a little, I've heard a bit of trepidation around, okay, we need to know when we're not going to be overfished or not going to be depleted.

So I'm trying to do the little mental pairing of the, of the concepts. So one of the questions that came up was, because probably more for Alan, but just to let people know that there was a question around the previous estimate of the depleted spawning biomass. Somewhere around 79-90 million pounds was the memory, and so the question was, has this been reestimated with the new OM? Because Alan had showed us last week how the performance measures had changed in terms of some of the probabilities of reaching them. And so I think there's some questions about how the new OM is going to be affecting estimates of depleted and what are the values, what's the ETA for getting those.

More broadly, I think questions around depleted, recovered, overfished, rebuilt is how do these concepts interact? How do they fit together in an MP? Do we actually need both of them? How would they work together? Would we maybe use one quantitatively and another one more qualitatively?

Which seems like, like to me where the commissioners are sort of thinking about depleted, and that's why they might be modifying the MP right now. So that's all the pieces on the depleted/recovered/overfished rebuilt. Should I just pop right into catchability or—? Let's go then. Keep going?

Okay. In terms of catchability, I talked to Ray last week around previous comments that came from SRB around incorporating bottom complexity into the state-space assessment. The advice at the time was that the value for effort wasn't really worth it. And so, but what I'm hearing from the Canadian Commissioner is that they're interested to explore this a bit further, because I'm hearing questions along the lines of, well, are there other things that would be worth it? For example, the effects of tides was one that came up.

I think, but I think this one is maybe less around area-specific and maybe around overall catchability Because I think I heard a comment in terms of like some areas do backfill more than others, but generally they see fishing that's better on building tides.

And maybe a different way of getting at this, as opposed to, you know, how about this thing, how about this thing, how about this thing, is maybe a broader question of, well, how different would catchability need to be in different areas and locations for it to make a difference to what the commissioners see in front of them, and how much of the difference would that be? I think basically broadly I'm really hearing that there is recognition that FIS plus space-time plus assessment model is the best available science, but I'm also hearing questions like, well, how do we make it better? So that's sort of mostly catchability, although whale depredation sort of ties into this. So I had a question around whale depredation, and we know that in 2023 SRB recommended not including whale depredation because the change to TCEY would be really small. And so I think what I'm hearing from the commissioners is that they're seeing continuing increases in whale depredation.

So thinking forward, what amount of whale depredation would result in a large enough change to the stock assessment that would warrant the inclusion of whale depredation? And then maybe looking at the current trend, if we extrapolate forward, about how long do we actually need to wait before we get to the point where will depredation impacts would be big enough to be warranted. And sort of, to me, this sort of ties back into the questions that I was hearing around, um, tides, bottom complexity, and generally moving away from the assumption of equal coast-wide catchability. I'm sort of hearing that each individual change on its own is pretty small. And so how do we balance the cumulative impacts of these multiple small changes on the stock assessment outputs that the commissioners see?

And that added uncertainty that we get when incorporating all of these small changes. So I think that's more of a broader question. Aging methods, I think overall I heard pretty— the commissioners were really interested and happy that you guys added— asked for the comparisons of the 3 aging methods at the table, and I think they're going to take a peek at it. Because Dave mentioned that it's in SRB materials. I think one thing that popped to mind for me was we had also talked last week around the AI policy and needing to be— that AI work needs to be verified.

And so I think one question that I had is, how would the checking or the verification of AI aging be done? Like, what would that involve? And that is It. Okay, thanks, Annemarie. Do we want to take a second to get a pulse of Canada for response?

It might be useful, even though we're going to address it a little bit later as part of the presentations, just so Annemarie has that initial feedback as well. So maybe if we go around and each can respond to those comments. Yeah, Alec, can we start with you about how Whether the depleted reference point has been reestimated. Yeah, sure. It has not.

That's part of the program work for this year. And it's a high priority piece of program of work. But as you'll see in the presentation, I'll get later that you have a sequential process of developing, operating, all that stuff in front of that, making these types of analyses. So, I'll be presenting you, uh, later today, the whole framework for the MSE, how that's performing, some initial results, and then the next step will be doing these types of analysis. Um, and I guess she also mentioned about how these all tie together, and I think that's part of the discussion that we'll be having, um, and how best do we bring these concepts into the harvest strategy policy.

Sounds good. And Maria, I'm going to leave that one there unless you want to respond to that. Nope, that's great. Thank you. Okay, Ray, can we talk about some of these covariates on catchability and the fish smallfoot incorporating bottom type and tide state?

So we did, as you probably recall, Olaf, we did a few meetings ago address that with the SRB, as Marie mentioned. I meant to find your notes on that, but you basically stated that this wasn't worth pursuing upon comparison. It was not something that we've done anywhere else.

And the value, given the amount of work that it would take. But most of these things have come up over the number of times over the 20 years I've been here. We know the catchability varies. A lot of these things are very hard to quantify or incorporate into the analyses.

I'd like Ian to offer how he might take the competing assessment.

We've explored some of these covariance units. Great mention. Bottom rugosity as a covariate to catchability. Yeah, it's probably there, but this is not something that anybody goes with. And it's— it seems like it would be even more of an issue for trawl.

You know, we— bottom type and depth is already in our models. So, actually trying to have a covariate that was the rugosity of the bottoms. Would be pretty far down in the weeds. And the SRB agreed with us last time when we provided a little more broad summary on that topic. With regard to tides, we've actually looked into tides quite a lot because it's, it's completely correct.

In any given location, there's probably an optimal tide to fish on, but it's going to be a different tide in almost every location. Some locations fish better on lots of tide, and some locations fish better on weak tides. Or the incoming or outgoing. And so getting— using a tidal cove area at a broad spatial scale, it's kind of a huge undertaking, something that maybe we could look at on a local level. In that case, then, we're looking across years and across vessels, and we're quickly out of degrees of freedom.

So, although we've looked at this some in our survey data, we don't have multiple repeat sampling at the same locations across different tides that isn't confounded with time in the summer and vessel and other things. So this is, while it certainly affects catchability, it's not something that we want to be digging into. I think both of these are probably most relevant for survey standardization itself outside the assessment. In terms of within the assessment, as I'll talk to later on, we already have time-varying age-based catch selectivity. We have to have that because the gear interacts with the fish and there's a length-based component there with changes in growth over time.

Turn that into an age-based selectivity, obviously, so selectivity has to be able to change time to account for dimorphic growth and changes in growth over time. So in some regards, we already have quite a bit of flexibility in terms of how catchability in— or selectivity in the survey and both catchability and selectivity in the fishery.

I'll leave it at that. Sounds good. What about including whale depredation? That was the—. Yeah, and I'm sorry, I haven't had a chance to go back and reference the specific notes.

I could provide you with the reference to the specific SRB meetings that we've covered this topic. Spent 2 or 3 years ago, we spent 2 or 3 or 4 meetings looking in depth at this. We looked at observer records, and we also looked at logbook records. We compared the frequency at which whales are observed during fishing. Activity in the commercial fishery.

And then we used the survey data and the presence or absence of whales and the change in catch rate standardized through the geostatistical to try to estimate an offset in catch when whales were present or not present, present and likely to be depredating. And then we went through the exercise of applying that rate, even though it doesn't, it's not an exact match between the survey fishing and the commercial fishing. We said, what if the survey rate were a good proxy? Applied it to the commercial fishing, and then we basically did the calculation of how much, how much volume of fish would be potentially consumed by whales in a particular year. And then we actually took that all the way through.

So we added that mortality into the Stokke system as additional mortality, assuming that it would have roughly the same selectivity as the commercial fishery because they're eating off the commercial fishing lines. And then we recalculated what the quota would be if you accounted for that mortality in the stock assessment. And so the way that works and the way it's been done in, like, for example, for sablefish does this assessment, so you take the mortality out of the assessment and the net effect is generally the biomass is a little bit larger because you had to have a few more fish to support a slightly larger catch and get the same trend. So, and then, and then so the biomass is a little larger, so the potential quota is a little larger, but then you have to subtract off what you think the whales are going to eat in the upcoming fishery, and you basically end up very close to back where you started. Where this breaks down is if there's a rapid trend, because you're not, you know, catching the trend properly.

And essentially where we got to is that we don't have a reliable estimate of the amount of fish that are being eaten by whales, because it's something we can't observe. We can use the survey as somewhat of a proxy, but there's a lot of reasons why survey fishing doesn't make a proxy to the fishery. We do small amount of gear and then we go a long distance, set another small amount of gear. The fishery, sometimes they'll set a whole bunch of gear in one area and work on that gear for days. And so after a lot of years of working on this, we're basically at a dead end in terms of our ability to estimate whale depredation.

There's a lot of reasons why we don't think we're getting complete, a complete picture of whale depredation in the logbooks, for a variety of reasons. And so we, without any avenue to get a reasonably precise estimate of the volume of fish, making a quantitative adjustment to the stock assessment is kind of shaky. We're basically going to have to be just guessing, educated, an educated guess about the mortality associated with whales, and then doing this, add it in and subtract it back out. Where we focused our effort is, and I think we might hear a little bit about it in the research discussion, in the long term, we think that the best solution is to try to reduce whale depredation, because if it's not happening, then we don't have to estimate. Or if it's happening at a low level, it's going to have a smaller effect on things.

So, in the long term, we're thinking it's more productive, rather than trying to come up with better ways to estimate it, it's going to be more productive to come up with ways to reduce it. And that's where the catch protection study that we just wrapped up last year came from, and a lot of other thinking and work toward reduction in whale depredation. To be honest, this is a topic that's come up virtually every year for the last 15 years that I've been here, and we don't have a good avenue for how we would get more information on the amount of fish being consumed by whales. It's something maybe we can talk about a little more in our research, in our broad sort of horizon scan research discussion. Ways we've brainstormed things like tagging fish and releasing them in the presence of whales to at least get at the whale depredation that could be occurring after the fish are caught.

But even, even that doesn't really work for fish that are being captured on the— Maybe I should back up a little bit. The whale depredation is primarily sperm whales and killer whales.

Both depending on where you are in the Central and Eastern Gulf, it's sperm whales primarily, although a little increasing amount of killer whales, and in the Western Gulf of Bering Sea, it's Chum Island killer whales. And they really do take a large percentage of the fish when they're present during fishing activity.

This is one of the topics that I don't have a good suggestion of where we should go here in terms of new estimates or changes to the approach. That was helpful. Thank you guys for bringing this up again. I know we've had a lot of discussions about it.

Ray, can you just remind us in the PIS when there is depredation, what do you do with those sets? Are they removed from the analysis? Right. We have some criteria. It's no more presence at all, but it's enough to have to remove because you don't want to see what they're doing.

Walkers, we have to have some— what is it? Lipsomi. Lipsomi. Thank you. Lipsomi.

Further evidence that walkers are present. For those of you that are generally— we have about 4% of sets— 4% of sets affected by any issue that causes the data to be invalid. Very used, and about half of those— sorry, most of those are due to oil degradation.

Um, aging methods, I think we'll cover this. I mean, most of Basia's presentation is about verification of the API, so I think we're— no one is using this blindly without Verification. That's a great contribution.

Annemarie, anything else that you want to bring up here? Nope, that's good. Thank you. Great. Thank you.

So for the US, Pete had emailed and said they had— the US Commissioners had nothing specific to ask the SRV. And so I was just going to highlight some of the comments that the US Commissioners made at the work meeting last week. Anne-Marie has covered off with all the posts, so I don't think I need to reiterate, so I'll just leave it there.

Okay, with that agenda item closed then, Chair, we can move on to agenda item 4, and this is where we really step through each of the core work areas. And we'll start with 4.1, which is research, and specifically the biology and ecology, which Giuseppe's going to present. And as the reference document, we've provided Information Paper 01, which is the Program of Integrated Research and Monitoring for the next, next period. We have moved away, as you know, from specific 5-year plans just to make it a rolling program. The SRB, as always, is invited to provide additional feedback or ideas for improvement at that plan.

Throughout the course of this meeting and also intercessionally. If there are any modifications or changes in our work streams, we would then bring them to the Commission at the interim meeting and annual meeting for potential amendment of the workstreams themselves. So that's just there as reference, Information Paper 01. And with that, we can bring up the presentation for Paper 5, and I'll pass to Giuseppe to lead us through it. I just ask one question.

The observer, so where you have EG scientific advisor, so all those, those four issues, where's— if that is information coming at some point from fishers, I'm just trying to understand how it gets here. So Annemarie, for example, is not an observer on fishing boat. No, that's a— How is that information collected? Yeah, so, well, to step back, so Anne-Marie and Pete are both the science advisors for the two governments, and so while they're not board members, they're listed as observers. In turn, they bring any information specifically from their delegations, whether it be commissioners or other advisors, to the delegations to this meeting and ensure and that's communicated, and then also take the information that is shared at this meeting back to their delegations.

And that's sort of standard sort of RFML practice. In terms of actual data collection from observer equivalents, so we have our Fishery Independent Setline Survey, and you're going to hear a little bit more about that, how that information is collected, how it's brought or shared with the Secretariat. And then processed the next day or two. So it might be better. Okay, sure.

