 Charlotte DeMott is a senior research scientist in the Department of Atmospheric Science at Colorado State University and she will be talking about diagnosing sources of tropical SST drift in coupled forecast models. Charlotte, we're looking forward to your talk. All right, thank you Judith and thank you Anish. So I think we've pretty much covered my first slide of the presentation. So I will go on ahead and move on and I'll, if I'm rattled today, this talk is being given during the intermission of our NSF-RU student symposium today. So I've been listening to talks, I'll give a talk then I'll go listen to some more. So anyway, yeah the topic of my talk is really focused on SST drift in coupled forecast models. And so it's just a way to motivate this. We can just ask the question, why consider the ocean in terms of its contributions to S2S forecast skill? And this is just a very simple picture that I've shown in other talks that talks about how SST variability is communicated to the atmosphere via surface flexes. And the point of this figure is to illustrate that really a variety of processes can contribute to surface flexes, processes in the atmosphere associated with cloudiness and processes in the ocean that can influence the sea surface temperature. And so there's a couple of questions that you can ask in terms of S2S forecasting. And one is what ocean processes affect the atmosphere via the flexes? And another reason that you might care about ocean processes on S2S time scales is that ocean weather in its own right can be important for biological species, commercial activity, etc. So there's sort of two components, the ocean can affect the atmosphere. And as we've heard from previous talks, the tropical convection then regulates whether globally via teleconnections, but also ocean weather can have its own impacts without really even considering its feedback to the ocean. So this is a figure from the citation shown here. It's just a very nice illustration of some ocean processes that can affect the air sea exchange. And so in most of my talk, I'll be considering the region of the Indo-Pacific warm pool. This type of figure can be considered anywhere on the globe. But if you consider it in terms of the warm pool, you have the upper ocean mix layer. Some of the processes that can affect the sea surface temperature is vertical mixing across the mix layer base, either by Kelvin-Helmholtz instability, mix layer turbulence, entrainment at the mix layer base and heat flexes from the mix layer to the deeper ocean. There are also circulations that take place within the mix layer. So these are ocean processes. The atmosphere then of course interacts with the ocean. So for example, cloudiness regulates the amount of solar heating that impacts the upper ocean. Winds contribute to mixing, which can extend down throughout the depth of the mix layer. And then all of these processes together lead to sea surface temperature variability, which is then communicated back to the atmosphere through surface flexes. So how might ocean—I'm sorry, this is a poor title—how might ocean variability affect S2S forecasts? So there's a couple ways to think about this. Two of them are really related to the coupled model itself. Two of them are more of natural variety. So you can imagine if you have a coupled forecast model, both the ocean and the atmosphere have to be initialized. So how well both of these components are initialized may affect the coupled feedbacks that take place during the model integration. There's also this tendency for coupled forecast models to want to drift to their preferred climatological state. And I'll show an example of this in a couple slides. So those are model specific considerations. But there are also more fundamental issues, such as the nature of ocean coupled feedbacks themselves. And so when you initialize a coupled forecast model, some things that might matter for how your forecast proceeds are what is the initial state of the ocean? Are you initializing your model over a region of warm SST anomalies, cold SST anomalies, how might those things evolve over time? And then also how might the upper ocean itself evolve during your forecast integration by some of those processes that I showed in the previous schematic illustration? So my focus today is really going to be on some of the model aspects. And so I will also be focusing how some of these model initialization and drift aspects will affect the MGO, since the MGO is a good thing to talk about in a workshop devoted to S2S processes. So as we all well know by now, the MGO is a source of S2S predictability. And it has been shown in many studies that including ocean coupling generally improves MGO forecast skill. So this slide is intended to give an overview of sort of the two main aspects of model, a couple of model forecasts that I want to consider here in the context of the MGO. And so the main point to note is that we know that the mean state moisture and anomalous winds are really key to the MGO eastward propagation. And so what we're seeing in this plot is the November through April mean state SST given in the shading. And then the mean state total column water vapor or total column water shown with contours. And you can just look at this with your eye and see that the two are closely related. And indeed the spatial correlation of these two variables over the domain shown is about 0.85. So for the MGO, I'm sorry, so we have a high correlation between these two variables. And furthermore, SST anomalies about this mean state help regulate the boundary layer energy, the buoyancy, this can affect convection, and also wind anomalies that would be part of the MGO. So just to illustrate this visually, if you have a MGO convection here in the eastern Indian Ocean, that is associated with low level wind circulations as shown with these blue ellipses. So you have cyclonic flow in the northern and southern hemisphere to the west of the convection and then you have a more broad region of anti-cyclonic flow to the east. And really it's where these two circulations meet up and you have regions of strong poleward flow. This strong poleward flow affects the high mean state moisture near the equator