 Runners up who are also going to give a student model or talks and the next one is Zach Flaming who will be talking about a hydrological model for prediction reanalysis and capacity building and each of these entries where we're phenomenal. We're pleased that they're providing Talks with us because the student talks are only 15 minutes We sometimes don't take any questions like we did with Julia so So if you take all 15 minutes no questions if you end a little early, you'll get one or two. So, thank you All right, thank you So I'm going to talk about EF5 which stands for the ensemble framework for flash flood forecasting This is a hydrological modeling framework We developed at the University of Oklahoma and collaboration with the National Severe Storms Laboratory The idea was that we wanted a hydrological model that was suitable for producing flash flood forecasts At high resolution which is for us one kilometer over the entire US and we wanted to be able to produce these forecasts very rapidly So in response to a rainfall that was occurring So we were using a radar based rainfall that occurs every two minutes And so that was kind of the driver for development of this whole system What we did when we were developing this is we actually made it very modular and so we Designed different modules here and so that you can actually plug and play in the code or and in the configuration files to Enable and disable the different modules. So we have modules for precipitation So we can load in precipitation in different formats of things like that and we can convert on the fly between formats We do not reproject on the fly though So it's only a conversion process the same goes for the evapotranspiration We've also added in kind of snowmelt models and so you can see kind of how the flow goes if you add in snowmelt That the precipitation goes to the snowmelt model which then goes to the surface runoff model For the surface runoff models, we have kind of three options We have crest which is basically a Vic based model We have the Sacramento model and then we have what we call hydrophobic Which is basically there's no infiltration into the land surface and everything runs off And then we couple those two couple different routing options the primary one being kinematic wave routing Which is basically just a very simple You know going downhill down slope routing scheme We use that we can then produce a variety of outputs where we can look at things most commonly stream flow We can also do some Recurrence interval work if you have a historical run of the model simulation that you can compare against We're playing around with inundation and so we're trying to do different things there So we can get water depth and water extent and we're also looking at soil moisture We've also coupled this with parameter options optimization with dream if you want to calibrate the hydrologic model Although that's not actually how we tend to do things Just kind of going forward. What does this all look like? Well, this is written in C++ It's about 20,000 lines of code and about a thousand of that is actually with the water balance models There's another about a thousand lines of code That's actually with the routing models and the rest of that is really what I call kind of glue code It's code that's just making it all work making it so that you can read in the configuration files You can have all these different options. You can specify the parameters You can specify the inputs and outputs and things like that I'm kind of the point of showing this here is just showing you that you should really kind of adopt these frameworks because You're getting a lot of code for free when you do that and that your models are actually pretty simple compared to the overall Framework of reading files reading data and doing conversions and things like that So this is the system we built For doing kind of a flash flood forecasting It's called flash which stands for flood locations and simulated hydrographs. You can see on the left here We're using the MRMS which stands for multi radar multi sensor and it's the kind of q3 Which is qpe version 3 the rainfall observations You can see that these are at a scale of one kilometer every two minutes So we're trying to build a hydrologic model that can actually take these Rainfall estimates and then propagate them forward into kind of a stream flow forecast produce forecasting flooding And so to do that we built the storm field distributed hydrologic modeling framework Which is the f5 which is kind of explaining there and then we're kind of taking this further now and we're going And we're saying hey, okay, we've built this what can we do with this? Can we predict actually impacts can we you couple this with something like the agent-based modeling and things like that? We kind of tie this all together to see what will the humans do when we issue a flash flood warning and how will they Interact will they actually not drive through flooded roads if we tell them roads are flooded and things like that And that's kind of where we're trying to get so if you've ever had a wireless emergency alert on your phone for a flash flood Warning at three in the morning, you know how kind of annoying that is and you know how it probably wasn't relevant to you So that's kind of what we're trying to fix right as we're trying to give tools To the National Weather Service that they can actually issue better flash flood warning So they don't have to alert people when they don't need to that the people that they do alert can take it very seriously If you're familiar with the national water model, it's just something that's also come online very recently You're probably wondering kind of what's the difference here And this is just a brief rundown of kind of the differences So the national water model has 2.6 million forecast points. We have about 10.2 million We have multiple different water balance models Which I kind of described the national water model only uses no ampere right now the national water model likes to calibrate and We don't we just look at a kind of a priori parameters We don't do any calibration of the hydrographs in our model We do that all driving the parameters for them estimates of kind of the land surface and soil surface On the biggest difference though is the actually of the cycling of the model and how frequently they're run The flashes run every 10 minutes with a 12 hour forecast compared to the national water model Which