 People They'll see that they'll see this okay Yeah, so I'll just pass it along and hopefully they can hear me. I can hear myself Okay, welcome back everybody on this beautiful day So we thought we would wrap up the meeting with a session that involves a kind of roundtable discussion Okay, from my point of view This is breaking up as I talk is that really happening? No, it's okay. Okay So we thought we'd wrap this up with a roundtable discussion and I've asked our panelists our distinguished panelists to spend a few minutes each answering one two or three of questions that we thought had had come up and the first was to Speak to themes that have emerged from this workshop that they think are worth You know what they found interesting found intriguing would like to highlight would like to identify as cross-cutting themes or important themes And the second is to if they found any themes that it didn't quite You know become really obvious themes from this things that may have been overlooked or things that may have been You know mentioned been passing but didn't come back around. I thought they could have mentioned some of that And then the third question is what are the next steps for this community for these two communities as we move forward? And so part of this is for us to hear from the panelists part of it is for us to have a discussion Among this entire community and part of it is to help inform the writing committee That's going to be preparing some sort of report document paper or for this community to go forward Okay, so without further ado, I will just pass the mic along. Can you everybody introduce yourself and then So my name is Yolanda van Wieck and I would like So I prepared a statement based on what I heard in some of the breakout groups that I thought that You may have an opinion about I want to Comment on So here it is so the two communities who are out two communities have been largely decoupled in the last decades Which has resorted into the parallel development of numerical modeling approaches So what we heard here a lot in the last two days were the problems that are still Very much alive in both modeling communities So the problems with the code so the codes are still very much in development So one could make a statement saying that this might not be the right time to start Coupling these codes and move for as a combined community However, because tectonics and surface processes are coupled at least some degree You could also say that code development in both communities should probably not continue to be parallel It should be continued to integrate And then I would like to quote one example of one of the breakout groups that I was in that illustrates What may have gone wrong in this parallel development of the codes So if you study the drainage system development and sediment infill of a rift basin People noted that it was found that for example the dip and depth of the border fault of a rift basin plus The shape of the hanging wall block of the rift basin Controlled to a large extent how your sedimentary basin fills up with sediments and how the drainage system develops So in geodynamic models as we just saw in John's talk The dip of a fault whether for this list trick what's depth it's actually roots Those are difficult to control results in geodynamic models But they matter very greatly in Service processes and landscape evolution. So in other words, it matters a lot in one community Maybe not reproduced well at the moment in the other community So the path forward could be more communication and collaboration between these communities. Maybe more Meetings such as this one and maybe something even like a community research initiative Should we break this up with question response or should we just each march through and give a statement any initial and initial thoughts on that Well, I just add to Yolante's that there's also then the step beyond beyond model group one and model group two You need it's in order for the yeah, you have to use it you have to really talk to it. Yep. Like a rock star Like the rock star you are I would just add that there's also the Observation Community at some point there that in perhaps compassed in this in the Surface processes modeling group, but you can't ignore So, thank you With the caveat that this is an impossible task to come up with a few seemingly intelligent things to say There's a few notes here. So one thing that came across very often in the breakout sessions and I think it's actually why we're here first place need for cross education Right so that each community needs to know well the other group. What do we know? What don't we know? Really like to know really like to be able to predict or what sensitivities we want to understand for our competition It's really essential you'll know what where's the common overlap? Problems that can be worked on the usually beneficial and Right, so I think we're kind of just getting to that point now Surface process either more from what the challenges in the tectonics models are So in a way, we're beginning to get ready to have that Conversation, but I guess one thing I'd like to add is I'd like to see the base Speaking of the data have the base and analysis community drawn into this group or Since many people have spoken to other sedimentary record is you a longer term Preservation of what the surface is Right back to even tens of millions of years. So that really should be part of this conversation And way if you think about at least in my mind the world of coupling surface processes and tectonics Way started the Chris Beaumont and his group circle other Peter Coons and his group were doing it sort of about