Yeah. Just to make those two distinctions, the observers on the schedule here, on the agenda here, refer to non-members but very important advices to the process. Yes, makes sense.

Cool. Okay, just have a look in here. Yeah, thank you. Yeah, good morning, everybody. Yep, for the benefit of the new members, particularly Dr. Fleming, who's first meeting, as Dr. Wilson's just indicated, all our research activities are contemplated in that research and monitoring plan.

And in terms of the biological and ecological research, there's actually 5 areas that are contemplated in I'm going to go one by one, just providing you some updates, large, to previous meetings. But the, the 5 areas are summarized in migration and population dynamics. The second one would be reproduction. The third one is growth. The fourth one is discard mortality and survival assessment.

And the fifth one is what we call fisheries technology. So, I'm going to go by one by one and present it to you with some updates regarding this first area, migration and population dynamics. Over the last few years, we've been presenting to the SRB. The bulk of our work has been related to the delineation and population genomic structure of Pacific halibut. This work is now well under progress.

In fact, this work is mostly completed now. We still have two manuscripts that we're working on. One manuscript is describing really the bulk of the work in community population structure. And as you know from the last SRB, SRB28, there was the description of the new bioinformatic method that was developed developed in-house to process all the low coverage whole genome sequencing data. So this manuscript is now being submitted to Bioinformatics, to the Journal of Bioinformatics.

This paper is entitled FST-GL: Efficient Performance-Oriented Estimation of FST from Genotype Likelihoods, and it's currently in review. So we hope to hear sometime soon about the decision paper. One of the other areas where we've been working on quite actively over the last year, uh, has been the development of an epigenetic clock for age-specific housing. So this is a project that has received funding from Alaska Sea Grant, uh, and, uh, as, uh, as methods we've been using, uh, reduced representation by solvent sequencing, uh, which, which, uh, entails the sequencing, the bisulfite sequencing of samples from Pacific calibre of known ages, and then methylation levels are established. And then using predictive models, mostly ElasticNet type models, we derive age-correlated CpG sites that are used to build a clock.

And this clock is assess for performance regarding accuracy, precision. So what I'm going to be doing is just give you an update of where we are in this project right now.

Just a little bit of background of what the type of samples we would use for the development of the clock. We have exploited samples that have been collected in the Fisher's Independent Satellite Survey. For 4 years from 2021 to 2024. And these were samples that were double-aged, so the, the ages are as accurate as you can actually get them. So we drew from a pool of samples from fish aged 6 to 30.

We actually selected 5 males and 5 females for each of those age classes, so a total of 250 49 samples. These samples are geographically distributed. Here you see on top, on the top right, a table with 4 years of collection and the numbers of samples that were collected per regulatory area of the 8 regulatory areas. As you can see, most of the samples are concentrated within the area where we consider the center stock, which is PHA regulatory areas 2B, 2C, and 3A, but they're spread out throughout all 8 regulatory areas. And this, this plot here just indicates the proportion of samples by year of collection, which were fairly uniformly distributed.

So here you can see that from ages 6, from the left, to ages most likely 25, we have really 5 males and 5 females. Females are on the top, males are on the bottom. And then, and particularly in older ages from 28 to 30, we had— we still had 10 individuals per age class, but the proportion of males and females was not equal because of the difficulty in getting samples at those ages.

So, we've completed the sequencing of all those 249 samples, and these are the initial results that we obtained. The controls, methylated, unmethylated, are— seem to be correct. The number of reads are sufficient to move on with this project. We have developed a bioinformatic pipeline in-house. To process all this data.

And these are some of the initial data that we have generated in terms of the number of CPGs per age class. We have about a million CPGs per sample. We filter all those reads, those reads that are at least with a 10x coverage or or higher. And we've also identified the number of CPG sites that are common to all samples. Now, I will have to say that between the publication of this document, that was about 30 days before the meeting, we identified that there was actually a bug in the bioinformatic pipeline.

One of the trimming softwares that we used had an error for processing our RBS data. So these numbers have now changed slightly. So the 39,000 has now decreased to about 21,000. So 21,000 CPGs that are common for all 249 samples. So this is where we are right now.

We are now doing preprocessing filtering steps that are required prior to modeling. Are now— what we are is in the process of constructing the matrix that will go into elastic net models that will be selected to identify those CPG sites that are most closely associated with— Thank you.

One of the requests from the SRB-28 was to delineate plans for cross-validation. I mean, and this is post, um, uh, testing for the performance of the clock. One of the first things that we're actually going to be doing is the testing accuracy using mean correlation, and then precision estimating the mean absolute error and the reproducibility. But the cross-validation of the model is something that the SRD requested, uh, for us to start thinking about. And this is what we're proposing initially to do an in-sample cross-validation.

There are several steps to that. One is starting with an out-of-area type validation. So it's leave one area out. In fact, here you can see in this plot that you could choose all areas except 2C, the one in blue, train the model with all those areas except the one that was selected out, and then use that area. In this case, this is an example, 2C as part of the test set.

Alternatively, you could do a leave one year out approach in which you train the model with all years except one. And then you test the model with that year of data. This case will be 2022. That would be the blue in the column you see. Or you could do an area-year combination in which you train model with all areas in all years except a combination area and year.

So in this case would be be all years, all areas except 2C in 2022. That would be again— So also, we thought about the out-of-sample cross-validation. This was actually presented a couple of meetings before. So we went out and selected samples outside of the collection of samples that are used for many models. So we selected samples from the commercial fishery, not from is from different years, from 2017 to 2019, a total of 46 samples.

And these samples have actually now been sequenced.

And we have the initial sequencing results. These results are actually fairly new. And the metrics, at least the initial metrics of controls, methylated, unmethylated appear to be correct. And the number of reads that we have to move the samples through the pipeline are also seem to be, seem to be sufficient.

Another request from the SRB, and that was— sorry, another— I think it was— that was a recommendation. Sorry, it was recommendation 02. That was to create a comparison of the different aging methods. And for this particular— I'm not going to show the table here, but you have that in the document. I think it's table Number 5, in which there's a description of the characteristics of the epigenetic clock, the development of an epigenetic clock, the advantages and disadvantages, a very initial estimate of cost, obviously, with the understanding that those costs are going to be changing.

And the method that's going to be used eventually for implementing this clock is most likely going to be different. Than the method that has been used for the development of the block. So hopefully we'll be incorporating that table into the broader picture of different aging models that we're testing here at the FGC Secretariat.

The second research area is that of reproduction. And that we've done quite a bit of work on this area with the two main objectives, and one was the revision of the maturity schedules for female Pacific halibut, and the other one was to start looking at fecundity estimations for the first time. The maturity, the new maturity estimates that have been derived from histological assessment is now a paper that Colin is about to submit. It's prepared for submission, and I think Colin has already started submission, but I will pass the word to Colin for more updates on the maturity and plans for the program.

Yeah, so most— I don't have a lot of new data to present today. Mostly it's a review of stuff we've done in the past since we have another new SRP member today. But this is just kind of give you a little overview of, of what we've been doing over the past few years regarding reproductive research for females. So just to give some historical perspective, so previously the IPHC has used visual maturity estimates on our annual setline survey to to create maturity ojive for the stock assessment. And in more recent years, we've gone more towards histological methods to update those maturity ojives.

So starting in 2022 and to the present day, we've collected histology samples on our annual setline survey. And so we created new coast-wide maturity ojives using actual weighting of regional abundance estimates from Ray's FIS space-time model to create a new coast-wide maturity ojive for histology. Then using those same fish from 2022 to 2024, we created a visual estimate as well and then conducted a calibration between the histology and the visual to give us a new coast-wide maturity ojive that we can go back in time even further, all the way back to 2002. And the reason why we chose 2002 was because that was when the IPHC changed their primary aging method from mostly surface age reading to break and bake method. So we have now created a historical time series using calibrated visual maturity estimates from 2002 to 2024.

And that's what you see as the green line on the figure here. And then also, as you can see, the red line is what our previous estimates were, and these aren't direct comparisons because the previous estimate was only taken from fish sample in regulatory areas 2B, which is Canada, and 3A, which is the Central Gulf of Alaska, whereas the green line is a total coast-wide estimate. From all regulatory areas and all biological regions. And as you can see, our A50 estimate dropped by 0.6 years from our previous estimate. So we have seen a slight shift in the left of the curve to younger maturing females.

And then the— oops, sorry. And then the last is that we also truncated the curve to zero at age 7, So our previous indication was that we thought females did not mature any earlier than age 8 based on visual, but histologically we've actually found females to mature at age 7. So we have not seen any females, so we dropped it to 0 below 7 because we have never seen a mature female below that age. And so this just kind of gives you, and then this was all incorporated into last year's stock assessment as a an update to the maturity of OJIVE and SSB estimations in the assessment. And if you want more further documentation on this, refer to document 6 from SRB 26, and it'll give you all the information that you need there.

Yeah. Joanna, one of the things that wasn't immediately apparent to me is that these decisions about the lower left of the curve, they do matter because there's the biomass. So it seems like a very small difference, but it's multiplied by how much proportion of the biomass. Yeah, especially in today's environment where we're seeing a lot of young fish in the Pacific halibut stock right now. And we did, as part of the full assessment last year, we actually did a sensitivity.

Yeah, yeah, better assessment. Yeah, so then Another thing that we looked at, because we now have a very long time series of 23 years, we wanted to look and see potential shifts in maturity through that time. And as you can see here on the figure, these are individual years of coast-wide ojives that were constructed from 2002 all the way to 2024. And what you can see is that there has been two distinct shifts during this time series. From 2002 to about 2016, there was a shift to older maturing females, and then at about 2016 to 2020, 2021, there was actually a large decrease in showing a lot younger maturing females over that time series.

Rather than trying to decipher all these colors, what we did was we created a plot that just shows the age 50 values, or age at 50%, percent maturity and what the, what those shifts look like through that time series. As you can see, there's two distinct lines there. The one on the top is what the visual estimate is giving that's not calibrated by the histology. And then the solid black line is the calibrated visual estimates through time. And as you can see here, from 2002 to 2016, there was a, decrease over— or an increase over time to older maturing females and an increase in the age 50 and then a sharp decrease down to 2021.

And then more recently, it has slightly leveled off. So one of the requests of the SRB that we've discussed over the past year was potentially looking at either environmental factors or other things that might be driving this. So We've just begun to look at some of this stuff and would love some feedback from the SRB, whether it's now or discussions later on tomorrow or later in the week about how we might be able to look at some of this stuff a little bit better. So what we've done is— and these by no means are the best way to visualize some of this stuff, but we just wanted to overlay some potential covariates or factors that might be potentially driving the coast-wide maturity at age. So this is, you see that solid black line is the calibrated visual maturity A50 values, and this blue line is taking the fish that we use to create that black curve and looking at the mean net weight by year through that same time series.

And as you can see here from 2002 down to 2010, there was a decrease in the mean net weight of those fish sampled. And then starting in about 2016, there was an uptick in the size of those fish. And if this, like, to put this in perspective, like, if this were to hold true, this potentially could mean that as fish are smaller, they're maturing later, and as they're getting bigger, they're maturing sooner. Now, obviously, there's different ways that we could break this down, looking at size and age and how that might impact this curve and other things to potentially look at. We also looked at near-bottom temperature, just focusing on the FIS Water Profiler data.

So that data goes all the way back to 2009. So these are just mean near-bottom temperature data taken at stations that had a water profiler dropped on the survey. And as you can see here, from 2009 to 2020, there was an increase in near-bottom temperature. And then, and then in more recent years, there has been a decrease. So we're trying to decipher to see whether or not maybe there's some temperature effects or how to, how to maybe look at that a little bit bit more, in more broad-scale terms.

Obviously, there's a lot more environmental data out there that we could use from Alaskan waters, Canadian waters, but this is just more specifically from stations that were tested for near-bottom temperature on the, on the survey. And then also, we overlapped the Pacific Decadal Oscillation Index values. With the Coastwide A50 maturity. And as you can see, this does trend pretty well with the increase, as well as the decrease in the Coastwide A50 values. So, this is something that we could definitely look at a lot closer to see whether or not, you know, because the PDA— PDO is a lot more broad scale, you know, like Pacific Ocean Basin effects.

And whether or not it potentially could be more broad-scale effects that are driving these changes rather than potentially more localized environmental impacts.

Does anybody have any questions with that? That's all I had for that. We can discuss that a lot more in detail. Colin, do you have a maturity suite Are those more stable than the maturity age?

Sorry, say that again. Maturity versus weight.

Percent mature versus weight. That is not, I don't think that's something that we have, we looked at that. We did it for length. Yeah, we haven't done it for, we have done it for length. So the length relationship is a lot lot, it's not as good.

That's why age is used versus like length, weight. Age is a better predictor of maturity for Pacific halibut. One of your hypotheses there was that it's changes in weight at age. Yeah, yeah, yeah. That's why I think, I think, I think if we broke this down more at size at age and potentially looking at different I think that might be a little bit better way of looking at this rather than focusing on like the weight itself.

Yeah, I mean, I'm trying to think about that pathway of like PDO affecting weighted age and affecting maturity. Yeah. What are the best, best tests for seeing whether that pathway is going? Yeah. Yeah, I think that's definitely a discussion that we should have.

