 So I'm going to be talking about the influence of 21st century climate change on Southwest precipitation, and in particular California precipitation, and how internal variability changes rain and snow, and with it the predictability. So California is in a drought, and it's a big deal for us in California. It's one of the most severe droughts on record, largely attributed to record lows and precipitation, and records in high temperature. That's led to very stringent water restrictions by the governor, and that's affected our everyday lives. It's also affecting the agriculture industry, which of course affects the state's economy. So you hear about this every day in the news in California, and you hear about it probably every week in the national news. And you can Google it, California drought, any day, and you'll get some new news story. I read that a couple of nights ago, and I got some story in LA Times. This one in particular, the mandate on California water cuts slated to continue if drought persists. And so what this particular article says is if the drought continues on through January, the water restrictions are going to be extended another six to eight months. And so the reason why they think it might end in January is because of the big old Nino that's coming. But so the point is this big news in California, but California's experienced drought before in the past, and the Western U.S. has experienced drought before in the past. So here's a plot, a reconstruction from Ed Cook's paper 2004 from a collection of tree rings throughout the Western U.S., showing the percent of the Western U.S. under drought, where the higher values indicate drier than normal conditions, and the lower values indicating wetter conditions. And this reconstruction goes back to 800 up to the present. And you can see that around 1,000 years, you start getting a lot of high values associated with the mega droughts. And then over time, these mega droughts don't seem as prevalent, and you start getting more dips in this time series associated with poluvials. And then if you actually focus on this last interval, last 100 years, you can see that maybe there's a slight increase that's occurring too. So if you have a lot of mega droughts that occurred around the medieval climate anomaly and you have a slight increase that's occurring in the last 100 years, you might want to make an assumption maybe that these periods where the northern hemisphere is really warm or associated with large mega droughts or drier conditions in the Western U.S. and maybe over California as well. And that seemed to be the case for the CMEF-3 projections in the 21st century. But then CMEF-5 came along and said, no, no, no, not so fast. And so what's applied here is the wintertime precipitation for the RCB 8.5 scenario. So this is a collection of 78 different models. So it's the 30-some models that go into the IPCC projections with all their ensemble simulations. And what it shows is, so the blue color show wetter conditions, and the reds and oranges show the drier conditions. And you can see that on average they show during the wintertime, it's wetter in the 21st century, if you believe the models. So you might look at this and say, okay, this is great news for California, great news for California residents, and definitely good news for the water resource managers that have to deal with maintaining the water supply. But that might not be exactly the case. So here's a slide that came from a talk from the AGU Chapman Conference last spring on California drought. This is from a talk given by Janine Jones, who's one of the top water resource managers in California. And so she lists certain questions that the water resource managers aren't asking, she's presenting to a bunch of scientists. So I'll just go ahead and list them. So how much water will it take to end the drought? Can we close the water budget? What caused the drought? How can drought monitoring be improved? And anything containing the word drought index. And so the scientists didn't really like this slide very much. Reason why is because scientists love asking these types of questions when it's related to drought. So then you might ask, okay, well, what kind of questions are they asking in the water resource managers? So what they're asking for is, at the beginning of the season, how much water will we get this season? And in the middle of the season, how much more water will we get? So they're really looking for better predictions. So California precipitation is very seasonal in the sense that it doesn't really rain or snow at all in the summer time. You know, a little bit of rain and snow in the fall and the spring, and then we get blasted with rain and snow in the winter. And it's really the wintertime snow up in the Sierra Nevada mountains that we really care about because it's the water infrastructures built on the wintertime snowpack. So for this talk, I'm mostly going to be talking about December, January, February precipitation over California. So in a very simple sense, you can break down wintertime precipitation in California between a predictable component and a non-predictive component. So the predictive component is something we've all been talking about the last couple days. Pacific sea surface temperatures and maybe to a lesser extent, Atlantic sea surface temperature influence on California precipitation. The very, the most common and well-known relationship would be the El Nino or Enso relationship. So