poleward. And this is a primary mechanism that contributes to the eastward propagation of the MGO. So the key point is that it's really the mean state moisture, which is strongly tied to the mean state SST that interacts with anomalous winds that we really want to understand for MGO propagation. So moving on to some forecast model results, we can think of two ways that the SST forecast skill might affect MGO propagation. The first is the sea surface temperature initialization. If you have a good initialization, you might think that your sea surface temperatures have a reasonable interaction with convection in its development and subsequent wind anomalies. But you can also think about this SST drift. And this is the mean state SST drift that will affect the mean state moisture, which is the other component that affects MGO propagation. So what I'm showing here is MGO forecast skill as measured by the biobariet correlation coefficient for it looks like nine different models in the S2S database, but it's really five. The top row, these are three different versions of the Bureau of Meteorology model. These are three separate models, and these are three different versions of the ECMWF model in the bottom row. And so what I'm showing here are we compare the SST anomaly patterns that the model is initialized with to those from observations. And we compute the spatial pattern correlation skill and find the top 25 and bottom 25 most skillful initialized states. Okay, so this says when you have a model that's initialized with sea surface temperature anomalies that agree very well with observations, this is the top 25, less good agreement at the bottom 25. And what you can see is that in some models, there's really virtually no difference in MGO prediction skill, according to whether you have a good ocean initial state or poor ocean initial state. For some models, you do see a larger difference, but I will point out these differences are not statistically significant. If I were to draw error bars on these curves, they're just huge. Okay, so the SST initialized state, at least the anomaly does not seem to be the leading order cause of MGO prediction skill. So now I want to go back and turn my attention to the SST drift, because the hypothesis is that as goes the SST drift, so goes the mean state moisture drift. And what I'm showing here is for the same nine model versions is the 30 day SST drift across the Indo-Pacific warm pool. And so you can see that some models drift colder, some are fairly neutral, this version of the model is actually flux corrected, so we expect it to have very little drift. And then some models drift actually positive SST in certain locations. Now just to show you to give you a better idea of what the strip looks like, if we look at the SSTs averaged in this box here for this model, what we're showing in the lower left panel, this is the lead dependent SST climatology. So the black line is the true climatology, and then as we go from blue to red lines, this goes from a lead day one to lead day 30. So you can see during the first half of the year, the model drifts warm, during the latter part of the year, it tends to drift cold. So you can, this is the same picture, I've just subtracted the black line from each of the colored lines here. And you can see that at times the lead dependent climatology is nearly one degree C. So over just 30 days, your climatological average SST departs a full degree Celsius from climatology. So that's pretty significant. Now if we want to understand what is causing this source of drift, we're just going to consider this example from the ECMWF. And to do this, we're going to look at the SST tendency, which here is the model drift. And you can consider that there are two basic processes that contribute to this. One would be drift in the net surface heating, which would directly contribute to SST tendency. And the other is drift in ocean processes. And the way we're going to analyze this, I'm just going to do this because my keynote to PowerPoint translation was a little bit odd. There's a few steps. So I'm going to try to break it down. The first thing we're going to do is compute the SST lead dependent climatology. So that's what's shown here. This is from zero to 30 days. You can see that for this region in the western Pacific, the SST increases by about a third degree Kelvin. From this, you can compute the SST tendency, which is shown here in the blue curve in the second panel. Then we're also looking at the lead dependent drift in the net surface energy flux. So heat flux into the upper ocean. You can see that these two curves tend to follow a similar trajectory. And in fact, if you regress the tendency onto the net surface, you can then predict the amount of SST tendency that is directly correlated to the net surface heating. And that's what you get in this curve shown in orange. So the surface flux predicted SST tendency really accounts for most of the variability in the blue curve. Everything else, the residual, we attribute to ocean processes. Most forecast models really only provide you with SST. So all of the ocean processes have to be inferred unless you're lucky enough to have ocean output from your forecast model. And then finally, we can say, okay, it's the net surface flux that's really contributing to most of the drift. If you like, and I won't talk about this much further, you can then drill down and assess which of the four terms most contribute to the flux drift. And for this model, the net surface shortwave and latent heat fluxes roughly equally contribute. In general, the long wave and sensible heat flux terms are small. They usually offset each other too. Okay, so I want to go back here. This is for one region of the ocean. And for this point, we can see that by and large, it is the surface flux drift that is responsible for the SST drift in this region. But you can say, well, what's going on here? So if we repeat this exercise for every single grid point in this domain, we can quantify how much of the SST drift is driven by surface fluxes, how much is driven by ocean processes. So this is what I'm showing here for models. You have two minutes to wrap up. Okay, perfect. So the way this is plotted, I'm