 It's my pleasure to welcome our next keynote speaker, Marika Holland. Marika is a senior scientist and the section head for the paleo and polar climate research section of Climate and Global Dynamics Lab at NCUR. Thanks again, Marika, for accepting our invite and look forward to your talk. Thank you. Yeah, thanks a lot for the opportunity to speak here. So I'm going to share my screen. So hopefully you can. Yeah, it's nice. Okay, perfect. Okay. Great. Yeah, so I'm going to talk with you today about sea ice predictability so I'm a sea ice geophysicist and a climate scientist and so I normally often think about things on very long time scales kind of century time scales. But more and more with the changes happening in the Arctic. I think a lot of people in my field have started thinking about sea ice predictability predictions on shorter timescale so I will be talking about initial value predictability. It's sort of seasonal and maybe a little bit longer time scale. Probably not sub seasonal, I won't touch on that too much. But I'm also going to bring in a climate angle to this, and talk a little bit about how we expect these initial value predictability characteristics of sea ice to potentially change in the warming climate. And I'm sorry, hopefully that someone just started mowing the lawn next door, hopefully that noise isn't filtering through too much but what can you do. Okay, so that's kind of the context for my talk and I think a lot of the interest in seasonal predictability and the Arctic has grown in the last decade or so because of the long term changes happening in the sea ice cover. So this shows the time series of September sea ice in the Arctic from 1979 here to 2020 and you can see over this time period over the modern satellite era we've had this traumatic loss of Arctic sea ice. You know, values around 7 million square kilometers or more in the 1980s. This year we were less than 4 million square kilometers for the month of September in the Arctic, so the change is rapid, it's dramatic. It's affecting the entire Arctic system. On top of this of course you can see that there's also this very large year to year variations and sea ice and so I've just pulled out two years here which are kind of the extreme year to year variations you might expect but September 2012 here which was our record and 2013 so the next year. And you can see that you know the difference in sea ice here is dramatic. And if you're thinking about marine access in the Arctic. You know this 2012 conditions had extensive open waters north of Alaska and bearing straight, whereas in 2013, there was much more ice cover there so this has real implications for people accessing the Arctic. And there has been a lot of changes. In terms of things like shipping in the Arctic. So, this just shows some some ships within the Arctic this one's actually named Maria, which was great I could find a vessel and that has my name. But these ships are actually going through the Arctic they're going through the Arctic more frequently. And so there is a real need for these kinds of predictions on seasonal time scales for these marine access considerations. There's also a need from them in the context of indigenous hunters. And so, for example, there's a sea ice for walrus outlook, which is basically to support indigenous hunters in the region her hunting walrus that pull out onto the sea ice so they're happening, they're happening rapidly and it's brought this sort of new interest in being able to predict sea ice on these seasonal time scales to have safe marine access in the region. So in the context of that, there's been a great deal of research in the last decade or so on the topic of predictability of sea ice on seasonal time scales. And I'm going to give you a little bit of background on that. And some of the major things that have been found. And I'm going to put that in the context of this autocorrelation plot that was published by Ed Blanchard Ricklesworth back in 2011 so about a decade ago. This shows just the lagged autocorrelation of the northern hemisphere ice area so here you can see, you know, January at the top correlated with January so the but the values on the diagonal here are going to be one. And then in following months just to get a sense of how persistent sea ice anomalies might be etc. So what you can see from this is that there is a persistence timescale in the sea ice area of several months. So, no matter what month you look at here, you know, it's, it's highly correlated for the next two to three months so that suggests some predictability in the sea ice on that kind of timescale. Additionally, there's an interesting structure here that there are anomalies that reemerge in the sea ice cover. And so, in order to explain that I'm going to put it in the context of the ice area annual cycle which you can see, this is just a climological annual cycle that shown for two years to illustrate why we get this reemergence of anomalies. So you can see that here I'm highlighting it in the circle here where some of this reemergence occurs. So say may see ice anomalies. You get very little correlations kind of, you know, October time care but then those correlations increase and become significant again. So what's happening here is during this sort of ice retreat season that these months represent, you can see on the ice area annual cycle the sea ice is undergoing a very dramatic reduction so it's melting back into the Arctic basin. And as it does that it can leave ocean heat content anomalies behind. And rapidly the sea ice melts back can influence things like how much solar absorption there is in the ocean, which can lead to ocean heat content anomalies that then the sea ice doesn't feel for several months because it's melting back into the Arctic basin. But when the ice regrows in the fall and winter it can reencounter those ocean heat content anomalies and give a predictable signal. So that leads to this reemergence of predictability at this time of year. We also see a reemergence of these anomalies from one summer to the next often. And the mechanism for that is actually quite different. So here again it's being illustrated on the plot of the annual cycle. The predictability mechanism that drives that kind of summer to summer predictability is associated with ice thickness anomalies. So ice thickness anomalies can arise in the fall freeze up time period depending on atmospheric conditions etc. And those ice thickness anomalies are long lived ice thickness