[Automated voice] This conference will now be recorded.

[No audio]

[Deana Crouser] Yeah, go ahead.

[Will Fennie] So, thank you, Deana, for the

introduction. And thank you all for being here today.

I'm going to talk about some of my work, from my postdoc at the Southwest Fisheries Science Center

with the CalCOFI program. I'm looking at how oceanographic conditions shape

Rockfish survival in the California current. And then at the end of this I'm going to talk

about how I plan to translate this work, to walleye pollock, in the Gulf of Alaska

potentially. I want to acknowledge my co-authors; Noah Ben-Aderet, Garfield Kwan,

Jarred Santora, Isaac Schroeder, Steven Bograd, Andrew Thompson. And let's dive in.

In that spot. There it is. Ok.

So effective ecosystem based fisheries management relies in part upon our ability to understand

how oceanographic conditions influence the number of young fish that survive,

to recruit the adult population. And this can be particularly challenging to understand. Especially

with species that have long pelagic-larvae and juvenal stages, that suffer high mortality.

This is true of rockfish who are characterized by really

long life-stages or sorry, life-spans. And are characterized by episodic recruitment,

where they occasionally will have these really large recruitment events. But the norm is really

low recruitment. So there's a lot of work to try to understand how the oceanographic

conditions young fish experience through early life, influences their early survival.

And there's a lot of work to understand what drives mortality during these vulnerable early

life stages. And we've come, to consensus that predation due to or, mortality due to predation,

starvation, and advection are the main drivers of loss during these vulnerable

early life stages. And the critical period that determines your class strength can extend

from yolk-sac larval stage in some species through the juvenile stage in other species,

including walleye pollock. So there's a lot of work to try to understand what, improves survival.

And there's avenues of research that look at the traits of young fish, as well as the oceanographic

conditions they experience during development. And then finally whether or not the characteristics

of their parents influence the quality of larvae produced and their subsequent survival.

So when we look at traits of larval fish that improve survival,

it's generally thought that being bigger is better, because you can escape eight

limited predators. Growing faster improves survival because it allows you to swim faster,

better capture food, and avoid predation. And developing more rapidly increases your

survival because you spend less time with these high mortality windows.

And so together these are traits that are thought to really improve survival during

this vulnerable life stage. And concurrently there's a lot of work looking at how the oceanographic

conditions young fish experience during these vulnerable life stages, and can

contribute to their survival. And I was fortunate to be working in the California

current system during the marine heat wave when our understanding of how oceanographic

variability related to live and juvenile rockfish abundance completely collapsed.

We thought that there was a lot of, strong correlations between sea level pressure,

and the level of juvenile rockfish abundance, which correlated well with recruitment to the

fishery. And then we had this huge pulse of warm water that hit the system, similar to in Alaska.

And our relationships completely fell apart. And during this time Isaac Schroeder and his colleagues

looked at how deeper water masses, might be related to climbing juvenal rockfish abundance.

And they settle on two main water masses. It's the specifics of Arctic upper water mass,

which is highlighted with this blue arrow to the north. And this water mass is characterized by

being cold, high in oxygen, low in salinity and entrains lipid-rich large zooplankton.

It's thought that this led to improve feeding conditions for the young life stages of rockfish.

And led to really high abundance of [indistinct].
In contrast there's what's termed "spicy waters"

which are a combination of equatorial Pacific water, eastern North Pacific central

water that are warm, salty, low in oxygen and nutrients, all lipid poor zooplankton.

And when there's a greater influence of this water mass in the system,

it's thought that feeding conditions are poor. And few, fewer fish survive to the juvenile stage.

But these offers also thought that potentially it's the conditions that the females experience,

when they're in gestation that may have set up their larvae for success. So when there's more

of this PSUW water in the system, females may have better feeding or lower metabolic

cost and that led to improve quality of their larvae and higher survival.

And then finally there's a lot of work that's looked at traits, that adult fish possess,

that may influence the quality of larvae they produce and their subsequent success. Steven

Berkeley, from 2004 looked at how the age of female Black rockfish influenced the amount

of lipid [indistinct] they gave to their young and the effects that that had on their survival. So

what they showed is that older female rockfish provided higher lipid reserves that translated

into faster growth rates with their larvae and increased resistance to starvation.

