 student presentations from the different student projects that were done during the colloquium in the past three weeks. We would like to thank the student mentors again for all the projects for all the work that went into the preparation, especially the hard work put in by all the students for getting impressive results just in a short period of time. So, we had posted an order in terms of the presentation of the projects, the first group that will be the climate group would present up Pauline will you I'm getting an echo maybe if you reduce your speaker volume echo might reduce I think it's a bit better yeah let's try and okay do you see my screen yeah I can see your screen I'll give you a eight minute warning okay okay great I'm like we would be several to to go through the slides and we'll start with the screen okay perfect thank you very much. So our group was working on as to us verification using climp read, which is a Python package that was developed by our facilitator Aaron spring, and there were seven students participating in this total. There we go. So we divided ourselves into four subgroups and just a brief project overview for each of the groups. innocent and I were looking at certain stratospheric warming over the United States. I mean I was looking at state dependent predictability of two meter temperature depending on the nao phases. Jan and Matt, we're looking at state dependent predictability, particularly the influence of the MDF phase at initialization on European forecast. And Pauline and Shinja, we're looking at estimating prediction skills in precipitation hind casts. So the question that innocent and I asked were do SSWs lead to a better two meter temperature predictive skill over the United States. And so we looked at three different models, the WACCM CESM to the ECMWF models, each of these were initialized weekly. We looked at the United States and two meter temperature. The forecast time that we used was 1999 to 2020. And during this period, there were 16 SSW events. And the metric that we used to evaluate was the difference in root mean squared error between the SSW events and the non SSW winter time, which is December And so this difference in root mean squared error can be plotted as a function of lead time as shown where a positive value means that SSW is actually led to a worse prediction and a negative value means that SSW is led to a better prediction. And so going through each of these three models, the WACCM model for lead times zero to 10 actually had worse predictions during SSWs. So at lead times about 1530 days ahead of time had better predictions during SSWs. CESM to model, there was very little difference in predictions between SSW events and non SSW events, and the ECMWF, which is the green line for all lead times had better predictions during SSWs. And just to note that the bumpiness in these curves, we think is likely caused by there being relatively few events that were analyzed. And so general conclusion from our subgroup is that SSWs can lead to better prediction skill, but that's dependent upon lead time and also model. So I'm presenting the results from Irina, and she looked at the NAO and the impact on two meter temperature in two different models in one with normal top, the CSM2, and then one with a higher top WACCM. And she found that for, you can see here the top two, top four figures for lead time of 15 days, there's higher skill for forecast that are initialized in NAO positive for the region of Europe and also most of Asia. And she doesn't really see a difference between the two models so the high and low top doesn't seem to play a high role for this region. And she also looked more specifically for this Eurasia region, more at the temporal evolution of the forecast skill in the two models, and she can see that the forecast skill decays as you would expect fairly quickly. And that for forecast weeks three and four, she sees a higher skill for the Eurasia region using the higher top model. For me and Jan, we're looking into how the MGO phase affects forecast skill over Europe. So we decided to focus on phases two and three and six and seven. The standard patterns which are shown on the left of this slide, as these tend to give the strongest telecommunication pattern towards Europe. So the results of this were that after about two weeks when initiated in MGO phase six and seven, forecast seemed to become much worse than climatology. But this isn't seen in any other phases. I guess to find out how why this happened, we, on the next slide, looked at the geospatial pattern of Ruben Square. So on these maps, the red regions are forecast better when the MGO is in phase six and seven and blue regions are forecast worse. We can see, especially in the weeks three and four, that in the North Pacific, there is a much better forecast skill, but this better skill doesn't make it across Europe. And I guess following up on that, we've been wondering how this relates to the increased bias of the ECMWF model over Iceland in the same MGO phase. So estimating prediction skill of precipitation and after it is corrected by quantum mapping. And as you know that post-processing is necessary before a model forecast can be practically applied. So quantum mapping is a very useful tool for post-processing ensemble model forecasts. And this method because precipitation is not normally distributed. And so simple bias correction is not very applicable. And the theory behind this quantum mapping is to match the CDF of raw forecast forecast and to the CDF of observations. So we applied the quantum mapping in same thread and to the hand cuts of