 Thank you very much, Andrew. Yeah, this is a broad specter here. I'm going from one extreme to the other so I will talk a bit about flood forecasting and how we have implemented two systems at least in WF one European system and one global system and I will focus on the European system because that has gone further in its operational development and we also have a study there on seasonal forecasting so the The outline of my talk today will be first an short introduction to hydrology for those of you who may not know as much about hydrology, so Give some specific. Yeah, okay So I will talk about the two systems we have European system and global system and then I will show a bit on a Case study we have done looking at at S2S in hydrological forecasting as well. This is still work under development So we just started looking at taking hydrological forecasting to the to the subseason of the seasonal scaler as well So flooding is really really a global challenge This is a picture from something called the Dartmouth flood observatory and It's a US system where they have looked they're using a satellite imagery to to monitor floods And these are the reported floods in 2010 So the picture shows here you have you basically have floodings all over the globe So it's really something that is abundant all over the world, but there's Difference also in the frequency of where this happens This to work so Some countries are more flood prone than others obviously so Here's another picture showing the number of occurrences of flood by country From 1974 to 2003 so it's a 30 year period and you can see some countries showing up in red showing that they have more more abundant floods South America Southeast Asia and the US are very flood prone But no matter where you are in the world you can you can be hit by the devastating floods So it's very important to try to capture the Possibility to forecast these these floods these events. I don't know if this is out of battery or not This is not really Okay, I Used the fingers here anyway. So what causes the flooding? Well, there can be many things in the Nordic or in the Snow-fed catchments we have a lot of snowmelt runoff So in the spring we can expect to get a flood but sometimes this is more more larger than others times and of course rainfall Obviously Either a short very intense rainfall or a long prolonged season like the monsoon season can give dodge floods You can also have obstructions in in the river discharge only in the river That causes it ice jam or another tree that the course of the diamond then you get a can get a Devastating flood from this you have also coastal storms You have urban stormwater runoff when you have a large intensity precipitation event hitting a city for example You also have dam failures So you have the course dam failures very difficult. You can't really forecast it, but all the other ones We are pretty good idea about the forecast and maybe not the ice jams and destructions But all the other things we we do have a capability of for forecasting these events and the way we do this is Through a chain. So we are starting off with the metrological input which here is shown as the EPS So that is the ensemble prediction system. You can also use a deterministic model But today most centers are really using the probabilistic forecasting in in hydrology And you can use a preprocessing a calibration of your of your input data And then you fed it to feed it through a hydrological model and hydrology model is basically taking the precipitation and temperature and evapotranspiration routing it through the soil groundwater and Producing discharge you also get other things like soil moisture groundwater recharge Lakes and other things as well. So it can be quite a complex model very similar to a land surface model is is basically a Type of hydrological model you can use those four for flood forecasting as well And then you get the output and usually you do some post processing And then you compare these two for example a climatological mean and you come up with a warning depending on the situation and Then you can also use verification tools to to feedback into a model. So this is this is what we our system looks like And because we're not perfect. So the two main reasons why forecast fails is that We don't have enough information on initial conditions As I talked about yesterday the poor coverage of data is a problem in a global level. So we have We can have also errors in the data simulation in our in our model So we can have a lot of problems with initial uncertainties. Basically, we do not really know where we're starting from But also the model is always just a Description of reality and we might not capture the right things we are The true atmospheric phenomenon that we want to capture. So we also have a lot of model uncertainties as well and We are working in a chaotic system. So we can't really forecast these events on a very But we most of the time but we do have a have to be consideration that we get a lot of forecast fails as well So that was a very quick hydrological introduction. So now we're going to move into the two operational systems that we are running a DCWF first off is the European flood awareness system and It's a transboundary system. So it's Gives has 50 partners in our network that receives these forecasts and they provide observations and also feedback on on the performance of our warnings And we do have a large collection of hydrological data as well. So there's a very it's really a two-way interaction with our Partners we're given the forecast. They give us data and feedback and we have a constant dialogue on how to improve the system as well And it's basically it's covering most of Europe. We are extending now eastwards as well. So Turkey and Ukraine and Belarus will be included in the system as well and we have a number