anomalies then can influence the melt out the next summer. This reemergence of those anomalies from one summer to the next associated with those long lived ice thickness anomalies. So there's different aspects of predictability on these different timescales depending on the seasonality, but these are some of the kind of major background of what we think is going on with sea ice predictability. I'm showing this from this one study, which was one of the first studies to sort of document this but this has been found in numerous other studies so it's not just based on these auto correlations based on a lot of other evidence as well. Okay, so there's a lot of evidence that we do have predictability and sea ice on seasonal out to perhaps inter annual timescales. And that that is an initial value predictability aspect of the system. But of course, you know, those, the system is chaotic and the degradation and that predictability is associated with the sensitive dependence on those initial conditions. And I think this, this plot that's actually from a paper break brand stutter and tang nicely shows though that there might be a different predictability characteristics depending on where you start what your, your climate conditions are say. So for example here on the left you have, you know, a cloud of initial conditions say, and over time those don't diverge a great deal, suggesting that there's predictability out for quite a long period of time. Whereas the one on the right here you can see that those initial conditions that error in them grows dramatically, such that over the same period of time you might have a much less predictable system. So, these kinds of studies suggest that the predictability characteristics can be dependent on the initial conditions and of course that's been shown for a number of aspects of the system. And because the climate in the Arctic is changing so dramatically it begs the question as to whether cold thick ice covered Arctic might have different predictability characteristics than a warm, thin ice covered Arctic. So I'm going to explore this question a little bit. And there's reason to believe for the sea ice that this could actually be important because sea ice processes are actually climate state dependent. And some of the sea ice processes that we know are important for this initial value predictability are climate state dependent. So this is ice growth. So, thin ice with less snow will grow more rapidly, subject to the same portion. So, if you look at the schematic on the left of the floating ice covering the Arctic, you know the ice is made up of regions of different thicknesses, there can be open water areas, there's snow on top of that sea ice and all of that sea ice is in motion. The ice grows primarily at the base of the ice pack, and that growth is associated with the conduction of heat through the sea ice and snow, and that heat then being lost to the atmosphere. So basically how much ice growth you have is associated with that conduction, and that is very strongly related to the thickness of the sea ice. If you look at, if you look at just a plot of ice growth versus ice thickness, you see that thinner ice will grow more rapidly. This relationship is also nonlinear device thickness again actually goes as one over the thickness because of that conduction relationship. So what this means is that this is actually a negative feedback on ice thickness. So if the ice spins, it'll grow a little more and counteract that thing. And that is actually a feedback that is stronger in a thinner ice regime. So I'll come back to this when I'm talking about the predictability results. There are other aspects of sea ice processes that are climate state dependent and in particular the open water formation that happens during melting in the summer. So again, if you go back to this schematic on the left and you think about melting the sea ice vertically, which is how it primarily melts. The areas of open water are going to preferentially occur in these thin ice regions because they can completely melt out. That thing can have a big impact on the surface albedo feedback and how much shortwave you can absorb in the ocean. So we do see that the amount of open water formed per centimeter of ice melt is more effective in a thin ice regime. And again that there's a nonlinear function of that open water formation efficiency with ice thickness. So we have these processes in the sea ice system that are climate state dependent. We have a rapidly thinning and declining sea ice cover. So it begs the question of how Arctic sea ice predictability characteristics might change in a warming climate. Okay, so we set out to investigate this by using the community or system model as a tool and performing sets of perfect model predictability experiments. So basically what we did is that for each decade from 1980 to 2030, we initialized ensemble predictions on January 1 in which we took restart conditions from the CSM one large ensembles we had a perfect knowledge of our initial state from the model. We applied a very small perturbation to the round off level perturbation in the air temperature, and then we ran one year predictions. For each climate state each of those decades we actually picked four initial states that sort of sampled across the range that we would get from the CSM one large ensemble. And we performed 15 ensemble members for each prediction state. So I'm showing an example of that here for the Arctic Ocean ice volume predictions, just to illustrate the method so the dash lines are the CSM one large ensemble range in this property this is taken from 40 ensemble members of a free running climate model. So the spread across them is really a measure of internal variability in the CSM simulated climate. And then I've picked four different initial states from which to run my predictions, and I've run 15 ensemble members each and you can see that after 20 or 30 days those ensemble predictions start to diverge. And that divergence becomes actually considerably larger when they reach the melt season. Okay, so this is shown for Arctic ice volume. And we can quantify some metrics of predictability and here I'm just using something very simple I'm just looking at the spread across my prediction ensemble, which is shown the average of which is shown in blue the individual ensembles are shown in great. And then the internal variability. And so if this prediction ensemble is much smaller than what we expect from internal