precipitation in different countries from ECMWF model, and we use several different metrics to estimate the prediction skill. And here we take the continuous ranked probability score as an example. And you can see that after the correction by quantum mapping, the skew and improved obviously, and because the lower value of the CRPS, the better skew of the model gives. And we also extended the verification by differentiating it's about into different phases of different processes. So you can see the underlying correlation coefficient that depending on whether we are in different phases of for example NaO or XO or NaO. And so far we didn't see that much difference in skew between the general layers. So we did this just with the first analysis and we hope maybe to find more results with other temporal accumulations of precipitation for example, or by focusing on a substantial intensity of precipitation for example in streams. I hope you'll hear me properly. It's echo, I can hear an echo falling but it's fine, yeah, yeah, and yeah you have one more minute. I'll be quick on the tech messages. So, I'm going to show that generally the presence of certain ones can be to a better model, sorry, and I'm giving a highlighted that there is a better skew for to mature temperature during the position in NaO phase. So I'm going to show that there is an increase in focus error if the focus is initiated in phase C, C7, and of energy. So we investigated, we went through before processing with the help of our own with the development of the Pantanamati in Indian CREB and we analyzed the improvement of the petitions here. And with this quick analysis we didn't notice any difference in focus error depending on different phases. So we thought about some positive elaboration between the groups, for example in San Juan and the Union are working on two meter temperatures and maybe some investigation on compound effect of this C7, which is very long and also a compound effect on two meter temperature prediction can be an option. Also to further the analysis of the impact of energy, BMI, or precipitation, so we were working, and also more generally the analysis of the precipitation can be applied to every method that are in the other groups. So we'd like to thank you very much for your attention. Thank you very much for your time. Great. Thanks Colleen and thanks all for great results and thanks to think about. So we'll move to our next group presentation we'll come back to question answers and discussion, either after all the four discussions but we also have more time this afternoon for longer discussions and presenting more research so thanks. Our next group will be the PiWR, PiCPT. My name is Kyle Lessinger and I'm here to briefly introduce you to our current research into weather typing using the Python package PiWR, developed by our leads and Helen Munoz and Andrew Robertson. Additional members of our research group include Danny Do, Pedro Herrera, Kyle, or me, Kyle Lessinger, Kelsey Malloy and Kyle Nardi. So in mid July 2021, there was large scale flooding that occurred over Western Europe, due to heavy sustained precipitation, and then the flooding had a large impact on both human lives and damage to property. With this event in mind, and our selection of the PiWR research group, our research goal was to understand specific weather types over the North Atlantic that cause extreme events. Ultimately, we'd like to increase forecast scale on the S2S time scale for these extremes. So here we see a time series on the left of 500 hexascals you have potential height between July 1 and July 22 over Western Europe. We see these low height fields and they're pervasive for several days and then they transition out. And then also on the right, we see the same pattern on the July 14. And so this is, and these aren't in the exact coordinate boxes, so they're not meant to be exactly the same, but we can still see that same pattern. And these are the kind of the types of weather types that we're looking to understand a little bit more so we can find out if they have any sources of predictability. So previous research has already identified that weather typing is an established method for identifying recurrent circulation patterns. And our methodology includes first applying a multivariable EOS field to both 500 hexascals you have potential height and integrated water vapor. Next, we ran a canning clustering loop for four different weather types to identify recurrent circulation patterns. We finally analyzed various sources of predictability from local sources and teleconnections. And our data include in separate analysis for both global geopotential height and integrated water vapor and CPC precipitation, the global unified gauge based daily precipitation. So we use this to identify the land based extremes. The first set of images for weather typing using a coupled EOF for the first row is geopotential height. The second row is integrated water vapor. And then the third row is precipitation. So we coupled the geopotential height and the water vapor. And this represents about 80 to 90% of the explained barrier. So we see these four weather types in the summer over the North Atlantic. We have an EO, Scandinavian blocking and even a ridge like pattern. And without coupling of the geopotential height and integrated water vapor. Then we wouldn't have seen a strong signal with weather type three, which is the Scandinavian blocking with the low pressure and so this is kind of the weather type that we're looking for for extreme patterns that kind of affected Europe. Kind of similar pattern. And so we're going to