of partners giving forecasts to the Member States or the member organizations and so the benefits why we're doing this Well, this is an added value to the national hydrological services Usually they have their own information on a very local scale, but we are trying to give Forecasts on the longer lead times up to 15 days and probabilistic forecast that we are really trying to push into this As far as we can into the subsystem But is this with the system Currently, we only use in the 15-day forecast. So we can't really say we're doing in the seasonal forecast yet and We are providing also novel tools to new techniques to use a data for example satellite data or ways to Calibrate the model and make it better and of course we also have a very important role to serve the European Commission because in a crisis they want to have comparable information across Europe and has been used in two cases quite recently both in the in the flood In Europe until the 13 and the Balkan flood last year this system showed food very useful For them it's a tool to anticipate the management during a crisis. They can contact the civil protection agencies and See what they want and what they need. They can also Use other kind of services for example the mapping service in Copanicus so Copanicus is this large project that is using satellite imagery to really benefit the European countries So this part this is a part of these Copanicus services. So the way we do a forecast or prediction. It's I explained it briefly before but just going into more into detail. So we're starting off with a large set of historical data. We have maps we have different layers of soil type vegetation and all that is fed into a flood database and From this we produce a hydraulic model and Forecast typically looks like this for a typical occasion. So you have your starting conditions here Here's time on this scale and here is discharge in cubic means per second so we have the The proxy observations, which is the modeled water discharge and then we can reach come up to the forecast date here and then we feed through our Ensemble predictions systems. We also using a number of different models. We are using these in WF Ensemble which you can see here in box plots We also using a terministic Ensemble or the term is a model for me simply efficient red one the black one is the German weather service D2D model And they all give a discharge Curve but depending on the input precipitation and temperature This is then transferred into Information on whether we have a flood situation or not into an interface And then they the partners send an email out to the network if there's if they need to And the way we a forecast to cease this of course we have this is what we have a multimodal system So we have a large number of time series. We have the termistic models. We have probabilistic models and The way we try to simplify this is that we have From the climatology we can calculate certain return periods. So in green is depicted the one and a half year term period In also the two-year term period in the yellow another one half The yellow is a two-year return period, which means that on average The discharge would reach this level every two years The red one is a five-year return period and then we also have a purple for a 20-year term period So these are depicting levels in the river which are More extreme than the normal conditions. It doesn't really say that they actually are damaging clouds just showing there's something's happening And of course we can we can Tabalize this as well. So if you have just an isic model going forward in time We can say that we can color code it whether it hits a certain threshold for example We we define events which can last over more than more than Number of days, so we're looking at the peak threshold The peak thread the maximum threshold. So yes, exactly. Yeah, but it can happen over more. Yeah, exactly So we're looking at the maximums. Yeah, well, we we do use term periods for that's what the forecast is usually are used to using So it's it's really a matter of Tradition the way people are using these forecasts Yes, you can you can you can do the same thing as well looking at percentiles for a minute, but we are it's just What you call a protocol to use really? You mean the system you Well, it is the system is really the whole The whole chain So it's the hydraulic a model fed by the the ensemble prediction system and then So the it is a hydraulic a model where we're with the soil moisture and the ground water and the evaporation is taken care of inside the model as well So it's similar to a think of a land surface model if you're not works with hydraulic before So in a land surface model you would produce runoff and that runoff can be routed to the rivers So that's what we're doing. We're just routing the runoff to a river Well in the hydraulic a model. We usually did not do use data simulation We are using in just precipitation and temperature and then over time you would develop the soil moisture. There's a problem in Data simulation for example in an operational land surface model. You will have Satellite data being simulated every day And that but that is more the purpose to have a good feedback to the atmosphere from the land surface model But the problem is that it will Destroy the water balance in the in the model So you would either add or subtract water as you go along to do Because your your goal with the data simulation is not to produce a perfect runoff is to do Better energy balance to the atmosphere Yes The absurd rainfall that we're using the that's from observation on that work in Europe So it's it's something that we have gathered from all the on the member states It's monitor right up to the to the forecast date exactly so we if you look at this Figure here everything up to here is basically monitored Observation