So our goal here was to try and marry those three avenues of research together,

by looking at one trait of larval
fish. And that's their core width,

which is a proxy for their size at hatcher extrusion. We want to use that to look

for and figure out if core width is a good predictor of larval growth and survival. We

expected that larger core widths which indicate larger size and extrusion would lead to faster

growth rates and higher survival. We also want to look backwards and see if there are

oceanographic drivers of maternal investment for larval quality. So we expected that there'd be a

negative relationship between the amount of spicy water in the system and core size of these larvae.

And we expected positive relationships between the amount of PSUW in the system

as well as distance from port and core size. And distance from port we're using as a proxy

for fishing pressure. The further you are from port, the more likely there would be

older larger females out there, that might provide, produce higher quality larvae. So

to test these hypothesis we took advantage of CalCOFI's winter ichthyoplankton surveys.

And we collected genetically identified rockfish from 1998 to 2013. So we used

the eight most common species in total. I want to
say we pulled, I mean Noah Ben-Aderet pulled almost

1500 total out of these larvae and measured their core width; which is indicated in orange here.

We also measured the recent growth, which is the last two increments. And

the distance between the last two increments and then enumerated age to figure out how old

they were which we're using as a proxy for survival because mortality is really

high during these early life stages. So if you have higher average survival in a given year or

higher average age in a given year, we're assuming that indicates higher survival.

And then we align, oceanographic conditions, at the regional, the basic scale regional and then

station in that specific scale; and use partially squares regression analysis to look at how growth

and age were related to the oceanographic conditions that these larvae experienced.

We also included, you can see my pointer here, material provisioning indicator of their core

size. Used the winter oceanographic conditions and looked at North Pacific Gyre Oscillation,

Pacific Decadal Oscillation, Oceanic Niño Index at the basin scale.

Regionally we looked at meriodional winds and Biologically Effective Upwelling Transport Index.

And then at the station and depth-specific scale, we looked at the amount of PSUW in the system and

the spiciness of the system. And these were calculated for the upper water column where

larvae hang out in winter. We also measured port distance. When we wanted to look at how

the conditions females experienced during the fall influence the core size of larvae

that they produce, we look at fall oceanographic conditions at the same basin and regional scales.

But at the station and depth-specific scale, we use PSUW and spice at the depth range for

the adult rockfish for each species. And just to give you a quick idea of what that looks

like here's our different species of rockfish and the bars indicate their depth range. And

then the iso pigment layers are the horizontal lines. We use the ones that best align with

where the fish hangout. And before I jump into results, I want to forward you to my figures,

a little bit, so partially squares regression analysis is pretty similar to principal

component analysis. That it breaks down the variability in our data into two components.

The first captures the majority of the variability and it's loaded on the x-axis and variables that

contribute positively, to that first component will be in the positive x direction. Variables

that contribute negatively will be in the negative x direction. Variables that contribute

significantly to the variability of that first component will be highlighted in red. And ones

that are not significantly contributing are in blue. Then our variable of

interest will be highlighted in orange. Okay so let's dive into what the growth rate was also.

So here's our eight species of rockfish. We'll note that S. mystinus didn't have

very many fish that were older than 3 days old. So we couldn't really get an estimate

on their recent growth. For everything else, things look a little messy for the top species,

but as you get into the bottom you'll start to see that core size is overlapping with row three,

and is significantly correlated with that first component.

So what we found is that our first component explained 8 to 54% of the variability in recent

growth. And for five out of seven species core size was a significant contributor to

that first component indicating a positive relationship between core width and growth rate.

When we looked at age, which is our proxy for survival, these patterns become much more clear

as core size is almost overlapping to age in all of these species. We found that our

first component explained 15 to almost 60% of the variability in age. And that for all eight

species core width was a significant contributor to that first component,

indicating that you have a larger core size you're likely to, you're more likely to survive. And

finally when we looked at the fall oceanographic conditions that females were experiencing during

gestation, how that influenced the core size of the larvae they produce,

we found that our first component explained 10 to 46% of the variability in core size.