explore a little bit more about this specific weather type. And then you can see at the very bottom precipitation, it is a composite of precipitation for each day's weather type. And so that we do see some are more likely to cause more or less precipitation over time. So with all these patterns in mind, we can start asking questions about weather type and their prevalence and predictability. So when we just saw summer and now this is the winter EOF, coupled EOF, and it's pretty much a different picture, but we do see a stronger signal for the patterns. And so that's pretty much the only mention for this point. So we then explored extreme wet and dry events that were associated with those specific weather types to identify when excessively wet or dry conditions occurred. So we had our own definitions for what a wet and dry event is. And you can ask us that later if you'd like, but for this one we want to look at summer, the top part, we look at weather type three. So across Western Europe, it does display more extreme wet days and less extreme dry days. So then for winter, we want to look at weather type one. And then this is the NAO positive, and it shows the highest percentage of wet days across the Western Europe, and as opposed to a lower persistence of dry days. And then I will move it on to Kyle. Alright, thanks Kyle. My name is Kyle Nardi. I will present the second half of this talk. So the next step is to explore some of the sources of predictability for these weather types. So first we explore sources of summertime predictability and one potential source is the East Asian monsoon. So here we show anomalous probability based on a specific East Asian monsoon state. We separate into weak, neutral and strong. And we show anomalous probability of a particular weather type in the days leading up to and the days following one of the particular East Asian monsoon phases. So what's noteworthy here is we have several areas where there are anomalously low or anomalously high probabilities of a particular weather type occurring in the days after either the week or the strong East Asian monsoon phase. Indicating that this could be a potential source of predictability for these weather types during the summer. So a potential avenue of future research is to understand why this is occurring physically. So these are just some preliminary plots showing the geopotential height pattern and the geopotential anomaly patterns for the specific weather types over the hemispheric domain. And our goal in the future is to understand whether or not there's some sort of dynamical response due to the East Asian monsoon in the summertime. So now we explore a predictability during the winter and one established source of predictability is the MJO. So this is a plot of anomalous frequency of occurrence for different NaO phases, NaO positive and NaO negative in the days before and the days after a particular MJO phase. Any MJO phase shown here is high amplitude and the RMM amplitude greater than one and then we have a phase zero here which indicates a low amplitude MJO phase of any of the one day phases. So what's noteworthy here is that we have an area of a propagating MJO signal especially in NaO negative. And this is something that has been shown actually in other studies. This was a figure that was shown a few days ago in a talk. This is from Casue et al 2008. And it indicates that the phase of the MJO can modulate the phase of the NaO in the week days and weeks after the phase occurs. So in comparison to what we found, we found very similar results namely that essentially five to 15 days after MJO phases three and four we get an anomalously high probability of occurrence for NaO positive. And then in the five to 15 days after phases six through eight, we get an anomalously high probability of occurrence for NaO negative. So again, the MJO produces a potential source of predictability for the NaO during the winter time. Another piece of the puzzle is exploring how some of these weather types have changed in frequency over time. So this is analysis from 1871 to 2008 from the NOAA 20th century reanalysis. And it's exploring how the probability of occurrence for these four weather types has changed over time in both the winter and summer. So this plot here shows the relative trend magnitude of each weather type over this time period for both the winter and the summer. And any area here shown in color has a statistically significant relative trend magnitude and 90% confidence. So what's noteworthy here in the winter is that weather type three, which we found to be a wetter weather type, increases in frequency over time during this time period, whereas in the summertime, weather type three actually is less frequent in terms of probability of occurrence over this time period. So the opposite occurs for weather type four, which we find to be the drier weather type. During the winter weather type four becomes less frequent, and weather type four during the summer becomes more frequent, indicating that these weather types are changing in frequency over time and as a result, the probability of these extreme precipitation events could also be changing over time. All right, thank you. So our conclusions are the main goal of the study was to connect extreme wet and dry events over Europe to certain weather types over the North Atlantic, and we showed that extreme precipitation events are favored under certain North Atlantic weather types. And we also are interested in exploring the predictability of these weather types and we found that specific phases of the East Asian monsoon during the summer and the