temperature run through the hydrological model and then at the starting date we switch to the Meteorological models So we're not taking into consideration that they're biased in the hydrological models that in the meteorological models at the moment So that's that's another issue as well that you could do a calibration of your input data, but in this system currently we are we are Just feeding the meteorology models into them Yes, excuse me, sorry no no it's not We're using them for for references for example using snow war snow what snow equivalent data Snow water equivalent, and we're using soil moisture to compare with our own model So in the operational system you can look at satellite data as well, but we are currently not using it in the system Basically for the reasons that in hydrology it has not had the same impact is in the simmetology is improving the results There's still a debate whether how much how you should use the data and How it will affect the results one problem is it doesn't penetrate deep enough So you can only see the top soil level So you can't really see that the water below ground and that is very important for hydrology But they'll interact with the rivers as well, so Well in the EPSS system they are using satellite data so in the precipitation from the EPS systems They will have used satellite data to for their initialization of the model. Yes So this you have to be Separated the how the EPS work and how the hydrology work So in hydrology model, we don't do not do any satellite data simulation But of course in the operational metrology models that's very important issue There is a You mean within the Greek sub grid scale variability sort of yeah Yeah The system is set up basically on the it's in a five kilometer grid covering Europe So whatever happens in that one we treat as a uniform distribution of snow so we don't do any sub grid variability of snow and We also do not issue forecast for catchments smaller than 2,000 square kilometers So we are not the system is not designed for very high Resolution forecast is really for more of the larger rivers and as an overview But we'll try to go down in resolution and we will Improve the models continuously, but as the moment we are doing a five kilometer grid This is a rainfall runoff model, but I will also show an example of when we can use a land surface model for that So I'll come back to that one Just to see how we do the the simplification of the time series for the for the ensemble prediction system We also have information on the on the probabilities So we do not cannot only detect the the warning level that you're you can Breach but also the probability By just counting the ensemble members above a certain threshold So we can put this all into sort of one diagram and you can see that the evolution in time as well of how many of The ensemble members predict a certain threshold. So here we're looking at The codes here are for our ensemble model above the high level and and the severe level and also so you can see continue this one. So here's Forecast time and Then you can see at which point the models predict that that you will go above some threshold You can also put this into a two-dimensional diagram where you have the forecast day on the horizontal on the vertical axis and then you have the the forecast dates on the on the Horizontal horizontal axis so as you progress in time you will have More information sorry and also you can see the jump in this of the forecast as well So you get closer to the event you can see how many of the models predict a certain event happening or not So you have some examples yesterday from a 26 day Lead time that's quite impressive But you would also need to a signal like that could appear quite early in the forecast But then you want to see that signal continuing for the next following forecast is way before before you issue a forecast So consistency is quite important in this context and the system has been evaluated using different for skill scores, we also look at the number of warnings be sent out and what actually happened and this is the Account of the number of watches and alerts be have sent out and The yellow line here is the note in number of hits and the blue line number of false alarms and the orange one Is not known really so you can see that we do have a larger number of hits than than false alarms, which is good for system We also have been quite active in the last two years issuing forecast This is due to the two major events that was mentioned earlier the European flood in 2013 and the Balkan floods in 2014 They were massive events and they covered a large region which meant that we sent a lot of warnings out a lot of alerts but they may be counted only as one event or country-level event and Just mentioning these two events that happened. This is one in that hit the Northern side of the Alps. So basically it was was a weather type or weather Circulation was well known. I think it's called 5B where the water picked up moisture from the Mediterranean and then went sort of a loop eastwards and then it came on from the Alps on the northern side Also picking up soil moisture as it went along and this this triggered an extreme Precipitation event on the northern side of the Alps and If you remember this situation, there was a large area. So we're never hit by by this and Unfortunately, we did predict it but not the severity enough. So we did underpredict it quite heavily That has to do both with the problems in the hydrological model, but also in the metrolocal model. So we We should have been about 10,000 So we had less than half of what we should have had or one model said about half of the actual discharge It depended on certain stations, but where they actually had the first hit we did underestimate The rainfall and the discharge So we went back and looked at the