Yeah, yeah, about what— maybe some— they're just— I just wanted to— this is just something that we've just begun to start looking at. So I think it's kind of a new avenue and kind of fun to see now that we do have a really nice discrete historical time series of this information. And all of these are across all areas.

They, they are. Now, one thing to note is that the, the solid black line is a 3— each, each of those years is a 3-year rolling data window. And the reason why we did that is because in more recent years, we've had reductions in the survey, and we haven't been able to get samples across all 4 biological regions. So we've then lumped them by 3-year rolling averages to get every biological region represented by year, basically. And so that's how, that's how this curve is ultimately generated.

So that's, that's just one thing to note about potential differences in what's presented here versus what's presented on that actual black line. And that's just because we wanted to make sure that every region was accounted for, for every year that we estimated a coast-wide maturity estimate. Sure, let's also put a note to revisit that because there are some more sophisticated gap-filling approaches that might be overkill but might be helpful. Okay, yeah, no, that would be great. Yeah, in fact, the work that Ray was showing with the spatial model is a great way to tackle that.

Yeah, Ray will be presenting some of that tomorrow for you guys and see where we're at with that. Yeah. So I'll just move on to fecundity. So now that we've revised our maturity estimates for the assessment, the next step is looking at fecundity estimates. Now this is something that has never actually been done for Pacific halibut.

One of the questions out there that a lot of stakeholders are very interested in is what's potentially called like the BOF theory, which is Big Old Fat Female fish. Are they producing more eggs and outputting more into the water column than maybe smaller females are. And so, in order to address that question, we're going to be using what's called the autodiometric method, and that's depicted here in this figure. And basically what it is, is we collect a whole fresh ovary weight out at sea, and then we take a subsample of that ovary and place it in formalin, bring it back here to the office, and then with that preserved sample, We take a subsample of that subsample, weigh it, and then basically get a mean oocyte diameter for that weight subsample. And then that's what allows you to build this curve that you see on the right, which is called the autodiametric curve, which is basically on the x-axis you have mean oocyte diameter and on the y-axis you have oocyte density per gram of tissue.

And it allows you to then, once you get a value, can then extrapolate it. You can extrapolate that numbers of grams per tissue using the mean oocyte diameter out to the total gonad weight. And that gives you potential anaphrochondry or total number of oocytes for that individual, individual female. Um, the, the nice thing about this method is once this autodiametric curve is built, you can then skip the actual weighing of the subsample and just get an oocyte diameter from a subsample of oocytes, usually typically between about 250 to 300 oocytes from that ovary, and then extrapolate that out to number of oocytes per gram of tissue, and then using the whole gonad weight, get It allows you to just process samples a lot faster once this autodiametric curve is built. And so that'll be a nice thing for us because it's very tedious to count individual oocytes and ovaries, and it takes a very long time.

So to do this, starting in 2023, we've gone out and collected fecundity samples on our setline survey. So in 2023, we specifically just targeted Biological Region 3, which is the, basically the eastern, central, and western Gulf of Alaska. In 2024, we collected samples in Biological Regions 2 and 4. Regions 2 is basically southeast Alaska down to the west coast of the US, and Biological Region 4 is the Central Aleutians towards the Bering Sea. And then in 2025, we were actually able to collect samples across all 4 biological regions on the Settling Survey, which was great.

And then also in 2024 and 2025, we did what we called a Ficunity special collections, which were actually outside, post-fish, outside of the Settling Survey. To try to get later developing females. Because in order to build that autodiometric curve, you have to be able to get a whole gamut of oocyte development, whether it's the early mature females, all the way progressing through vitillogenesis through oocyte maturation towards fatter, bigger oocytes later in the year. So, we did collections in biological region 2, uh, mostly in northern British Columbia.

And we collected those samples anywhere from, I think it was like mid to late August through late October.

And then in this year, once again, we collected fecundity samples on this year's setline survey in Regions 2, 3, and 4. And then actually currently, Currently, right now, we have 2 vessels on the water that are doing another fecundity special collections in biological regions 2 and 3, specifically targeting regulatory area 2C and 3A. And that will hopefully give us a little bit more spatial and temporal coverage in our fecundity samples.

Remind us of the seasonal cycle of egg development. So those, those August to October samples that should be spawning. Yeah, so our initial histological study was collected in the Portlock region, which was basically the central Gulf of Alaska. That basically found the annual— that was our first estimation of the annual reproductive cycle for females. And what we do tend to see is Vitillogenesis usually starts in May and then runs through August, September, potentially October.

And then once you start getting into the fall months is when you actually start seeing oocyte maturation for a lot of the females. So finishing vitillogenesis and progressing through maturation, and then beginning in the winter is when they start spawning. Yeah, that was that initial study that we did back in 20— I think that was 2017, 2018. That gave us an idea that the FIS sampling window was a good window for us to collect histological samples because we knew that females were beginning to start maturing at towards the beginning of the FIS and then running through this summer season. And then, as of For those first 3 years, we've collected over 1,100 G2 samples.

So a G2 female is a visual classification, basically, of a developing female in the field that you can visibly see, like, egg development. And those are the only samples that we're probably going to be able to use for autodiometric method because a spent regressing female is not going to be developed far enough to be able to use the autodiagnostic method for. And then these are just the age and weight distributions of those G2 maturing females. You can see that the majority of our samples, this is across all 3 years, were collected, the age is mostly between largest proportion of ages is 11 and 12 and 13, and then the weight, most of the weight, a high proportion of the weight is between like 3 and 10 kilograms, and so this just kind of gives you a distribution of the samples that we've collected.

And then lastly, I'll just go over some of the sample collections for this year's FIS. So as of the deadline for this presentation, which we submitted on 4 September at the beginning of this month. In Region 2, we were able to collect like 277 histology samples, only 5 fecundity samples. A lot of the females that were collected in Region 2 were classified as G4s, which means that they weren't progressed far enough along. Um, so they were only collecting fecundity samples on those G2 females that were classified visually in the field.

Um, Region 3, we collected almost 480 histology samples, 29 fecundity samples. Um, and in Region 4, we collected 427 histology and 38 fecundity. Some of these numbers will increase a little bit in Regions 2 and Region 3 because of, as of this date, we did have a boat still sampling until the very end of the fishing year, until 15th of September, so those, some of those numbers in those two regions will increase, but those Region 4 numbers are, are pretty final because those, that boat was done, I think, by July or early August.

Yeah, so that's all I have for for this SRB meeting. I will just note that I do, I do anticipate a pretty large update for June of next year regarding both potentially 2025 histology samples, maybe even 2026 as well. And then also a lot of preliminary data on fecundity estimation, which is kind of the next big go-around as we start gearing up for the next major assessment update, which will be in 2028. So, yeah, I'll be in the lab a lot.

All right, the next research area is that of growth, and this is a really important area. Really impinges on size at age, which is a critical issue for stock assessment. And most of our previous work has been related to delineating the effects of temperature on juvenile growth. Having identified that the juvenile stage is probably a critical stage for determining environmental effects on growth patterns, and In this case, we completed the study. We published a paper last year summarizing all those results on juvenile growth temperatures and demonstrated that juvenile growth is in fact highly plastic in response to temperature.

And we identify a number of molecular mechanisms that were involved in this growth responses. But now we're actually moving into identifying potential drivers of changes in size at age. This is, this is an area where you benefit from the SRB's input. And, and as it was in the last SRB 028, this was one of the biggest topics for discussion during our planning session. So this, this year you will see that size at age is again in the picture, and we really need to identify some paths forward regarding this area.

The next one is mortality and survival assessment, and this area is mostly completed. We had two tasks initially, and one was to identify discord mortality rates in the longline fishery. That was completed. We provided those results. We published those results in three different papers.

And more recently, we've been working also on discard mortality rate estimation in the charter recreational fishery. And now we have a manuscript in preparation that summarizes most of Claude's work on that area. So hopefully we'll have that one out fairly soon.

But the final area is the one on fishing technology, and that's one of the two areas where I'm going to let Bob show you the results, is investigating new methods for whale avoidance or deterrence to reduce whale depredation in the Gulf. Bob, go ahead.

All right, so this is just a, just a brief update on the work from our last meeting.

Joanne, in particular for your benefit, we'll be talking about some techniques that we've been trying to use to protect the catch from the whales eating it. So this is a schematic of kind of what happens typically as the gear, longline gear, is being hauled off the ocean floor. Whales are in the area and they generally are stripping the fish or taking bites out of the fish as they are headed up to the vessel. And we've been working with industry on a couple different methods and settled on this one as having the most promise. It's an underwater shuttle.

It's deployed by sliding it down the ground line or the main line that hooks are attached to. It travels down the line just by gravity and hooks with fish on them enter the device. The fish is released from the device on the inside. The hook hook continues on the outside of the device. Eventually it encounters a stopper that's pre-installed on the ground line, and then it rides up to the surface with all the fish inside the device.

And so that's kind of schematic there. Once the device gets up to the surface, it's hoisted up using a crane, and then the fish are unloaded all at once out of that device. So, we did some testing on that last summer in the presence of depredating whales up in the Bering Sea and Aleutian Islands area. And we had video footage, we had cameras connected to that shuttle device, both on the outside and inside. And since our last meeting, I've had time to go through some more video footage.

So, the table on the left shows that it generally successfully collects the halibut that are going inside of it. There's a small amount, about 8%, that pass through it. Those are generally small halibut, but the majority of them are being retained. On other fish that the fishers are interested in, particularly sablefish and less or so peacock rockfish, we had poor retention rates, but that was largely related to the way that stoppers fit into closing the device off. Those are fairly easily remedied, but for this study, we didn't have it fitting tight enough, and so that would be something that would be recommended correcting going forward.

As far as the size of Pacific halibut being retained, see the graphic on the right, the control is— and the shuttle are producing very similar sized fish. We're not seeing a lot of large fish being sloughed off to the outside of the shuttle. Or being too big to enter the shuttle itself physically.

And our conclusions thus far that the shuttles can be safely deployed and retrieved by vessels that have a picking boom. They've got good retention as far as numbers go, size of halibut, and there's continued interest from the fleet that are suffering from depredation and how to use it.

Yeah, yeah, just this is our last slide is just to show you the projects that we have currently going on. One is the Alaska Sea Grant, the one I mentioned, that funds the development of epigenetic method for aging. And we earlier this summer, we submitted a research proposal to the North Pacific Research Board to fund some of the work. So this is pending decision.

And that's all we have. Take some additional questions now or take it forward.

Maybe some helpful background and to justify the focus on growth, weighted age variations seems to be the dominant driver of changes in productivity. Ian has some good figures on this. It's enormous, uh, decadal changes in productivity of the stock, and it's not driven by recruitment as much as it is by changes in wage structure. So for me, that was a bit of a revelation. I'm used to employment recruitment-driven stock analysis, but this is mostly a punctual—.

The only comment I had is having worked on the epigenetic age clock for Atlantic halibut, we had to separate the males and the females because of the differences in growth rate to get a better clock. So we did it separated by sex. The other thing we did is we used the elastic net model to like reduce the parameter space. So, um, drive down the number of CPG sites, and then we went to more— a more basic— we actually ended up using a GAM where we could have sort of nonlinear effects of those methylation sites because some of them clearly had nonlinear relationships with age, and that seemed to work pretty well. So I can send you that.

Yeah, no, we actually have the copy of the paper that it's in. I think it's in review again now, so yeah, but we have a version of it that's probably 2 months old. Yeah, so that seems to help us get it.

On the epigenetic aging, I saw that there some retractions recently of papers from Australian group working on aging. Is that something that's worrisome for the field as a whole, or is it isolated? No, I'm not particularly aware of that retraction. I know the work of the Australian group, and I think what they're been trying to do over the last few years— correct me if I'm wrong— is to use pan markers for so cross-species. And I'm not sure if that retraction is because they were using a relatively small number of species.

There were, you know, 100 maybe, 80 to 100. So I'm not sure if that's one of the issues, cross-species validation. But I think one of their aims was to generate a pan-Phileaus clock.

Not a concern. Yeah, we didn't find it concerning for the purposes of doing something particular species.

From my understanding, I think there's, there's still a pretty good reason to develop species-specific models. Yeah.

Anna and Mike, I just want to give you guys a chance to ask any questions on the biology section before we move on.

Yes, I would— I have a— shall I go ahead?

Okay, so I was wondering about this, like, it seems that maturity age is, is coming down with kind of improved growth conditions, but I'm not sure how to interpret this Pacific Oscillation Index. So recent years it has been negative. So what kind of weather that kind of, or general climatic conditions that implies?

Yeah, that you're right that the, in more recent years we have been in a negative regime of the PDO.

Unfortunately, I wish we could go way back in time like the assessment and the MSC do with the PDO back 100 years. I don't, I don't know if we were to see way higher PDO values if that changed the A50 or not. I don't know, but that is a good— it is a good point. Yeah, my understanding is that this index is negatively autocorrected. Correlated.

So that, of course, explains some of the kind of peaks up and down, but the general trend across years seems to be declining.