MGO during the winter could contribute to windows of opportunity for forecasting some of these patterns. And we also found that regimes are changing frequency over time, and that potentially is going to lead to changes in frequency of these extreme weather events. Our future work, we feel there are many avenues of future research from this topic. We're interested in better understanding the physical mechanisms driving these extreme events under certain weather types. For example, we've done analysis of atmospheric rivers and we found that atmospheric rivers are more frequent under certain weather types, so that could be one source of precipitation from these weather types. We're also interested in evaluating model depictions of these weather types and seeing if the subseasonal seasonal teleconnection patterns from the East Asian monsoon and the MGO also are prevalent within the S2S models, and we're interested in exploring some of the relationships governing the weather types, the occurrence of these weather types and the modes of climate variability, whether it be NAO, MGO as we showed, or PNA potentially even lower frequency modes of climate variability like QBO and so on. So, thank you for your attention. This was our group. It was a great group to work with. We definitely want to thank Andy Robertson and Anhil Munoz who wasn't able to be here today for their guidance. And if anyone has any questions, we'd be happy to discuss during the longer form discussion. Thanks. Great. Thanks, Kailin. Thanks all again. Impressive work. Thank you. So our next group will be the hydrology group. I know you're ready to share screen and start. Can you see my screen now? Yeah. Okay, cool. So, our group looked at hydrology as to us focused for hydrology applications. And we were led by John and Andy, and our group members were Eric, Kahang, Funing and myself. And we actually started off with something different. So I'll present the first part of our work. And then Eric will take over for the second investigation. So we initially started with looking at an, basically an optimization problem for a dam in Japan. And we were using runoff from era five and optimizing for power generation. G here using this differential equation where I is like river runoff. W is flow rate through the turbines and R is the relief flow. And the two particular constraints, depending on the level of the water level and flow rate w. So the constraints were on W and head height H. And, right, so if you focus on the top left here, if you look at this dark orange box, the smaller box, that's the grid size for the era five data, whereas the lighter yellow box is the S to S database grid size. So we already see some issues that we would face here. And if you look at the time series, the top time series on the, on the right, we see when we incorporate the ensemble forecasts on around 120, we see there's some ensembles, which are negative. So there was some issues there. And another particular reason for this start day was we had an extreme event extreme precipitation we wanted to see how that would affect the optimization. And so we tried to apply some bias correction, and we use the CDF matching, but we matched the S to S CDF to the era five CDF. And then when we applied that bias corrected ensembles we get this bottom bias corrected time series full runoff. And these, this is a time series for runoff. And then we use the optimization to obtain a time series for a few values so we had inflow a head height in orange green was the flow rate. The head is the relief flow and the purple line is for power generation which we optimizing for. And so the first, the top figure is showing just one ensemble member before bias correction was applied. Around, actually when we start to be incorporating the ensemble number forecast, and around a 120 approximately. We see that we get some negative values for generation so we see that this ensemble is causing some issues with the optimization model. And at the bottom is the bias corrected ensembles so these ensemble means for these particular parameters. And we see we get some relatively useful values. One thing, since our direction of a project change we don't show these the RMSE and comparison with the climatology and the resistance. So this is as far as we've got for the first part of the project. And I will let Eric, I need to go to Eric. Thanks as you know can you keep going along the slides for me. Thank you. So our second project involved using as to us data to inform ensemble stream flow prediction. And I'll get into all without that all means in just a second, next slide please. Thank you. So, the idea is, if you want to predict stream flow in a basin, depending on the base and you're in you can actually get a lot of information just from the initial land surface moisture conditions. And what they found out is if you can use, if you use historical weather and force it through a hydrologic model and take the ensemble mean of that, as long as you have the initial land surface and moisture conditions, then you'll have a pretty accurate representation of the flow into the future, as long as you're, you know, in the right time of time of the year. So, in spring, if you know, you know how much snow and waters and you're in your watershed then you're going to know basically how high your piece going to be. And so that that idea is called the ensemble stream flow prediction. But what what it's lacking is any actual information of future weather and climate. So what we wanted to do is to use future weather and climate data to weigh different ensemble members and provide an alternate