model and see what happens if we increase the resolution so we're looking now over a box of covering the Alps and here is total precipitation accumulated total precipitation For the 72 hours that this event lasted over and the black one is the observations So that's that's based on observational data over this period on the on the vertical axle here You have millimeters of precipitation so currently the green line is our High-resolution is a different model as you can see we we do under predict the precipitation quite heavily so the So we've run the model with different resolutions and as you can see that they this will be improved with the high High-resolution as well. You will be able to capture more of their graphic Precipitation you will also maybe come to a situation where you can have completely a convective Not a convective precipitation Fully resolved convection possibly They need to go to a higher scale or high resolution, but you're approaching that with a five kilometer But that's not the only thing as well we also looked at how we can improve the physics of the of the of the cloud physics of the actual rainfall generation in the cloud So here's some examples from the very high-resolution model Top four lines here with that just an improved the cloud physics in in the model in the And together with the increased resolution So the top blue line there of the purple line is really getting close to the observations And we are not very far off from from reaching this resolution in the model We are going to increase to next year to eight kilometers That would be released sometime next year the ensemble will be followed as well at up to 16 kilometers. So Maybe in five six years time maybe up to ten years time we can we can reach the five kilometer resolution That would be a major improvement for for the these kind of events And here's another example from the UK where you can see how you use this kind of Table when you're reaching Another flood event so you can see you can monitor an event have being picked up by the ensembles quite early But you won't do anything until you get down to where you have a more of a clear signal in your system So you will issue a warning when you are three four or five days ahead But you would see the signal already maybe ten days ahead most of quite often We also have a flash flood system. I do not talk so much, but it's just mentioned that we are Doing flash flood forecasting as well on a very small scale and this is based only on precipitation and soil moisture a combination of those two Because we have got the feedback the many countries want to have information on very very high intense trend for So moving over to the global scale them We also do have a global flood awareness system, which is not operational at the moment. It's a Research product, but it does give you Similar kind of information as with the European system, but now on a global scale So this is what the for the interface would look like and you can also add things layers like precipitation from the model as well so you can have all kind of of information together with your your Forecast information as well So these dots here or this these triangles and dots they denote whether there's a flood event forecasted or not, so they if you have a pointing upwards triangle you will have an increasing discharge As you can expect that the flood event has is still In progress or it will happen soon the red one denote where you have an ongoing flood Or the sort of the circles the note when you have an ongoing flood and the downwards pointing is when the flood is receding So and This system has been developed From our land surface model. So here. We actually are using our own in America by the prediction model We take the output from the land surface model. I mentioned before the problem with data assimilation. So that's still in there So you should be a bit careful when using this one, but it is routed through a network to produce floods Probabilities of flood events on the global scale In in this one, we are using our H-TESL land surface model Which is coupled with the IF so it's a part of the whole integrated forecast system So it's a of these operational Land surface model as used in in the ECWF. Yeah So the distinction is just that The basically they come from two different scales I would say from hydrology you develop catchment models on a very small scale and they have been now developed into sort of Continental scale hydrological models Whereas the land surface models are going the other way around they started on the sort of the global scale Even though and then they can now be used also for our catchment based information So but they have been developed with two different purposes The land surface model has developed to to take care of the so the Interface between the land and the boundary boundary layer so to get a good Interaction with the atmosphere So what happens with the soil moisture evapotranspiration and All of that so it's not designed to give a good response in terms of runoff So in the hydrological community the runoff rainfall runoff models that have been developed They are very focused on producing a good rain runoff signal very depending on there's some hydrological models that look very much like a land surface model And they are just a bit more complex in terms of groundwater. There is hydrological models Which are basically just a black box So you can have everything in between as well So you can have they can in hydrological model can contain much more of statistical relationships So it's not everything is described in a physical way They are more distributed models. Yes, they also contain a lot of parameters that you can play around with as well They do need some tuning as you call it the color or calibration to perform well In this in our case we have four soil layers Yeah, but