 So this is breakout session three, mitigating groundwater model uncertainties. We have four questions. So we're going to do, we have 50 minutes, so about 12 minutes per question. So the first question is, reflecting on model frontiers discussed during the panel, what are the most important groundwater modern uncertainties to address? And our rapporteur is Loreline and she's going to start typing up there. So if you don't like what she typed or you want to add something, please tell her. So we can just do it with one pass. So what are the most important groundwater model uncertainties to address? To address? I think it's a loaded question. I mean, I guess it's for anything. So I guess to estimate flow and recharge. Estimate flow and recharge, obviously. Yeah, what are the questions we're trying to get at? Yeah, that's what I was thinking. The first thing that the answer is TB, like what do we want to use our models for? Flow and recharge, so. No, but I mean, I think that, no, but I wanted as a point up there. Yeah, that's- The uncertainty that we need to address is what are our goals with the models? Because we have a lot of different goals and we need to address what it is we want to get out of them first. Since the topic of the workshop is flow and recharge. So I would just say flow and recharge. But flow and recharge to like deep confine systems, flow and recharge as relates to. Everywhere, all the time, at all scales. That's an open-ended question. Yes, I agree. But I think adding scale in there would be really relevant because a lot of these continental models compared to the scale of local decision-making is quite challenging. And in it to Laura's point of why are we making these models, the sort of the end of the discussion on the panel about different geologic settings, being modeled so completely differently and that we perhaps do have a bias towards alluvial aquifers. I mean, I thought that was sort of an interesting takeaway from my perspective. Okay, even if we put in the scales and the purposes, I would say the most difficult groundwater model uncertainty is probably figuring out the subsurface stratigraphy. Would that be reasonable or properties? I mean, stratigraphy is one. Yeah, more generally, yeah. Subsurface hydrogeological properties, right? If you're a model, that's true. If you're interested in Tony's question, that's totally not relevant in a way because you really just need to know how much the people are pumping. If you don't know that, it's a waste of time, right? And what's the response to that pumping? Those two things are the most critical. So we could put subsurface properties and like human. Human factors, human factors, yeah. Anything else? Yeah, and I think the human factors is kind of like two parts. It's the human use, but also, we talked about this yesterday, infrastructure, so that you could actually understand like what are their other options if you wanna do any sort of modeling, not just like how much they're pumping, but actually like what are the wells they have, what's their availability, the other sources. But remember also the second question asks you specifically what kind of data, proxy data and approaches could be helpful in minimizing model uncertainty. So I think we want to focus more on what is the uncertainties and then we want to say the specific things like, I guess that would make sense, right? Okay, so I guess I'll rephrase. I would say the uncertainty is how much groundwater we're using and what are viable, what controls are alternative uses? Or like what are the factors driving decision making? No, I'm just kidding. What did I say? The uncertainty is how much groundwater are we using and what are the factors that drive the decision to use groundwater to pump? Yeah, what is the storage capacity of the aquifer, right? But then beyond that if you say how much of it is fresh water, how much water is actually potable versus water that may be too saline, deeper in the aquifer? We didn't talk a lot about, I know in the list of panel questions was biogeochemistry and the movement of contaminants through groundwater. I don't know if we want to speak to that. That made me think of that when you said quality of water. I think that we probably do it in number two. But if we get specific to the data sets, I think that should move to the second one. Yes, please. From online, we have 11 possible orders of magnitude in hydraulic conductivity as mentioned yesterday. This is a major source of uncertainty since it could magnify the effect of other uncertainties propagated from other components of the water cycle. I realize this doesn't fit under this heading, but it doesn't fit anywhere and maybe it doesn't matter since our sponsor is NGA, but one of the things that isn't anywhere is sort of how we convey uncertainty. And I don't know how that plays in here, but I do think that's a really difficult thing when we're thinking about stakeholders. I think how you convey uncertainty, I think it's a reasonable question. I agree, I mean, different models have different ways to characterize it because I think uncertainty is input uncertainty, process uncertainty, parameter uncertainty, and that all results in output uncertainty. So I agree, that's a valid question. Who the end users are, yeah. Let's go to the second one and start listing the data where I think the ones on subsurface hydrogeological project properties. Can we go to the second question, Loreline? Thank you. I mean, it certainly seems like age of water is a key one for. Oh, yeah, absolutely, yeah. Quality of water too. Let me, as a sub point on quality. We could also do better using the data we've already acquired. For example, Stacey's remark, and this is something I've talked with some of your colleagues over the years, you're collecting all this data and no one's analyzing it. There's a lot of information. He's continuous water level records about system behavior, system response to stresses, and we're not harvesting that. So