variability. It suggests that there's a predictable signal that the initial values are giving us some predictability of the sea ice volume. And again I'm running these for a year so you can see that how that variance grows in our prediction ensemble throughout the year. And you can see it's always much less than internal variability suggesting that for sea ice volume or CS thickness. There's a very highly predictable signal for an entire year. This is seen in many other studies as well. And I'm just showing this for the 1980 ensemble set to illustrate the, the design of our study. Okay, so what about ice area, because there's a lot more interest from stakeholders in ice area than an ice volume because it's much more relevant to marine access considerations. I'm showing basically the same plot of ice area variant or variants and the prediction ensemble and blue, the internal variability and red, and I'm just showing this for the summer. All of these were initialized on January 1 so we're looking out, you know, six to eight months here, six to nine months and now to December actually. And you can see that this actually looks really different than the ice volume. The prediction ensemble variants across the ensemble members is indistinguishable from internal variability. And it really suggests that for 1980 at least, there's no predictability for summertime sea ice area, based on initialized forecast in January, but we see something quite different as we move into the future with a thinner and warmer sea ice cover. We actually see that we start to have a predictable signal in the Arctic sea ice area. And the initialization prediction spread from our initial prediction ensemble is considerably less than what we get just from internal variability. And this is particularly true at 2010 so I'm showing this here just in the simple kind of variance metrics but we see this also if we look at something like the anomaly correlations as well. So, it looks like as we move into a warmer, thinner Arctic system, we have enhanced predictability. And this is true when we look at these atmospheric metrics. It's also true if we look at maps of sea ice concentration and the Arctic. This is just shown where this prediction skill here would be higher in the darker blue colors. And again, when you look kind of across the Arctic. It looks like there's enhanced predictability and that predictability is particularly strong in this 2010 time period. Okay, so why is this the case. We know based on these previous studies that ice thickness is a very important predictor of summer ice area so we can use that to think about what causes a loss of predictability in the sea ice system. So we know, I sickness is important because those thickness anomalies are long lived and thickness anomalies at the beginning of the summer can affect how much melt out we get over the summer. So that we can relate to that then to the loss of predictability by looking at these two different factors. What's the growth of our ice thickness anomalies from their initialized state. So this would be an ice thickness error growth metric. And basically, this is just shown here from one ensemble set. I'm not showing it in terms of ice volume but that's just the integrated metric of ice thickness, and you can see that over time those ice thickness. The simulation start to diverge in their ice thickness such that by early summer. When they can affect melt out you have a considerable spread across them. So we can just quantify this by looking at the spread across our ensemble members. The aspect of this loss of predictability is how these ice thickness anomalies at the beginning of summer affect the melt out over the summer and ultimately affect the September sea ice area. And we can quantify that by just looking at the regression of September ice area on the July ice thickness. And that's what's shown here, and you can see it's quite linear so we are just going to use regression metric for that to look at the melt out sensitivity to these thickness errors. Okay, so what does this look like when we look across our simulations and the changes over time. So here we have our January initialized conditions where the error in ice thickness quantified as the spread across our prediction ensembles is very small, and then you can see it grows into the months into the future. You also see that that ice thickness our growth is smaller in the warmer climate. And that's because of this ice thickness ice growth rate relationship, which is climate state dependent, which means that anomalies in thin ice are more strongly damped. So it's related to this ice growth ice thickness relationship and the fact that that is nonlinear. In terms of this melt out sensitivity to ice thickness errors. We see something actually quite different we see that that's actually larger in the warmer climate. So again I'm just computing this based on a regression of September ice area on the previous July ice thickness anomalies. So you can see that from 1980 here that value is quite low, but out in the warmer climate of the 2020s and 2030s that value is quite large. So that melt out sensitivity to ice thickness grows in a warming climate. This is consistent with there being a lot of thin ice that can depending on the weather conditions in the summer can easily melt out or if it's a little colder can actually stick around. For the summer months. And that's related to this influence of ice thickness on the open water formation efficiency associated with summer ice melt. So we have these two competing aspects that affect the predictability in the system. We have this growth of the ice thickness anomalies from their initialized state that's smaller in the warmer climate. And we have the fact that those ice thickness anomalies can affect summer melt out, which is larger in the warmer climate. So these two things compete. And it actually means that there's this kind of sweet spot for predictability that's in these moderate conditions, which in our climate model simulations are for our 2010 ensemble state. And so in this case the September sea ice area predictability is highest because they have not those simulations have a modest ice thickener thickness error growth. And those ice thickness errors at the beginning of the melt season, have a modest influence on the September sea ice area. So that's why we have the largest September sea ice predictability for that, that state of the climate. Okay, so just to conclude, and Arctic sea ice area has