 Hello everyone from wherever you're joining. Thank you for being here today for this NCAR Explorer Series lecture, predicting future climate. What can we expect in the next decade with Dr. Steve Yeager and Dr. Isla Simpson? My name is Dr. Dan Ziehlo and I'm an education specialist here at the National Center for Atmospheric Research or NCAR, which is a world-leading organization dedicated to understanding our system science. And that includes our atmosphere, weather, climate, the sun, and the importance of all of these systems to our society. I'm really glad to be with you all today to learn more about how scientists like Isla and Steve use models and observations to make decadal climate predictions. So for this lecture, we will take questions at the end, but please definitely submit any questions you might have during the talk using the Slido platform. So if you scroll down the webpage, you can see the Slido window just below where you are seeing the live stream video of this event. And if you haven't already, go ahead and click on the green Join Event button and then you can ask questions on the Q&A tab and answer poll questions on the Polls tab, both of which are found in that blue bar across the top. And definitely be sure to join Slido to add your thoughts to our Work Lab question. How do you imagine the climate will be different a decade from now? Because we're going to get to that real soon. This lecture is also being recorded and will be available on the NCAR Explorer Series website. Now with us today, we have NCAR scientists, Dr. Steve Yeager and Dr. Isla Simpson, both from NCAR's Climate and Global Dynamics Lab. Dr. Yeager focuses his research on the role that the ocean plays on modulating our Earth's climate on time scales from seasons to centuries. He currently co-leads a working group for developing Earth system prediction applications of the community Earth system model and also has leadership roles in other international programs. Prior to joining NCAR, Steve was a math and physics teacher in Fiji while volunteering with the Peace Corps before completing his PhD in atmospheric and oceanic sciences from the University of Colorado, Boulder. Dr. Simpson studies large scale atmospheric dynamics and its representation in global climate models. Specifically, she uses models to understand the mechanisms at play and the variability of the large scale atmospheric circulation and its impacts on regional climate and hydroclimate. Before joining NCAR, Isla completed her PhD from Imperial College of London followed by postdoctoral positions at the University of Toronto and the La Montdority Earth Observatory at Columbia. Steve and Isla, do you wanna quick turn on your cameras and say hello before we check out our workload? Hi. Hello. Glad to be here. Thanks, Steve. Great. Now, Brett and Paul, would you be able to pull up that workload for us to see our audience's thoughts? Great. So when folks were imagining the climate in the future, the biggest three words I see right now are hotter, warmer and worse. And that seems to be kind of the themes across all of here. I'm seeing drier, extreme weather, larger variations, continued warming, disastrous, much worse, storms, chaotic. Steve, how are these words resonating with you, maybe in terms of what climate might look like in the future? I think they're spot on. Awesome. And so with that, let's dive in and hear from Steve and Isla. I'm gonna share my screen. I'm gonna get us started today. So thanks for that intro, Dan. I'm gonna give us an overview of some of the sort of conceptual framework of how we think about this problem of predicting climate and describe some of the tools that we developed to answer this question, what can we expect in the next decade? So the focus of this talk is really how can we predict climate a decade ahead? And then Isla will take over and present some actual decadal climate predictions for Western Europe and the Southwestern US. So just to get us started here, this image in the background that you see is an example of a decadal climate prediction. The globe in the center here shows the observed 10-year winter sea ice trend that was measured by satellites over the decade from 1991 to 2001. And the red shows where the sea ice expanded and blue shows where there was sea ice retreat over the course of this decade. And then the two globes on the left and right are climate prediction systems that use the CESM model developed at NCAR. We can see right away that the prediction system on the right does a much better job of predicting the trends that were actually observed in nature. And so over the course of this talk, we'll learn about how we develop these prediction systems, how these two systems are different and why the one on the right is much better suited for answering the question of what to expect in the next decade. So just to start things off with the premise that prolonged changes in climate can impact lives and livelihoods. Here's a well-known example of a decadal climate change that occurred in the 1930s. A prolonged period of very dry conditions in the Midwestern United States. Dust Bowl resulted in mass migration out of the Midwestern Plain States because once fertile farmland turned into arid wasteland. Another example that will be familiar to many adults is the drought that occurred in the African Sahel region in the 1980s. This resulted in severe famine in countries like Ethiopia. And a prolonged climate change that we're expecting this century is associated with anthropogenic global warming. We don't have images yet of what sort of climate impacts we can expect from global warming, but these computer generated images give us some sense of what we can expect if we allow the planet to warm by three degrees Celsius by 2100. So what causes these types of climate shifts from decade to decade? The answer is it's very complicated. But climate scientists like to unravel that complexity by splitting the drivers of climate change into two categories. We think about forced variability and change. And that's associated with drivers such as changes in greenhouse gases, volcanic eruptions, changes in solar cycle. These are considered external forcings. They're external to the normal workings of the coupled ocean atmosphere system that we live in. And on the right hand side, there's the variability that just comes from the internal chaotic motions of our climate system. These are natural modes of variability that take place across a range of space and time scales from turbulence in the ocean and atmosphere that we otherwise know as weather to longer time scale coherent modes of coupled variability like cycles of El Niño and La Niña in the tropical Pacific to longer time scale variations in things like ocean circulation, in particular the Atlantic meridional overturning circulation that is believed to be very important for climate change on decadal and multi-decadal time scales. So as an example of how we understand regional climate change on time scales of decades, let's consider this analysis of climate change in the US Southwest. This top plot shows several curves. The green and gray curve here that I'm highlighting is a Southwest drought index which basically measures how wet or dry the soil is in this large box covering the Western United States. And we can see that there's a lot of decadal variability in this drought index. And during the 30s, it was a period of prolonged dry conditions in the Southwestern United States. These other two curves plotted on top are sea surface temperature indices. And what that means is we average the sea surface temperature over these box regions. So the AMV is an index that stands for Atlantic multi-decadal variability. We average SST over this large box. Niño 34 is a measure of how warm or cold the tropical Pacific is. And we can see by eye that these SST indices have many wiggles in common with the Southwest drought index. We say that they're correlated with each other. On the bottom, we see statistical regressions. These quantify how changes in SST index translate into changes in drought index computed at each point over the continent. So for example, when AMV is warm, when the North Atlantic is warmer than usual, there tends to be drier conditions over most of the US, but particularly in the Southwest US. On the left here, we see the regression with global mean SST. So this shows that when globally averaged SST gets warmer, the Southwest tends to get drier. So what this kind of analysis tells us is that there's an important role for ocean variability as a driver of regional climate change. If we think of AMV and Niño 34 as reflecting internal variations in the climate system, those certainly play a role, but