I think we could do a lot better on that score and I think that's something we should emphasize. Stacey, is all the data which GS has collected, is it all available on publicly viewable databases? Yes, it is. Now we have a policy that started in 2016, all data that used for figures or tables must also be publicly available with a DOI identifier. So everything we publish is now without data. So you have water level data for all over the US? Yes, yes. But there's a climate, like there's a, go ahead, Laura. Well, I just want to point out that I think the USGS is really focusing on making this better. So you don't want to throw it at the USGS. And you do have a lot of data that is available, but it is not necessarily making your data. I mean, I can publish my terabytes of model output and say they're published and you can use them, but actually making them really useful and usable for modeling is different than having all of our well logs that you could look them up and find them somewhere. And so I do think that USGS has done a really great job with stream flow and with groundwater observations to are actually really easy to scrape from the USGS website. But I think some of the stuff like talking to Burke about like the SkyTem data, things like that, like how is that going to be available as part of like the national map or in ways that are consistent and don't require an entire separate project to try to stitch it all together and get it in a consistent way. Right, and I think that's true for even for stream temperature data, like water quality parameters too. So I think it goes back to this comment about, right, so we collect a lot of data, but I think the synthesis of that data in usable frameworks or modeling frameworks. And I think that also not just that, but I think that a lot of large scale continental global data sets are the same way. I think that they need to be available. I think about surface water, like on the catchment scale basis rather than kind of gridded outputs I think would be. So I think that that's a broad comment in general. I think it's a really good one. So I think that's how I'd maybe freeze it. And the great thing is like, you already have a policy of publishing it all. So it's not like you have to fight the fight of making it open. It's just like there's like a next step needed to make it really usable. Exactly. Yeah, I mean, there's a lot of information about subservice hydrologic properties embedded in those continuous one level records. For me, it was, yeah, I mean, USGS is fine and US is fine. I'm looking beyond US. And Stacey, so all of the data sets which USGS publishes, and this is my own personal interest, it's all in an easily searchable database. That's what Laura's point is. Some of it is like, I think Streamflow is, yeah, they're trying to get better at that, but it's not right. But groundwater, yeah, you can do a data pool, but right in terms of whether it's perhaps usable out of the box might be. Or like if I wanted to build a subsurface for the whole US, I could go through and find every USGS model that's published. I mean, the USGS has built mod flow models all over the place. I could go through, find all of those. They're all published, they're all accessible. I could dig through their inputs and use that to improve what I am starting from, which is a national data set. But that's a giant undertaking in itself. So I think with the USGS is working on a national groundwater model, so I think that will actually address a lot of that in the US. Globally, that's even a way harder challenge. Right, yeah, so in theory, yes. But I think, right, the two points that are raised is that there has to be a better analysis and synthesis of our data to be usable at scales relevant, you know, either global or sub, or continental or subcontinental. So I think that's kind of captures it. Yes. Yeah, I agree. I think it's good to emphasize the data standards in US and say this is a model. Like, you know, I always say that Kansas or the Ogallala Aquifer and Central Valley, they're great examples because that's the role model. You don't get any better than that. I mean, it's difficult for us to envisage right now, better than that. And USGS for all over the US. But the end story is going to be global. And I would add to the second one, data simulation, because it says data and approaches. So we can add data simulation to the second one. But Venkat, just as a side note for USGS, that does not relate to the model is there is now, if you read a USGS paper, there is a DOI identifier that has to have all the data that was used to build the tables and figures, which is great from an authorship perspective, and it's always super, because a lot of things you have to pump. But anyways, you can access that now, maybe not in a usable format, but I just wanted to mention that. And Venkat, I hate to say this, but I think you're missing the point. And the idea here is we can develop insights from the data we have in the US that we can then take to other areas. And I think that's a very profitable approach that we should not ignore. Yeah, because you could create a data sparse data set from a rich data set, right? And see what methodologies might be usable then to apply to a data sparse. Well, also for modeling, right? The calibration, validation, or models, you need a rich data set to go to begin with. So then you can apply globally, you know? That's what I said, that California and Kansas and the aquifers in the US are models, are role models for other players. So if we go to a different country and say, if they say, what do we have to do? And we can say, do this. Venkat, let me just correct you. California has no pumping data, just this. Oh, Kansas, Kansas. Oh, there you go, thank you. Okay, I go on the record to say Kansas is the best. Read my lips. No, I agree. I mean, I think that USGS as an agency who has coordinated groundwater data in US, including putting all those things, I think taking it to the next level