 It is my pleasure to introduce Mike DeFlorio, who will be giving the next talk. Mike is a research analyst at the Center for Western Weather and Water Extremes at the Scripps Institution of Oceanography at the University of California, San Diego, and he is interested in a wide range of S2S, sorry, of regional topics, including the S2S predictability of atmospheric rivers and precipitation, mid-latitude, teleconnected responses to tropical forcings, aerosol, cloud interactions, global coupled climate modeling, and California extreme precipitation events. Mike, welcome. Thank you, Judith. I will share my screen and go into full screen. So yeah, I just wanted to thank both Judith and Anish for all the hard work of organizing this and for inviting me to speak, and also Jaime and Rich for such great talks preceding me. So yeah, as Judith alluded to, I'll be talking today about S2S forecasting and some underlying challenges associated with forecasting atmospheric rivers, ridging, and also just precipitation more generally over the western U.S. So this has kind of been a theme of the workshop, in the last few weeks, but we know that many end users in the applications community are interested in improved S2S forecasts of atmospheric quantities. So this is a sort of flowchart described in that from White et al. 2017, and at the top there we have this kind of demand from end users which starts the interest in improving S2S forecasting. So there's a reliable and actionable information that is needed for decision making. And then going further down, we see that there's a variety of lead times and a variety of actions or decisions that can be influenced at those various lead times. And I'll show kind of an application of this chart to water management in the West in a few slides. But this is just, I thought, a nice articulation of why there's been such a push really in the last five years or one of the primary reasons there's been such a push in aiming to improve S2S forecasts. It's been a long-standing problem, right? This is, you know, especially on the seasonal timescale. It's been many decades of research that have led us to where we are now. But it's this fundamental demand from end users that has driven a lot of the recent concerted effort in improving S2S forecasting. So in the West we're in a particularly unique region for trying to make progress on this issue. And fundamentally the challenge that end users have to deal with here is that we experience in our region the largest inter-annual variability of wintertime precipitation relative to average conditions. So this is a map that's articulating that. It's showing the coefficient of variation in year-to-year precipitation normalized by average precipitation. So the take-home is that if you look, for example, in the eastern half of the U.S., the dark blues and light blues indicate a low coefficient of year-to-year variation, meaning that relative to the average conditions, there isn't much inter-annual variability in wintertime precip. Whereas in the West, and especially Southern Nevada and Central and Southern California, we have relatively high inter-annual variation in this quantity. So just going back for a second, this is really, if you were to articulate in one slide or with one picture, why water resource management in the West is such a challenge, this would be a primary demonstration of that. And we're experiencing that now in California and in the West more broadly with an inter-annual period of extreme drought, which I'll touch upon in a few slides. So just looking, I showed that diagram from the White et al paper, which is really a general depiction of end user needs as a function of lead time for better S2S forecasting. If we apply that to the West and for water management, this large inter-annual variability that we looked at in the previous slide really increases the demand from water managers in our region for better S2S forecasts. So it's just because there's so little year-to-year persistence in the sign of the precipitation anomalies that we get, it really makes it a harder problem for water managers here. So we created this figure, which was, this is an adaptation of a figure that was published in EOS, an article that we wrote and was published earlier this summer, just outlining the specific lead-dependent decisions for water managers in the West as a function of lead time starting from weather out on the left side of the diagram here, all the way out to climate. And in the S2S time scale here, there's a variety of decision support needs that water managers have, and then modes of variability and other physical processes that we know impact predictability. And again, just going back for a second, three of the modes of variability that I'll really be focusing on in this talk and that we heard Haimi and others touch upon are the MGO, QBO, and so modes of variability. So I just threw these slides in this morning. A colleague had just sent over a PDF summary that NOAA and USDA and others have put together, one of the more recent updates for the US drought monitor over the Western US. So this is a map that was released last Thursday. And so you can see that really across the entire region and especially near the upper Colorado River basin into Utah, Southern Nevada, the pipeline from the Colorado River that feeds a lot of our water supply in California, we are experiencing the highest intensity level of drought. So we are in the exceptional drought category in many of those locations. Extreme drought is the second worst, and that's covering all of the other areas in the state of California that aren't in the exceptional category. And some of you may have heard in the news recently, for example, in Southern Nevada that Lake Mead, which is a primary local water resource for that community, is approaching its lowest levels in the historical record. And it may have reached it. I haven't looked at the most recent update in the last few days, but