so does this externally forced warming of global mean SSTs. We have to take all of this into consideration to develop an understanding of the climate of the Southwest. When we broaden our perspective to think about larger spatial scales, the global scale, then it turns out that changes in atmosphere composition, gases and aerosols that make up our atmosphere, those are important drivers of climate change from decade to decade. So on the bottom plot here, we see the observed global average temperature time series from 1850 to 2020. We see this steady increase in global surface temperature that we associate with primarily this increasing concentration of CO2 in our atmosphere that's been measured at sites like Mauna Loa in Hawaii. But this is just part of the story. There's other changes in our atmosphere that we also have to consider like periodic volcanic eruptions that inject aerosols or dust into the stratosphere, changes in other kinds of pollution like sulfur dioxide. And so these also play a role in helping us to understand this observed time series. So if we go back to this plot of what drives changes in climate, we can identify some drivers that are potentially predictable. These include the changes in forcing. So changes in atmospheric composition, but also changes in the slower modes of internal variability that are related to the ocean. These are also potentially predictable drivers of changes in regional climate and we can harness that for developing decadal climate prediction systems. There's always gonna be some component of variability that we can't predict. We call that unpredictable noise and that will give us an inherent limit on our ability to predict climate on the time scales of years to decades. So up to now we've been looking at observations but they're very limited and sparse. And so we use global climate models to supplement our understanding of how the earth system works. These climate models are computer programs that simulate all the components of the earth system based on physical principles. So what we do is we run these computer programs on grids like the one shown on the left here. These split the earth atmosphere and land and ocean into small regions or grid cells. And then we run these programs on large computers like the Cheyenne Supercomputer at here at NCAR. And so these computer simulations are basically virtual earths. They allow us to understand past and predict future climate change. I'd love to share with you a video of what a virtual earth looks like. But in the interest of time, I'm just gonna point you to this YouTube site that has a great collection of videos of what the CESM model looks like. So how do we model earth's climate? We take our computer programs and we run them with observed or projected changes in gas and aerosol concentrations. So we've seen that we have this observed time history of increasing CO2 up to present. And if we wanted to run our models into the future, we have to make some assumptions about what that concentration will be in the future. These emission scenarios are based on estimates of what energy sources will be in the future, what population growth will look like, et cetera. So for example, this red curve is a business's usual scenario that assumes heavy fossil fuel use into the future, whereas this blue curve is assuming that we're gonna choose a sustainable development pathway. And so the concentration of CO2 will start to taper off. Now, when we run our models using these four things, then we get curves like the one shown here on the upper right. These lines here are model simulations. And we see that the climate is quite stable up until about 1950. And then we see this dramatic increase in global mean surface temperature into the future up to 2100. It's hard to see, but there's actually a red curve in here that represents the observed global mean surface temperature. So the model is able to do a good job of reproducing the observed changes in global mean surface temperature. Now, a great thing that we can do with climate models is we can run multiple Earths at the same time. We call this running a large ensemble. So this gray shading here is actually 30 different gray lines plotted on top of each other. These are different ensemble members and they're represented sort of by this diagram on the right. Each one is a different version of Earth. And doing this allows us to study the forced and internal components of climate variability. All of these ensemble members start from nearly identical initial conditions and then they diverge because each develops its own internal variability. The real Earth that we know about from observations is then treated as just one of many possible histories that could have occurred. Then we define the forced variability as the variability that all ensemble members share in common. We get that by taking the average of all of these Earths. The internal variability is the variability that's unique to each ensemble member. That's represented by the spread of the gray lines here. When we project global surface temperature into the future, we find that the forced component that's coming mostly from anthropogenic greenhouse gas emissions is much larger than the internal component. It's hard to relate to global mean temperature. So let's see what that means for regional climate. This busy plot here is showing winter temperature trends over a 50 year time period from each of our virtual Earths. Down here on the bottom right is the actual observed winter temperature trend over that 50 year time period. We see warming over most of North America. But if we look at the virtual Earths, we see a wide variety of temperature trends that could have occurred. Some members show cooling over the US and some show even more dramatic warming than was observed. So we conclude that internal variability probably contributed a lot to past temperature trends in the United States. Some of this trend that was observed can be attributed to the external forcing, but some of it was probably related to just internal variability. When we do the same exercise and look at future trends of winter temperature over North America, looking into the future now, we find much less variability from member to member. They all show warming and they have temperature trends that are more dominated by the forced component of variability that's coming from anthropogenic CO2 emissions. So what does all this have to do with prediction? Well, to design a good prediction system, you have to know which type of variability you're trying to predict. Are you trying to predict internal variability or are you trying to predict forced climate change? And this diagram here gives you an idea of which matters more as a function of spatial scale and time scale. We live down here in the lower left-hand corner. We live in cities and our lives go about from day to day and we're most familiar with internal variability, weather and internal variability is what matters most for weather prediction. That's why you don't hear much about global warming when you tune into your local weather forecast. Forced variability is what matters most for predicting future global change. And this is particularly dominant when we look at time scales of a century into the future and consider the entire globe as a whole. But in the future, the forced component of variability is going to start to dominate this cube and we're going to find that it will become more and more apparent to us even on our local scale, even on the city scale. If we've lived in one city for several decades we might have already noticed that there's been an increase in temperatures. I've lived in Boulder since the late 90s and I've noticed that the summers are getting hotter, the winters are getting hotter. So this is something that we can start to already perceive in our lives. But we're only right here on this curve of global warming and it's really future generations that will be most affected by the forced component of climate change. Now the topic of this talk is decadal climate prediction. We're trying to predict things on time scales from a year out to about a decade and on spatial scales from about the size of the state of Colorado out to the planetary scale. And you can see that this cube intersects both internal and forced regions of this diagram. And that means that both forced and internal variability is important to the problem of decadal climate prediction. Before we get into the nitty gritty of decadal climate prediction let's get one common question out of the way which is if we can't predict weather beyond two weeks how can we say anything about what will happen in the next few years to decades? The answer is we