Colin, do you want to wait until we have our kind of open free-form discussion to talk more about your next steps for that? I think that would be good. Yeah, let's do that. I think that could be a good discussion.

So yeah, Mike, do you want to jump in now? Oh yeah, it was very similar to Anna's question. I was just— with the maturity growth part of it, does that mean that growth You've actually been seeing a recovery in size at age in recent years, that they are actually starting to get bigger again, or did I misunderstand something there? Ian might be able to talk to this, but I do know in more recent years, some of the younger age classes have shown a slight increase in size at age. Now it's not to the historical trend that we saw like back in the '70s or whatever.

But there has, if you look back at, I don't know, when was the last, I don't remember when Ian presented that figure last, but there is a historical size at age figure and in more recent years, the younger age classes have shown a slight uptick in size at age. Okay, cool. Yeah.

Thanks, Mike. Sounds like we'll be able to talk about that more at the MSC presentation.

Okay, I think we're ready to move on to stock assessment. Yeah, you can either take a—. You're scheduled for a break if you need one. Um, I don't know if you need one or— it's up to you. Sure, let's, let's take a, uh, 15-minute break.

Sounds good. Right. Be back in 15 minutes, those listening online.

All right. Thanks, Pedro. So we have a pretty good chunk of time allocated for the stock assessment. I don't think I will take the whole bunch of it. I think we had left a little extra in case there were things to tie up, but Alan will probably appreciate having a little extra time for his section.

Uh, we'll kind of get started. So what I'm going to go through, I'm going to do a quick review of our assessment review process just to orient us of where we are in the multi-year process. Uh, I'll go through the SRB request from the previous meeting. This is pretty light just given where we are in the cycle. And then I'm also going to step through something that we do for the commissioners actually just last week, which is go through the top research priorities, try to identify which ones are the most important.

And hopefully it will help also tie in some of what you've just heard from Josep, why we prioritize certain things and how potentially important they are for the stock assessment. And then I'll just briefly step you through the steps that that will be taken, the process taken after this meeting, leading to the management advice for this year. Our— maybe, Joanna, for you, our process is a little bit unique in that we have a two-part review, June and then September. The data won't be available to— the final data won't be available to me until the 1st of November. And the results are needed by the Commission basically by the 3rd week November.

So what we, what we do essentially is we finalize the models now, and the car is running with the doors open. We throw the data in, close the doors, and it's done for the year. So we don't make major— we don't make any model changes after this meeting. So, um, in, in, in a normal or in a, in a full assessment year, we might have a laundry list of things that be looking at at this meeting to make sure that they're sort of good and final on them for management. Because this is an update year, speak to in a minute, we have less to go through.

And this presentation is sort of out of character in that it's more looking forward to what we're working on in development, even though that's not generally what we look at in September. I mean, the only stock assessment I know of, one of the 2026 assessments, that's 2024. 6 Days. Yeah, so we like, we have, we'll have ages from both our survey and this year's fishery. We'll have the index from the survey, all the updated mortality estimates.

This is sort of a real-time stock assessment, which is exciting, but we, it's a lot of work in the fall. It's like 9 business days between data and assessment results, but because of that we make We're very tactical in making model changes. We make sure everything is tested and robust before we make a change, because the last thing we want is the wheels coming off the bus in November when we don't really have an ability to change. We used to actually have kind of what we called the ripcord, which was the ability to call a special SRV meeting in December if something bad happened and we needed to get some work done quickly because something unexpected happened. After a number of years of not having anything bad happen in the fall, we haven't— I don't think we've— other than a conference call, maybe 13 or 14 years ago, we haven't invoked that thing.

But we could, in theory, if we got data that were just wildly inconsistent with our stock assessment model, we could convene a call or something in the late fall before the management action would take place in January.

Perfect. So just to orient you to the process, it's a multi-year process, and maybe Alan will speak to this a little bit in his presentation as well, but it's kind of in phased lockstep with the MSE development. So we generally do a full stock assessment about every 3 years. And when we do a full stock assessment, we take everything right down to the framework, look at all the model assumptions, we rebuild data sources, we re— you know, We look at the inner workings of the geostatistical models and survey standardization, all that sort of thing. And that, that, so that occurs about every 3 years.

We've done that. This is the last 4 that we've done since 2015. It's taken us a while to kind of dial in exactly what this process looks like.

Then in the intervening years, and including this year, 2026, we do an updated stock assessment. And the goal for the updates, if something really important happens, obviously we'll do it. But the goal is to make relatively minor or necessary changes as something becomes apparent that we need to change. I'm trying to think of a good example of things that we've done in update years.

There's minor sort of upkeep on models, I would say, are things that we do between. Within, so that's the multi-year schedule. Within the calendar year, as I mentioned before, our June SRB meeting is generally research and development. So we try to do in this winter and spring after the management decisions are made, we try to do the bulk of the research and development on the stock assessment. Present things in June that, like, hey, we'd like to try this this year.

Maybe a few years ago we looked at, here's a new way to weight the— so we present that in June, we dig into it, and what this allows us to do is it allows the SRB to make— and sorry, I used the wrong word here— it would be requests primarily to the Secretary, not recommendations. Um, and that then we have the summer to go through those that actually do them and explore them. And then in, in a full assessment year, we would be coming back at this meeting in September and saying, okay, here's the list of recommendations or requests we had from the spring, and here's what we found. And then we would go through it. Okay, this one, for example, the weighting of the ensemble, we included after a summer of work.

Well, we're not quite ready to jump in on that yet, so let's put that on— let's table that for now, I did not use that as a stock assessment this year. But it gives us this buffer in the summer to go through that process. And then in September, we would be, like I said, we'd be finalizing the stock assessment for this calendar year, and then we'd also be planning for next June. So there's almost always a recommendation or two for, okay, now looking forward to the next process, these are a thing or some things that we'd want to see next June to talk about this research technology. So, you know, I think we want to clarify, we, we want to have a full and open discussion about the stock assessment, and there's no like tasks too big, but we, we will just be cognizant of the schedule as we set requests with the timelines for, for Dylan, because this is not, this is not a period where he can make big changes between now and, and November.

But as things come up, we should certainly build them and just, just start with them for later. Yeah. So, with that, this was the request from the June meeting, and this was actually rolled over in similar verbiage from previous meetings. And it's basically to continue development of a state space model for specific elements. We're currently using 4 stock assessment models as, as our management ensemble that creates the assessment.

Each of those models is, is equally weighted in the ensemble, and they're all maximum likelihood-based models, basically penalized likelihood with time-varying deviations being treated in a likelihood framework, but not a full random effects state space. So, this has been something that we've been looking at for some time. And actually, this summer was the first time that we've actually really made some good progress on this. It was, it was not something that we were able to include in the full assessment last year. We had things like the updated maturity curve and several other pressing things that we're working on.

So this, this, the request from the SRB was to plan to have this ready for the 2028 full stock assessment that's coming up, but to provide some checkpoints along the way. So this is, this is the first real check. And I'll note, in fact, that because the deadline, the document deadline was fairly early for this meeting this year, actually got quite a lot done after the document deadline. So some of the stuff that I'll show you in the presentation, what didn't make it which I think is okay in this case, because it's really— I'm just sort of showing you a progress.

So, to recap what I presented in June, in June we were really scoping, looking at, well, what platforms are out there that are providing a good state space platform? We've had, over the last couple years, we've had quite a few discussions about the pros and cons of writing a halibut-specific model versus using a platform that is more generic and used for multiple stock assessments. I think that's an ongoing discussion that probably will be ongoing forever. Even if we write our own model, we'll be discussing whether we should continue using that or using a platform. We, I will say, sort of historically, we've had a mix of both.

We, until 2012, we were using a model written in AD Model Builder that was Halibut-specific. And I think that that led us to some stagnation in our model development because it was not a generalized platform. It was somewhat unwieldy to add, subtract features and explore new things. And so it tended to just sit and that led us to— challenges, which we then in 2012, we started using StockSynthesis as a platform, and we have been able to build not only the 4-model ensemble, but we've been able to explore a wide variety of different structural assumptions within that platform. It's provided for a little bit easier review.

It's capable of being handed off. It's not a, you know, it's not a one-off model, so anybody who would familiarity with StockSynthesis, can pick it up and get it. So, through several reviews, it's been good. However, both AD Model Builder and StockSynthesis are in their sunset years. And so, we are— we've been, as part of the development of which models we're going to explore, we've been in touch with the Tuna Commission and been sort of looking around the world at what everybody's doing as they're converting some of the old AD model builder models to more modern stock assessments.

And as I presented in the June document, these three platforms, and I confess, I hate the acronyms that go with people's stock assessment models. I wish people would just come up with a name for their model rather than a dumb acronym. So the three models here, WAM, SPOC, and SEATTLE, We, all three of these are good contenders. They all have the basic features that we would need to mimic a halibut stock assessment. Some of them are a little closer than others.

WHAM, which is heavily used in the Northeast US, doesn't really try to have sex-specific dynamics, but has kind of a backdoor route where you can set up multiple stocks and treat them and treat your sex-specific data as multiple stocks. The developers of WHAM, similarly with SAM in Europe, have no interest in adding sex-specific dynamics. So they're not going there. So that, that's not terribly well appealing, but it is pretty well used across a bunch of different places. And I think we can, we can create a model there that's— that may mimic the how of it, dynamics, so we're not taking it off the list.

The next one on the list is Spork. This is being developed at Juneau. It is really mainly focused on spatial dynamics, which makes it appealing for halibut, but it's fairly new and currently only being used for sablefish. So I would rank this one— WHAMM is number 3, Spork is number 2 on our list right now, and I think it will be— I've already— we've been in touch with the developer, Matt Chang, and he's willing to work with us on this. So this is going to be something that we tackle a little bit next year.

But what we started on this year was a Seattle model. This is being developed and maintained by Grant Adams here in Seattle. Its primary focus is actually multi-species modeling, but it's, you know, all these models have kind of the same backbone in them. They have a, of basic age-based dynamics and a reasonably rich set of features to create age-structured models. This one's actually being used for multiple North Pacific stock assessments, including some of the Pollock assessments this year.

It has, it has the ability to have multiple sexes in there. And one of the things that I think is really interesting and why I chose Seattle to work on first is that it's now fully integrated with Dynamic Structural Equation Modeling, which allows you to have multiple environmental covariates and, and basically create a linkage map. So you no longer have to worry about multicollinearity and picking the best one. You can create a conceptual model, estimate the strength of the linkages, and do and integrate that as well as the uncertainty right into the stocks. Deals well with missing data.

It's really— I think DSEM is the the way covariates and stock assessments are going to be done. As far as I can tell, it's what we always want for dealing with covariates. So that's pretty cool that they now have DSEM completely integrated into Seattle. We're not there yet. We're not ready to get to covariates.

We're still building the basic framework. But this is— that's what will be— that's why we prioritized Seattle. Right off the bat. Also, it's nice having Grant right here in Seattle, so— and he's been very responsive. We've already worked back and forth on a bunch of different features there, uh, and, and made some progress on, on replicating things that we needed for, for Halibut.

And I don't want to discount, uh, Matt Chang in Juneau, who's also been very responsive and, and has offered to help us transition into some features using this So, anyway, that's the background on why we chose to start with Seattle. I would hope that by the time we get to 2028, we actually have a candidate model in all 3 of these platforms, perhaps multiple models in all 3. That's, that's the goal. And so that's why we're getting started now a year and a bit out so that we can have some functional models going forward. I just wanted to clarify, you don't have environmental coverage yet in Seattle, but the existing assessment.

Correct. Yeah, so yeah, we haven't, we haven't done anything with that feature yet, but having it there is, um, so the, the goal, the, the way we tackled this, and I think this is what we've sort of agreed on over the last couple SRB meetings, is rather than jump right in and try to do a bridging model and a complete map of every feature we have in the current stock assessment model. We're taking kind of a different approach, which is to start from scratch and try something very simple and say, what essentially is the minimum model we need to mimic halibut dynamics reasonably well, and build up from there, and then add complexity as we go until we get back towards something that's going to feel more like reasonable stock assessment that we would put forward to use for management. So there's a little more detail in the document, but we started with a pretty simple setup in Seattle. We've just— we aggregated everything into one fishery.

We've got a single survey index. We're using just 20 ages, and we're— this is essentially like the short coast-wide model, one of the four models, Part of this is in realizing that when we go to a random effects approach, we've got a lot of parameters to deal with. It's a long time series, may be challenging. So we're trying to start with kind of the— we've fixed natural mortality at the values that are used in the short post-wide model for now. Because it's a short model and we've got 100 years of fishing, we do have to have essentially complete, completely free initial conditions, both the numbers at age and in postgroups, so that the model can just initiate wherever it needs to, recognizing that we already have 100 years of fishing leading into the beginning, the year 1 of the model.

So, starting in any kind of an equilibrium doesn't make any sense. And does Seattle have numbers at age random effects?

It currently does not have numbers at age random effects, but it has recruitment.

Just the recruitment. Just the recruitment. Okay. I believe it does not yet have full— the full matrix numbers at age random effects. But there are no— Yeah.

Tricky. Yeah, I'm skeptical.