it's a situation that's worsening day by day. And in California, it's triggered the declaration of emergency drought measures from the state, the governor's office. And this is just another way of looking at that, the drought in this region. So this is a percentage of normal precipitation for the last year. So going back to last July and up through last week. And so again, especially in Southern Nevada, Central and Southern California, you're seeing the percentage of normal precipitation values all the way down to 20 to 30%. So just highlighting a current ongoing situation in the West that is really relevant for the objective of the community to improve S2S prediction of precipitation. So this is a really nice schematic from the map effort, the S2S prediction task force put together. I think it's probably been presented already in the summer school. But I thought it was a nice introduction to some of the topics I'll be talking about over the West. So it's really just depicting the symphony of physical processes and potential predictor fields that we're interested in over the Western US to try to better the prediction of precipitation and atmospheric rivers and ridging and all of these related processes at S2S time scales. And there's a key model data listed in the upper right hand corner there, which I'll touch upon again in a couple of slides. So I now have a couple of slides of background on potential sources of S2S predictability that are particularly relevant for the Western US. So I've broken it down into seasonal and sub-seasonal time scales. And this is not an exhaustive list of the potential modes of variability that we are interested in. These slides won't necessarily touch upon all of the ways that the individual modes can interact with each other and synchronize with each other to enhance or de-enhance predictability. But we'll start with ENSO, the well-known mode of inter-annual variability in the tropics. And if we just focus on how ENSO is impacting the Western US, this is a really nice simple summary from yet at all 2018. And the left plot here is just showing Z500 and SST composites during DJF composited on warm and cool phases of the El Nino Southern Oscillation. So you get the canonical El Nino pattern composite in the left and the canonical La Nino pattern on the right. So the key take-home here is that this enhanced troughing and ridging respectively can influence the teleconnected response of precipitation conditioned on El Nino and La Nino in our region. So again, this is just the canonical response. There obviously are different flavors of the SST variability in the tropics that will impact the extra tropical circulation response in a mean sense. And of course, because there are other confounding factors in the climate system, ENSO will only be one such component of the climate system that affects the seasonal anomalous distribution of rainfall in the West. But it really is a motive variability that has caused issues for scientists in the West from a PR perspective because the canonical relationships don't always work out. But communicating that to stakeholders when it doesn't work out is a real challenge. And so we should just keep in mind that it's still a foundational component of many of the prediction systems statistically more looking at dynamical systems that can help us make progress on this issue. So shifting to the sub-seasonal timescale, so thinking more about like two to six weeks lead time. So ENSO, because it's slower varying, you wouldn't expect it to have as much of an impact in isolation on predictability on those timescales. So as Jaime mentioned in her talk and gave a really nice overview of, two of the potential sources of predictability that we're interested on these timescales for the West are the MGO and QBO. So they're interesting in dependent of each other and particularly interesting as lots of research in the last five or six years has shown when we combine them and look at their their phase locking with each other. So this was a figure from a really nice paper out of Elizabeth Barnes's group from Colorado State, Tom's et al. 2020. And it's just looking at the fraction of inter-seasonal variability. So it's a squared coherence metric for I believe it was 250 hectopascal geobtential height patterns. And the top plot is showing the squared coherence conditioned upon all QBO phases. And then the bottom is showing if you do the composites just based on westerly or easterly QBO. And so the point being that the QBO, independent of the MGO, can have a large impact on, you know, modulating large scale circulation across the globe really, but in particular across the Western US. And in a couple of sides, I'll go a little bit more into detail about the interplay between the MGO and QBO and some studies that have shown how they interact with each other. So we'll start that sort of train of thought now. So this is looking now at impacts first on seasonal timescales considering Enso. So this is just from observations. There's no prediction system involved in this plot or in this analysis. And this is just demonstrating the well-known observed canonical correlation between wintertime leading principal components of tropical Pacific SST variability and seasonal precipitation anomalies over the Western US. So this is just using, you know, observed data and looking at what the typical or what the correlation is between those leading PCs and the precip anomalies. And what comes out of it is a very well-known sort of dipole pattern where the warm phase of the El Nino Southern Oscillation corresponds typically to wetter conditions in the Southwest. And you get sort of the dipole pattern that sets up with a bifurcation point around Northern California where you get the opposite sign correlation in the Pacific Northwest. If we look at sub-seasonal impacts, again, just considering observed data, if we focus first on impacts of the MGO and the QBO, which we outlined sort of the