in general, during El Nino conditions usually leads to rainier winters over California and La Nina conditions usually relate to drier than normal conditions in the state. But then there's also internal variability, which can be quite large for California. So one good example of that would be in 2010, the winter of 2010, 2011, very strong La Nina in terms of the Pacific, tropical Pacific SSTs. But over the last few decades, that was actually one of the rainiest winters over California. So this non-predictive component can be quite large. So the question is, as the 21st century gets warmer and maybe California gets wetter, how do these two components change? So you can imagine over the course of time, this internal variability maybe get reduced and that might help the predictive component. In other words, that would be good for water resource managers. Another possibility is the internal variability might increase, which would make these seasonal predictions even worse. So we'll take that last scenario, increase internal variability, and use that as our hypothesis. So our hypothesis would be changes in the 21st century will cause an increase in atmospheric internal variability in terms of California wintertime precipitation. So how are we going to quantify that? And so this will be very much a review for what we've heard mostly. So imagine we have six ensemble simulations, we're going to use ensemble simulations to quantify the internal variability over time. So imagine we have six ensemble simulations and this is results for the six different members showing projections in winter precipitation. Reds indicate drier conditions, blues indicate wetter conditions. And then you can quantify the forced response by taking the mean, the ensemble mean, the trend analysis. And if you want to calculate the internal variability of one of these members, say this fifth member right here, you take that member, subtract out the forced response and you can quantify the internal variability. So you look for this particular case, internal variability is quite high over Northern California, which is a region that we care about for our problem. So it's really just the variability around the forced response. So imagine we have a variable, this can be temperature, precipitation, whatever, and over time you have a linear trend in the 21st century. So that would be your forced response. So imagine you have one case where you have three different ensemble simulations and there's an ensemble's disagree a little bit, or they disagree a lot, a beginning portion of the 21st century. Then over time they start to agree a little more over time. Then you have a second case where the opposite occurs, where they agree a little bit more towards the beginning of the time period. But over time they start to differ quite a bit more. So that would be more consistent with what our hypothesis would be, an increase in internal variability. So the method we're going to do is we're going to take two different chunks of the time series, two different chunks of the 21st century, take the early 20 years, so the first 20 years of the simulation, it's actually 2006 to 2025. And then we're going to take the surface forcing, the SSTs and the CIS conditions from RCPA 8.5 simulations and run a series of AMIP simulations. So we're going to use an atmospheric general circulation model and force it at the surface with sea ice and sea surface temperatures. And then we're going to do the same thing for the last 20 years and then look to see how the variance changes for precipitation over California. Okay, so the model that we're going to use for this is called the global spectrum model. It's actually called isoGSM. It's actually got water isotope traces in it. But I'm not going to be showing any isotope results. So it's a T62 resolution, run it at about two and a half minute time steps. And there's about 28 vertical layers in the atmosphere. So it also has a land model, it has the NOAA land model. And it's been, one of the good things about GSM is there's been a few studies that have validated the precipitation over California pretty extensively, mostly with precipitation and using the isotopes that are built into the models. So we have a pretty good confidence that the model does pretty well over our study region. Okay, so like I said, we're going to take the first 20 years and we're going to choose two different models. We're actually going to, in the future, what we're currently doing is using more CMF5 models to do these experiments with. But the results I'm going to show today are just for the GFDLCM3 and the IPSL CMF5LR simulations. So again, these are RCP 8.5 scenarios. We're going to run it. We calculate the seasonal cycle, SSTs and CIS distributions. And so it's just a repeating seasonal cycle over and over in these simulations, these AMIP simulations. So we're going to do three different runs of each one of these cases, so each one of these models for the early 20th century. And do it, run it for 50 years and we're going to do it three times. So it's really 150 years of modeled simulations. Let's see. And then we're going to do the exact same thing, but for the last 20 years, for these same models. We're going to do it both with and without the CO2 and methane and greenhouse gas emissions, greenhouse gases in the atmosphere to quantify how much of the variance is due to the actual surface forcing and how much of it is actually related to actually putting the greenhouse gases