And we're a little shocked to see how well our hypotheses matched the data

here. Where we had a negative correlation between spiciness

and core width for five out of eight species. And positive relationships

with PSUW in distance for six and seven out of the eight species, respectively.

So in conclusion we found that core size actually does do a really good job of predicting the

success of larvae. Where larger core sizes lead to increased growth rate and higher survival.

And that when there are spicy water masses in the system or increased spiciness of the system

that leads to larger core sizes of larvae and when there's more PSUW or females are

found further from shore that leads to higher size of cores, for their larvae. We wanted to

dive a little bit deeper into that to figure out what could be explaining those patterns. So this

figure shows the average of the five highest years of PSUW across the system on the top.

And then the five lowest years on the bottom to give an idea of what the conditions were

that females were experiencing in this region during those types of conditions. So what I

want to draw your attention to is these cross shelf gradients in PSUW in this top left figure

here. So there's really low PSUW generally nearshore
and it increases as we move offshore.

It's been in contrast in the low years there's really low PSUW nearshore and

then that stays pretty low as you move offshore. And this leads to drastically

different conditions that females are experiencing during gestation. So when

you look at temperature you see the strong cross shelf gradient of warm water nearshore. Much

cooler water offshore. High PSUW years, there's less of a gradient during the slow PSUW years.

Similarly we see low oxygen nearshore and increases offshore. And there's less of a

gradient in these low PSUW. Finally when we look at nitrate we'll see that there's a lot

of nitrate nearshore and extending out into the offshore are these high PSUW years. And

there's less the gradient in these low PSUW years. So this shows that females

especially in these offshore regions are experiencing colder water and higher

oxygenated water. And we think this has important consequences for the larvae that they produce.

Specifically we think that because females are hanging out in this higher PSUW water,

which entrains lipid-rich zooplankton, that the food web is more nutritious during these years.

That leads to better feeding of the females so they can better provision their larvae.

Also because oceanographic conditions are colder and there's more oxygen in the system, we think

this reduces the metabolic cost of females. So they can provide more energy to reproduction,

to their larvae. And finally we think possibly that when there's more oxygen in the system

it increases vertical habitat available to females. They can find or access improved habitats

or improved areas of higher habitat quality that may improve their reproductive success.

So what we think we've done here is add a little bit of a piece to this puzzle,

figuring out how early mortality affects these life stages of fishes. And when oceanographic

conditions influence females, that affects the amount of [indistinct] nutrition they can

provide to their larvae. And that may have downstream effects to their susceptibility

to starvation. And so some next steps here, trying to figure out what the actual mechanism

is between what females experienced in the core size of their larvae, and what

that translates to in their larvae and then trying to link that to recruitment.

This is something we struggled with a little bit, in southern California because the sampling of

ichthyoplankton occurs in southern California, a lot of the pelagic juvenal rockfish sampling

occurs in central California. So there's a mismatch between where they're sampling. That

makes it challenging to align larval traits to the juvenile traits. And that's one of the

things I'm really excited about when I get to translate this work to the Gulf of Alaska,

with the EcoFOCI Program. So I'm really excited to link observed and modeled oceanographic data,

for the Gulf of Alaska to look at how, what the adults are experiencing translates

to our awesome time series of larval and walleye pollock in Alaska.

And seeing if that will allow us to better understand variability in the abundance

of age 0 and recruits. And some additional ideas I have are trying to drill down and figure

out what those mechanisms are for explaining why core size might vary from one year to the

next. And to do that I hope to collaborate with MACE to collect eggs from reproductive

females in their winter pre-spawning surveys; fertilize them at sea and look at a range

of ages and sizes of females and then compare the core sizes of the larvae that they produce.

And also I'm really excited about the opportunity to use age 0

pollock otoliths, which are the next life stage after the larvae

to look at whether or not their real life history traits align with larval traits,

or whether or not selection based on early size, growth rate or the timing of spawning

which Lauren Rogers shows is highly sensitive to environmental variability. So with that I'd

like to acknowledge the whole CalCOFI team, who I couldn't have done any of this without.

