 So, I think by now you all realize system CSDMS is the community of communities and we have little nuggets of each of these communities. We heard the human dimension and agent-based modeling and integrated assessment modeling earlier in the day. To some extent we're going to hear more, but this next talk represents back in the brain community for all of us oceanographers and we have a great Nick, Nicky Lovendurski, another Pollock. Who's going to talk about the carbon? Okay, thank you. So, hi everyone. I'm Jill Lovendurski. I'm in deep Polish. Yeah. And I'm going to talk to you today about ocean carbon and acidification and I'm going to bring this talk in the context of this question that's up here on the head or slide. Can we predict the future of these things? And I imagine you already know the answer to the question. Who thinks we can predict the future? Okay. Well, there's a little answer to that. It's like, no, I think we cannot predict the future. Okay, so most of you, we cannot predict the future. The answer to this question is no. But we're not going to just stop at answering that question. We're going to try to understand why we cannot predict the future of these things. We're going to try to look at what drives the uncertainty in our prediction. And the way that I'd like to think about this is that if we understand what drives the uncertainty in our prediction, then we can work as a community to minimize those forces of uncertainty in our predictions. So I'm going to begin by showing you an animated bar plot. It's not a particularly exciting animation, but it's an animation nonetheless. And what you're going to see on the left is the sources of carbon for the atmosphere. And on the right, the sinks of carbon in both the atmosphere and the ocean. And you're going to see this in its kind history. So starting in 1765, up to the present day, which is, I think, 2011 in this figure, what you're looking at here are the sources and sinks of amphibrogenic carbon. So we humans have done a lot of amphibrogenic activities that release extra carbon into the atmosphere. And those activities are either deforestation, land use change, or the burning of fossil fuels. And what you'll see is that early on in the 1700s, the primary source of extra carbon, amphibrogenic carbon, to the atmosphere, was deforestation and land use change. But then starting in the mid-1800s and onward, the dominant source of amphibrogenic carbon to the atmosphere is through the burning of fossil fuels. And you'll also see on the right that the amphibrogenic carbon is being partitioned into both the ocean and the atmosphere. So let me start this exciting, sarcastic video of this part of us so we can see what happens. So we begin, hopefully that part of the bottom will go away, with the source funnel on the left in the 1800s. This is a deforestation source right here. And you can see that that extra carbon is being distributed evenly between the ocean and the atmosphere at reservoirs. And then as we move into the 1800s, and now about 1850 and beyond, you start to see that there's an extra source due to the burning of fossil fuels. And again, that's being fairly evenly distributed between the ocean and the atmosphere. And then as we move into the 1900s, the fossil fuel source just starts to take off. And actually the net land, bar, you'll notice, will shrink if you're shrinking now. And the partitioning between the ocean and the atmosphere is no longer a 50-50, but the ocean is still playing a really important role in the flattening of the extra carbon. And so the point of this bar thought was to demonstrate that we've emitted extra carbon, and the ocean has done us a huge favor by absorbing a lot of that extra carbon. If not for the ocean, a portion of the extra carbon, there would be a lot more carbon in the atmosphere today, and it would be significantly warmer. So the ocean has done us this great favor. And for those of you who are unfamiliar with ocean carbon uptake, I thought I would tell you that the carbon uptake of the ocean is not evenly distributed. It's not that everywhere in the ocean is slaving up the same amount of carbon. There's an uneven distribution of the carbon. So here, another video, this is a time-lapse. Every 10 years, you can look at the total inventory or storage of the carbon in the ocean, or notice if I can get this to play. And what you'll notice, beginning in 1910, moving into sort of the present day, is that the ever-magnetic carbon accumulation is likely to be occurring in two main hotspots. The first is here in the North Atlantic, where you'll notice with every passing century there's an additional accumulation of the carbon. The second big hotspot is here in the Southern Ocean, where you'll notice between about 30,000 and about 60,000 the mass of the accumulation of the carbon as well. And so this is the present day picture in sort of where the movie ends. And then the point here is to remind all of you that the ever-magnetic carbon is not evenly distributed. Some places are taking up more than others. And that uptake primarily has to do with the fact that the physical circulation allows the uptake. Here in the North Atlantic, we have one water from the surface moving north to the Gulf Sea. As it moves northward, it loses heat to the atmosphere. It cools you, too. From the outside, gas is more soluble than cold water. And then this water is subjected through North Atlantic deep