that can differ as well between different models There's not set that you have to do it this way just the way that they have decided to do it But they don't usually did not have a connection with the groundwater So whatever is Discharged out of the land surface model is just you know Disappearing where it's in a hydrological model. You can have also interaction with this with the groundwater So you have much more complexity in how the water is transferred In the In the soil in the ground usually For the hydrological model. Yes Yes Yes, but we will develop the global one as well But we just set it up as it was so it was there We had a global system which gave us runoff. So we thought we'd just use it just in With the routing scheme to produce a discharge output So it that's why I say it's still a very much a research project But it's been proved to be quite useful anyway in some cases. So We have looked at Yeah Sorry the grid size The The routing is done on a 0.1 kilometer a 0.1 degree. So it's about 10 kilometer They are very similar they just have differences in in how they Have set up their system. So the good thing about this one is it's integrated into the into the focus directly Whereas week you have to run by having some sort of forcing but here we get it from directly from the operational Uh, yeah, I can show the next slide here. We've actually done some calibration. So moving on to that one. It's You can we haven't really calibrated the the physics of the model That's yet to be done if you want to treat it as as an hydrological model The problem is if you do that, you will also influence the the uptake of the plants and the vegetation and the vapor Transpiration. So we have to show that whatever we do in the land surface model. It's not deteriorate the actual forecast. So It's a bit tricky what we're done here is we take in the model and run it offline forced by by Interim on the global scale So in that case, we can just focus on the how the model performs without data simulation just in any river So we do get a good signal basically in the larger rivers Um on a smaller scale, it doesn't really work performed that well But on the amazon, for example, we get a very good signal to to work there But that's that's not so difficult to do to be honest because it's very driven by the seasonal cycle anyway, so But there's there's a big challenge in I would say in calibrating this model So you're hitting a very good point there that we are we need to calibrate it to Make it more Hydrological useful But we have used it and it's used by the the world food program for example when they are When there's an oncoming or ongoing flood event They can see whether they need to dispatch extra Help to the people or to Do something so they can use this system to see what they can anticipate in certain areas And it is available for any money to use as well. All you need to do is go to this website And and register and for the course of this for the for this training now For in this workshop, you can use this this username and this password It's it's a global password for a global username So if you log in and use this and change something next person logging in will have those changes So Be aware that this is just for for this week now if you want to look at the system and what it produces What it can be used for as well is to to validate your your It's an independent evaluation of the forecast as well It's for this week, but but if you want to just To friday. Yes. Yeah, so if you just go ahead and use this But if you want to register Send this so you can look at this website and you can register as a user and then you will get to know your own Password personal password So you don't have to use this one, but this is for for this purpose of this training course now you can use this one So We are developing the system and we are mainly developing the european system at the moment We will put a lot of effort next coming years into the global system as well And some of the things we have introduced is a mold we are looking at the multimodal hydrology not only in the Driving data, but also in the hydrological model But I will show now some results from from When we are trying to take this onto the seasonal scale as well could be more interesting for you to see that that study so We are doing now with the european system up to 15 days flash flood and flood warnings And we want to see if we can use it for applications on the longer time scales the obvious Customers for this information with the hydropower management You can also use it for spring flood predictions Maybe for low pro predictions for navigation agriculture water needs. So there's a lot of things you can use Seasonal hydrological model output for And we're getting an increase in in the the scale from the new wave combating prediction model So we think that we can actually try to make this step now go to this To the seasonal timescale as well So the aim of this study is to use the we're starting off by using our seasonal forecast. So the the seven months European seasonal forecast model To to provide probabilistic forecast for for beyond 15 days And we set up the system in a way that we um be comparing our seasonal forecast Which is for the right here. That's the we're using the high the full hind cast system Of the seasonal forecast We're running it through our list flood model, which is the hydrological model. So we are not using the operational Land surface. This is this is still hydrological model And let me come up with a seasonal ensemble prediction But we also want to compare that against How what happened if you use observations? So we're also using um observational data, which we pull out from this database So at a random order. So we are producing a climatological Ensemble which will run through our hydrological model to produce something we call