As well as the NOAA crew that go out and collect these samples. CalCOFI Surveys.

My funding sources. Emily and Deana, for running the seminar. And if you

have any ideas, or would like to collaborate, please contact me at will.fennie@noaa.gov.

And if you have any questions, I would love to answer them.

Thank You.

[Applause]

[Deana Crouser] All right! So yeah, look at, just look at the comments while I switch over

presentations. We'll keep an eye on the comment section and we'll

see if anyone's got comments online or anybody in the room.

[Will Fennie] Yeah, Lisa.

[Participant] I have a question. Did you look,

so you looked at nutrients. Did you look at phytoplankton? Like chlorophyll or any

satellite information, to kind of do that link between, you know...

[Will Fennie] I did not.

[Participant] Primary production? And then zooplankton.

[Will Fennie] Yeah. I'm trying to remember. So there was, there may have been some coreval data

from the CTD cast, and we initially looked at our CTD data and then we decided we want to

work specifically with PSUW [indistinct]
for some of that,

but it could be cool to go back
and try to keep that apart.

[Participant] Thanks.

[Will Fennie] We'll leave that for another postdoc. Maybe I'll keep that in my back pocket.

[Participant laughs]

[Deana Crouser] Ok. Yeah. Mostly the questions are about the sound quality

which I think is a little messed up. So we're just going to start projecting

and we're going to be really quiet in the room, so that way we can all hear properly.

[Will Fennie] Well, if there are no more questions

at the moment please email me or contact me directly.

[Participant] So this is, all right can you hear me,

this is Rich Bell online. I have a quick question.

[Deana Crouser] All right, so thank you so much. Well that was a great

presentation. Our next presenter is Al Hermann.

He is a CICOES and PMEL
affiliate. And he's going to

be talking about modeling in the past, present, and future. Thank you so much, Al.

[Albert J. Hermann] Thanks. Okay so I'll try to talk loud. Since that's been requested here.

So, yeah this will be somewhat of a whirlwind tour of some modeling

activities we've been engaged in over the past 30 plus years now.

Great! So, past, present and future. Some aspects of our biogeochemical modeling and

how it's being applied to fisheries oceanography.

So I'll start with some acknowledgments. Right up front, it's been a huge number of

people that we've collaborated with over the years. My apologies if your name is

not here and you think it should be. But, it is a very long list. Many,

many fine colleagues we we've collaborated with over the years.

So as far as the end users of this work, they include the use of our Bering Sea hindcasts,

for yearly status reports, and field planning. We try to update our hindcasts to the present,

twice a year and offer them to the to the public;

more on that later. Both the Bering Sea and Gulf of Alaska projections are used for a variety of

fisheries modeling of single and multispecies full food webs, as with the Atlantis model.

Projections of essential fish habitat and management strategy evaluation. But,

not to be left out at the bottom there, individual based modeling has been an

ongoing activity, uh, for a couple of decades now with individuals both here and elsewhere.

So way back in the 1990s we used a model out of the Rutgers Shop

called the Semispectral Primitive Equation Model to model the northern Gulf of Alaska, 
Shelikof Strait, in particular, and Phyllis already had some moorings in place that she

was maintaining for multiple years there. Made a great resource while comparing the model and with the observations. The observed currents and you see that on the right there, comparison of the two. We got a pretty good match and kept that work going. Ultimately we expanded into a different array of models and a different array of model forcings.

This is a master flowchart put together by colleague, Liz Dobbins who's now at UAF, way back in the 1990s, early aughts, regarding our use of global models as forcing to multiply nested Regional circulation models, in some cases driven by multiply nested atmospheric models as well. Those produce physical variables which then drive biogeochemical models, all of that fused with data wherever we could find it. More and more emerging data from the EcoFOCI program and remotely sensed assets.

And these days including saildrones and in addition as noted in previous graph we feed that into individual based models, as well as Eulerian biogeochemical models. Here's an example of an early coupling of circulation with NPZ, and individual based models.