not going to show the details. It's a nice exercise by a paper by Daria Halkines in 2015. It's a balance factor. Essentially, the more blue shading you see, the stronger the drift is controlled by the ocean, the more orange or red, more strongly the drift is controlled by surface fluxes. And the real take home message here is that somewhat surprisingly to me, there's a lot of regions where SST drift is not really well correlated with net surface energy flux drift. There's a lot of variability from one model to the next, one region to the next. But it's very hard to make any broad statement about why do models experience SST drift. And so I'm going to skip my next couple slides, but I just want to come back to this point about MJO propagation. We're looking at drift and usually when we evaluate MJO forecast skill, we look at lead depend, we subtract out the lead dependent climatology, right? But the point in looking at the drift here is that as the model is integrated, the model equations itself are operating on the moisture that is there, the SST that's there, the moisture that's there. And especially for the MJO, these wind anomalies are acting on the mean state moisture. So I think it's kind of impossible to completely remove the effects of SST drift on this MJO propagation problem. So I think I will just close with the plug for the S2S database, which does include some ocean output variables that could help better understand this blue term for various models, for example. And I will close right there and take some questions. Thank you very much. And sorry, I did not mean to rush you. No, that's quite all right. I really very much liked your talk and the points you're making. Arun has a question. Arun, go ahead and unmute yourself. Thanks, Sharla. It's a nice great talk. Can you summarize again, how did you quantify the quality of SST initialization? Yes, okay. So you are talking about this plot? No, it was at the very beginning when you said some SST initializations were bad and some were better. Oh, yes. Okay. Right. I will try to find the slide. Yeah, so we was just a couple of slides down. Okay, we compared the SST anomalies in the model initial state to the OISST anomalies. So there are other ways you could do this, you could use like a more other datasets, but it was sort of just a simple assessment. We compared the SST anomalies present in the model initial state to those that you would see in the observational record. It's just like a pattern correlation across the same domain that I showed in most of these plots. All right. I would assume that correlations differences are pretty small in grid and band. They generally are and although larger differences from some models than others, and I confess I don't have a slide that shows this nor do I recall the exact details, but I can look that up and maybe let you know. Okay. Thank you. Hemi, you were next. Thanks for the nice talk. Charlotte, you showed at the beginning that the SST is closely related with the moisture distribution and this is the key. So how does your SST bias link with this process? That's right. So I actually did not even touch on that. You know, some of your work has shown you can have initial state biases in column water vapor. And you, yeah, so you could also have initial biases in SST. So is the SST and the moisture bias similar? Did you find a similar pattern? I have looked at that. Yeah. So in, yeah, I've actually looked at these, the correlation, the temporal correlation point by point in the SST drift and the column water vapor drift is very high in certain locations. And some reasons it's not. And so one of the slides I skipped over was like this Colne rule. If you try to narrow this down to regions where the ocean appears to be driving the atmospheric variability versus vice versa. Yeah, the drift is actually really high. And I think that's because convection operates very quickly. When there's a drift in SST, the convection responds very quickly. And whatever convection is going to do to moisten the troposphere, that process is very fast. And I think that's probably why the two are so strongly correlated in regions where the SST drift could influence the atmosphere drift. Okay. Thank you. Thank you. Magdalena. Yeah. Hello, Charlotte. Thank you very much for the talk. I think it's very important work. I mean, very relevant for model development, this piece of work. I had a question regarding with the surface flaxes. There is quite a lot of uncertainty on surface flaxes. Do you know or do you have an idea whether the uncertainty in flaxes could affect your estimation of what are the relevant processes? That's a great question. And certainly they can. I've been doing a little bit of work with a surface flux intercomparison project that by Caroline Reynolds is helping organize. And so really, I mean, even just which bulk surface flux algorithm is used in a model can really have a pretty substantial impact on how much energy is transferred from ocean to atmosphere at various phases of the MGO. And we're kind of finding through some other work using the DOE E3SM model is that these algorithm-specific differences in the surface flaxes are not uniformly distributed throughout the MGO phase. So it's really interesting. So it's not just that the flaxes are either uniformly too high or too low. It appears if you change one bulk flux algorithm to another, you will see, for example, a greater surface flux from the ocean to atmosphere when you have strong convection and a more reduced surface flux when the convection is suppressed. So it could widen the distribution of flaxes. But yeah, that's something we're sort of just getting into now. And it's really interesting to think about how, two things, I mean, how the flaxes can affect the convection, but also how those flaxes might affect the patterns of SST drift and SST anomalies in these simulations. Thank you. In the interest of time, I think we should go to the next talk. Jacqueline has typed her question in the chat, so maybe you could respond to it. Great. So thank you so much, Charlotte, for finding the time between other commitments to give us this really interesting talk about such an important topic. So thank you so much.