Of random effects on numbers at age, except where you really think you have a mechanism. You know, I personally, I don't think that fish just appear spontaneously at age 12. They're doing well. But if you're doing a small area and you potentially have immigration and emigration, there's actually a paper that Tim and others just published recently showing that under moderate— it's a simulation study showing that under moderate levels of migration and immigration and emigration, they can kind of mimic that. So I think that's a great case.

If we're doing a coast-wide model, I'm not so convinced that we need that. But numbers— but we can have random effects on natural mortality and selectivity and other factors that I do think have a logical time varying component. But it is interesting to compare across these models because they all— they're all similar on the COVID but you get into the pages and there's some important differences there. And these ones are available. Also, I have to admit, over the last 6 months, watching the development of AI and model development, having a missing feature is is not nearly as difficult now as it was a year ago because it's pretty easy to add things and test them quickly.

So I'm not so worried if there are individual features that are missing from these platforms. I'm really not that worried about it because, like, you know, I can— we can do it ourselves, or we, you know, I could contact whoever's maintaining the code. It's on GitHub, and, you know, collectively we can have a feature added. It's really— it'll be interesting to see. Yeah, it will.

And actually, as you'll see, we're surprisingly close to a result that looks like the stock assessment with almost nothing in this model, which is kind of interesting, kind of encouraging. Actually, pretty cool to have a very simple model that produces internally logically consistent dynamics that we think are reasonable.

So again, you'll note that there are some— a bunch of things that are added here that weren't available when we did the document. And so this, this current version of the model actually has random effects for recruitment deviations, which we had for the document, but since then we've added random effects for fishery catchability. We've got random effects for time-varying parameters on the selectivity curve for the survey. And in fact, I've got a model running that has it for the fishery yet, but it's not the one that made it in this presentation. For this presentation, we had— this version has time invariant fishery selectivity.

We've got a couple different shapes of fishery selectivity. The fishery selectivity itself is challenging. In the current stock synthesis framework, we essentially have unlimited flexibility in how we can set that up. And as you'll see, the challenges are that we have sex-specific age composition data only for a period of years. And so in the, in the current implementation, we're using a random walk on the offset of male and female selectivity.

Basically, the apex male and female selectivity curves has a random walk, but it's, it's a time block because we don't have any data on the early part of the time period. So that's, you know, you think about that now, we've got a random walk, we've got deviations, but we've also got a time block in there and it's an offset parameter as well as having the ascending limb moving back and forth. We've got an offset between males and females. So that's a lot of complexity in the selectivity curve. We don't have quite all those options implemented yet in Seattle.

So this, this version I'm going to show you doesn't have any— it doesn't have time-varying selectivity yet, and that's why it's not really a model that we put forward. It's just an example of what we can do. It does, it does have time-varying age-based selectivity for the survey. And again, the reason why we need time-varying survey selectivity is because we know there's a length component selectivity. The gear doesn't select small fish, it's a 60-knot hook, and there's a, there's a gear, there's a contact selectivity curve associated with that.

We've got changes in size at age, and the interaction of those two things leads to time-varying age-based selectivity. This is something that we've had to model for the last 20 years in the stock assessment.

We're going to need it for the fishery side as well. So this, you know, take this for what it is, which is just an example. In this, in this early example, I was using a nonparametric selectivity. It turns out that the nonparametric— I'll call it convergence resistant when you try to apply it to the whole time series because you're now basically estimating a parameter for every age for every year. And when you have, as I'll show you in a minute, when you have Periods of years where you don't have sex-specific information, it's very poorly informed.

The offsets for all those parameters between males and females become very poorly informed. So I'm working with Grant to come up with some ways and time block and have flexibility where we need it to fit the data, but not years where we essentially can't estimate two completely separate selectivity curves for each year. From periods where we only have aggregate male/female data. And unfortunately, until 2017, we don't know what the commercial— the sex ratio of the commercial catch is. One of my big surprises when I got here at the Commission 15 years ago was we have one of the most incredible data sets in the world, but wait a minute, we don't know what the sex ratio of the commercial catch is?

And it turns out it's about 80 to 85% female. So that makes a huge difference in the female spawning biomass having most of that coming from it, having most of the commercial catch coming out of the female portion of biomass. That has a huge effect on the scaling of the population biomass. And the fish does have sex though? It does, and yeah.

And the selectivity, I mean, is it mostly contact selectivity? I mean, can you use the selectivity information from the PIS to constrain the fishery selectivity, or are there enough differences that—. We haven't tried to use a prior from the survey selectivity. The route we went was just getting sex-specific information. We can actually look, I think I have the plots of both compositions, you can sort of eyeball and see what the percentages are.

The biggest challenge is that the fishery is the landed, so there's the size limit that cuts off all the young fish and that's where all the males are, so we have a much higher fraction male in the survey catch because it's all sizes, where the fishery is the age comp from fishery is the landings. So in the current stock assessment models, we actually separate the sublegal discards for the commercial fishery, and we treat them as a separate fleet, and we have a separate selectivity curve that is dome-shaped, that it's small fish, males and females. In this model, I've just aggregated all the catch. I'm just saying there's catch, and I'm just using the age composition position from the fisheries. It's kind of an exploration of— you even need to— commercial fishery.

At some point I can add another fleet back in. We can do the same approach in Seattle. I just haven't done it yet. Starting with a simple one, one fishery fleet. I'm using all the mortality, so bycatch, sublegal discards, recreational, it's all just piled into the fishery.

And I'm just using the commercial fishery age comps because that's the bulk of the fishery. But it's definitely an approximation. It's not an endpoint model.

So this is the random effects, and don't worry too much about the trend here because this is going to just depend on what's going on in this particular example model. But I just wanted to show that we can propagate the variance. The whole idea of going to a state space model is that we do a— we'll do a better job of propagating the variance in the time varying deviations because we don't have to fix the variance parameter. We can actually estimate it. And it seems to be working reasonably well.

We're propagating the— both the time series, the random walk variance from the fishery catchability, but that's also translating the other model quantities as well. So that seems to be working fine in this application.

So we're fitting, you know, this particular model is fitting the time series quite well for both the fishery independent setline survey and the aggregate fishery, noting that we model the fish— the survey is modeled in numbers per unit effort, and that's because That's the fundamental unit that we're catching them in. We've had— we used to model them in weight, but then recommendation of the SRB, at least 10 years ago or more, we realized that we were modeling it in weight and then you put it in the model and converting it back to numbers, taking it out of the right numbers at age, and then converting it back to weight again to fit. So it just made sense to model it in numbers. And so that's why the trend you see varies so much. When we plot the survey catch of legal size fish against the fishery catch rates, they actually line up really well.

We get a very similar trend from the logbooks that we get from legal survey catch. Again, don't get too hung up on the fits. I'm just illustrating that we're actually mimicking the, the trend in both of these indices pretty well. Again, of course we are for the fishery because it's our primary catchable So, we're going to try to figure out what's not fitting.

Um, so this is the composition information. It's a little small up there. Females are the red curves on top. Males are the blue curves on the bottom. This is the survey data and the gray bars behind are the raw data.

And the colored lines are the fit to those data. And I just— I'm showing this just to show that we're actually mimicking the distributions reasonably well. We're not quite at the point where it's time to look at the residuals in detail and actually dial in, you know, is the data weighting exactly right? And do we need a little more flexibility here or there? But just to show that with some time-varying selectivity, we're able to mimic the fact that even in the survey, the bulk of the catch is female, but you can track your classes through these data.

And, you know, with time-varying selectivity, we can fit both the trend and the comps reasonably well. And I think that's the important piece. With time-varying processes, with just time-varying selectivity, we'd expect that we could fit one or the other pretty well. But the fact that we can fit them both is actually encouraging. You can mimic the process pretty well.

And this is really where the, the sex ratio information is anchored, is by the survey. The fishery, as you'll see, we only have 2017 onward. In this particular run, because I don't have time-varying fishery selectivity, we can fit the index quite well, but we can't fit the comp data. And in fact, the fits to the males are horrible. It's predicting a similar fraction of males, but you can see that the males in the fishery in this particular run— can't get my cursor, here we go.

Can't see it, that's right.

Yeah, there we go. You can see that the fraction male in the fishery, if you just look at the gray bands in some of these lower panels, it's that, that's, you know, 10 or 15% of the catch is coming out of the males. So this is, it's pretty important to get this right, and this will actually have some effect on scaling. We found when we first got this sex-specific information was that it changed the scaling of the entire population estimate because we were taking a lot more females out of the population than we thought. And so we had to have more of them out there to still have any left at the end of the time series.

So it actually— this was in 2019, the full stock assessment in 2019 was the first time we had enough data to actually capture this. And it actually had a pretty big effect on the scaling of the overall stock assessment. So the challenge here is in figuring out how to get the right level of complexity. So we need some change in fishery selectivity, even when we don't have sex-specific data, because we know it's still going on in that time period, but not so much that it's not estimated. So that's really as far as we got with this development, was trying to find the right amount of flexibility in the selectivity curve.

To get in the ballpark here. And then again, once we get in the ballpark for fishery selectivity, then, then we can sort of kaleidoscope out and start saying, okay, well, what if we break off the recreational fishery as a separate fleet, treat that separately, how does that change things? What if we add back— we've also in the current models, we have the commercial fishery, the survey, we have recreational fishing, and subsistence are a separate fleet. We have a little bit of composition information to inform the recreational selectivity. We also treat bycatch and trawl gear as a separate fishing fleet.

And again, we have some composition data to that. So those will all be things that we can add back to this model as we go. But I am quite curious to see what we get just, just the full aggregate model once we get this flexibility in. So the big thing that we need to add here, I'll just go back to here. We need to add the ability for a parametric shape to take a different, to have a different peak selectivity.

That's the one thing that's really missing here. In the current stock assessment models, the males will asymptote to not full selectivity compared to females. And that's the, that's the big thing that's missing. And once we have that implemented, then I think we'll have a better fit to the survey data, and then we can also explore letting that change in the fishery. And that will, that will fix this issue of having way too many males in the, in the fishery composition data.

That's to the, the sex aggregated data before 2017, surprisingly good, given that it's a fixed selected. Yeah, I know. I was pretty surprised that we were as close as we were just with such a simple model here. I should put simple in quotes. It's got a lot of— it's got a lot of constrained random effects parameters, but it's really a nice straightforward version.

Can you speak to that a bit? Like, just Grant Adams, who he is, what this is all available on? Yeah, yeah, like the—. Sure. Yeah.

So this is a model. The Seattle model was first developed by Kirsten Holzman, used in the Bering Sea. And the goal— it's actually a multi-species model. So you can have N species in this model and model them all simultaneously. And it's got the ability to have diet and consumption data.

So there's— so Grant, as part of his dissertation, I was actually on his committee. He did some work looking at the Gulf of Alaska. So Kirsten had done some work in the Bering Sea, Grant worked in the Gulf of Alaska, and he had a 4-species model where he had arrowtooth flounder, pollock, Pacific cod, and halibut in the model, and he had predator-prey dynamics included. And he actually looked at what— he did an MSE on the multi-species approach to see what would be missed by not including multi-species dynamics. But in the process of building it all out, he basically built platform, fully random effects platform.

That was his, his addition above what Kirsten had. She had written Seattle in ADMB for the Bering Sea, and he rewrote it in TMB for the Gulf of Alaska, which allowed him to random effects. He's since rewritten it in RTMB. So this is actually, it's actually an RTMB. And like I said, they're actually using it for, well, you would have seen a bunch of them last week.

I think 3 or 4 assessments this year in the Gulf. Yeah, that's pretty good. And then 1 or 2 in Bering Sea. So at this point, we're not really interested in any of the diet stuff. That's, but that can all be turned off.

Doesn't run unlike stock synthesis. You're not running every feature in the model every time you do any feature in the model. These, these models in TMB, you can, when you turn something off, it's just not even part of what's running.

So, yeah, I just want to say that the acronym stands for Climate Enhanced Age-Based Model with Temperature Specific Trophic Linkages. Energetics. So that gives you an idea of where it wasn't developed as a stock assessment per se, but it's been sort of evolved, had all those really advanced features behind the scenes.

And, and then Grant came along, has developed it, put in— they've been able to run it as a single stock assessment. But it's now being heavily developed for single species stock assessment as well as multi-species stock assessment and all the trophic linkages and things like that. And something I learned at this meeting I was at a couple weeks ago, it's actually, they're keeping it in TMB, not, Grant originally wrote it in RTMB. Yeah. But they found, I think it was the DSAM stuff.

Yep. This, there's some big matrices that have to be worked on and TMD runs it in less than a minute. RTMP takes almost— so it's that one thing they found where the RTMP versus TMD. Thanks for that reminder. He's got it in both.

Yeah, it's, it's there both, both in RTMP and TMD. But you're right, I forgot about that. And it's, it's developing at a pace. I think since I put we put the document out, it's like 7 versions down the road. It's all GitHub.

You know, you can download all this stuff, and it's, it's many versions down the road. Actually, in testing out several of the features, I let Grant know there were some inconsistencies in some of them, and they were fixed by the time I, you know, got back to the office the next day. So it's a fun time to be involved in model development. And again, I don't wanna shortchange Matt Chang with Spork because he's do— he's, he's got the similar level of development going on in that model. So I have no doubts that when we get there, both of them have done a complete map from several different stock synthesis models.