directly into the atmosphere. And then eventually what we're going to do is we're going to calculate the variance, simulated precipitation variance and we're going to use an F test of variance to test the, well, not the variance actually changes. Okay, so just to give you some sort of sense of what the model does. So here's the results of the E21 GFDL simulations. So like I said, there's three different runs and they run for 50 years. And this is just calculating DJF precipitation for this box region right here over California. And you can see that the model for these three different cases, the mean precipitation is roughly about the same. It's about 3.9 millimeters per day. But there's quite a bit of variability. In fact, you see all these large variations. There's no changes to the surface forcing. The only changes is really in the essence of the initial conditions are different. So you get a lot of variability about this mean. But you can actually treat each one of these simulation years as a different model realization and put these onto a histogram. And so that's what's done here. So here's the histogram. The thin line here is actually the histogram. And then the thick curve is the Gaussian fit to the histogram. So it's really the, but it's really the shaded region that we're going to focus on. So the shaded region is the variance about the model mean, I guess. And so the variance here is 1.7 millimeters per day. Okay, so that's good. But when you compare that with the late 20th century, you can see the distribution shifts a little bit. So in addition to a slight increase in the mean, there's also an increase in the variance. It goes from 1.7 millimeters per day to about 2.3 millimeters per day. It's about 35% increase in variance. But when we performed an F test on each one of these two sets of data, we found that the probability is only about 0.07. So it's not quite at the 95% confidence level. But if you put CO2 and methane and the other greenhouse gases in the atmosphere, we find the variance actually increases quite a bit more. In fact, it goes from 1.6 now to 3.1 millimeters per day. So almost a doubling of the variance. And when we performed the F test, it was significantly different at the 99% confidence level. So that's a GFDL case. So what we found for the GFDL cases, there was an increase in the internal variability. For the APSL case, here's the early 21st century. What we found relative to GFDL is the precipitation is higher and the variability itself is actually higher. But when we actually force it with the last 20 years of the 21st century, we find that the variance increases for both with and without the CO2 and methane gases. So what that's saying for both of these cases is that something going on in the surface forcing that's causing the internal variability to increase. And then sometimes, or at least for the GFDL case, adding the greenhouse gases actually increases the internal variability even more. So I'll admit I haven't totally diagnosed why this would be the case. Obviously, it has something to do with the surface forcing differences. So what's applied here is the SST differences for the winter time for the two cases. Left shows the GFDL. The right shows IPSL. Oranges and yellows show like the regions where there's warmer SSTs. There's warmer SSTs everywhere because it's the RCP 8.5 scenario. But if you compare these two, you don't see in terms of the Pacific Ocean temperatures, you don't see a pattern that's really that consistent. Maybe you have something coming off the Eastern North Pacific, but there's really not a consistent pattern. One thing that really stuck out to me maybe was that they all have these sea ice laws. So maybe this has something to do with Arctic amplification. And there have been studies that have looked at Arctic amplifications and how that changes the jet stream and how the changes in the jet stream might affect things like weather and climate extremes. So when you look at the jet stream, so here's just the general geopotential height at 200 millibars for the GFDL case early 20th century or early 21st century. And you can see that there's a permanent ridge right along the coastline and then there's permanent troughs on either side of it. And if you subtract this out from the late 21st century case, you can see how this changes over time. And you see that this ridge, of course, right here actually gets even stronger. And there's a dip to the left and a dip to the right or to the east or to the west and the east. So what this shows is that what jet stream is actually getting wavier. And maybe this is has something to do with increases in internal variability. But this is all sort of I'm not totally solving this quite yet. But anyway, back to our hypothesis. We have hypothesized that changes in the 21st century will cause an increase in atmospheric internal variability in terms of California wintertime precipitation. And we haven't done a complete set of simulations, but our results thus far would indicate that this is true. But going back to why we even care about this to begin with, will this affect California water resource managers? And our results say that seasonal precipitation in California might become more difficult to predict in the 21st century just based on our results. So there are some limitations in our approach here. Very simplified AMIF simulations. But our results would suggest that the seasonal predictions will become more difficult. So I'll leave it there.