ensemble stream stream flow prediction So that's our base flow or the other base system to compare against and The other question was how should we um look at the scores and dissemination of this one So we decided to look on weekly catchment averages Now we moving beyond 15 days. We have to go a bit larger in time and space And we're looking at seasonal changes DJF Well winter spring summer and autumn And we're concentrating for the moment on the lead time eight one to eight weeks. So what happens on these first two months? And we are comparing against our our model climatology in the run by the perfect forecast Which is the observed precipitation and temperature And the strategy of evaluating this month. We are interested in this well to see how much of the metrologo forcing Versus the initial conditions matter So if you have just the perfect forecast Running here, which is this this here run by observational precipitation and temperature If you then attach to that your climatological forecast or your seasonal forecast, you will get a stream flow prediction that will differ in time over time And the the observations may look go somewhere like like this Now i'm using only the climatological forecast. So this is not the seasonal forecast. So this is this going to look like this in in If you just take 15 years from the past and run through your model And this is the way that This has been done in in in the past quite often This is to just to see them what we can expect starting from the correct initial conditions and then running through the Climatological past Now if you just do the other way around is that so we pick up 15 years from the from our Climatology and we run that by the perfect metrologo forcing You will see that in the beginning you will of course have a big disparity in where you're starting from but in time you will Catch up with it with a with your perfect run as well And you can compare these two and see where you where do you this? Variation in the initial condition becomes less than the variation in the metrologo forcing So you can see how long on what lead time does your initial conditions really matter So the results from this study some this is looking at comparing now the seasonal forecast Which is the dynamical model the system for from eastern wf in green and the blue is the climatological forecast On average both systems do produce a skill up to eight weeks and you can see that This skill in the first two with the first weeks here is basically A lot to do with these initial conditions So even if you take your model and just run it with historical data You will have a skill just because you have a good starting point a good initial conditions But you will have A little bit of the advantage using the seasonal forecast over time Some catchments will be better than others of course. I will show that on on the geographical scale as well Yes But it is more accurate than just using the climatology if you want to beat the climatology obviously otherwise There would be no point and this is looking over europe and europe is not very good in terms of seasonal predictability Unfortunately, we won't expect to have a lot more than beyond this four to eight weeks And if you compare then with the the reverse ESP where we looked at With disturbing initial conditions, but then have a perfect forecast That's shown here in purple. It will start off quite badly But then after after a while you will Beat the the other systems because you're not forcing it with the perfect forecast So you can see here that it takes up to two weeks before you have lost the the signal from the initial conditions Or where without the really matter So you can see where where you have the the breakpoint between initial conditions and much longer forcing Oh, sorry, key g is uh, it's a skill score. It's called the clean gupta skill score. It's looking at the correlation and the the Also the mean mean, uh, mean error of the forecast Clean gupta It's just it's something that was it was a development of another score called Nash Sutcliffe Yeah Yes Yes, it is Yes, definitely Yeah, I mean sometimes catchment doesn't have any memory at all. They just are completely rainfall driven No, exactly. Yeah but This is showing a bit of of difference between seasons as well. So in the summer we really have much shorter memory in the system Uh, they are quite equal in terms of skill And uh, but in the winter time we do get a much better Uh, we have a better forecast in the winter. Sorry in the winter you have a longer memory No, sorry in the winter you have a higher skill in the forecast. That's what I should say I will talk about the memory later, but in the winter we do have a higher predictability in in the in in the rainfall And that is shown in other studies as well that for winter time in europe We do have a better skill in the summertime because the winter time is more driven by The the large scale circulations in summertime you have a lot more influence from local conditions this This is done by this is the seasonal forecast So I don't think the seasonal season is not it's not an integrity database. So this is the No, this is this is the this is going up to seven months So this is using the seasonal forecast. I'm just concentrating on the first eight weeks now at the moment But it's it's run by the seasonal forecast Yeah It's not a ticket database. It is a seasonal forecast of of european center Which is in the s2s as well. Is it or no, okay This is uh, this is not the ticket database. So we haven't we are going to we have implemented the the The monthly forecast as well. We haven't yet analyzed the results from that So this this would probably get a higher skill as well We have a the the monthly forecast. We should update much more frequently So this is just looking at the first shot at what how much can we get from just using