aim for decadal climate prediction not decadal weather prediction. To understand the difference consider this pinball game. If we drop one ball we can consider that one day's weather. There will be a lot of randomness as to where that ball will land in this slots at the bottom. But if we repeat that ball drop many, many times then we will build up a distribution of where balls tend to land. Most of them will fall in the middle that's called the normal or the mean climate. There will be extremes on either side of hot and cold or wet and dry but those will happen less frequently. And this is what we mean by the climate of a particular region. Now variations in sea surface temperature like we saw the AMV index tilt the climate game for particular regions like the US Southwest. So they cause the distribution to shift towards warmer and drier or colder and wetter. And climate prediction is about forecasting changes in these multi-year statistics. We're not trying to predict where individual balls will drop. We're trying to predict where they will tend to drop. Anthropogenic greenhouse gas warming will tilt the climate game further than we've ever seen. And so as climate scientists we feel a little like this frog on the left. We know that large changes in climate are coming down the pike. But as a human population we haven't yet experienced severe impacts. And so it's hard to communicate the urgency of responding to this climate threat. Now let's get into the details of how we do decadal climate prediction. What we do is we initialize the climate model from our best estimate of observations. That's represented by the black line here. Each of these red lines represents a model simulation that's been initialized from the black line. Then we do that and we use radiative forcings that come from observations. And we do that over and over again and generate a large collection of what we call hind casts. These are predictions of past change. We evaluate how well the hind cast do it predicting what happened in the past. So we can see that for example, this hind cast set predicted this downward trend. If our collection of hind casts show skill at predicting past climate change, then that gives us confidence that when we do a future forecast that it will be credible. Now this initialization step from the observed state of the earth system is what helps us in predicting the internal variability in the climate system. And using observed and projected radiative forcings changes in the concentration of gases in the atmosphere. That helps us in predicting the forced variability and change. So we've seen now two different climate prediction systems that use the CESM model. The first kind of prediction system we call uninitialized and we have a name for this one, the CESM large ensemble. In this type of prediction system, the skill comes from correctly simulating the forced variability and change. And we expect skill for long time scales and large spatial scales. So we use this type of system for projecting future global climate change. The second type of system is an initialized system. We call this one the DPLE, the Decadal Prediction Large Ensemble. Here the skill comes from correctly simulating both forced and internal components of variability. And for this system, we expect better skill for short time scales and small spatial scales. So let's now look at an actual result from our initialized decadal prediction system. This plot shows predictions of five year average ocean temperature. And the top row shows the skill of our initialized prediction system. Red means we've got high skill, blue means we have low skill. And the different column shows how far in advance we're trying to predict the truth. And we can see that if we do a prediction, even nine years before we know the actual truth, we're getting high skill at predicting changes in ocean temperature over most of the globe. Now, a lot of this skill is coming from the forced warming of the oceans because we've got increasing CO2 concentrations, the oceans are warming, and we're able to predict that warming. So we also wanna ask the question, how much better is initialized than initialized? Do we really need to synchronize the internal variability? And when we look at the results for that question, we see that we get some improvement, particularly in the North Atlantic, but also in the Southern Ocean. And this increase of skill in the North Atlantic helps us to predict that AMV index that we saw was important for some regional climate changes. Here's another example of a prediction result. In this case, we're trying to predict five year average summertime rainfall. Here's the skill of the initialized prediction system. We're not seeing a whole lot of skill in the US Southwest and that's a topic of research. We don't quite understand why, but we do see that we have good skill over the Sahel region of Africa and the skill is much better when we initialize the model. And so if we focus in on this box region here that is the Sahel, this is what the time series look like, a five year average summertime rainfall. Black is observations. And you can see that during the 80s, during the severe drought in Africa, the rainfall was below average. Red is what we get from our initialized prediction system and we see a pretty good match to black. But if we try to predict with an uninitialized system, we get much less skill. So the take home message is that if we had DPLE back in the 1970s, we could have predicted the 1980 Sahel drought and possibly been able to forewarn people or give some advanced warning of a severe economic hardship to come. So that completes my half of the talk. Next, we're gonna have Isla who's gonna present some decal prediction case studies that will highlight different components of predictability. European winter precipitation will highlight the skill gain from internal variability and then US Southwest hydroclimate will emphasize forced change. But before we have Isla, I think we're going to go into a quick Slido poll to quiz you on how well you've paid attention, I guess at this point. Okay, I guess we'll give everyone a few seconds to answer the question. Excellent. So most of you are saying one degree for how much has the global annual average temperature increased over the past 50 years? We'll see the answer in a second. We'll go to the next question. Okay, so the next question is, how much will global annual average temperature increase in the next 50 years? Well, no one's saying zero, but we have a variety of answers. Okay, we'll move on and I'll show you the answer. Okay, so here's a figure that's trying to illustrate what the answer to that question is. So this is showing global mean surface temperature. And so the black is what we have observed in our historical record and then the red is our model simulated change. And so for those of you who answered one Celsius for the past change over the last 50 years, that is correct. That's what we've observed. And you'll see that it's also what our model has simulated. For the projection for the next 50 years, we don't really know the answer because we have uncertainties. We don't know what we're going to do in terms of our greenhouse gas emissions. We don't know for sure that our model is doing things correctly, but we have some confidence because we have simulated the past okay. And of course, there could be some internal variability that would change the answer slightly. But what our model is suggesting here, and this is under the highest of the emission scenarios. So if we don't do anything about our greenhouse gas emissions that we will warm by about two degrees Celsius. So I think the majority of you got those answers correct. So we'll go on to some case studies now. And so we've got two of these and they're using the prediction systems that Steve has just described. And so the first of these is a case where the internal variability is really important. And where we're able to get some skill in predicting things if we can initialize properly. So this is using the initialized prediction system. And in particular, the key player here is that predictability that Steve mentioned in the North Atlantic. And so this is looking at predictability in the North Atlantic jet stream and the implications for precipitation in Western Europe. So just to orient you with what the North Atlantic jet stream is, what I'm showing here, this is the climatological average. So this is the average of a very long time period about a hundred years of what we call the zonal wind which is the wind from the West toward the East. So this is the speed of the wind in that direction. And this is about two and a half kilometers above the surface. And so