water formation and brings with it anthropogenic carbon into the interior of the ocean. Similarly here in the Federation, we have mode and intermediate water formation which seducts anthropogenic carbon lead and water into the interior of the Southern Ocean. So the physical circulation of the ocean is setting these patterns of anthropogenic carbon. Okay. So this anthropogenic carbon uptake by the ocean is great for those of us who live in the United States and want this water and don't want it to get too warm. But it's not particularly good for those organisms that live in the ocean. Some are particularly the ones that try to build shells out of calcium carbonate because it leads directly to ocean acidification. Parmodioxide is a gas just like any other gas. It's valuable. It depends on temperature. But it's different from other gases in that when it enters the water, it actually reacts with the water to produce a number of carbonate species, their carbonic acid, bicarbonate and carbonate ions. And then doing so it's shed hydrogen ions. I don't know when you all last took chemistry, aqueous chemistry, or even high school chemistry, maybe it wasn't high school. I'll remind you that these are equilibrium reactions that can essentially run in both directions. And the way that you can think of it doesn't seem like a proper system. If you add CO2 here on the left, this reaction gets pushed to the right and essentially produce more carbonic acid. This reaction gets pushed to the right and essentially produces more hydrogen ions. This reaction gets pushed to the right and essentially produces more hydrogen ions. So the reason what happens fundamentally is when you add excess carbon to the ocean, or when the ocean takes up excess carbon from the atmosphere, you increase the concentration of the hydrogen ion into water. And between three and three times in the present day, we think that hydrogen ion concentration has gone up by approximately 30% in the surface ocean because of this uptake of the atmospheric current. Also, you may remember from high school chemistry that the pH is negative one times long, base 10, of the hydrogen ion concentration. So that when CO2 goes up and hydrogen ion concentrations go up, the pH goes down. And so we're moving pH of the ocean around eight. So you're moving from the basic ocean to a less basic ocean. This is otherwise known as ocean acidification. And on the right, you're looking at a map of the actual change in the pH estimated, roughly estimated, from the 1700s to the 1990s. And this is a slightly misleading map. Every color on this color bar is a negative number. So what this suggests is that everywhere the ocean pH has dropped, but in some places it has dropped more quickly than in other. And you can see these hot spots of lower pH stand out as places where the previous slide, the ocean storage in the atmosphere during the carbon, is occurring. So ocean carbon uptake is good for the atmosphere, bad for the ocean, especially bad for the organisms in the ocean that are trying to build shells out of calcium carbonate. This acidification is challenging to those shell builders. So I'm highlighting experiments that were done in the laboratory where they simulated ocean acidification for two lower choices of organisms from the top. This is a coca-lipid form. This is single-celled USWI algae. It lives primarily in the surface of the ocean and it produces a calcium carbonate shell coca-lip. These little things. It's kind of a beautiful organism when you look at it under a microscope. And you can see what happens when you expose this organism to the lower and lower pH water that the shell just begins to dissolve. And then on the bottom you're looking at the shell of a terracotta. This is, I like to think of these as little tiny swimming snails in the ocean. They're rule-planted. And this is just the shell, but they're actually very beautiful. And they build a shell of calcium carbonate in the form of a lagonite, which is more valuable actually than the calcitic structure of this shell. And you can see that the shell is almost completely dissolved when exposed to acidified water. So maybe you're thinking, why should I care if some shells dissolve? Does it really matter? It doesn't really matter because these organisms have only so much energy to go around. They eat or they photosynthesize and they collect energy. And then they have to decide how they're going to expend that energy. And if they are in an acidified ocean, they're going to need to expend a lot more energy to build and maintain their shell, which means the thing of having a lot less energy to grow may have a lot less energy to reproduce. So that's why we're concerned. Ocean acidification, not just because individual organisms might rob a shell, but because the community as a whole may change because of this change in acidification. So I hope I've at least motivated enough why we might want to understand how much ocean carbonate has preceded and how much ocean carbonate has likely to change in the future, why we might want to predict this. And so now turning to the main process is talk, which is how much carbon dioxide will the ocean absorb? Between now and the middle or the end of the century, can we look at prediction of how much carbon the ocean will absorb? And what you're looking at here are estimates of this ocean carbon flux for each model contributed to the fifth couple of model intercomparison