is run every hour within 18 hour forecast So you can see that you're getting six runs in an hour with flash And so if you have a flash flood that's about to occur is occurring You can issue a warning much quicker because you'll have the information into the hydrologic model much quicker the other thing we've done with flash as we've looked at doing a Sorry with EFI as we've done a hydrologic reanalysis. So with the MRMS precip We actually have a reanalysis that's been derived from that Discovering the time period from 2001 through 2011 which is the time period there is basically single pole radar So this data set is very homogeneous and so that's kind of why we're looking at it So we re-ran that we put in the five-minute data, which was what was collected at that time We put that all in and then we kept just four values as the output from the hydrologic model We kept the maximum stream flow the time of the maximum stream flow and then the maximum stream flow normalized by the drainage area And then we also kept the minimum soil moisture and the idea was to use these values to then kind of develop Climatologies of where flash flood was happening and to see how they're the model was performing with that So this is just an example of output from kind of the hydrographs from the stream analysis So this is a flash flood in Arkansas that killed 20 people in June of 2010 It occurred at a campground where people were sleeping and the courage kind of overnight You can see the black dots here the observations And then we have kind of the three model runs from the three different water balance models So you can see how they compare we're a little bit early in this gauge But overall like you'd prefer to be earlier than late. So that's okay. We have a different Gauge here that was located nearby and kind of a different basin So you can see we're much better on the timing here and you can see that the models the two Infiltration based models tend to underestimate a little bit and the hydrophobic model overestimates a little bit And this is kind of what you would expect you want to be able to encompass the observations with your models Look at kind of a larger scale validation. This is the correlation coefficient across the US So all these dots are different USGS gauges. You can see how kind of the correlation goes there We're pretty good out in the east We don't deal with snow when we did the reanalysis So we do poorly in the West the radar coverage is also poor in the West And so you have a significant drop-off in correlations there, which is kind of what you'd expect You can also look at this in terms of the same information just in terms of now basin area and also kind of bias So you can see kind of presenting the same information here We're going up to what we consider a flash flood basins which are about a thousand kilometers squared And so you can see that across that range we tend to do pretty good You'll see kind of the low red dots where we again in the mountain West We don't deal with snow and the radar coverage is bad So our rainfall estimates are bad so you can see overall We're pretty happy with the performance of the system that actually seems to be working pretty well Again without any sort of calibration We are making all this data available. We have kind of a website here You can go you can actually browse the data and then you can download the data for individual days It's very useful. We're doing some other analysis here So kind of the bottom left plot with the yellows is the maximum unit discharge of that occurred over this entire time period in the bottom right you have the Minimum soil moisture that was associated with flood events We're looking at in flood events was the soil saturated before the rainfall that caused the flood Or was it kind of a rain falling on dry soil that then was just the rainfall intensity was so great That it overwhelmed and produced a flood can also look at the timing of the flood Which is what you see in the top right as a function of seasonality there, too if you look closely at the plot you can actually see and kind of the spring which is the March April May time frame you can see things like MCS coming off the mountains in Colorado and propagating downstream And you can see that reflected in our model simulations as well So one of the other projects we've been doing with the F5 is actually a capacity building So we've been working with their various agencies Particularly with NASA with their severe program to look at how we can use the F5 in Africa to do kind of flood forecasting there To be able to scale up so you can see the workshops. We've done over the past few years here We've done five of them now Various countries and so it's been kind of a big success and we've been really focused with the F5 I'm making it very usable for the users So we have hopefully nice documentation We've made nice kind of power points that go through the training and to do kind of all those things for you So that people can actually learn how to use it so they can take this and they can use this in their own countries On all these workshops, they're in specific countries and specific places, but they invite Representatives from kind of all the surrounding countries There to learn as well and then they can go and they can take that back to their national governments As well and so it's kind of a nice thing and it's a nice thing for us because we've actually fixed a lot of bugs in the Model as we've been working on different things that people have been trying different things. We're like, that's funny That's not working right and we'll dig in the code I'm like, oh, yes, we made a typo there and so you can kind of fix those things as part of this process So it's good for everyone And this is my last slide just to kind of summarize everything again We have a EF5, an ensemble framework for flash flood forecasting, which is a really nice system We're doing hydrologic modeling now We have a reanalysis that you can go download if you're interested in that and we also have kind of the capacity building We actually have videos that are online on YouTube as well And so you can go check those out if you're interested in kind of learning how to use it and walking through the training EF5 is all open source. The source code is on GitHub So you can go there if you're interested in downloading it and modifying it and contributing That's all I have for today. Thank you. We do have time for questions. So well done