what you see is on average you have these strong westerly winds. So the flow is going in this direction. And it kind of is tilting from the Southwest toward the Northeast here across the Atlantic. So this is the North Atlantic jet stream. And so what that means is the prevailing flow in the wintertime is kind of going like this. And this is important for the climate of Western Europe because our weather systems or our highs and lows that give us our weather in the mid latitudes, these are strongly coupled to the jet stream. So the jet stream is kind of guiding these weather systems. The jet stream is also bringing in moisture from the ocean toward the continent. And so wherever the jet stream kind of intersects the continent is kind of a stormy and a wet place like the UK is. It turns out that in the late winter, so March in particular, and this is really going to be a story about March, the jet stream has varied a lot on low frequency timescale. So kind of decadal to multi-decadal timescale that's varied a lot over our observational record. And so to show you where it's varied, what I'm showing here is the standard deviation of that zonal west to east wind speed for 10 year chunks over the observational record. And so if you don't know what standard deviation means, just think of this as being a measure of how much has the wind varied over the record. And you can see that it's varied a lot towards the west of the UK and it's varied a lot over North Africa and the Iberian Peninsula, kind of to the west of those regions. And so if we just take an average of our wind speed or westerly wind speed in this box here to the west of the UK, this is what it's looked like in the observations. We've got a bunch of different lines here. These are a bunch of different observation based data products and they have different start points and end points. But you can see that where they overlap with one another they're kind of agreeing. And what you see is that like from year to year, the average wind speed in March varies a lot. But you also see you have this kind of low frequency variability in addition to that higher frequency noise. So for example, in the 80s and 90s you had kind of stronger westerly winds to the west of the UK and then in the 50s and 60s you have weaker westerly winds. And this matters for the climate of Western Europe and I'll just illustrate that to you by showing you two different time periods. So I'm taking here 31 year averages. So on the right here this is showing the zone of wind for the average from 1980 to 2010. And then on the left is for 1935 to 1965. And you can see that this is showing you the structure of that jet stream and it's varied a lot. So in the more recent decades, it was strongly tilted. It's kind of heading in toward the UK whereas in the earlier period it was just kind of heading straight across the Atlantic and it was more intersecting the continent kind of at the Iberian Peninsula or North Africa area. And this has implications for precipitation. So here I'm showing the precipitation averages over those two different time periods. And so on the right, what you see is that for that kind of 31 year period centered on 1995 Scotland was very wet. You have about 10 millimeters per day on average precipitation over Scotland. And then the Iberian Peninsula is kind of dry and it doesn't even make it on the color bar there. You have less than two millimeters per day on average of precipitation for this period. But if we go to the left here and we look at the period centered on 1950 you see a pretty different picture. Scotland was not as wet as it was in the more recent decades and then the Iberian Peninsula was more wet and they have pretty similar precipitation averages both coming in at kind of about five millimeters per day. And so this was surprising to me because I grew up in that purple blob in the 80s and 90s and it's a pretty wet climate. And it's hard if you come from there to imagine that there was a whole kind of three decades where in March the precipitation average in Scotland was kind of similar to what it was in Portugal which is the Iberian Peninsula where you think it's a much drier climate but that's what happened. And we can understand that why that happened and that's because of these changes in the jet stream. So in the more recent decades the jet stream was kind of bringing the storms and the moisture directly in toward the UK whereas in these decades it was bringing it more toward the Iberian Peninsula and North Africa. And it turns out that this variability in the jet stream is highly coupled to that North Atlantic ocean variability that Steve introduced or the AMV. And so I'm showing that here. So in the red line what you have is this speed of the westerly winds in that box that we already looked at to the west of the UK. And then in the black line is minus one times the AMV index which is basically a measure of what the sea surface temperatures have been doing in this North Atlantic box. And here we're going back to 1850. And if you look at the wobbles from year to year there those two things are not really coming along together. There's noise in both of them that is anti not correlated. But if you start to look at the lower frequencies of this kind of down and up and down you can see that they kind of are coming along together. And so here we have the yearly values and on the right we kind of get rid of that noise by taking 20 year averages and overlapping them over the course of the decade. So we're looking at kind of what happens on 20 year time scales. And there you can see that these two things are really coming along together. And so this is in the sense that in the 80s and 90s we had colder sea surface temperatures in the North Atlantic and we had stronger West Julie winds to the West of the UK and Scotland was very wet. And then we have this period back in the 1950s where we have warmer sea surface temperatures in the North Atlantic and we have weaker winds to the West of the UK and Scotland was drier and Spain and Portugal were wetter. There are some caveats here obviously just showing that these two things are connected doesn't tell us what's causing what. But we do have some evidence and I'm not going to go into that in the interest of time but we have evidence for this connection representing an influence of the ocean on the atmosphere. We also only have a very short record here when we're talking about this kind of low frequency variability and we only have about 150 years of record. We only have a couple of wiggles here to see that these two things are coming along together. So it's unclear how robust this is and whether it's going to continue and only time will tell really. We wanna observe more of these wiggles to be sure that these two things are connected. But in the meantime, we have the potential here because we can predict this variability in the North Atlantic Ocean as Steve illustrated we have the potential for predictability on long time scales for the climate in Western Europe. And so here I'm showing some regressions onto that Atlantic multidicado variability index and this is basically showing how are these things connected to the AMV and what these connections are. So basically when you have sea surface temperatures here in the sub-polar North Atlantic that are warm like we had in the 1950s then the jet stream is displaced to the south and that leads to the UK being drier and the Iberian Peninsula being wetter and then the opposite is true for times when you have colder sea surface temperatures in the North Atlantic. Now unfortunately for some reason our model does not capture this connection between the sea surface temperature and the jet stream and precipitation and trying to understand why our model doesn't do this is an active topic of research. But even though we can't do it to capture all of this connection we can predict the sea surface temperatures in this region. And so even though we can't capture all of the connection this potentially can give us some predictability. So just to, and Steve already illustrated this but just to illustrate this in a bit more detail here this is showing the sea surface temperature is averaged over this region in the sub-polar North Atlantic and then these are 10 year averages over the course of the record overlapping and the black is the observations and then the red is our predictions from our initialized prediction system where we have both the external forcings and we have the ocean and the sea ice are initialized based on observations. And so what you see here in the red is that for the forthcoming decade we can predict what the average sea surface temperature is going to do. And in particular we