project, our CMF-5s, each Earth system model. And the y-axis here is the globally integrated flux. You take the CO2 flux in every location in the ocean and you integrate it into the whole globe, and so you get a number that's negative, which means that there's a net uptake of carbon by the ocean. CO2 is fluxing from the atmosphere into the ocean in every one of these models. And this is the time series beginning in 2006, which essentially is the start point for all of the CMF-5 model integrations, out to 2080. And what you can see is that for short production lead times, we have a pretty good sense of how much carbon the ocean will take up. It's 2.5 plus or minus a few 0.1 percent petergram with carbon per year. But by 2080, you can see that there's a big divergence in our prediction, though such that some model iterations predict that the ocean will be taking up about one petergram per year, and others predict somewhere around six. That's a big spread, right? The big implications for this indication of the ocean and has big implications for the organisms that live in the ocean, whether we follow this yellow line or whether we follow this big line. So our prediction, you see, is uncertain, right? We cannot predict the future with much certainty. But we can start to get at why that might be. And here I have a color code of these according to the emission simulation or the scenario that they will run under. So here in yellow, you see RCP 2.6. This is sort of the most aggressive in terms of what we do as humans to mitigate our emissions. And under the RCP 2.6, you see, we follow this nice trajectory where ocean carbon is increasing and then decreases. And so by 2080, the ocean is taking up less carbon per year than it was in 2006. That would be quite good for the organisms. But if we instead follow RCP 8.5, you can see that we're down here in terms of where we take up. So our prediction is uncertain, and then you can start to see visually that the emission scenario, at least in the global sense, is what's driving this sort of uncertainty. But even if you know exactly what emission scenario you're going to follow, you still have the monster in you. There's still some spread from yellow to yellow and from pink to pink here. And that's what I have to do with other sources of uncertainty that we're going to explore here today. I want you to have this picture in your mind because this is a globally integrated picture. And I want you to really compare that with this picture. This is now zoomed in to a small region. So in this case, I zoomed into the California current system and to really think and introduce the California current system in his talk yesterday morning. So you all have a sense that this is a region with an eastern boundary upwelling. It's a region where there's a lot of pictures, and so we might be concerned about climate change in this particular region. We can also say here that this region we know is particularly vulnerable to acidification. So we might want to know who won't. We might want to be able to predict how much carbon dioxide is going to change in this region. And what this picture just looks like, a big mess, right? It's just a big spaghetti diagram. You can see, actually, there's... This is the zero line, so above this line, carbon is flossing from the ocean to the atmosphere, and below this line, carbon is flossing into the ocean. And some of these models for some years of prediction are now collapsing, and some of the... They don't even agree as to the sign of the mean CO2 flux in this region. There's quite a bit of uncertainty in our prediction. So a lot of the uncertainty seems to be unaffected by how far out in the future you're trying to project, and uncertainty tomorrow is just as high as the uncertainty in 2080. And you'll notice that this emission scenario doesn't matter at all. If you follow RCPA.5, if you follow RCPA.6, it really doesn't matter, right? Your prediction is equally uncertain. It's a very different picture regionally than what we talk locally. So what we're going to do today is explore the various sources of prediction uncertainty. And I'm going to take a page out of our book, essentially a paper that was written by Ed Hawkins and Rowan Sutton in the context of looking at temperature. Temperature predictions in the atmosphere, whether we can predict with much certainty how much temperature in the atmosphere is going to change, and we're going to apply that exact method to ocean carbon uptake to see if we can predict the exact prediction uncertainty. And in that paper, they say that there are three main sources of prediction uncertainty for climate model. The first is what's called internal variability. It's also known as initial condition uncertainty. And that has to do with the fact that climate models develop their own modes of internal climate variability. An example of that would be in El Nino, La Nino, different climate models, or even the same climate model with slightly different initial conditions depending on the mode of different time, different units, and have a different atmosphere than the next generation. If the models inherently have this variability, and this makes our prediction of the future somewhat uncertain. The second source of prediction uncertainty is the emissions in air. We've talked a little bit about this already. But if we follow our Cp 8.5, the O2 in the atmosphere is likely to be higher than 2,100, then