weighting of ensemble matters, instead of just the mean, we'd have we put more stress on ones where we that's where we thought the future would more likely be. So in this study we look at the Buffalo Bill Reservoir in Wyoming. Next slide please. And so what we to do this project we had needed four pieces of the puzzle first we had observed stream flow into Buffalo Bill Reservoir about 40 years of it. We observed meteorology from era five focusing on two meter temperature and precipitation, along with S to S forecast of two meter precipitation temperature. And what we do is we create a rank table and what a rank table is, is you have your historical record of temperature and precipitation, you know, two weeks ahead. And you compare that value for every year so you know, if we're starting on April 1, and we have our S to S data for two weeks ahead that would be, you know, April 1, April 15, once the average temperature, how does that compare year to year. So the lowest temperature would have a low rank and the highest have a high rank. So if 2016 had a temperature of 25 degrees Celsius 2000 had a temperature of 20 degrees Celsius, then 2016 would be ranked higher on there. They're kind of ranked between zero and one. And so we do that for the S to S forecast as well as the observe meteorological values we have over the same time periods. And we build rank tables for both those and we also have these S to S watershed flow forecast, where we take a hydrologic model forced through with historical weather data like I described earlier. And then we have these ensemble members that say okay I have 40 different years of weather data forced through this this hydrologic model based on this initialization date. So typical, typical waiting. One more minute. Okay, I'll go quicker. Yeah. And basically we just wanted to check you know can we use S to S data to better inform our waiting. So next slide please. Next slide. Yeah, right here. So, first thing we wanted to check is which S to S data set would work best and we found that they all had similar spread so anything would be appropriate future work will be used to actually select the best set but for now and you and you're fine. Next slide. So that's a different waiting function so the idea is which ensemble members to weigh more and by how much it we found that the less emphasis we gave to any single member the better accuracy is we didn't want anything too wild because if you miss a peak, you know completely then you're going to have a worse performance and just a straight line. That's the next slide. And so between selecting our S to S data between understanding our waiting function we also introduced climate oscillation indices from and so and the Pacific North American oscillation to see if those helped at all and what we found is, if we took. If we took the S to S forecast of precipitation temperature and gave that half half the weight in value and the other half we gave it to the end so and P and a states at the time, then overall, if you look at the bottom summary stats overall we can just barely outperform the ensemble me. So graphically we're showing a very good case in 2016 where we beat the ensemble mean by almost point one in the modified clean looped efficiency and, and by almost point two national cliff efficiency so much better prediction than just the ensemble mean. However, overall we still have a lot of work to do before we, you know, can really claim we're better. So next slide please. The idea is as to us forecast and climate oscillation states do help with future forecasting. We, we have a lot of potential to do, and we have some future steps, including looking at what works and what doesn't different days we can start at, and what different and using climate oscillation transitions rather than their states. So we're looking forward to new directions and we're open to anyone's comments or suggestions thanks so much for going over. Great. Thanks. Thanks all really good work. Two different projects. So our next group is the US West Coast precipitation prediction group. I'm going to be presenting. Full screen. Yeah. Great. So hi everyone, my name is Deanna Nash, and I will be presenting on behalf of US West Coast tutorial group. The other student members of this group are genoc, Nico's Alex and bill. So today I'll be telling you about our project regarding S to S prediction of a series of atmospheric river or AR events that made landfall in the western US during the winter in 2017 and led to several extreme rainfall events, which damaged the mains bill way of the February 2017, which is pictured on the far left on the plot in the middle we see that the AR landfall frequency during the 2017 water year was higher than average at most latitudes. And the plot on the right from white out is showing the precipitation during 2017 water year was 300% of normal at the Oregon dam which is indicated by the black cross on the map. So the science goal of this project is to provide a hind cast skill assessment for AR activity integrated water vapor transport or IVT and precipitation prior to the Orville dam incident in February 2017. So to complete this, our group explored the hind cast skill through a variety of facets, including MJO connections, synoptic conditions, IVT and precipitation forecasts and AR probability forecasts. So first let's look at the MJO connections. The plot on the left is showing the RM index values for November 2016 to March 2017. So we focused on a particular AR event, which happened on February 5 as indicated by the yellow star. And this was the event right before the Orville dam incident. MJO was in