were able to predict here in retrospect that we would have these colder sea surface temperatures in the 80s and 90s and we have warmer sea surface temperatures either side of that. So even though we can capture this connection in our motto with the jet stream and the precipitation we have the potential for predictability because we can kind of do a combined numerical plus statistical prediction where we can say, okay we can predict the sea surface temperatures and we know how the sea surface temperatures are connected to precipitation. So then maybe we can come up with a skillful prediction for precipitation. So I'm gonna show you what the predictions look like for these two regions in the west of the UK and the Iberian Peninsula. And so on the top here is gonna be the predictions for the UK and the bottom is for the Iberian Peninsula. And here on the left you're gonna see the time series of 10 year average precipitation. And right now all you can see in these black dots here is the observed precipitation. In these panels what you're gonna see is the skill of the forecast. In case anyone is familiar with these measures this is gonna be the anomaly correlation coefficient and this is gonna be the mean squared skill score. Basically all you need to know is that if you have a value of one you have a perfect forecast and if you have zero or negative values then you don't have any skill in your forecast. So first of all in the light blue here this is what our model prediction gives us if we don't initialize. So this is from, this is the predictability that comes in just from the external for sinks or changes in greenhouse gases, the volcanoes, solar variability and things like that. And you see that you don't have any skill. Here is the, this is the precipitation that comes from our model if we both initialize. So we capture the ocean variability correctly, well to the best of our skill in predicting that and we have the external for sinks. And you see again that this doesn't have any skill and that's because our model is failing to capture this connection between the sea surface temperature and the precipitation in Europe. But here in red this is what we get if instead we use our model predicted sea surface temperatures and we use the observed connection between the sea surface temperatures and precipitation to come up with a prediction for precipitation in for the forthcoming decade in these regions. And you can see that now we have some skills. So we can predict this up and then the subsequent dime. And of course we only have a very short record here to be validating this skill. So we have to be careful but it at least shows promise here that we could skillfully predict precipitation over the forthcoming decade in these regions. So now we can try and make a prediction of what's gonna happen for the next decade. And actually since this was done a couple of years ago our prediction here is for 2018 to 2028. And so this is what our model is suggesting will happen to precipitation over the UK and Portugal. And so the predictions are that over the next 10 years it will be a little wetter than it was in the UK. It'll be a little wetter than it was in the early 2000s but not quite as wet as it was in the 80s and 90s. And then for Portugal it'll be a little drier than it was in the early 2000s but not quite as dry as it was in the 80s and 90s. So we can watch for the next decade and see whether this actually continues to have skill. So for the second case study that we're gonna finish up here with is a case study that's a little closer to home. And this is gonna be one that's looking slightly longer term not quite the next decade but more kind of the next three decades. And this is gonna be looking at the impact of forced climate change on the US Southwest. And if you're interested in reading more I guide you toward this report that came out recently on the drought that just is happening and ongoing in the US Southwest. And so I'm gonna kind of illustrate this in the context of what happened in 2020 because I'm sure we're all kind of aware that 2020 was extreme for a number of reasons but one of the reasons is that the US Southwest experienced a very extreme drought. So here on the left we've got the US drought monitor for October 6th of 2020 and you can see that vast portions of the US Southwest here were under extreme or exceptional drought conditions. And on the right, this is an image from Colorado Public Radio that illustrates the wildfires that we had in the region. And so this year was a very extreme for wildfire in Colorado and elsewhere. And so this is the Cowwood fire that burned just north of Boulder. And the title here kind of says exactly the conclusion that we'll come to at the end here is that what happened in 2020 was pretty extreme even if you account for climate change but it is unfortunately a sign of things to come. So we'll try and address the question of what role did climate change play and what we saw in 2020 and what should we be expecting for the US Southwest as we move forward. And so to try and answer these questions I'm gonna use these simulations that Steve's introduced in the first half of the talk. And here we're gonna be using our uninitialized simulation. So here we're looking at the impacts of external forcings in particular we're looking at the impacts of increasing greenhouse gases. And the data that I'm gonna show you here is actually not just coming from our model some of it's from our model built here at NCAR but here we're actually including 33 different models from modeling centers all over the world. And these are the simulations that contributed to the latest intergovernmental panel on climate change report. And so as Steve introduced we use these scenarios to project greenhouse gas emissions moving forward. And for this analysis we'll be using this red one which is where we kind of business as usual we keep on emitting greenhouse gases. So just bear in mind that what we do with our greenhouse gas emissions is an uncertainty here but it actually doesn't make too much difference for the next couple of decades where it really makes a difference is out toward the end of the 21st century. So everything I'm gonna show you here I'm gonna simplify down and just show you annual averages. So these are gonna be the averages from January to December and averaged over these six states in the US Southwest. And so what you have here is the observational record going back to 1950 on the left is precipitation and on the right is near surface air temperature and these are anomalies from the average of 1950 to 2000. And so what you see is that 2020 was really dry it was actually the driest year on record although we do have some years that were dry in the past and then 2020 was also really hot. It was not the hottest year on record but it was one of the hottest. And so if we look at the precipitation here there's no real clear long-term trend we kind of have this kind of slow up to the 1980s and then a gentle decline in over the last few decades and the feeling kind of consensus view at this point is that this is mostly internal variability related to those variability in the ocean as Steve described. But when we look at the near surface air temperature we see this very clear trend towards higher near surface air temperature. And so here we look at our model simulation and so each of these gray lines is a different model simulation. And so recall that we don't expect to capture all of these little wiggles in our observational record because some of these are unpredictable internal variability. What we wanna capture correctly is kind of the overall statistics of this variability. What we can do though is we can average over all of our simulations and that kind of picks out what's common in each of them and that's really what's coming in from the external forcings that we're putting in. And so what you see for precipitation is that there's no real clear trend in precipitation and the models at least the external forcings are not really doing a whole lot. When you look at temperature though there's a very clear trend toward increasing temperature as a result of the external forcings. And you can even see the effects of the volcanoes here in the model. So we've got Alchichon and Pinatubo eruptions here resulting in a slight cooling in the US Southwest. And so I'm gonna kind of illustrate what happened in 2020 and what we should be expecting for the next few decades by using two different variables that are relevant for drought and for fire. So where we look at precipitation which I'm sure is familiar to everyone but then this other quantity vapor pressure deficit may be not quite so familiar. So vapor pressure