Some work with Sarah Hinkley and collaborators and the individual based approach allowed us to get really detailed on pollock larvae, shown down there. The lower right hand corner, actually, nope I misspoke there this is not pollock but rather the zooplankton you're seeing there. This was just a snapshot of some individual based model output in the Gulf of Alaska. It's fun to watch movies of this in 2D. It's even more fun to render it in three dimensions if you had simple red blue 3D glasses you'd see this popping on the screen at you; we did several presentations that way. That was a lot of fun.

Expanding on that work we branched that into active stereo glasses, which are synchronized with an infrared emitter. And you display on a computer screen or a special silver backed projection screen. Now actually for this technology you don't need silver backed, that's a whole other realm. This technology allowed us to work in color and to easily use it for outreach. Here's a photo from our use of that technology at the Seattle Aquarium. To illustrate some points

regarding 3D circulation in the Gulf of Alaska and Bering Sea.

So over the years we've used a variety of grids and resolutions to address these issues in the Gulf and the Bering Sea and the Northeast Pacific, more generally. One of the main ones has been the one you see outlined in blue there. The Bering 10K model. We've also done a good deal of work with the Gulf of Alaska 3K model; 3 kilometer horizontal resolution. What that refers to 10K, 10k horizontal resolution. We have plans to refine down to 3 km resolution in the eastern Bering Sea as well. This figure from Kelly Kearney over in building four.

Basic features of some of our present models in the Northeast Pacific and the Gulf. We're using the regional ocean modeling system. And in the one-way nesting scheme we take boundary conditions for the 3K Model from the 10K model. We're down 42 for levels. We have explicit tidal dynamics, which create tidal mixing; very important for many parts of the shelf. We have an input of freshwater runoff and iron contained that there in the coast, very important for the Gulf of Alaska. Multiple plankton size classes and biogeochemistry and other details regarding forcing.

A fun way to visualize the output is by showing all the variables at once in a single animation frame by frame. I'm not actually going to show a movie today, but this is just to show you some of the fine detail you get down at three kilometer resolution. And one of the things you have to bear in mind when interpreting this and comparing with data is that the offshore eddies, will never be in exactly the right place. There's a whole lot of chaos in nature, you can never hope to capture it all with a model. Even a data assimilating model, which this is not.

This model runs freely, driven by observed, the observed atmosphere and the observed large scale ocean. Regional model itself, runs freely. But no matter how much data you assimilate some of the details will always be wrong.

Speaking of data though EcoFOCI has collected a massive store of data including tens of thousands of CTDs and model samples over the years. This is a treasure trove of information to compare a model output with,
calibration and validation purposes. So, not only those CTDs but also moorings which are shown in the large colored dots there. M2 and M8, in particular. You'll see, pretty soon in another slide.

We used assets like this to compare climatological model output. So here you're looking at October currents. At 40 meters depth from the model shown in black arrows. And a climatology of the, the mooring results, the moored results for that month; at 40 meters depth.

And some decent matchups, there even in some of the, the smaller canyons, which was nice to see. We can also use data like ocean station PAPA to calibrate and validate the bigger Northeast Pacific model. And so this is data line P. Ocean Station PAPA,

that is, end of line P. And you'll notice how things got warmer, in 2014; in the data down below there you'll notice the similar warming around 2014. In lot and you can see again things don't match perfectly because they never do. We're off by a degree or so at depth. But that's just life, in the modeling world. You're always left with some bias and you find ways to artfully deal with that, in real world applications.

Now moving to the Bering Sea, some of the unique aspects of this as probably everybody in this room knows, some of the people online maybe not so much, physically there's seasonal ice with formation in the north, with advection to the south. There is strong tidal mixing which sets up distinct biophysical regimes, in the inner, mid and outer shelf. There are plankton living in the ice and there's a strong benthic food chain.

So all of those things really need to be included in a decent biogeochemical model of that region. So, - oh fuzzy there -
I apologize. It got lost in translation. Well, you get the general idea though. Here's our Bering 10K model domain, spanning the entirety of the Bering Sea Shelf and Basin. And with the inner, middle, and outer shelf shown fuzzily labeled in orange, light green and darker green.