and ask other countries of the world for their models. So the models published in Europe are all their metadata and their data for all the tables and figures are all available too. Yeah, I mean, I think that's, well, I mean, because this is another can of worms, but I think that's the standard in academic publishing, right? Is if you're building a model and you are publishing it, that you're gonna use other data sets that are citable or you'll be providing access to your outputs. But there is actually, I mean, this is just kind of a tangent, but there's a huge issue in like actually what we do with large model outputs because we have all of these efforts to create better repositories for data. None of those are really set up for handling model data and it's not really clear what all of our model outputs we wanna be saving or making them, does someone need to be able to run your model? So I think that's something that the modeling community is just like currently struggling with in terms of documenting and making models available. But I think there's a big, I mean, in general, what I see from people building models in the academic setting is they're like very open and sharing. So I'll try to ask Matt to comment on this because NASA does a great job of it. I use NASA global models all the time. I don't run it. I use the outputs all the time. So Matt, do you want to talk about the DAC and how they, storing the model outputs? I mean, you store tons and tons of model outputs. I know that, so. Mike. NASA has the DAC, sorry. Mike. Oh yeah, sorry. NASA has the distributed active archive centers for doing that, so it makes it, I barely have to think about it, other than we set up a relationship with our DAC, which is the God Earth Sciences Data Information Services Center. And so we deliver our data there and they are experts in thinking about what metadata are needed. You know, restart files and what format, reformatting for people and allowing people to subset the data and that sort of thing. They're far from perfect, but you know. But that's also a huge, I mean, you've got like staff. Do you have people whose job it is to they all think about that? Like I barely even know about it because I just, you know, I send data over there and they ask me questions. I answer a few questions and they get it done, you know. But not everyone has that opportunity. Range storm and think of what can be done, similar approach to the academic community. I mean, is that, can USGS take some of those results and put them on their databases, Stacey, or that's not allowed because you are a government agency and you are not, because NASA DAC, can you take that data and, you know, data sets from these big groundwater models, especially if they compare with GRACE, let's say. I mean, as a site, then say, hey, I'll put it all. Now we're talking about, you know, our community where we want to preserve the data sets generated by you and your students and your collaborators. Well, go ahead. Well, I was just saying, I mean, we, you know, it's not a store all for everything. You know, I have to provide some funding. You know, it's just, I don't know what it is. You know, $15,000 a year or whatever to store the GLDAS output. And, you know, I imagine the big missions that are storing data in the DACs, they have an allocation of funding to pay the staff at the DAC. So it's not like you just say, like, here's a great data set that's really useful. Can you store it for us, you know? Yeah, I was gonna say, it's not that we don't know of places that we could put things and there's like the EarthCube out, there's a bunch of efforts. It's kind of like a bigger question of how we decide what, I mean, we'll have petabytes, like the scale of outputs when we're running really large models globally is just huge and figuring out like how much, when do we like draw the line on just recomputing versus storing versus, so there's the people who are like in DACs who are thinking about that, they're not the only ones, right? And there's a whole group of people who are thinking about that challenge. So I don't, like, I don't wanna like take us too far down that, but I would also say that like, that's the academic community. I think there's also a wealth of information in models that are not built within the academic community, especially at smaller scales, and a lot of those are built by consultants or just other groups of people outside academia and those are not often published or you might not be allowed to have access to them or if it's a government model, things like that, that there's a huge amount of models that maybe for academia it's a big problem if you wanna like try to stitch them all together, but at least you can like contact people and get information, but I don't think we really know all of the other models that are out there. I mean, you showed your figure where there were eight countries sharing groundwater data, like everybody still has some and is modeling it, we just don't know what's going on. Yeah, I wonder in the government, I'm sorry, did you wanna, in the government management side, there are interagency coordination efforts like the US Global Change Research Program that could potentially facilitate interagency hosting of data in this example, if there's a global change bent to the data set that could potentially be a way to integrate data and USGCRP then has international connections to kind of sister agencies or interagency coordination internationally. So in the academic community there's that discussion going but there could be potentially the federal kind of management level, there could be some opportunity there to engage with coordination, interagency coordination efforts, so. I do know when NSF has proposals to review, they look at data management plan and I know that I do not know if you have to pay money to quasi to host your data, I don't know the answer because that may be a place. Yeah, so we have a cyber infrastructure project with quasi and