Most of the features are there. It's just we use, we use so, so much of the richness features in stock synthesis that I think we have kind of a unique example of needing a lot of overhead to support what would be an exact map for our current starting system. Mike, I saw you had your hand up.

I've got a whole bunch of questions and ideas floating around my head, so I'm trying to decide where to start. I guess the first one is, have you— I want to go back to Olaf's idea about trying to use the sex, basically the information in the survey with the, to help inform the selectivity pattern in the fishery. And like, right now I'm thinking of it in terms of an age and length structured model, where if you had the length structure in there along with kind of bisects and you had the minimum size limit and you had your selectivity curve for the index, if you made an assumption that said the selectivity curve for the— basically the catching part of that has to be similar to what it is for the index, then I think you could do something with that. But right now you're kind of— leaving some information on the table by telling the model that these are two completely independent things when actually they're quite similar?

Yeah, so we've been down several of these paths.

Let me start with the survey information. So I said we weren't really using the survey information to inform the fishery, but in the current models, we actually do use the survey information to form the sublegal catches in the fishery. So we don't have sublegal age composition information for most of the, well, for any of the fisheries because those fish are discarded at sea. And so we use the age composition of the survey sublegal fish as essentially a prior on the sublegal captures in the fishery. And we do that in the current models by treating those as a separate fleet, because it's simpler.

We can isolate just that information, and we can then treat the fishery information, the fishery comps separately. And so that's the way we actually, what we do in the current 4 models. We use the survey as a proxy for the commercial fishery sub-legals. In terms of size-based dynamics, so we spent about 5 years working on a model that actually had size and growth information included in it. And we concluded that it wasn't worth the trouble for halibut.

And the reason why was because there's so much variability in growth, the CV of size and age is huge, so the length doesn't tell you hardly anything about the age of the fish, and we have time-varying growth. And so to model, to add in size dynamics, you now have another time-varying process that you have to deal with, which is time-varying growth. And so we were really getting a lot of overhead to add in any kind of size-based information. It was a huge amount of overhead in terms of both calculations and complexity in estimated parameters, and we really weren't getting anything back on that. That's— I mean, when I got here to the Commission, I was— I had quite a few years of using models that included length comps and age comps in the same model, and it just— after 5 years of working on it with Halibut, just concluded that it wasn't wasn't worth it for Halibut to try to deal with growth dynamics explicitly in the model.

I think it might be worthwhile in a research-style model, but yeah, that's the history anyway. Maybe it's time to revisit that again. You'll— there's so many different ways to try to get at this thing though. It's like you can have the complete age-structured model, but letting selectivity and size at age vary over time. And so, so there's other ways to get at it.

It's just, I'm, I'm just, well, I'm often surprised when, when trying to do something that I think should be mechanistically more realistic doesn't work out better. But it's not— this wouldn't be the first time that that's happened to me. So, yeah.

But yeah, also, it is perhaps a case to be made that the selectivity, especially in this model where the fishery is being aggregated to the coast-wide level, the selectivity is very dependent on where the catch comes from in a particular year. So we have big differences in size and age, in sizes and ages in the catch across the geographic distribution of the fishery. And so right now that's being catch-weighted up into an aggregate composition data for the whole coast, but that could be quite different than the survey where the survey is representing the entire population in proportion to the population rather than in proportion to the catch. So there's definitely some things to think about there in terms of using this survey as a prior. I'd be much more comfortable to do that on a region-by-region level and say, yeah, the legal size fish caught by the survey ought to look like the legal size fish in the fishery within a given region, then perhaps when it's aggregated up to the coast-wide level and the weighting is going to be potentially quite different in some years, especially in recent years, we've been taking catch out of some areas that's not in proportion to their biomass across the whole population.

Okay, that definitely makes sense. Yeah, some more, yeah, some more things to think about there. I guess my biggest conclusion, and I would encourage SRB to consider reaffirming the value of having sex-specific information from the commercial fishery. This is like fundamental information that we just started gaining in 2017, and as this time series gets longer, it's going to be more and more valuable. Is really critical ongoing monitoring.

It was, for the first few years, it was a research project and then we converted this into ongoing monitoring. And I really think that 20 years from now, whoever's doing the stock assessment is really going to thank us for collecting sex-specific information on the catch. And I guess, Brianna, just for you, and maybe you too, Mike, it's surprisingly hard to figure out what sex these fish are when they get landed at the dock. Dock. So the harvesters dress these fish at sea, they remove the gonads, and once those things are gone, it's really difficult to tell males from females at the dock.

We know all the big ones are females, but the ones that overlap, it's really, really difficult. We've done work trying to look at it through morphological features.

We spun up, just for historical reference, we spun up a project to actually have the fishermen mark their fish at sea so that the port samplers could then look at the marks and know males and females. It's very easy to look at the gonads and tell, but once they're gone, it's very difficult. So we actually spun that up because in 2015 and 2016, when we started this, it was like $17 a sample to do the genetics. And then over the 4 years it took us to actually get to the point where we were getting ready to to propose a marking, a coast-wide, okay, you got to mark your fish at sea, the price came down to, what is it now, 50 cents a sample or something, more or less. Way under.

I don't know what that percentage change is from $17 to 50 cents a sample, but at even a dollar a sample, we didn't feel like we could ask the commercial fishery to go through all that trouble. So now we're doing it in-house here. And actually, since, since we produced this presentation, Darren and the lab crew here, the interns and everybody else who's contributed, finished the 2025 information. And so you'll see 2025 here is aggregated, but I actually have on my computer, just finishing the bootstrapping for the sample sizes, we actually now have the males and females separated out. So I'll do the assessment this year with 2026 being sex is aggregated, but 2025 will be sex is disaggregated for this year's assessment.

So that will add another year to this time series. And we do that, we do, we have for the last few years done that sort of incremental stepwise. How much do, does it change the picture when we add just that single piece of information? We separate that out. And for the first couple of years, it made a really big difference because we had almost no information on it.

Now it's starting to— let's see, this difference is we had one here, but I still think that over the long term it's providing a really valuable time series of information on the sex ratio of catch. I remember from previous discussions that the, the otoliths are too clean to have any genetic material, so you can't go back with any— there's no, there's no biological information.

That was what I was just going to ask. Yeah, and we've thought about, since every, every bar in Alaska has a fish mounted on the wall, the old ones have skin in them. We thought about trying to go back and take samples out of those. Oh yeah. But not, maybe not a random sample of the population, so we wouldn't be generating age comps, although we might be able to look at some, some sex ratio information.

We do have some spot surveys that we we're done going back in time.

So we do have a little bit of sex-specific information from historical surveys, which you don't see in the survey plot here because I don't have— I'm only cutting it off in 1992. But we do have a little bit of information that goes back even as far as the 1960s, but it's pretty spatially limited. May you also be able to age those bar specimens? It's true with epigenetic aging. That's a good point.

Bar talks at every proposal. You've also explained why you didn't have it in for these harder— yeah, why wouldn't they have just done that routine? Yeah, and our first attempt on that was to just ask them, well, couldn't you just leave the gonads in and take the guts out for us? And the processors immediately came in and said, absolutely no way. Will you do that?

It turns out that anything— these fish are incredibly robust. They'll go 7 to 10 days in an ice hole and have no problem with feed quality. But if you leave any part of the guts or the reproductive tract in there, they will— they're no good when they get thawed. So that was even just leaving a little piece of the gonad in there was a non-starter with the processors. It was a biologist learning curve moment.

Like, couldn't you just do this for us? And the answer was hard no.

So where did I want to leave off? Yeah, again, don't get hung up on the results. As I mentioned in the document, we— I was kind of encouraged seeing the fact that as the signal for recruitment peters out at the end of the time series, Treating them as random effects, I think, actually propagates the uncertainty in deviations really well. And that's one of the things I most think is most appealing about these state space models. Again, not having to fix those variance parameters.

We go through in the current model, we go through this whole tuning process where we tune the variances to match the variability that's coming out. But we're not propagating the variance in the sigma. Parameters, processes. This model doesn't have a lot of variability in the spawning biomass estimates because natural mortality is fixed. And that's really going all the way back to the beginning.

That's another reason to get out of the penalized likelihood space and into random effects. I think we might be able to do a better job of propagating a certain natural mortality as well, but we haven't tried that in this particular framework.

So, to sum up this effort, which is still very preliminary and in development, we've got a model up and running in Seattle. We're actually in the ballpark of sort of biomass levels. It's not fitting the data to the point where we're really looking at this as a production-level assessment yet. But I think with a little more work on some of the selectivity features, we're going to be pretty close pretty quickly. And then we'll go through a stepwise process of building out some of the additional complexity to see if it matters and see which, and/or which pieces matter.

The key pieces are dealing with fishery selectivity, and I think with some time blocks and some constraints on the variability in between males and females prior to the good data information, I think we can, we can build something that that will work or get us to more or less where we are right now. In the long term, I'm really excited to get into covariates and using the DSEM. That's, that's really encouraging. And it's something, especially if we can develop a longer time series model where we can bring in the phases of PDO, that could be really interesting to do in this DSEM. Where it's all seamlessly integrated into the stock assessment.

The plan from here, or our proposal for the plan from here, is to continue refining this model and hopefully by next June have something that looks more like an actual assessment from this model. And then also begin, and I think we'll begin probably on Sporc next, because it's got the spatial dynamics. And I, Alan may bring this up in his presentation, but the nice crossover there is that we may be able to generate— Alan and I together may be able to come up with a model that we can fork off and have one fork being an assessment ensemble component, and the other fork maybe actually being a spatial model that can be useful in either as an operating model or as an estimation. In the MSE. So that's another place where we may have some good crossover from the MSE and the assessment development processes.

So I think we're probably going to put, at least between now and next June, we're probably not going to jump into WAM, but we'll hopefully at least have a couple of things to show you next spring that are a little closer to something that we might call a stock assessment. That's all super cool, Ian. I would, I would still encourage trying to do a little bit of work on WAM because like one of the things that's been a comment before is that your, your ensemble models all have fairly similar structure, whereas now between the 3 models you're talking about here, you can actually have quite different structure. If you have one that's spatial, one that's doing numbers at age random effects, and another one that's doing more of the kind of what I'd call process-based random effects. And those might end up being quite different, which would be really interesting to see what will come out of that.

I'm still really confused both by the literature as well as just trying to think through it, what the— what the real implications are of doing numbers at age random effects versus doing processes. I tend to agree with you that I like the idea of doing processes better. I don't like the idea of, oh, there's just all this added process variability that we're not going to try and understand where it comes from or what's happening with it.

But it basically— it should be more robust a misspecification of where we put our variability. Because I think one of the other things that's been shown with some of these models, especially once you get multiple time-varying processes in, if they're not the ones that are correctly specified, then you can get pretty far off. And also, even if they are correctly specified, I'm still not entirely certain when we have multiple time-varying processes that the models able to get those, tell which, which things are varying right some of the time. I think it really depends where the information's coming in from your data and whether that process should have some data that's almost independently informing it as part of the dataset.

I'm nodding my head, Mike. I agree with you entirely. And your, your point about where you're assigning assigning these things makes a huge difference. So if your, if your variability is coming from random effects on numbers at age versus coming from random effects on natural mortality, you're going to get your reference points, everything is going to be affected by that. You're essentially assigning it to different, different bins, different buckets in the process.

And that's going to have a huge influence, maybe not on the time series of biomass you estimate, but why you get the time series. Series and what that means for your reference points is going to be huge. Yep, yep. Mike, I was about to say the same thing about not discounting, uh, WAM. And Ian, when you mentioned before about, um, it not being quite as big a lift as it used to be adding new features, uh, you know, I think the main problem with WAM is the lack of sex-specific capability and And that might no longer be quite the kill point that it used to be.

Yeah, to maybe add that. Challengingly, there's a postdoc here on the West Coast who spent considerable amount of time on a fork of WHAM adding sex-specific dynamics. So that exists, but there hasn't been a lot of interest in bringing that fork back to the main program because they went to this multi-area, multi-stock WAM instead of that direction. So that's honestly that, and for WAM and SAM, which is obviously much more mature than any of these, it's— I candidly, I'm quite frustrated with the developers of both of those that they're not even interested in considering sex-specific dynamics because there are species in New England and in Europe that have highly dimorphic growth that really need that. And I think they just don't want to see what happens when you put it in.

Yeah. And they've done, at least the Mid-Atlantic, New England ones, they've done preliminary analyses that has convinced them that they don't need to worry about it, but I don't think those analyses are very informative because they're basically making up data and then saying, oh look, the data are the same for males and females, and so therefore it doesn't matter. And so I think it's a foregone conclusion by the way they have decided to try and disaggregate the sexes. But I totally agree with you. I wish they would consider it more.

I think we just got the first sex-specific assessment through the Northeast Center, I think, with the assessment of longfin squid. So maybe Dogfish is, but if, if not, then we've done the second one. So maybe we'll start moving the needle a little bit and they'll care about sex. I have to believe that time and reviewers will wear them down eventually. A lot of the reviewers come from Europe though.