deficit, it's basically the atmosphere's thirst for water. So it's the difference between how much water vapor the atmosphere can hold and how much it has. And so if you have a higher vapor pressure deficit if you have water available you're gonna see more evaporation. So higher vapor pressure deficit will mean more drying out of the soils and the vegetation, more evaporation from rivers and reservoirs. And so higher vapor pressure deficit is kind of a conducive to more fires and more extreme drought conditions. Now how much water vapor the atmosphere is capable of holding that is strongly connected to temperature. And so this is where we might expect that greenhouse gas driven warming might play an important role in what we see here in the Southwest. So I'm gonna illustrate things using both these quantities but it's important to kind of keep in mind that they're connected to one another. So if you have a year with high precipitation you'll likely have low vapor pressure deficit because that year will be cold and there'll be more water vapor in the atmosphere. But both of these can also be affected by internal variability and they may also be affected by anthropogenic forcing. So I'm just gonna guide you through this diagram which is showing on the horizontal here is showing precipitation. So we've got wet years on the right and dry years on the left. And then the vertical axis here is showing vapor pressure deficit. So we've got high vapor pressure deficit on the top and low on the bottom. And so what you see here for now is just the observed annual averages of these quantities. And these points are color coded according to the year. So we're going from the dark blues here in the 1950s up to the white in the 1990s. And these are the differences of each year relative to the average from 1950 to 2000. So you see this relationship here between precipitation and vapor pressure deficit. Typically a wet year will have a lower vapor pressure deficit and a dry year will have a higher vapor pressure deficit. Now here in the contours is our kind of model-based representation of the statistics of these quantities. It's basically showing what is the probability of a given year sitting at a given location in this diagram with the darker colors being a higher probability. So our model suggests that we're most likely to sit around here with very small anomalies in these quantities. We're less likely to sit out here at the edges of these contours, but we're really not likely to sit out here or here. It's not likely that we'll have a year with high precipitation and a high vapor pressure deficit. And so we can kind of take some comfort here that our models seem to simulate the statistics somewhat well. All of our observation points here kind of sit within this cloud of contours that our models suggest is kind of where they should sit. Now here is showing in the oranges and going towards the reds is the last 20 years. And what we've seen in the last 20 years is this clear shift towards higher vapor pressure deficit. And then you have 2020 is kind of out on its own here. It's got very low precipitation and very high vapor pressure deficit. And so we could ask, well, suppose there was no climate change and we were just persistently in the climate of 1950 to 2000, what would be the chances that we'd have a year like 2020 in terms of both precipitation and vapor pressure deficit? And so using our model statistics, we find that the chances of having a year like 2020 is about for precipitations about 2%. So we'd expect maybe one in 50 years to look like 2020 in terms of having this low precipitation, which is kind of consistent with the fact we've seen it once in our observational record. But when we look to vapor pressure deficit, our model suggests that in the climate of 1950 to 2000, it was impossible to get a year with as high vapor pressure deficit. There's over 2000 years of simulation going into this calculation and not one of them has vapor pressure deficit as high as what we saw in 2020. But of course we have climate change and the climate has changed over the last couple of decades. And so in orange here, this is our model simulated probability distribution, but now for the climate of 2010 to 2030. And so the forced climate change largely due to CO2 has caused our climate to shift up towards higher vapor pressure deficit, which is very aligned with what we've seen in observations. And if we now ask that same question, but in the 2010 to 2030 climate, and we ask what's the chances of having a year like 2020? For precipitation, it hasn't really changed. There's a slight decrease, but I don't think that's significant. But for vapor pressure deficit, it has now become possible. It's still very extreme. So even accounting for climate change, there's still only about 0.4% chance that you'd have year as high with as high VPD as 2020. But it's gone from being basically impossible in the pre-2000 climate to now being possible. And so now we can ask, what should we be expecting in the coming decades in the US Southwest? And this is what our model predicted distribution of these quantities look like. So this is for 2030 to 2050. And you see that we continue to move up towards higher vapor pressure deficit. And we go, when it comes to thinking about what's the chances of having a year like 2020? Again, for precipitation, there's no real systematic change, but we see this clear trend toward having more years like 2020 when it comes to vapor pressure deficit. And our prediction suggested within the next couple of decades, about one in 10 years would end up looking like 2020 in terms of vapor pressure deficit. So this is to kind of summarize what our projections are for the US Southwest in these particular quantities. We see real no systematic change in precipitation on average in our models. There's not really any clear suggestion that we should see more years with precipitation as low as what we saw in 2020 due to anthropogenic forcing. But that's not true for vapor pressure deficit and anthropogenic forcing has really allowed a year like 2020 to happen. It was really impossible or extremely unlikely to get it in the climate of the late 20th century. And it has now, even though it's still not very extreme, it has now become possible. And moving forward, we should expect about one in 10 years will look like 2020 in terms of vapor pressure deficit. So hopefully we've given you an illustration of the various ways in which we can try and predict the climate for the first coming decades. And so just a few take-home messages here before we take some questions. So really there's two factors that contribute to changes in regional climate from decade to decade. We have internal variability and then we have forced variability and change. And so our initialized decadal predictions, these have the potential to harness predictable internal variability by initializing based on observations and also including that forced signal. And these show promise, but they're less developed compared to our shorter-term weather predictions. We can see that the North Atlantic Ocean variability is a big player. That is a lot of low frequency variability and we're able to predict that. And it offers the potential for predictions in the continent surrounded the Atlantic on fairly long time scales. And then of course, the work is ongoing to find other sources of predictability and to improve our models to improve these decadal predictions. When it comes to the forced climate projections, the forced component is going to be an increasingly important factor. And I think we can already see that in our records in the Southwest. We do have uncertainties here and we may have glossed over them in the interest of time, but I'm happy to take any questions on these things, but we have uncertainties in the forced things. We don't know exactly what we're going to do in terms of our greenhouse gas emissions. We have uncertainties in the model response and I definitely didn't go into the details of that here. I took all of the models together, but not all of the models say the same thing and we're continually working to try and understand why certain models do some things and other models do different things. And then of course, we have unpredictable internal variability, which is always going to be a source of uncertainty and it's kind of irreducible. But I think in here in the Southwest, despite these uncertainties, we can have some confidence that things are going to get warmer and unfortunately that's going to lead to more eridification and more extreme drought and wild fire season. So with that, I guess we'll finish up and be happy to take questions. Awesome, thank you so much. So we