with HydroShare, I mean, quasi's will take your data but not if you have 300 terabytes of, they're not ready for that and it would be unfair anyway, like if you have huge outputs then like you need, you're gonna have to pay something somewhere to put them and to store them. I'm a commenter. People would want that one. Oh, sorry, go ahead. Yeah, people don't want that. Yeah, just as a comment, you need to figure out if there's a demand for this stuff, right? Every other person could have a model and petabytes of data and these are not necessarily maintained because they're three year research projects and there's a beginning and end date. So if there's no clientele, then it becomes a burden on a data center, I would say, right? So you need to have some threshold or criteria and to select which ones need to be for the benefit of the community or something like that. Yeah, that's exactly like kind of the point I'm trying to make. It's like, it's not that we don't, we have places we could put things, we have data standards. It's just the question of like, we're just running these huge things with huge amounts of outputs. We don't want, nor is it gonna be useful to have everything sitting somewhere. But I feel like this is tangential to our, like run a tangent here. Yeah, I'd like to add some things for data assimilation and link to the model outputs also, is that there's really no use to have raw data. You want systematically estimates of the precision, the accuracy of the data that you put out. And I think this is still greatly lacking also from the USGS, is that there are numbers that are put out but we don't know how accurate they are. It doesn't mean that the results need to be perfect. We just need to know how precise they are so that we can aggregate and merge different types of data. And that's lacking, I think, in most data sets. I haven't seen, like, we should have separate columns where you do have the number and then you have confidence intervals around that number. And I would love that to be systematic. I think that's an excellent thing to add. Yeah, definitely add to the list. Yep, yeah, it's a good point. Carly. From online. And we also need to include algorithms, analytics, and synthesis techniques, which are part of pre-processing and post-processing activities and also sources of uncertainty. So can I add the two things? One is if you heard the talk yesterday by Ed Bagley, he was talking about when SWAT is launched and they're gonna be producing over a terabyte a day of data and they're gonna be using the cloud, I think. So I guess NASA's already given up and said the DAX can't handle more than a terabyte a day so we've got to go to the cloud. I don't know what that means in terms of... Is the cloud free? It goes on at that much storage, I don't know. Is the cloud free? Is the cloud business free? No, it's not free, okay. Yeah, nothing's free. So just one point, like cloud may be an option, but I think more importantly, there may be petabytes of model output and people are sort of alluding to this, but you can distill that down to what people really want, which may be they don't need 15 minute time step that can use, maybe they would just want a daily time step and they don't necessarily need the flow between every single layer in the model, they just want to know what's the recharge and what's the base flow and what's the flow between larger grid squares or something like that. So if there were a set of desired output quantities from the model, some sort of standard set, then you could begin to ask different modeling groups, like this is what we want and it could be much more manageable than trying to just aggregate all the model output from every model. I also think, like speaking to Khamenei's point about how to convey uncertainty to stakeholders, I'm just wondering as we think about all of these large scale models getting developed and then the management side, how does a manager discern what model is best for their purposes, what guidance or standards, right? Can we provide so that, right, they know to use SWAT over mod flow, over some other type of model framework that suits their needs best, strikes me as a manager would, management would have a hard time under someone interested in those management questions would have a hard time. Knowing how to interpret the uncertainties and then know how to apply which model or which assumptions work best for their questions. Is that working? I was just thinking about, in the metadata specifying something around this, in the description of the models. Sorry, I missed the start of your point, sir. Yeah, basically what I'm asking is under which bullet point or do I create a new one? Yeah, how to interpret the output and assumptions of the model in order to apply them to the management questions that someone might have at hand, the specific management questions, yeah. Thank you. Sorry. Okay, so you want to take a stab at the third question. I know the third and the fourth questions were questions which were common for the first two breakouts as well. The third question is modeling ground water. What are some examples of existing or potential application of NGA resources? Vin Kat, can you overview NGIU resources for us? No, I cannot. No, I think the whole point here is, I mean, as I said yesterday in the breakout in this room is I think they're starting afresh, okay? At least in this direction. Now, I do not know the other directions because I've never done research in the other areas. So, it's not what they can give us now, but is there things we can ask them which they can attempt to give us in the future? Doesn't mean they'll give us because yesterday somebody said they want high performance computing and Matt eloquently put it saying that they don't even allow you into their facility with your cell phone so you cannot get an account on their supercomputer to which I said that maybe they won't give you an account on their supercomputer but they could say we'll buy you