physical processes in the previous slides, if we look at the impacts on Western Canada and US AR activity, this was a really nice paper from Mundank at all 2018, which first looked at these kind of lagged atmospheric river activity responses to different phases of the MGO and conditioned upon easterly and westerly QBO conditions over different regions along the Western North American continent. So just showing the areas in green, for example, corresponding to increased AR activity and the areas in red corresponding to decreased AR activity. So again, just appreciating that the phases of these two modes of variability, particularly when considered in sync, in synchronicity with each other, can have substantial impacts on the modulation of AR activity and also skill in predicting ARs. And then another way to look at this, this is some work that's being done by Zhibao Wang, who's a postdoc in our group at CW3, and she is looking at kind of a similar, you know, the similar idea of conditioned, coupled MGO and QBO conditional composites of, in this case, flooding, extreme flooding duration, frequency and intensity over California. And again, without getting too much into the granularity of the details, I know there's a lot of results in that plot, but just seeking to appreciate the large impact that the phase and amplitude of both of these modes can have on these, you know, Western US water-centric metrics. And that's kind of the point of both of these studies. So another quantity of interest in addition to precipitation itself, you know, that's kind of the end goal for water managers. But we know even on weather timescales that precipitation is, you know, is really difficult to predict. It's actually harder to predict. There's evidence suggesting it's harder to predict than IVT associated with atmospheric rivers. So one approach that is really interested water managers in the West is to look at the interaction of different ridge types and circulation regimes with the Western US AR and precipitation anomalies. And, you know, this is kind of a nice follow-up to Rich's talk, which focused on blocking. And he mentioned, you know, that there's lots of ways that you can characterize, you know, circulation regimes and blocking, you know, globally. And one approach that was led here by Peter Gibson in a paper that came out last year is looking at a combined EOF of the Z500 anomalies over the North Pacific and Western US, so the region that's shown there in Figure A, with the daily land precipitation anomalies over the Western US. So this is just seeking to identify the co-varying circulation and precipitation modes that are the most prevalent in the historical record over the Western US. And so in Peter's analysis, he pulled out three dominant ridge, or three dominant ridge types that are associated with significant drought over California. So that's, they're sort of summarized in Figure B. It's a, this is kind of a depiction of the ridge detection algorithm that was used in this study to do this. But basically, there is a North Ridge type centered in that top box, a West Ridge type in the lower left box, and a South Ridge type in the lower right box that show up as dominant modes of circulation associated with drought in California. And what Peter did, among other things in this study, was looked at composites of IVT in the left column here on the right figure, and also relative risk of atmospheric rivers and precipitation anomalies conditioned upon the prevalence of those different ridge types. So if you just look at the top row, for example, when we see days in the historical record with North Ridge type conditions, we tend to see enhanced IVT pluming into the Pacific Northwest and a deficit of IVT over California, which is kind of consistent with the precipitation patterns that are, I believe it's maybe EOF2 here that's showing that in the combined EOF methods. So the take home without going into each individual figure here is that there are many different ways to do this sort of analysis, but stakeholders are really interested in this combination of circulation regimes with the resultant IVT atmospheric river activity and precipitation anomalies in the Western US with the hopes that the circulation regimes themselves may have more extended prediction skill than just the total precipitation by itself. So that's, you know, kind of a conclusion of the background of some of the physical processes, the predictance of interest that are impacting water management in the Western US and that are kind of motivating water managers in the West to invest in this research and really collaborate with researchers in designing these applied research projects. And so a key component of, you know, making headway on this sort of issue is that once we design, you know, these prediction systems, we need to evaluate them. And so without spending too much time on this, because I know a lot of the previous talks have gone into detail on this, but we can do this evaluation in really three fundamental ways. We can use dynamical ensembles. So we've heard a lot about the SubEx dataset, S2S database, and also NMME for more seasonal time scales. You know, those are dynamically rooted ensemble systems that we can evaluate the skill for predicting any of these metrics. We can create statistical and machine learning models, which we'll hear more about later in the week in a session. And those could involve, you know, canonical correlation analysis, EOFs, and then also emerging machine learning techniques like deep neural nets and random forest. And then another way that we could tackle the problem is to do sort of hybrid approaches. And one method that our group at CW3 has looked into is training machine learning models, you know, some of which are listed in the second column there, on large ensemble climate simulations. So with the hopes of generating a larger sample size in the historical record over which we can do this evaluation. So in the