have a couple of great questions lined up already in Slido. And again, definitely if you have any questions for both Mila and Steve, pop those into Slido so we can get to them. I'll start off with a slightly more easy one that hasn't been asked yet. How did the two of you get interested in this type of work? You want to go first, Steve? Well, when I first came to NCAR, I started working on the problem of trying to reconstruct ocean variability over the past decades. And that was basically trying to kind of, it was the germ of trying to address the problem of initialization, right? So we had never really done initialized decadal climate prediction because of the problem associated with lack of observations. We don't know much about what has happened in the ocean because there are so few observations. But with our models, we're able to generate reconstructions. And once you have a reconstruction of the best estimate of what was happening in the ocean in the 20th century, then it's kind of natural to try to initialize your couple of climate model and see if you get any skill. And I think we were all very surprised to find that we actually do have skill on decadal time scales when we do that. I'm a little embarrassed by my answer. I guess how I got into the field in general is I studied physics and then I didn't really know what to do. And then the day after tomorrow came out and I find it very exciting. And so then I was like, oh, that's interesting. And so yeah, I started working in the field and then came to NCAR and met Steve and got into decadal predictions as well. Thanks for sharing the stories. So diving into the audience questions. So our top rated one right now is from Mark and I'll rephrase it just a little bit. Who's wondering, I guess going back to the idea of initialization versus not initialize this. And if you use data from a past event like the African drought in the 80s to initialize or train a model, is it obvious that that model should then predict that drought and also maybe tack on like why is it important to know if or if not your model will hind cast a known event in the past? So like, should I go Steve or... Yeah, go ahead. I guess this is a question that often comes up and I think it stems from, it's very difficult to explain what climate models are. And I guess that they're actually not trained on historic data. They're based on physical principles. And so certainly some things we look at like the global mean temperature. If we don't capture it to some degree, then we start to worry that something's wrong in our model. But something like the Sahel drought is not at all included in the training of our model. We start from the physics and then once you're getting those physical processes correctly, then you can start to simulate things like the drought, but the observations have not gone into that rather than in the initialization which helps you to get the factors that are important. Great. So our next question comes from Mateus who's wondering if you have linked your climate predictions to wildfire occurrences, air quality and other societally relevant aspects for the Southwest and Colorado in particular? Yeah, so definitely there have been studies that have shown very clear links. Not my studies, but other people have shown very clear links between vapor pressure deficit and wildfire occurrence and then subsequent I guess air quality impacts. There's very clear connections there. And yeah, it's something that people are working on, but yeah, definitely it's been done and it's continuing to be done. Great. And our next question comes from Michael who's wondering what are the prospects for decadal forecast of acute climate hazards like floods, tropical cyclones, heat waves, et cetera. Is there a chance of skill here or will we mainly only capture the means? Yeah, so I can take that one. So I think everything we showed today was focused on the mean statistics but as I illustrated in my schematic the entire PDF shifts and so you would expect skill in the extremes as well. And there have been studies that have shown that we have potential to predict the frequency of occurrence of Atlantic tropical cyclones in particular related to our ability to predict Atlantic sea surface temperature. Other work has also looked at heat waves over Europe. I think showing sort of modest success but as Ila showed, we do have this problem in our models of connecting ocean variability to impacts over land that is ongoing work. And then there's the question of resolution. So these are very expensive ensemble simulations. And so it's hard to get down to the regional scales where we would be looking at things like floods or kind of more local impacts until we can start to do these experiments with high resolution climate models. And the next question from Nick maybe actually ties in a little bit into that discussion about resolution and they're wondering what are the biggest challenges in improving the Cato forecast? You talked already a little bit about unpredictable noise like atmospheric and oceanic turbulence is there? Maybe any other challenges or limitations we're running into? I think the resolution could be a big one. I think we can see that we are not getting some things right like the connection between the ocean and various aspects of the circulation and it could be that we need to go to very high resolution to capture those things properly. And then we have the challenges of the computational expense. If we wanna run 40 members and many thousands of years go into that then it becomes impossible at this point. And then Steve has more to add. Yeah, I was just gonna mention that there's a lot of interest right now in what's called the signal to noise paradox which is something coming out of initialized prediction experiments like we've discussed today which suggests that the real world is more predictable than our models which is a very curious result. It suggests that there's something flawed in our couple of climate models that is limiting predictability more than it should and that if we could improve our models we have actually great potential to predict climate on these timescales. Which maybe also segues nicely into the next question from George who's wondering what are the error bars on predicting future climate? And Steve, you talked a little bit about that in your section. Yeah, the error bars are huge. So there are many error bars here to consider one being uncertainty in the scenario as Ila was highlighting at the end of her talk. So we don't know what atmospheric composition is gonna be like in future decades. We have to guess. And then there are error bars that are large associated with our imperfect knowledge of the Earth state, right? We've only had global coverage of observations of the Earth since satellites went into orbit back in the 80s. And so our knowledge of where to initialize our models from is very flawed and is full of all kinds of assumptions. So those are just two big error bars on the prediction problem. I guess one thing to add is that how Steve illustrated the error bars will also depend on what kind of spatial scales and time scales that you're thinking about. So if you're thinking about long-term averages for the global average, the errors are probably quite a lot smaller than they are if you're thinking about a local region for the next 10 years. All right, that makes sense. So our next question comes from Ed who's wondering if folks can find your forecast for the next couple of decades? If they're published anywhere or are they linked into something? Yeah, so our forecast, the DPLE and the LE that we talked about are available. If you Google NCAR DPLE, you'll find a website that links to the experiment description and the data. But I also included a link in the Explorer Series material that will take you to a webpage of the World Meteorological Organization, Real Time Decadal Climate Forecast. That's a operation run by the UK Met Office that gathers these decadal climate predictions run by centers around the world into kind of a single place so that you can actually interactively generate maps of what we're predicting for temperature precipitation from the next decade. Great. And this next question from Matias, I think it's a great one because I'm sure many scientists have thought about this. Is there a key question that has alluded you so far? And do you wish you could finally figure it out? So this is like, which question? Yeah, what is the- Which question do you really wanna know? Well, I really wanna know why we don't capture the connection between the sea surface temperature variability in the jet stream because I feel like it is highly connected to the signal-to-noise paradox