time on the IBM's computer somewhere. Something else. I mean, obviously not in their facility, I agree. So the same way, if they do start collecting data, somebody said, can we collect high spatial resolution data all over the globe? Who knows what they will do? I don't know at this point but in the next five or 10 years that's the kind of the horizon they're looking for. They may make things available because this community and groundwater requires it. You know, I mean, again, we should not, we should not try to do it what they have now but what we want. So for modeling, what will be the best thing they can provide? Because we have already done this thing with characterizing groundwater aquifer first. So I guess the ones in yesterday's breakout had some things about stratigraphy, 3D, 3D shape and everything. But for modeling, what would be the things which they can provide? Laurel. Yeah, so my data dream would be to have really, really high resolution satellite imagery that you could read and really highlight water infrastructure. So if you could know where each individual pumps are. So very, very fine scale imagery and you could recognize this and differentiate between types of agriculture and infer also where you have or charts that are irrigated or not. That would be the dreams that you can infer better human recharge to groundwater. Okay. Right now, yeah, absolutely. I mean, absolutely. I mean, high spatial resolution will always help, especially if you're going to work in a small area, then having one meter or sub one meter data will be much better than having 30 meter or one kilometer data. I agree. So. Just a comment in this, in that context, the previous comment, maybe you can request them to deliver a product to this community. They have the high resolution data, which is not just publicly available, but they could do an analysis maybe and answer that question. I think it came up yesterday that we talked about like proxy data and more soft data. I think in addition to the satellite imagery you talked about of really seeing what's happening, if they have some additional social data that could help us infer when there are crop failures or when we're needing to drill deeper wells because the groundwater has fallen below the levels of existing wells, things like that are electricity usage that could really help us understand the systems too. Are they interested in keeping your modern output? That's hosting the model outputs. But I would guess for their projects, so it wouldn't be for a project funded by some other agency. So yeah, that would be reasonable. Yeah, that could be a point that to hosting outputs from global groundwater models, I doubt if they would be helpful to host the outputs for US based. Well, they've opened the question, so just ask, make another question. How are they gonna host them if we can't get on their systems? Again, it's the same issue, right? Are they gonna pave someone else to host them? Maybe that's possibly, but they're not gonna host them on their system for us. That's true, that's a good point. I'd like to say it's more than just like, we need somewhere to put things. I mean, if you wanna facilitate water managers being able to, or maybe not even water managers directly, but local water planners being able to say like, well, I'm gonna grab this global groundwater model and start from there and building my model or adding in local stuff. It's not just that the outputs have to be sitting there. You have to have tools for, I mean, it's like what you talked about, Matt. Like you have to have intelligent tools for sub-setting and converting to data formats that people want and generating those as inputs to the other kinds of models that people use. So it's really like, you're really looking for more of a community platform, which it just doesn't really seem like that would be, this would be like what they would be into. Like it seems like better that we should ask for data out than like being the gathering place for that type of community interaction, but maybe. You know, high resolution imagery from geostationary satellite strikes me as we could get at a lot of issues in terms of timing of irrigation, potential recharge mechanisms, so that I think that's something they could help us with. Excuse me. Well, I mean, even just photography, high resolution photos of these areas I think we could do a lot with. I imagine they can do pretty well on the resolution front. Actually, there is commercial visible imagery, worldview, iconos, they're meter, submeter. I mean, you have it already now. I'm sure. Say they're not geostationary, yes. I'm sure these people can do sub one meter imagery without a problem. Of course, I agree. I was just gonna add, can the NGA do RFPs like for basically institutions to develop tools that would make it easier? So like RFP could be another way where NGA doesn't have to be the repository. They could maybe put out an RFP to have people already working on these. So NGA has BAAs, broad agency announcements, and they have it open for five years. So it's online. You can go and any university can respond to it. But in this BAA, I don't think there is RFP for storing data. So that could be, yeah. Not just storing data, but developing the tools to make water managers. I mean, I don't know if that's probably maybe not something in your, you'd be interested in doing. Yeah, I think like it's a nice thing. I just don't think that it's really useful to this conversation. Like I think that's just like really a different group of people that would do that and that like that's not, I don't know what NGA's resources are specifically, but it seems like if we're asking them for things, we should be asking them for something that relates to their resources. And like hosting model outputs and community platforms, I just feel like that's a little like, we could talk about the ways to do it and how they could or couldn't, but it just seems like why don't we ask them for things that they