phase five and six during this AR event and had a really high amplitude above to RM index value. And this really aligns with what we know from previous research. And so our plot on the right is showing the five day rolling mean of daily anomalies of AR landfall frequency in California broken down by the active MJO phase and different lags from zero to the 35 days. So if we look at the 20 to 25 lead before our AR event, the MJO was roughly in phase one. And if we reference the plot on the right, we would expect to see below AR landfall frequency in California, which doesn't really align with what was happening. So, or what happened 20 to 25 days later. So for a 20 to 25 day lead time. This particular event is not in accordance with our long term statistics, though other work has shown a really strong connection with AR and MJO. So next we'll explore the synoptic characteristics leading up to this AR event. On the left are showing the weekly 500 hectopascal geopotential height anomalies where the forecast values are shaded and the observed values are the contour lines. A progressive pattern characterized by an equator word shifted an extended jet, along with cyclogenesis in the central Pacific was observed leading up to the Orville dam incident. The surges off East Asia, by the week of the event increased the tropospheric meridional temperature gradient, thereby enhancing and retracting the north Pacific jet that led to enhance cyclogenesis in the central Pacific induced by deep western north Pacific dropping over the February 1 through seventh week period. So this is a cyclone activity aided in the development of downstream ridging over the Western US and subsequently over the Bering Sea via poleward latent heat transport. So this ridging or blocking or Omega regime conditions were conducive to multiple AR events in California. On the right are showing the forecasted daily geopotential heights with a three to 21 day lead, and we see that this observed Omega block in the contour lines is only accurately forecasted up to seven days in advance. So here we have the forecast skill exploration of IVT and precipitation. So the top row plots show the forecasted precipitation for February 7 at four different lead times. And the correlation values, which are a little small in these plots tells that precipitation in general is poorly forecasted, even with a one day lead. So we can actually use IVT as a proxy for ARs and precipitation, which is shown in the bottom row of those plots, where the forecast is shaded and the observed data is shown in black contours. We have a weak pattern correlation of 0.39 for the first forecast that's 12 days out. But as you get closer to the actual event or the lead time decreases the forecast becomes much more similar. We analyze the skill of the forecast at different lead times for the period January 1 through May 12. And the top panel on the right is showing the mean of each ensemble members spatial anomaly correlation in black and the anomaly correlation of the ensemble mean in blue. It shows that there's a bit more skill for the ensemble mean but both drop below 0.4 within 10 days. The bottom panel is showing the root mean square error of the ensemble mean compared to the spread of forecast. Spread follows very similar to the root mean square error and increases rapidly with lead time prior to 10 days, and then it levels off. So if we focus in on the forecasted IVT in the grid cells or the area around Orville Dam, which is indicated on the map as the black dots, we can see that the model skill is dependent on initialization. And the colored lines in the figure on the left represent the ensemble means for various initialization times, and the black line is the observation. So for example, the orange line is the ensemble mean for the model initialized on February 2, and the forecast skill around the peak in IVT on February 21 is greatly decreased. So that particular initialization time doesn't do well in that second peak. But if we compare this to the cyan line, which was initialized a little bit later on February 16, it does a better job representing that peak in IVT on the 21st of February. If we look at ensemble spread, the figure on the right is just the same as the figure on the left, but the shaded regions represent the each ensemble spread. And this is just telling us that certain initializations have more spread than others. And then last we looked at the number of ensembles that detected an AR for a two to three week leap. And the shaded contours and the figures on the left are showing the odds that an AR was detected at that grid cell for that time, and then the black contours indicate an AR was detected by the observation data. So for day to day probabilities, the model has some skill at predicting spatial location of an AR up to a 10 day leap. If we look at the number of ensembles that forecasted a detected AR within 17 to 23 day lead, roughly a three week lead, the model predicted that the AR frequency was going to be much higher than the climatological average, which is quite accurate. It doesn't necessarily catch the spatial scale of the AR, but the general frequency for that week was well done. So in conclusion, we did make some progress on answering our question on S to S model skill for AR events leading up to the Orville Dan crisis. However, like any good research project worth pursuing we discovered that our one question led to several other questions. For example, can we improve our results with a bias correction or what kind of skill do we have in predicting multiple AR events in one season or high intensity AR events compared