There, it's still an echo chamber. It's true. I mean, for us it's non-negotiable with 85% of the catch coming from females, you can't do this without sex-specific dynamics. It's just not possible. But I realize that not every species is as extreme as Pacific halibut.

Well, that argument should hold sway for summer flounder as well, and it doesn't, and it hasn't. That's a long historical thing. It's a topic for another venue. I was going to say though, I mean, the main thing that has given me pause cause with WAM is the no percent age random effects in the projection years. We've really struggled with that because we've got no constraints on them when we're projecting out.

That is less of a concern here because you're running the assessment every year.

So, one of the main drawbacks of WAM is not really relevant to others using that WAM considering, at least as part of UNSOLVABLE. And I, yeah, I don't want to give the wrong impression. It is still in the three. We're still planning to develop all three of these platforms. We're just giving you the rationale of why we prioritize one, two, and three.

And again, you know, we have, we have until 2028, so we're trying to get started here and get some, get some functioning models before it gets to the time where we're actually trying to refine them to production-ready quality. I was just going to ask about that. So you're thinking 2028 there will be state-space models of the system? Is it too early to say? At least for review.

I guess I won't say that we'd necessarily have them in the ensemble yet, but I think it'd be nice to have them ready to review and compare. And if we have some that are— that we can match up that we feel are as good or better than the ones we have in there now, I wouldn't see why we can't replace them or add them or some combination. It's what we have right now with the two-way cross of time series length and data aggregation is kind of a clean way to do it without worrying too much about weighting. When you get into models where we have any kind of nesting, We need to be very tactical in thinking about how we're going to do weighting across hypotheses. If we have 4 models that have different ways to treat natural mortality, but they're all basically the same model, and then we have another model that's, as Mike mentioned, is spatial, and then we have a model that has random effects on numbers at age, are those 3 hypotheses or 5?

Or, you know, we're going to have— there's going to be some thought that needs to happen there, which is why I think it's important to build the model sooner. And then we can get through some of that other stuff. You know, I'm also thinking forward with, with a platform like WAM, if we reverse engineer something using the multi-stock approach, trying to think forward, we can, we can reverse engineer a lot of the dynamics, but then when it comes to calculating reference points, things that get interesting in a hurry. So we want to get to the point where we have a model to work with before before we're actually trying to figure out what the reference point calculations look like. Because we are doing things like dynamic depletion levels in a particular year, which is really easy to do in Synthesis, not so easy as other platforms.

Seattle actually has an unfished, an annual unfished biomass calculation that provides the same kind of shadow population that you'd calculate a dynamic reference going against. And I believe that Smoke may have that as well, although I haven't checked that features. Anyway, it's pretty exciting time to, uh, be working on these models because unlike back in the late '90s when we were all doing this in ADMB, everything's on GitHub now and everybody's able to just add features like crazy. Much— I don't know what you think, Mike, but I find it much less frustrating than it was back in the late '90s and early 2000s. Well, the frustrating part is I have to learn to code again.

I was so comfortable with ADMD and now I have to learn new languages and problems.

May I ask about the natural mortality? Please, yes, go ahead. Yeah, so in the stock assessment document, I noticed that you had used values like 0.15 and 0.30, but were these just like starting values, or were they— how were they estimated, or are they just dummy values to run the model somehow? Yeah, thanks for the question, Anna. So in the current 4-model ensemble, 3 of the models estimate natural mortality, and 1 of the models— so the 2 long-time series models and the short Arias' fleets model all estimate natural mortality.

And it's estimated as an offset. So males and females have different level of natural mortality. Ages 0, 1, and 2 have elevated natural mortality compared to 3+. That was a, that was something we added 2 full assessments ago. So I could reference you back to the 2022 stock, full stock assessment for rationale on increased natural mortality at the younger ages.

When we estimate the values, we get values that range between about 0.17 and about 0.21 for females, and males are slightly lower than that. Interestingly, what we found when we, when we first got the sex ratio of the commercial catch was that that had a pretty big effect on the estimate for males. So we used to estimate a pretty big offset between males and females, that males had a much lower natural mortality than females. And then when we got the, um, when we got the sex-specific information, it actually brought those two estimates closer together. And I don't know if I could get the mechanism for that right off the top of my head again, but essentially how much you catch out of each sex dictates what that offset is.

So the, well, the way to think about it is that we used to think that we were catching a more equal sex ratio. So in order to have more older males, we had to kill them off slower with natural mortality. But once we realized we actually weren't catching very many, we didn't need natural mortality to be as high because they lived that long even fishing. So it was an interesting feedback between knowing the sex ratio of the catch and getting a better estimate of natural mortality. Mortality.

And that was actually really hard to explain to the commissioners why all these other parameters were changing. We got the sex ratio of the commercial catch because our understanding of the entire population dynamics hinges on what we're taking out versus what we see down the road. So we estimate in those 3 models, we're estimating values for females between, like I said, 0.17 and about 0.21, with males say in the 0.15 to 0.18 range. We have had historically, we've used fixed values. This was a huge discussion in the 2022 stock assessment.

The value for females in the short coastwide assessment is fixed at 0.15, and that was a historical decision that we traced back to basically wanting to have at least one model that was somewhat more conservative. Like, if we're wrong, we better have one that has a slightly lower natural mortality. Some of the rationale for lower natural mortality is that the PIT tagging study that we did in the early 2000s, if we did an independent estimate of natural mortality from those PIT tag recoveries, it was more in the range of 0.1 to 0.12, which is quite a bit lower than the estimates we get out of the assessment. So We have maintained one model that had a fixed natural mortality for females at 0.15. In the short coast-wide assessment, we estimate the value for males, and it comes out to be about 0.13.

And so we've, we've maintained that with the caveat that it sort of bothers everybody that we're still using that assumption, and we don't really have a good basis for that assumption.

In this— yeah, natural mortality is probably the estimates of the model predictions are really sensitive to the natural mortality. Absolutely. I mean, so we show that every year that if you— if you— it's almost a linear scaling. If you increase natural mortality, the biomass estimate goes up, and if you decrease it, it goes down. And so we show that sensitivity every year in the short model.

Ideally, we would like to be able to estimate those values in all 4 models, so they're consistent with the data. This short coast-wide implementation, we've not been able to do that yet. I'm curious to see— that's why I have it as one of the bullets here— curious to see once we get a sort of a full random effects approach, whether it's estimable. At some point, this time series will be long enough, we'll be— should be able to estimate it.

10 Years ago, we couldn't estimate it in the coast-wide, in the shore areas as Fleets models, but we've been doing that since 2022. So, you know, this time series, the survey information starts in the early '90s. And when I got here in 2012, it was still actually a relatively short time series, but we've added another 13 years since then. So, before '14 this year, I guess. So, at some point it will be estimable.

And as you'll see when I get to the research recommendations, this kind of estimability of leading parameters is one of the main technical research recommendations for the stock assessment over the next few years. Yeah, I would add there that it would be also really interesting to see the variability of the natural mortality because basically, It's possible basically that natural mortality stays roughly in the same level, but for instance, the variability about the average increases as a function of time.

Absolutely. I mean, I think we— yes, there's no doubt that natural mortality may be varying over time. And we may be able to come up with a model configuration that could allow some variability in natural mortality over time. Yeah, and especially like the time-dependent variability, that was it most consistent earlier and now it's maybe varying more from year to year. Quite possible.

As an interesting sidebar, I often get asked, what is natural mortality for Pacific halibut? Is it disease? Is it Whales? What kills big old halibut? Is it senescence?

And honestly— Is it mortality? Predation? I don't think we know. Yeah. I honestly, between disease, predation, and senescence, I don't think we have a good idea of what would kill a fully grown Pacific halibut.

Certainly whales eat a few, but is that really the dominant source of natural mortality? It's hard to say.

Senescence could be something interesting to look at, like especially if there's information about the relative fecundity of the really old big ones. Like does it bend down at some point or level out or—. Well, that's going to be when Colin finishes fecundity and skips spawning, senescence will be the next big question.

Yeah, and I've been trying to get a grad student to work on senescence in fish forever, and nobody wants to tackle it, probably for good reason, because it's going to be really hard to get any real data on it. But do we really know whether senescence is a thing in fish? I don't know. It is in a lot of other animals, but maybe not in fish. Yeah, I had a student who worked on that with, um, Wenders.

The freshwater little fish, and we managed to find quite clear evidence of senescence.

You can see it histologically, we just have yet to see it. Yeah. I mean, halibut, of course, are a little different than most fish. They don't ever stop growing, really. They just— the growth curve is kind of linear through their whole life.

That's one of the potential side applications of the epigenetic clock development is the difference between biological age and chronological age. When you start seeing those two diverging, that, at least in some wildlife species, has been shown that that's a sign of senescence.

Biological age shortening relative to chronological.

Yeah.

I think maybe now I'll transition into the last part of this presentation. Sounds good. And I just want to make a note, when we have our free-ranging discussion later on, I want to hear more about the tagging study that resulted in this. But I'll ask Ray to do it. Sure.

Yep. Ray has a good paper on that called Pacific Halibut on the Move from 2014 or '15. But yeah, we can talk about that later. Oh, this is interesting. It's got a little cut off.

I got converted into PowerPoint. I'm sorry about that. So these slides are actually not really generally for the SRB. These are slides that we usually present to the commissioners at our work meeting, because a few years ago they asked us to— we were presenting these laundry lists of research priorities. Their question, and I think this was actually— the SRB encouraged us to do this as well, was don't just tell them what you're going to be working on, tell them why and what's going to matter out of those research things.

This has really been part of the evolution of our prioritization of our research, which was that instead of just having a 30-bullet list of research priorities, we've gone to having— that list still exists in the full stock assessment, prioritized into several categories, which I've actually dropped the categories here, but we have the top 2 or 3 research priorities in each, each category and the rationale of why we're doing that. And so I've— we've moved these slides over from what we presented to the— when Alan presented to the commissioners last week, but I thought it would be interesting for the SRB to see it because I also wanted to just sort of show you again how we are generating some of the research priorities that you see from, from Giuseppe's group, and this back and forth of what we learn there and how it informs the research, and then why when we do sensitivities in the stock assessment, how it then informs what our priorities are. So I'm not going to spend a ton of time on them, but these are the prioritized stock assessment research priorities from our 5-year plan of research here at the Commission, the current plan. So the first one being just highlighting the importance of this ongoing monitoring of fishery sex ratio at age.

We did, you may recall, I'm not sure actually if you were here yet or not, a few years ago we looked to see We looked at this, the delta every year when we added the new sex ratio information to see how much it affected the assessment. And we also looked at whether we could potentially skip a year or skip 2 years between doing the genetic assay, basically trying to do a cost-benefit analysis. Do we really need to do this every year? And we concluded that although each individual year wasn't having a huge effect, especially at low stock sizes, it's really important that we didn't— that we continue to learn about the sex ratio of the catch. We didn't give up this piece of information just yet.

We only have 2017 to 2025 right now. So we— what we recommended was that we continue this critical ongoing monitoring for the time being, at least until we had a long enough time series to really anchor the stock assessment models in terms of that overall ratio. Sounds like it's not that expensive, but what's the total sample size? Total sample size, I think it's about 12,000. 12,000.

So you're talking per year. So it's, you know, that does not include technician time. The running of the sample is, I think the last estimate, 80 cents per sample. 80 Cents or something. That could have changed.

'Cause, you know, things get cheaper year by year. So, I mean, you're talking probably about $1,000 of reagents per year plus—.

The next one here is whale depredation, and we kind of already previewed this earlier. Based on the sensitivity analyses that we've done, we don't consider this to be critically sensitive as an assessment concern, but we recognize from the questions we heard from, from Annemarie and that we get from the stakeholders every year, this is really high on people's lists of things to think about. And even if we don't have a good way to estimate it, it's a, it's a big concern. And so our conclusion continues to be that the most fruitful avenue is to try to reduce the problem rather than just focus on how to estimate it. And so we're focusing our effort right now on ways to reduce whale depredation.

One— yeah, go ahead, Mike. I was just going to say one other thing though that intersects with some of your modeling plans is if you have time-varying M, it has the ability to start to account for not the perhaps the direct predation on the longlines, but the population level effect of enhanced predation by cetacean populations. Absolutely, and you know, part of what, one thing that people forget when they think about whale depredation, they think, oh, you're not getting the right answer because there's whale depredation, but the survey that we get every year, that the trend in the index doesn't reflect what we believe to be occurring in the water, it reflects what's really occurring in the water, like that really is the trend. And we've removed whale depredation by eliminating the stations that are contaminated by whale depredation. We should be getting an unbiased estimate of the trend from the survey, even if there's whale depredation that we can't see.

And so the delta between that and this, and the other signals we're getting, is what's going to tell us, just like you say, Mike, it's there in the data. And so to the degree that we can estimate an additional source of mortality, or, you know, people often ask, well, if you're not dealing with whale depredation explicitly, where does it come out? It comes, it manifests in the stock assessment as invisible productivity. It's recruitments that are smaller than they really are because they're, the decay curve of that recruit is just shifted downward because the whales are peeling them off at every age as they age. So we think we think that recruitment is lower than it should be, and we think perhaps that natural mortality is a little bit higher than it actually is if it's soaking up whale depredation.