if we follow our Cp 2.6, and that will, of course, be the O2 that developed into the ocean in some regions. And if you are interested, this is right there. That's where we are as of last year. So, unfortunately, it seems as though we are tracking our Cp 8.5 based on the predictions that were made by these scenario assessments back in 2005 and 2006. And then the second source of uncertainty is actually a big one. It's model structure. Different climate models are structured differently. So if you want to write a climate model or an earth system model, you have to write down the mathematical expressions that you want to solve. And then you have to decide how are you going to solve them using a computer? What numerical methods are you going to input? What is your grid size going to be? What is your grid point to be shaped like? How do you parameterize the things that you can't represent in the grid? All of those things are going to be different from climate model to climate model, and they're going to therefore affect the outcome. And even with the same emission scenarios, they're going to have a different result because the structure of your model is different. Similarly, different climate models have similar and different components. So here are some examples. This is the NCAR community earth system model, which you can see, and John Franspaw introduced with yesterday, has an atmosphere component, an ocean component, sea ice, land ice, and then they're all coupled together by a couple. This is the Princeton model, where you can see it has an ocean, an atmosphere, an atmosphere, but these two can seem to be much smaller than they are from the NCAR model. And then this is the French model, where you can see that the atmosphere model really is dominant, and the land becomes part of the atmosphere model as opposed to being a standalone component of the model. And so this is coupled together differently as well. And for those of you who are interested, this little gray circle right here is 1,000 lines of code. So each one of these bubbles have a big amount of code. So these models have a lot, a lot of lines of code. So of course, this is going to affect our prediction, right? How the model is structured is going to affect our prediction. Okay, so let's go ahead and do this. Let's assess the prediction uncertainty. We have two tools to do this. The first is what John Franspaw introduced yesterday. This is an ensemble simulation done with a single climate model. This is even the community earth system model, the NCAR climate model can be predicted to seem to uptake an ocean, change from year to year from an ocean scenario to a mission scenario. And I'll remind you how they did this. They took a singular climate model and they did a long control integration. Then in 1920, they applied round off level temperature differences to the atmosphere. Something like 10 to the negative 14K, as I remember hearing this yesterday from John Franspaw. And that's what it's a butterfly effect. It generated a completely different climate in each one of these ensemble numbers. And they did this 40 different times and 15 different times here. And so from 1920 to 2005, they ran these ensemble numbers under historical forcing what we've measured in the atmosphere. And then from 2006 until 2080, they ran these ensemble numbers under two different forcing scenarios. The medium ensemble here under ARC and P4.5, it's called medium because they're only 15 members. And the large ensemble down here which is run under RCPA 0.5 called large because it has 40 ensemble members. This is a really, really cool tool for looking at internal climate variability because you now have multiple realizations of climate variability. You need to reason ensemble numbers that you can really, truly analyze by a given prediction time. It's also a fantastic tool for looking at the influence of the emission scenario on the uptake of carbon and how that might affect your prediction. And so we're going to do both of those things with this model, but I should point out there's structural uncertainty with this model because it's just one model. So we're also going to contrast this with the results that we get from the CMIT5 archive where we have a variety of model structures we don't necessarily trust our measure of internal variability as much in the CMIT5 model because we have only one ensemble member for a lot of these models. But nevertheless we can do this analysis and start to get at how the structural uncertainty matters. What we're looking at here in a time series is the integrated, globally integrated carbon uptake. From the CSM ensembles, the y-axis is negative indicating the oceans taking up carbon. This is 2010. This is 2008. Each thin line is an individual ensemble member. The thick lines are the ensemble means. So we can contrast that with what you find in the California current system. It looks just like the CMIT5 structure, the spaghetti diagram I showed you earlier. Here, color coded by emission scenario, it's just a big, big spread in terms of carbon uptake in this Lincoln. And you'll notice that again, it doesn't seem to matter what emission scenario you follow until perhaps right here at the very end we start slightly to see a difference between the blue simulations and the red simulations or the pink simulations. Down here what I've done is I've quantified the actual uncertainty in this prediction. So this is what is