to lower intensity AR events. Can we leverage our knowledge of MJO, QBO and and so to improve our prediction skill. And so in all, we only really saw skill and predictions of this AR event, at least less than two weeks. And for more discussion, please come to our extended session this afternoon. Thank you. Thanks Deanna and thanks all in the group for a really, again, an impressive amount of work in such a short period. So, thank you. So we have set us, we had set aside 15 minutes, which originally was a break for student feedback, we still have three to five minutes and then we'll take a 10 minute break. Before the next plenary talk. So, does anyone have questions for any of the student groups. I don't see any right now in the chat but I had one for, oh, Zane, did you have a question. I see you're unmuted. I'm sorry, I just don't believe myself by accident. Yeah, should go ahead. Hello, hi. Can you hear me. Yeah. I have a question regarding this I'm Jill influence on the downstream weather both in Europe and atmosphere river. It seemed to be that everyone indicating the face, six and seven seem to have issues right so, in fact, face six and seven is when I'm Jill, and I'm index cannot tell you where the heating is. So, a lot of downstream weather impact depend on the convective heating precipitation. For instance, the connection large scale connection generated the recipe wave trend is in northern hemisphere. So if the collective heating of MGO in southern hemisphere, you would have very different impact. So, have you all thought about actually using the actual precipitation tracking, like we presented in the database is available to everyone is called large scale precipitation tracking that tells you exactly when, and where the collective heating is. So, I'll just speak for the, I'll just speak for the PWR side that's a really good suggestion we did not look at that but I think that would actually be really fascinating and it might, you know, increase the skill and maybe increase the anomalous probabilities if we can sort of fine tune it a little bit so that's a really good suggestion that will definitely take into account. Thank you. I guess the question also is, yeah, to the other groups. Yeah, I can go for the for the clip at your group. We did not consider to look into that but we also would find it really interesting to look at the different publications of the MGO and, and yeah, it's definitely something that is very exciting and interesting to look at. Thanks, Jan. We should go on getting that into Clifford. As part of the state dependence. Yeah. Yeah. The question is also to the US West Coast group that was looking at ARS and MGO teleconnection. Does anyone from that group want to respond. Yeah, maybe I can say so indeed we also didn't do that but definitely something worth looking at so hopefully yeah we're going to do it by time. Great. Yeah, the US West Coast is in particular because MGO heating if it is in northern northern hemisphere that influence your synoptic pattern, much more direct. And this is a place where the heating split really strongly either the northern hemisphere or southern hemisphere rarity on the equator. Great. Thanks. Thanks for the insights will definitely have a look at it. Thanks. Yeah. Thanks for your suggestion. So I had a question regarding the monsoon index that was used. And I'll be when you were presenting the slide. Did you use the seasonal mean monsoon? So maybe Kelsey, it was your analysis or was it the interseasonal component of the monsoon that you're looking at. Yeah, I can take this question. That's a really good question. Because we were focused on mostly I would say the sub seasonal scale. So we used two different methods as kind of preliminary ways to look at the monsoon and they're actually a lot of different indices you can use. I tested to one being the definition using u850 circulation. So we have daily data that's run with a five day running mean for u850. I calculated that circulation and then looked at the relationship between or I should say we looked at the lead lag relationship between what that indices was based on that circulation. And then I also looked at the circulation based on u200. So I looked at like upper and lower level ways to identify the monsoon phases. But I am interested in kind of taking that to the next step by looking at the boreal summer interseasonal oscillation or BSISO because that looks at it in terms of phases like the MGO. So I think that that could be also really interesting to look at since we did that, since we found stuff with like the MGO lead lie relationships, maybe same with the monsoon. That sounds great. Yeah, that was going to be my other comment and it's also related to Shuie's comment about a lot of these teleconnections are related to convective heating and less so to the dynamical forcing and some of it is dynamical forcing but a lot more to the convective heating right and just the wind indices may not capture that convective heating aspect of it. So, great. Yeah, very good point. Yeah, definitely. Okay, so I don't see other questions right now we'll come back to a larger discussion in the afternoon. Thanks again to all four groups who presented. Thanks, I thank you because I'm part of it is all the data crunching before a tutorial started and we use different databases but I want to say thanks to Abbey Jay who has prepared a lot of the data that went, especially into the materials, Unish downloaded, especially for the hydro and US West Coast. And so, thanks so much for all the background work in the background and in data pre processing and thanks to the tutorials.