That model, that we're running includes ice, single layer of ice which is sufficient for the seasonal ice formation. And as with the Gulf of Alaska explicit tides. It's very important for setting up the hydrography, of the shelf.

So using some of those wonderful EcoFOCI assets here at M2 and M8, a comparison of the observed temperatures. Ones shown mark data there, at M8 on top. And M2 on the bottom. With their model equivalent. And you can see a very nice correspondence with warm years versus cold.

If you look really carefully, you'll notice a persistent bias that we have grappled with and not yet found a really good solution to; for the years that is our mixed layer is shallower than it needs to be. And there are various reasons that may be, not time to go into them now but, we'll talk about them later.

Despite those biases, despite the imperfections, we still do quite a good job of replicating the cold pool. That is related to temperatures at the bottom of the Bering Sea Shelf. In warm versus cold years, observation shown on top from bottom fall surveys conducted by our colleagues in building four, Alaska Fisheries Science Center. And, are results from, extracted from the Bering 10K Model on the bottom.

Sorry about the fuzziness again here. This is the general structure of the biogeochemical model as it exists now. In the Bering 10K Model this is a diagram from Kelly Kearney. To see a nicely resolved version go to her paper from 2020. But the general idea here is multiple nutrients including, from bottom to top; iron, ammonium, nitrate. Ammonium in the ice and nitrate in the ice. Multiple sized classes of phytoplankton shown in green. Multiple categories of zooplankton shown in blue. And detrital category shown in Brown.

Now we have recently, in the past year or two, begun publicly serving the hindcast output from the Bering 10K model. And also our projections of the future downscaling results from IPCC Models. They're available through a server at, here at PMEL, through OpenDAP Service. There's also a Live Access Server attached to that. So you can make interactive plots like this, through a web browser. Or through your phone, if you wish. Oh dear, yeah, I'm sorry the resolution did not come through here.

Too many transfers I guess between Google and Windows and Macs. All right, the general idea here though in words, is you take CMIP6 models or CMIP5 models, and we even started this process back in CMIP3.
Feed them through a regional model of the Bering Sea, using different emission scenarios and different global models. As many as you can afford. That gives you regional projections of the cold pool bottom temperatures of zooplankton. Any number of indices that are modeled by the regional software.

Most significantly this gives you projections of bottom temperature. Which are very important to fish in the Bering Sea. The pollock tend to like colder regime which leads to bigger, fatter zooplankton. So there's a parallel with Will's talk. And conversely warm years can be not so good. Not so many big fat zooplankton for the fish to eat. So that's all part of the ACLIM Program.

With many colleagues in this building and Fisheries folks next door. Wei's paper, summarizing some of our recent work shows the different projections for the three global downscaled models under two different emission scenarios; low emissions and high. And of course the higher emission scenario leads to warmer temperatures in the Bering Sea. Warmer bottom temperatures as we go into the future. You'll notice there's an offset in any of these models between what they say about the present. And what the survey indicates about the present. That's the bias issue alluded to earlier. There are various ways of addressing that when you're using these results for fisheries models.

One of the things we noticed with the hindcasting work over the years, was certain atmospheric forcings, were correlated and coherent at wideband of frequencies, with the oceanic response. This is just showing a strong relationship between air temperature and vertically integrated ocean temperatures, on the Bering Sea shelf. In a particular polygon on the outer shelf in this case. And that among other things prompted us to try a multivariate analysis

where we look at a wide range of variables from the model. From the regional model output. And a wide range of atmospheric forcing variables to see what things tend to rise and fall together.

So again parallels with sort of the stuff that Will was talking about with the principal component sort of analysis. Except here we're doing principal components not on the raw variables but rather the EOFs, the dominant spatial patterns of those variables. So EOFs first and then do principal components, on the resulting EOFs.

And when you do that, when you do indeed find multivariate modes, patterns of different things in different months rising and falling together; as you examine this year by year, there's a whole lot listed on the left there. One of the things that emerged from this and Wei deals with as well, in her paper, is a shift in phenology in the future.