called the scaled uncertainty. It's the standard deviation divided by the main prediction. And it shows you what I think you can probably see with your eye in the picture that uncertainty, a prediction uncertainty grows with prediction lead time in the globally integrated case, right? So we trust our prediction here. We don't so much trust our prediction here. And because this is a scaled quantity, you can directly compare it with any region and any model and have the exact same y-axis. So right over here on the same y-axis, you're looking at the uncertainty in the California current system. And you'll notice that in 2007, prediction lead time of one year, this global uncertainty is higher than it is globally for almost all whole prediction lead times. You start with a very high uncertainty. That uncertainty doesn't grow much with time. It's actually fairly flat in comparison to the globally integrated case. So let's now dive a little bit deeper into understanding what causes the uncertainty. So here what you're looking at on the y-axis is the fraction of the total variance. So essentially what is the driver of the uncertainty in your prediction? And in the CSM on farmers, we have two sources. We have internal variability and we have scenario uncertainty. Those are two sources of uncertainty. Globally, you can see that for short prediction lead times the internal variability dominates the uncertainty. But then as seen as 2015, so that's like today, the dominant source of uncertainty is the emission scenario. But in the California current system, it's almost exactly opposite. In the internal climate variability dominates the uncertainty. And so out to 2070 in this simulation, it is only then that the emission scenario begins to matter. So by understanding this, we now have a better understanding of where we can potentially invest our resources to improve our prediction. So I've done the same thing with the CSM at five models and we can argue to the calcium bone about whether this internal number is right. I would say probably not exactly because we only have one bundle member, but nevertheless what jumps out at you is that the model of the lane in blue plays a really important role for short prediction lead times in controlling the uncertainty of our prediction. And then after about 2030 the emission scenario dominates but in the California current system, model structure dominates the uncertainty. If you can get the model structure right you can predict the future, right? Essentially that's what this says. If you imagine that the green lines aren't there, the randomly green lines on these from what we did on the previous slide. So it's a guess that our method is that we to a bunch from CSL and FAMAS to seamless five models that here you see internal variability is more important than scenario and here you see scenario is more important than internal variability. Okay. So we went ahead and did this everywhere across the globe. So now we're randomly globally integrated in one tiny region. We actually go over into 17 different biographical biomes and they're defined based on the main abundance of vitro plankton in the biome and the temperature of the biome and how much ice and theracism things that might matter for biology in the ocean essentially. And we calculated the uncertainty in the prediction in each of the biomes and we did this for every year all prediction lead times and so here you'll be at three time slices this is for a prediction lead time of four years in 2010, this is in 2045 this is in 2080 and the color the skilled uncertainty we are showing the color bar, the higher the uncertainty the yellow or the color and you'll notice that even four years out there are some reasons why the uncertainty is still uncertainty is larger than one it's larger than it was in the entire California current system. So you don't really want to try to predict the future in the left end of the trail pacific here but you're not going to do particularly well and also what stood out in this is the tropical Atlantic, the South Subtropical Atlantic zone, also very high uncertainty this might be useful to you if you're thinking about well I'd like to design a sort of observational campaign to go out and measure the sophistication in the ocean how it changes in time, but if you're going to do that don't go here because it's going to be tricky to interpret your results right, you might want to go somewhere that's not good and then you can see how that uncertainty changes with time over the course of the simulation and at the end you can see here that almost the entire world is yellow indicating that we don't trust your prediction in 2018 so when we went ahead and went into every biome and then partitioned the uncertainty ring this analysis of variance to see is an internal variability with an emission scenario and what we found is that the internal variability dominant to a certain point in any emission scenario takes over and so essentially what we see here is a map of when these lines cross, when exactly it does, the scenario take over the most important source of uncertainty in years, so 2040 to 2080 and beyond, so the dark color of the places where the scenario uncertainty takes over early and light color of the places where the scenarios that we think relate and then there's even some white areas here where forget it, the scenario uncertainty never matters, it's actually only internal climate