interest of time, I'll go through these pretty quickly. The next two slides I'll toggle back and forth are really just showing you an example of the modulation of sub-seasonal atmospheric predictions, atmospheric river prediction skill in the ECMWF hindcast system as part of the S2S database over, in this case, Central California. So what we did here is created a rock diagram plotting hit rate versus false alarm rate. And we created this evaluation system where hits and misses and false alarms and correct rejections were defined based on the relative position of observed and forecasted AR centroids using the Gwannon-Walliser AR detection algorithm. And this is just highlighting here. If you look at the triangles here, these are weak to lead time forecasts in the ECMWF prediction system. The black is showing the false alarm rate and hit rate that you would expect if you considered all forecasts. And then the red here is the hit rate and false alarm rate that you would obtain if you just considered hindcasts that were initialized during MJO phase eight conditions. And this is just highlighting an example of a forecast of opportunity in the ECMWF hindcast system where you see a 15 to 20% reduction in false alarm rate in week two when the MJO is in phase eight when you initialize your forecast. And if you go back to this slide, there's a similar picture that comes up in a previous study that we did which looked at subseasonal skill assessment in a little bit of a different way. But the point being that some of this early work on evaluating S2S database models is focused on identifying these forecasts of opportunity at subseasonal time scales over the Western US region. And it sets the stage for what we know and what we don't know, what are the prediction skill limits in a variety of these models over the Western US. It sets the stage for other approaches that we use later on. We've also used in this case five of the S2S database models to do an evaluation of subseasonal prediction of ridging. This is work that Peter Gibson has led. And this is again used five of these models for evaluating the ridge types that I introduced in previous slides. And I won't go again too much into detail here, but this is just highlighting another way that you can use the S2S database and the models therein to look at skill assessments at subseasonal lead times. And I will skip this slide. I will present more on this topic. So this is looking at seasonal prediction of precipitation clusters in a hybrid large ensemble climate simulation and machine learning model framework. I'll talk more about that next week during my lecture on Thursday with that. So to wrap up here, one of the end results here for Western US water managers is that they assist us in designing experimental forecast products at S2S lead times based on these research efforts. And this is just one such example of AR activity outlooks that have been created and designed with our stakeholders at DWR based on the AR hindcast skill assessments that were introduced in previous slides. So this is showing daily probabilities of AR occurrence from a sample forecast from this past winter in the NSEP system at week one and week two lead time. And then when we get into week three, you know, out more into the subseasonal time scale, we switch to a weekly aggregate of AR activity and produce this lower right hand plot here, which is an anomalous forecast of AR activity relative to climatological conditions. So we transition from more of a weather type forecast out into a weekly aggregate subseasonal forecast that relies on a hindcast climatology. We also have ridging outlooks that are based on the work that I described that Peter led. Again, these are on our public website during the winter. You can read more about the methodologies associated with them, the relationship to the hindcast skill assessments that were done previously before they get posted publicly. And in this case, this is just a subseasonal product that's predicting the probability of the three different ridge types that were identified in Peter's work as being important for drought in California. So just to wrap up, so more skillful as to us prediction of precipitation is obviously coveted by a variety of end users. We know that the Western US experience is very large swings in total precip, which creates a unique challenge for water resource management in our region. And Western US precipitation and AR activity, we also know, are modulated by the MGO, QBO, and so and, you know, other modes of variability on these time scales. And so we can implement a variety of prediction systems, both dynamically based, statistically based hybrid systems that can be used to evaluate the prediction skill of these quantities over the West. And then we can also use them to identify these forecasts of opportunity where skill can be relatively high compared to the average conditions. When we look at active phases of these modes of variability and also look at how they constructively or destructively interfere with each other. And the end game of this for Western US water managers is they really are seeking to benefit from experimental forecast products that have been vetted in this sort of hind cast skill assessment peer reviewed framework and tailored in a display sense, in a wording sense, and also in the quantities of interest that they're plotting to really tailor directly to water managers. So that's sort of an overview of what we do at CW3 and what a lot of the applied research in the S2S community over the West is focused on. These are some references that you can look at. And I ended here with a photo of Lake Mead, which I believe was taken a few years ago when the water level was a lot higher than it is now. I should have tried to find a before and after picture, but I'm sure if you Google it, you can find something like that. So thanks very much. And I'm happy to take any questions. Thank you very much, Mike. That was great.