would have, maybe. I guess I was just thinking if there's a need for it, then it's whether it's relevant to us or not. Then, but yeah. I understand. Being in the room with the NGA representative yesterday, he seemed to be saying that there was actually a really pro diary of things that they were doing. And in particular for geography data, where they organize workshops and exchanges of data. And I don't know if you understood better than me. I just think. No, but as I said, where we are at is the ground floor. I'm not saying that they would do anything which we suggest or they may do everything we suggest. So the question is if we don't put it forward, then we can never expect anything. And I think Stacy, your point is well taken. If some of these high resolution outputs, even over the United States, can help us learn lessons for global analysis. Well, that could be a small RFP. I mean, that could, I mean, it could be. But again, they may not want it. But that's, and let's go to the last question because people are already coming into the room from outside. So what are some of the examples of successful collaboration opportunities? What are promising partnerships that could help advance our understandings but from view of mitigating ground water modeling uncertainties? Manu, please. I'll propose this. The USGS, because she's here. Joins with the NGA and with the university community. So this is now super specific, like unlike yesterday, right? And we take the vast amount of data that is available to the USGS and other agencies in the United States. And we move to the world of artificial intelligence and machine learning, combining that with the kind of models that Laura is doing, where instead of running more and more simulations of this, we have a certain set of simulations that can be augmented by all of this data. And our goal here is to identify stimulus response models using machine learning and clustering of parameters that can then be used anywhere. So we move from this Newtonian idea to a modern venture where physics-enabled machine learning actually translates this vast amount of data to something useful because otherwise we are still churning in the old universe. And this would give the USGS an incentive to actually make the data into a format that people can do something with. I like it a lot. I think that would be very specific. Can we include NASA in it too? So that the relevant NASA inputs, because when you're trying to assemble this database, we want to be as encompassing to this workshop as possible. Again, I don't think we need all of the NASA DAC, but somehow, Matt, would you agree that that could be a reasonable interplay between USGS, NASA, and the university community who has produced a lot of output to be merged in so that it can be mined with artificial intelligence and knowledge discovery for questions. And maybe it can even start with just US-centric to start and then move outside because most of the stuff right now is US-centric. So that could be a, it would be a very easy no-brainer start if they would concur with it. Would you be asking NGA to do the analysis? No, I think the NGA would be providing the resources to USGS, NASA, and the university community to undertake it as a part of a team effort, but ultimately the benefits would go to them because they can see what kind of stuff can be done in the United States. And then, I hope I'm paraphrasing what you're saying. You're actually right. So you look at Africa, which is data-wide, essentially, but there are certain aspects that we can actually get information on in Africa and we have an analogue for those physical conditions in the United States. So as far as the physical aspect is there, we can nail it. Now, there are demographics of use and all that are still a different issue. I like that a lot. And then this could give Tony something to work to say, hey, I need these resources and, you know, again, it may go nowhere, but still, there's no wrong in asking, so. Certainly fits with the USGS water direction of integrated modeling and trying to achieve that. And I think having, I can't speak for the agency, but it seems to be in line. And I think also having an end product or a goal, right, a reason to integrate, a reason to do these things certainly makes it then attractive to want to engage. When, yeah, I think it makes it a lot more compelling case for us to want to, yeah. I do not have a fondness for traditional neural network models. I know that quite a few people in the groundwater modeling community with their local and regional models were able to approximate the results from a very detailed model with fairly parsimidist neural network model. And so, you know, we've improved quite a bit in understanding how to do that kind of thing now, compared to what I saw before. So this is hope. There is a gentleman, Paul Gruber. I just met at the break who's very interested in artificial intelligence, but unfortunately he's not in our breakout. But yeah, like I was just telling him about that's a frontier that, yeah, certainly worth exploring. So I want to call on Kamini, Ryan, and Helen. Do you want to add anything? Mike, Mike. The question is, I have nothing to add. I don't really have any additional bullet point, but I like Manu's suggestion a lot. I think one of the challenges that we run into as we try to extend US studies to other regions with, particularly with machine learning, is that the US has a limited number of, you know, it's got certain geologic properties and it's aquifers that may or may not be representative of other regions. So it's just something we'd have to think carefully about where it would extend and where it would be less applicable, because, you know, anywhere you go, you're dealing with complex geologic environments that are challenging to characterize. Okay, I think if we don't have any more questions, let's take a break. Thank you so much for this spirited discussion, not withstanding Jim Butler's inappropriate comments.