And that's problematic to some degree because it's not really natural mortality, but the net effect on the catch levels are going to be, it's going to be the same effect. It's this case of whether we subtract it off added on and then subtracted back off again, or it's just subtracted off by nature in this case. And what we're getting is the yield that's available after that's already built into the population dynamics. What's worrisome is when you have natural mortality fixed, like in the Coastwide Short Model, you're assuming it's constant. And if whale depredation is shifting over time, then it's a bad interaction.

So we've shown that, that if you have a trend in other people as well. But we've, we've, we've documented this sensitivities in our assessment, which is that if you have a trend in unobserved mortality, you can end up with the wrong— with the wrong trend estimated for your stock biomass.

I think I've seen this at some point in the past, but if you have a time series of the number of stations in the fish that are discarded below a whale? Is there a clear trend there? No, there isn't a clear trend. And it's actually quite low because we've set up the on-the-water design to try to avoid them. Set 4 to 8 skates and then we run 10 miles.

4 To 8 skates, run 10 miles. And if the whales show up, we take evasive action. They'll actually buoy off the gear and back off. Go and then come back and pump the gear. We do everything we can to avoid well degradation.

So because of that, it's not a good—. It's not a good index. Exactly. Yeah, exactly. And interestingly enough though, there's a lot of anecdotal information that this is getting worse and worse in the fishery, but there isn't a signal in either the logbook— logbooks or the observer data showing that.

And it's also anecdotally, it's this brand new problem that's just come up in in the last couple decades, but I mean, Claude, you found records that go back to the 1950s documenting oil depredation. So it's a hard one. Yeah, this is a hard one. I mean, the issues like this where science says it doesn't matter, or there's no trend, but the perception is that it's a big issue. I mean, I think we have to step carefully there because it can lead to sort of lack of confidence in the science when there's this observation out there on the water that's not being captured in the assessment.

And that's why I say here, this is really important for fishery efficiency. For somebody trying to get their gear out of the water with whales present, this is a huge problem. It— that doesn't mean it's a critical issue for the stock assessment, the whole— in the big picture. So certainly spatial hotspots for the— yeah, it could have effects on estimation. We're using coastline model.

Trends happen in certain spatials but not in others. Sorry, Captain. There are places on the coast where people just can't put gear down because they can't get the catch bag. Where are those spots?

There's some in each of the, the regions where they The 4A edge. In the 4A edge. Yeah, right. And a little bit in the— but particularly on the edge itself. So a little bit higher up, just below where it says 4A edge here, right there.

Yep, it's the worst of it. And Northern BC sometimes has some. And are these hotspots that didn't exist before? Like, are the wheels shifting or What did people speculate? Is the population increasing?

So the sort of common understanding is that this fishery was a derby until the mid-'90s, and it got down to be only 24 hours long. So the whole fishery catch came out in 24 hours. There really wasn't ability to have a big problem, and then they went to quota program, and now there are boats on the water with gear in the water from March to December. And so the thinking is that that's what sort of gave the whales the opportunity to really specialize in picking walleyes. So while they're schooling, they're students.

Exactly, and once they learn, they're good students.

So that's the thinking, is that that's really when this the problem became most visible. Although there are records going back to the '50s of whales depredating fish on Long Islands, including some species that we don't really see a lot of depredation on anymore.

From since the mid-'90s, it isn't clear that there's been a big increase, but as catch rates have become lower, I think for an individual operator, it may have become more and more important in their business because when you're struggling to catch fish compared to what you had decades ago, it's a bigger deal if you lose a few sets of reels. So again, I want to be really careful in not saying this is a major problem for an individual operator or even certain geographic areas, especially, you know, guys going out of Dutch Harbor trying to fish the edge. It's virtually impossible to fish the edge. And even up into the 4B edge, the guys in the Pribilof Islands, which is the island right below where it says 4C on the map there. Sorry, Mike and Anna, you don't have the map, and I don't have one in my presentation, but we're talking about the Bering Sea edge.

These are— whales are just a huge problem up there. There are definitely areas that are just not fished in. That's where we were kind of— that's why we developed this shuttle project to try and give them a tool to use when basically nothing else works. And we provide them just another— there's been fleets, other fleets like the Sablefish Focus Fleet, that have switched to other gear types like Slinky pots for overnight, which are like a spring-loaded, small form factor, easy thing to swap onto their same longline. And there was very quick movement to that by the Sablefish Target Fleet.

But then there was the perception amongst the longliners for Halibut that were still there that now the whale concentration, the whales concentrated on them now, because they weren't going after the same fish as us. And that's another thing that's in that feedback loop that we get when we hear about it. It's like, how much of this is actually increasing behavior? But certainly since I'd say the mid-'90s, it's been hard to tease out whether this is an increasing problem or it's that way, it's perceived to be that way, and it's part of the team. And as Ian alluded to earlier, there is information in the fishing logs, and the way that data is collected has changed a little bit over time.

But trying to understand that data or be sure that it's completed is a challenging thing because there's a lot of zero days where they didn't see them, so they don't bother to fill out that field. There's times when they didn't fill it out, and it's like, Is that because you were fearful of what the consequences might be by recording it, or did you forget, or did you, you know, what was the answer there? So, I think we have a better process now at the DOC when we're doing recreating fishing logs of sleuthing that out with our port samplers.

But it's hard to look at the time history, at least from IX Forward, and see what transfer Another interesting thing is, so we conducted a sentiment survey amongst stakeholders, and it tends to be like we hear about conference format, it's like a group of our stakeholders, they complain a lot. And when we look at the results, the results, like in terms of their perception, like, you know, what they assessed it, very quantitative way, was actually pretty positive. And all the comments they put in paper, they were actually Overall, my phishing was good when they were asked anonymously to provide the feedback how well the phishing went on their perspective. But then when you go to the forum, you hear a lot of these concerns voiced very loudly. So it's— there may be also the perception and what is presented to highlight other This is an issue and we, we kind of wanted to, to work on and understand it better versus like, in the end, maybe the fishing isn't so bad.

Well, I mean, it sounds like one of those issues that it doesn't affect most people, but it does affect you. It can difference between profitable and unprofitable. Now you suggest we put a pin in the wheel of variation just to be a bigger fail. For the purpose. Yeah.

Yeah. And I didn't, I didn't mean to open every can of worms with these research topics. Mostly I wanted to give you the overview of why we had these prioritized topics and show you some of the crossover to what you've seen from Joseph's group. So the next one is just coordination with the MSE. I think it's sometimes overlooked that, well, one, I already talked about the development cycle being staggered.

So I do a full assessment in 2025, and then Alan's going to show show you the reconditioning of the operating. And see, we're always one year out. When we first started on this 10 years ago, we're like, how are we ever going to get this coordinated? Because, you know, I get the data and I've made a bunch of changes, and there's no way you can then run full MSE in time, you know. So we realized that it's, it's always this one-year kind of lag, which makes a lot of sense.

But I think another important thing, especially for consumers of the is that we have this feedback loop that goes in both directions. So it's not just assessment to MSE. It's things that Alan learns in the MSE then going back to the assessment. And so this is a— it's a really important back and forth between the two processes. The other important— another important feedback loop here is that as the Commission has been evaluating the potential for multi-year management procedures.

If we went to a multi-year management procedure, that would have huge implications for what I could do in the assessment, because I wouldn't be doing an assessment in 9 days every year. We could do, we could do a lot more, for example, toward Bayesian models, because I spent a lot of time on that and then realized I can't do, at least currently, can't do a great Bayesian analysis in 9 days, not for multiple models. So that would open up some other doors. I don't think I want to spend too much more time on that, but just to highlight that this is actually a really important aspect of our research and development here. You've already heard about factors affecting size at age.

I think I'll put a full pin in that for Alan, because he's going to talk a lot about how important size at age is to the reference points. So let's not even do that in the assessment, but it's, it's relevant for both the assessment and the MSE. The last few of these, and the parts that I put in bold for the, the commission, is fecundity and skip spawning. So now that we have an updated maturity curve, fecundity and skip spawning are the next priorities, and we've been trying to highlight, as we did for maturity for at least 7 or 8 years leading up to last year, these can have a big effect on the assessment If we find out that big old females have a disproportionate amount of eggs per body weight, that's going to affect all of our reference points in our estimate of the spawning, the reproductive potential of the stock. So we're sort of putting people on notice that this stuff is really important and it could change our perception of stock status when we get the results.

Development of state-space models, again, here we're also putting people on notice. If as Mike mentioned, if we come up with some models that do a decent job of fitting the data and have very different structural assumptions, we could get fairly different results. And so if we add those to an ensemble, this could affect the results for management. And this is now a priority for the next full stock assessment. And I gather from some of the discussion last week, people are already wondering when and how big the change might be.

We were to add some new models then. So, you know, I think we need to sort of warn them that, yes, we're evaluating some new models and they could give us different answers. And we need to be prepared for that. Come up with a model, a good model that's ready for management. We could come up with something that affects the results.

And then the last one gets to this discussion about natural mortality in particular, leading parameter estimation. We've, as I kind of already went through, we're estimating natural mortality in 3 of these models, but not in the fourth. That assumption of fixed natural mortality for females of 0.15 is starting to get a little stale. We did a couple of years ago look at— we developed a prior on natural mortality based on the Pitaghi study centered around— forget it now if it was 0.1 or 0.2, and we looked at estimating it with that prior.

Included. We found for the, for the, for the 3 models where we're estimating it, we found that the prior didn't have much of an effect. And for the 4th model, we found that the prior didn't have enough effect. So, if we couldn't estimate it even prior, but again, that wasn't— that was in the Ken Lyons likelihood approach. So, we'll see, maybe we can get out of that box, state space, but We, again, highlighting for the Commission that if you change natural mortality, this has a huge effect.

It's a linear scalar on the size of your biomass.

Um, again, sorry for the formatting. We got some wrap PDF here. That's it for the research recommendations. Again, that's mainly for the Commission, but I just wanted to kind of show you the the feedback loop that we have. I think, uh, coming on that, we, we've talked, uh, at length about the, the value of freeing up your time, um, assessment.

So we've talked about so much that makes me think that, um, Commission really likes having an annual assessment because it just doesn't seem to get any traction. Is that—. Yeah, two, two, two things. Yes, I think the Commission really likes having— knowing that they're using the most current information. It's a hard thing to give.

I mean, it's pretty unique that they have real-time, a real-time assessment every year, and it's hard to give up, and I don't blame them. It's nice to know that your assessment reflects this year's information. It's as updated as we can make it. Um, and another issue we ran into is just when we were starting to really get down the trail of looking at a biennial or a triennial assessment cycle, We ran into 3 years in a row where we had to make reductions to the survey footprint. And so having an empirical rule to bridge us where we just use the survey trend to adjust the quota in between becomes less appealing if you don't really trust results as much as you did when you had a full survey.

So my crystal ball says that when we— next year and in the upcoming years, interest will increase again in, in having an empirical rule in between because our survey design will be back. Looks like it's going to be back up to full speed next year, right? And a couple years of that, I think it'll be back and we might be able to put it back on the table with an empirical rule. Our survey, when it's— when it's— when the design is full, we have the— what we really want for any design, it's as good as it gets. And we could easily use an empirical rule.

I try to tell people the assessment is just a smoother to take out, and we really have a great survey. It's just fitting the survey index anyway. So, we might get back there. That's why we keep it in here. So, we'll keep our powder dry for a couple years.

I think so. I'm recommending that before we bring it back up. I'm not giving up yet.

I just have a little bit more to go here. I've already kind of outlined this. This is just a laundry list of the information that will get updated. After this meeting, and it's the same things every year. It's just adding in the information from this year's survey, the information from this year's fishery logbooks, the information on the biological sampling from the commercial fishery done in— by our port samplers, including the sex ratio at age for the 2025 fishery, which lags a year, the biological information from bycatch and recreational sources, and then updating all the mortality estimates.

And as good data practice, we update not just the terminal year, but if there have been any adjustments to previous years, we rebuild the whole thing every year. So a lot of the work that, for example, this calendar year, Basia's team has done on cleaning up a lot of historical data, uh, richness is going to be, of course, They find— if they find missing logs from 5 years ago, those go into the database, they get included in the log.

So the recommendations are just to note this paper. Again, this was not as long a laundry list of requests as there might be in some years, and that this year is an update stock assessment for 2026. So we're not proposing any changes from the 2025 model structures, just adding the new data this year. And then any particular analysis you'd like to see next spring as part of the Plan 2027 update stock assessment, noting I've kind of already gone through what our R&D plan is between now and next June.

Thank you. Perfect. Perfect timing. Close to lunch. Yeah.

Great. And we have, we have a few minutes after lunch for additional questions, so we can do a few questions now while we're waiting. If we're not waiting, we can easily do them after lunch or whatever. We have at least another half an hour for questions. If you want to think about things over lunch and follow up, that's also good.

I think that sounds good unless anyone has, uh, Any questions right now?

All right, we're going to go ahead and break for lunch then, and—. All right, we'll come back at 1:30. Thank you.