variability that's dominating the uncertainty, so definitely don't go out here and try to up here and fix up with this model but interestingly what we see here are these dark regions are exactly the places where we saw, like eight slides ago, the ocean anthropogenic carbon is being stored, right? So here in the North Atlantic and here in the Southern Ocean there's hot spots where ocean anthropogenic carbon stored are places where the emission scenario matters and that actually makes sense, right? These are places where the ocean's accumulated carbon will be expected to accumulate carbon so the emission scenario stands out as the dominant source of prediction when we think of these regions before it does in other regions. So we then also did this in the CMIT 5 model, so we're not 100% trusting our internal variability, but we're going to assess model structure and I want to point out that this color by is more from 0 to 2 on the previous slide it was from 0 to 1 so overall there's just a heck of a lot more uncertainty in the CMIT 5 models than there is in the CSM on samples that probably has to do with the fact that there's structural uncertainty when we look at the CMIT 5 that wasn't there we'll see yet then and then you see that there are places where I mean most of the global we can't trust our predictions even 4 years out so CMIT 5 models are we are challenged to look at prediction in these particular biomes on the biome scale for ocean carbon uptake even 4 years out and then certainly really doesn't doesn't shrink at all it just continues to grow in most of these biomes. And then we went ahead and did the same sort of analysis but here our concern is when does the model structural uncertainty get taken over by the emission scenario of 30 when does the blue line cross the green line and white means never which means that in those biomes the model structure is a new dominant source of uncertainty for all prediction lead types and we can see that that is almost everywhere there are few places where that happens and where it does happen it's fairly late in the time period so I'm running out of time and there are my conclusions I hope that I convinced you that predictions of the future ocean carbon sink I fought with on certainty this is particularly true at regional scales the three sources of prediction uncertainty can vary with prediction lead type they can vary with facial averaging scale and they vary from region to region some regions are more inherently more certain than others and then finally if we want to produce reliable predictions on regional scales I would argue that we're vastly reducing model structural uncertainty we're not going to tell you how we do that just saying that we probably shouldn't lessen that if you think about emission scenario development the political science movement is really not super responsible for developing emission scenarios so we have what it's saying how that goes and also internal variability is instead of an aleatoric thing that's always going to be there we can't necessarily correct for that emission condition uncertainty but we could perhaps reduce uncertainty and model structure so that's all thank you for your attention to work with the definition of the uncertainty and model structure is that what you mean you know what I mean that's what I would say so in investigating so evaluating model adequacy is one of the things I do most and and even a much simpler model it's not easy but if all you're using are scenario representations it's really tough to get to things and these models take a long time to run so there are a whole bunch of methods that are very rarely used these are a little bit what I understand in global climate model that might be of some assistance and we haven't thought about like can we rank the models according to their skill and then carve it up and use that to help us reduce uncertainty in the model structure or things like that can we evaluate the skill in a broader sense the first step is sensitivity analysis what matters all these processes just from fiddling with them you have some ideas what matters but if you analyze that in a more structured way there might be some surprises there so that's usually time on the first step and then you can just go from there and start looking okay what are the data that you're done with yeah right so one of the problems is that in some biomes a model doesn't work really well and then you go to an open biome and it walks out and so how do you how do you assess skill when it changes from major to region exactly so there's some layers in there and so trying to put it in the same way that uses two models and yet does each of them have I think it's probably possible there's four questions about my ignorant process in the in the low emission scenarios the uptake rate decreased right and less negative at the end is that because I guess my question is does that have anything to do with temperature and is that actually a good proxy for acidification does the acidification rate really slow down or is it more but it's not the driving force the driving force is that in those in RCB 2.6 we actually get negative emissions by 2100 yeah that's really ambitious and there's that aggressive scenario like we really get our stuff together and we we make sure we reduce those emissions because if there's a lot less carbon they're to the lower end of the ocean and that's what we do that's what we do