that Steve mentioned and it has all sorts of implications for not just for prediction but also for thinking about long-term climate projections. If we're missing some key process that in our model, yeah, how that would change our climate projections is, yeah, we won't know until we can figure it out and get it into our model. And funnily enough, Nick's next question kind of ties into what you were just talking about, Isla. How can we better predict ocean temperatures given they are important to predict the cable precipitation and temperature changes? I think that might be more a question for Steve since he's an oceanographer, but- Yeah, so this is coming back to how do we improve the prediction system and resolution is one key area of research. We're trying to do these sorts of experiments with ocean miles that resolve ocean eddies. So this is the ocean turbulence that I put up in one of my slides. So if we have ocean miles that are really much more realistic in terms of their variability, does that help us predict some of the atmospheric impacts? That's a big question that we're exploring. But also to better predict the ocean, we need to better observe the ocean. So our ocean reconstructions, we know are flawed and we need some knowledge of what's going on in the deep ocean, which is terra incognita, right? We really don't know what the ocean, what happens in the ocean below 1000 meters because we have so few observations of the deep ocean, but there are suggestions that important things are happening that impact the surface climate. Great, and our next question comes from Gad and I'm assuming it's referring to the plots at the end that you were showing, Isla. Why is there such a large discrepancy between precipitation and vapor pressure deficit despite there being a large correlation? Yeah, that's a good question. So it's really because the vapor pressure deficit depends on two things. It depends on how much water vapor there is in the air and that's very correlated with precipitation, but it also depends on how much water vapor the air can hold, which is dependent on temperature. So it's really the fact that we're getting pushed into a new state in terms of temperature, which means we're kind of coming outside of what we normally see in terms of that relationship between precipitation and vapor pressure deficit and moving us up to just to a higher vapor pressure deficit because of that temperature effect. And then on top of that, we will still see that correlation between precipitation and vapor pressure deficit, but it's because it VPD depends on both temperature and factors that relate to precipitation as well. Awesome, and our next question comes from Jane who's wondering, so we don't expect more drought in the next 20 years in this region and Jane, I'm assuming the Colorado Front Range here, but do we expect more extreme wildfires? Well, first of all, to clarify, I think we definitely do expect more drought in the next 20 years, but it maybe it depends on what your definition of drought is because there's really those two effects. So there's the effect of precipitation and we don't see any clear trends in precipitation, although of course we could have a year that has low precipitation, but there's also the effect of temperature and that increasing vapor pressure deficit and that is going to lead to droughts being more severe. So even a less severe year in terms of precipitation could become more severe in the future in terms of drought because of that, those rising temperatures. And so coming along with that, we do expect more extreme wildfires. The wildfires are very correlated with vapor pressure deficit. And so I think we should expect both more drought and more wildfires moving forward because of that temperature effect. Yeah, thanks for clarifying that. So our next question comes from Krishna Kumar. Does climate forecasts overestimate extreme events, especially over the tropics or like statistical models, do they underestimate the extreme events? I guess it depends on what's meant by extreme events. If we're talking about cyclones, the models that we use underestimate cyclones because they can't simulate them properly. But I don't know if Ayla wanted to chime in on this one. Yeah, I agree. I think it probably depends on what you're looking at. I think something like heat waves, depending where we are, we probably do quite well at simulating extreme heat events, but yeah, tropical cyclones, we don't capture them because of the resolution. We also have some issues with extreme precipitation as well because of resolution. So it depends on which phenomenon you're talking about. I don't think there's a clear rule one way or the other. And our next question comes from Melody. So when initializing from observational data, the models give better decadal prediction. Have you compared the model skill with statistic methods like machine learning? I have not. I don't know. No. No, I'm not aware of any study that does a direct comparison of skill between sort of a full dynamic prediction system sort that we presented and a pure statistical method. I mean, I think statistical methods can be extremely skillful for very focused sorts of predictions like tropical cyclone frequency in a particular region. But the obvious benefit of using a full model like we're doing is that we can look at the whole host of processes that take place all over the globe and we're not kind of limited in terms of what we can look at for prediction. Right. So next question comes from Clay. How much does La Nina and the Pacific play into these models for the Southwest? And I love questions like this because it really gets at how interconnected our earth system is. Yeah, so definitely, well, maybe Steve, do you want to speak to the prediction system itself or? Well, I was just going to say that, you know, in the Seager and Ting analysis that I presented, there is a clear connection between drought in the US Southwest and El Mino. But that's sort of an inter-annual phenomenon. Our models can predict and so maybe a year in advance and then the skill degrades. So we're not getting much in terms of decadal prediction skill from being able to predict and so. Right, but it does, I guess there are long-term variations in the Pacific that are very strongly like the decline in precipitation that we've seen over the last couple of decades is thought to be linked to cooling of the tropical Pacific. And so if we could capture that in our prediction systems, there is a chance of having skillful predictions of those kinds of long-term trends as well. And going back a little bit to maybe some of the limitations in terms of doing large-scale computational models, Skip is wondering, what do you think about the rapid advancement in supercomputing has had on increasing the accuracy of these forecasts into the future? It's been instrumental, right? So we, decadal prediction as a field kind of was born in the 2010s and it was directly related to increase in supercomputing power that allows us to run these large ensembles. So we're now in the 2020s entering an era when we can start to try this using high resolution climate models. And we're just waiting for quantum computers to finally appear so that we can really get down to very, very local scales with these types of experiments. Great, and we have time for this last question. And then I have a final question after that for y'all. So Nick is wondering, since image was shut down at NCAR, is there still statistical analysis of NCAR's modeling taking place? Yeah, I mean, I think there's less of the kind of pure statisticians around but yeah, as climate scientists, we're doing statistical analysis on the models. Yeah, but yeah, that group is not here anymore. Yeah, and just to wrap up, for all of our students that are out there watching this lecture tonight, what is some of y'all's advice for folks that are maybe interested in giving into this type of work? Steve, do you wanna go? Go to grad school. Yeah, find a good graduate school advisor who's interested in these types of problems. That's the cleanest and clearest path to getting a job like we have at NCAR. Yeah, I agree. Great, and with that, I just wanna thank both you, Steve and Ila, for being here today to chat with us about Decado climate prediction. I also really wanna thank the team behind the scenes, Paul, Brett, and Malia for supporting our event today. And while we're done with the 2021 Explorer series, if you're interested in past Explorer series events, definitely check out our website for recordings and stay tuned for the announcement of our 2022 series. And so with that, I hope to see y'all in 2022 and I hope you have a great rest of your day. See ya.