predictability on seasonal and longer time scales. And there are different mechanisms that drive that depending on whether you're looking at summer or winter sea ice conditions. So with the summer ice area predictability. It's related to long lived ice thickness anomalies that can affect melt out. And both the longevity of those anomalies, those ice thickness anomalies and how effectively they impact melt out does depend on the sea ice state so because of that we expect predictability characteristics of sea ice to change in a warming climate. And I touched on this here but I think it's been mentioned. Another talks and is also a very active area of research, but these predictable signals in the sea ice can also impart predictability to the atmosphere, and to the ocean, and to the marine biogeochemistry. These are really interesting topics that, you know, are worth looking into and hold some promise for increasing our ability to predict on the sort of seasonal time scales, not just for the sea ice but for other aspects of the system as well. And so with that, I'm happy to take any questions if there are some and I have to throw a polar bear in because I took this photo, and I'm an Arctic scientist. Anyway, thanks for your attention and I'm happy to take questions if there are some awesome. Thanks a lot for great talk on the Arctic. I see Judith, you have your hand raised. Yeah, thank you very much. I feel the ice component is something that came a little bit too short in this workshop, but there will be future workshops to address that more. My question is on the uncertainty associated with unrepresented subgrid scale processes and subgrid scale heterogeneity and I was wondering if you could comment on that. And if you think it plays a role for these initialized forecast you have been doing. I think it does. So, so there is some evidence. That our climate models, these kind of perfect model experiments might be too predictable. And I know this is quite different from other aspects of the climate system like the NIO where it looks like our models might not be predictable enough. But there's some evidence that sea ice is too predictable. And I actually think that there may be some issues with how we represent subgrid scale processes in the model, things like wave sea ice actions which can add noise and you know ruin our predictable signals. Things like subgrid scale heterogeneity of snow on sea ice which we don't represent well at all, but which is really important for things like that ice growth ice thickness relationship. So it's another source of noise in the system that I actually don't think we capture well. And that's true for the Albedo to so all these things that can impact these predictable signals, I think are not terribly well represented in terms of their subgrid scale heterogeneity. And I think they could definitely impact. So one of these things that we see are important for the sea ice predictability, like how rapidly ice grows, how rapidly ice growth might kind of reduce anomalies in the system those kinds of things. So yeah, I think it's really important to improve those aspects of our models. Thanks to it and thanks Marika. Chudong, you have a question in the chat. So one wonder what is the role of the MIS in the predictability of solid sea ice so when you talk about the ice, the ice thickness I assume you're talking about the solid sea ice but then between the solid sea ice and open ocean and have the MIS MIS means the marginal ice zone there's a lot of ice floating ice and open small open water so it's the mixture of ice and water. I mean, for the sake of students. So I wonder, what's the role of MIS in the prediction predictability of the solid sea ice. So I think the, the size of the MIS, especially when you're moving into the summer months is going to be associated in part with how thick that the sea ice is from the previous term so how rapidly can melt that out how rapidly can break it up with wins and and other things that actually affect the MIS. I think there is some relationship between like the size of the MIS and the winter time sea ice thickness say if you're looking from winter to summer months. The MIS itself being able to predict say that the size of that the location of it is of course a really important topic and again there are things that we don't represent well in the MIS wave sea ice actions is a particular particularly important thing that we don't include in our climate models yet, but it's coming. So I think that'll help improve things like the prediction of the MIS and how rapidly the MIS grows and how large it is and those kinds of aspects of the sea ice system. Thanks Chiram. Thanks Maika. I had a question Maika in terms of the future prediction you showed that as we get into warmer climates the seasonal timescale predictability increases or the internal variability across the ensembles reduce right. If we look at sub seasonal timescale or shorter timescales, would you have a comment on how that would change especially with, for instance if the jet stream shifts north and we have more variable weather like exotropical cyclones or like arctic cyclones impacting sea ice melt, those are probably less predictable on on short timescales right. Yeah, so I think and I think the sea ice is also more responsive to those in a warmer climate. And so I would expect I mean and I haven't looked at this but my expectation would be on sub seasonal timescales, we would see a loss of predictability in a warming climate. Both because there's a potential for increased storms, say, and also because any individual storm has more potential to break up the sea ice. And our sea ice is more able to move and be impacted dynamically by the winds. And so those kind of aspects are likely to lead to less predictability on sub seasonal timescales and a warming Arctic. Great. Again, I guess like a sub seasonal timescale information would be useful for like navigation in the future article operations there. Yeah, and there's a lot of work going on in that area as well I'm just less involved in it. You know, in terms of sea ice motion. And there is a lot of predictability actually on sub seasonal timescales for the sea ice. So there is a lot of persistence. So, so that's good news. In terms of the sub seasonal timescales, although again I think it's likely to be impacted by the long term climate change that we're undergoing so yeah. Yeah, thanks again for your really great talk on the Arctic and polar climate. Thanks.