That as you go into the future, the bloom of phytoplankton happens a bit earlier. And the blooms of things further down the food chain, also starts happening a bit earlier. The multivariate EOF analysis not only shows us what things rise and fall together, with what spatial pattern, it also provides a means to extrapolate these dynamically downscaled results to a much broader array of global forcings.

That is we can now take dozens of different IPCC Models apply them to this new set of statistical rules, and infer what the regional downscaling model would have obtained had we been able to afford dynamical downscaling for all of those cases. So applying that procedure you get the set of patterns shown on the right, there.

The inferred increase in bottom temperature under low emission scenario ssp126, on the left the high emission scenario ssp585 on the right. And that gives you some indication not only the overall magnitude but also the spatial pattern of the expected change. Same thing on the bottom for euphausiids. One of those nice fat zooplankton that fish tend to like. And how they are expected to decline under ssp585, on the outer shelf. If you want some of the gory details on that recent work there's Wei's paper on the left, and my paper on the right, in Deep Sea Research.

And finally some of the most recent work we're trying to expand into the realm of machine learning. And, sorry about that. It's cut off a little bit. What it basically says, is machine learning LSTM, that is Long Short-Term Memory method, is used to replicate the behavior of the downscaling model. So we are using machine learning to relate monthly anomalies of the atmospheric forcing to monthly anomalies of the regional response. And the takeaway from the curves there is the blue is the actual downscaled result, from a particular global model.

The red is the LSTM replicant of that result trained on an entirely different set of models. So as in machine learning you train it with one set and then apply it to something completely independent for what it was trained on. And we're actually able to do quite well. So this, this seems promising and we're intending to pursue it further, some of that NOAA support we obtained this year.

So to summarize, and speak to the future just a bit, we've obtained over the years many useful hindcasts and projections for the Gulf and the Bering Sea. There's always room for improvement.

There's always a better model out there, that you could be using. The details will always differ from reality but you can produce plenty that's useful to end users along the way.

We have public access available for some of this output, not all of it. But a significant chunk is now on the public servers here at PMEL. And we are in the process of transitioning from our ROMS based set of models to MOM6, under a new NOAA initiative called Climate Ecosystems and Fisheries, CEFI,

which is just now spinning up. It's intended to operationalize some of this work which has been done in dozens of different soft money projects over the years. Kept it going. And hopefully the future will hold hardline NOAA support for this work. And finally we've done some work lately with nascent use of data assimilation to improve our hindcasts.

And that will be it. Thanks!

[Applause]

[Deana Crouser] All right. Let's see.

All right. Well, we did great on timing there. Let's see if we have any questions?

Let's see. All right we've got one question for Al.

What is the time scale for the changes you show in the ssp126 and the ssp585 scenario?

[Albert J. Hermann] That was basically comparing this decade versus the end of the century. So that's yeah, that's up to the end of the century versus the present.

[Deana Crouser] Nice!

Awesome. All right. Looks like that's all the questions online. We have anything in person? Any questions?

I have a question for Will.

Will, can you talk a little bit more about spicy water?

I think I missed what spicy water was.

[Will Fennie] So it's a term for these two water masses that tend to come from kind of the equatorial Pacific or the eastern North Pacific water.

And they termed them spicy because they're warm, salty and low in oxygen.

[Deana Crouser] Oh, all right. That's...

[Will Fennie] As opposed to "minty". Which are cool, fresher and iron oxygenized.

[Deana Crouser] I see, minty, I like that. [Laughs]
I haven't heard that before. Thank you.

Any other questions?

[Participant] Yeah, just add to Will's comment, usually those water mass have the same density, but their temperature and salinity configuration is different. Even though they do tend to have the same density but you know one is fresher, one is saltier, it's temperature salinity compensated changes. 
Sorry might be a little bit more.

[Deana Crouser] No, that's perfect.
[Will Fennie] [overlapping voices] Thank you.

[Participant] Sure.

[Deana Crouser] All right. Well, we don't have any more questions. It looks like we're right on time to wrap it up.

Thank you so much for participating. Thank you to our speakers Will and Al. It was a great chat. And yeah, we'll be here next Wednesday!

[Applause]