the seasonal forecast But we will the same kind of analysis with also with the monthly forecast. Yeah No, this is the raw raw output Yeah So as adrian pointed out, there's a big difference in in where you are as well. So Now we're looking at the crpss. Uh, so it's looking at At which week in the forecast Through the crpss, which is a skill score looking at the whole probabilistic forecast Go below zero So if you have a very White color here, it means that you lose skill quite quickly. If you have a blue color, you have a higher skill Longer Up to eight weeks here. So we can see that some catchments Especially in the winter time. We have a quite um Long lead time of skill so we can have a skill up to seven eight weeks actually in the forecast In the summer the situation is much more locally driven. So we we are losing a lot more skill So it's we can see that it's much more useful for the winter winter season You would yes, but we are looking at sort of lead averaging over a week and there's this some rivers in europe that had that kind of Concentration time of longer than a week, but they're not very many. So most of the rivers will have shorter than one week response But it does depend as well on the How long it takes to the water to reach the mouth of the river? Yes, it does And that can also be affected by if you have reservoirs or lakes, for example, that would influence results and Then comparing the the seasonal forecast against our our base base benchmark forecast the the extreme The ensemble prediction system with just the climatology. We can see here looking at the skill score of CRPSS if it's above zero we we can say that we have a skill if it's below zero we are Worse or acts bad as the climatology You can see here that the we have a skill for up to three to four weeks from any catchment This is averaging over all the catchments You see also this influence of the initial conditions in week one that both system are really performing equally good in week one but then you get To week two you actually get to Increasing skill due to the fact that you are you are picking up of the metrological Predictability in a week two and if you compare low flows with high flows the this is looking at The rock score which is another skill score that we are using as well counting the number of hits and misses in a in a forecast and system. We're looking now at the Flows below and above the 95 and five fifth percentile So the green one here Is the low flow so everything is below the fifth percentile in your in your forecast and you can see that the So this is the seasonal forecast so the green the Triangles is the fifth percentile and the Round one is a 95th percentile. So this is the depicting the low flow and this is the high flow And you can see we have a better predictability of the low flows than the high flows So we do have we can have a more More trust in the low flows of the system And we can see also that the Seasonal forecast is beating the climatological forecast of most of these lead times And then finally a slide here on what what is the Where does this predictability come from so we're looking now at a measure called critical lead time which is Basically when the various variability in your climatology is larger than the variability in your Forecasting system. So if you have a very White number here, you have your variability of your Climatology is much more important. So you have very little Effect of the of the much longer forcing so the effects will come all from the initial conditions in the model And if you have a high Number here very blue You have a longer Gain from from using your system from you have a much more influence from the initial conditions Sorry, so if you're blue you have a much more influence with initial conditions if you have a white everything is decided by the metallographer forcing So these areas here in winter time is Do you have a much more influence of initial conditions due to that? They are groundwater fed during winter time So they they are not so affected by what happens in terms of the Metrology forcing Snow is it's rather rather important, but also during winter time. These are quite low flows. Anyway, so they mostly run off that also restored in the in the groundwater Mostly frozen on the on the soil, but you still have running water beneath it Whereas if you look at the the western side of the of the Nordic countries, you have much more westerly winds So here the the metallurgical forcing would be much more important than the So here here the catchment that agent talked about the small catchments which have low memory And in the Mediterranean you also have the same thing in your winter time You have precipitation driven flows during winter And some areas that come up like Spain in in summertime you have the the drier moisture conditions that are important than the forcing itself So if you have a very dry soil, they can They can hold the water much better. They have much more capacity of absorbing the precipitation And then you have finally these snow melt driven discharges during spring and summer for the Nordic countries as well So there's the large difference in variability variability between these catchments depending on how they If the if the initial conditions or the metrologal forcing is important So the message from this one is that you you have a gain in using the seasonal forecast from I would say one to four weeks of the time especially in winter time And you do have a better skill in the low flows Than they are then the high flows, but we do have some have an advantage using the seasonal forecast from the climatology So we get a gain in skill and the uncertainty from the initial conditions is um Yeah, you have a transition between hydrological states as well, which is a crucial part of this process as well. So Okay, so that was my talk. Thank you