the same time but for Chris Beaumont anyway came from a Idea of trying to understand that stratigraphic record. That's actually our It's our beginning. We probably should get back to it So one question that I feel like we May be partly because we're just beginning to complete the beginnings of that cross education Other than the breakout groups, we haven't talked that much about what do we want a couple models for? What are the question? What are we trying to accomplish? So where are the opportunities to gain new? understandings on on both sides, maybe there's some surface constraints really help nail down some parameters or uncertainties which approaches are best for the long-term tectonic models and vice versa and all of us of many have talked about difficulties of actually in both sides the surface process and the tectonic side you got a wide range of timescales Play a role or the server record And how do you manage that right even just the tectonic models? We're just hearing how do you manage? Model that maybe wants to be able to understand co-seismic displacements and longer-term mantle convection It's probably actually different models How do we tackle that anyway, we're having a conversation at dinner Thinking about what do you need one of any different categories of questions you might ask So we know that tectonics affects surface processes has an influence on So as we started off with a nice talk by Allison There are really good cases where you might want to use a tectonics model to drive a landscape I'll get better information how the surface responds. So in that case you want fully resolve landscape evolution Be able to take on board that Intensity of shaking or whatever it is pattern of up and pattern of substance On another class of questions Still uncertain whether surface processes really do influence tectonics like they're a two-way coupling A lot of us think so Most of our models predict that there would be so and it should be strong been hard to prove But in that case maybe the only things that matter for you can have a very simple surface process model because the effect on your Upper-crustal stress right, so maybe getting the mean topography right But great Tucker pointed out to me last night that Got to be careful because what if your observable data you want to use to test that model is At a higher resolution of data so that you need to know more details about that Apography or the sediments in the lake or whatever it is Depending on what data you're going to use to test those models might need more resolution or an intermediate class like the Dinas models looking at subduction zone Where the amount of sediment coming out in the nature of the sediment coming off might change the rheologic properties of all Rambling a little bit there but It's important to think about what the nature of the question. You want to ask a couple model help guide What's the perfect way to try to Classes so Guess only last thing just to one of the sort there are potential opportunities where surface observables help constrain Models say mantle dynamics one the ones that our communities talked about a fair bit Is the dynamic topography predicted by a man's infection models? Or different mental convection models predict completely different Histories of dynamic None some a lot very different timescales Those are surface observables that we could nail down say the up with history of the Colorado Plateau Might be able to start figuring out what are the Many of the good to think about What do we want a couple models for what are we trying to explain and where the opportunities? So does that mean the need is for collection of more data or for a synthesis of data that is already exists or for somehow consolidation of data and then ways to link it to the models Where there are new data sets that need to be collected well as a very general answer I would say that both fields are ultimately data limited can do all kinds of modeling you want but in the end That models got to be try to explain some data one of the biggest challenges Often there's a lot of uncertainty in that date like the case. I was just talking about that head new histories of Regional uplift that might be Attributed to dynamic topography mantle convection Then we could do something but most everywhere where that's hypothesized is great debate and great uncertainty of whether it up with it Or it's certainly landed up. Yeah, we need the data. We need the modeling capabilities to test against that data So I I'd like to start out with something that will never go out of fashion and not just specific to this field, but I good quality numerical methods and software so there are a lot of Opportunities there for improving the accuracy robustness numerical methods and for being able to use those methods and Software to formulate the inverse problems. And so one area that I think is relatively overlooked is Developing gradients of misfit functions with respect to parameters. So there's often spatially distributed parameters There's just a lot of parameters that we don't know and we choose them somehow we choose them based on some observations and some intuition but it Being able to take the gradient of some kind of misfit that says how well do we agree with data? That's something where there are good mathematical tools and we could write software in that way Sometimes it constrains the choices that we would make for numerical methods So if all we're doing is a forward prediction based on a known initial and boundary condition We have a lot of choices of methods and if we want to make a prediction whose derivative is not noisy then we have a subset of those Methods and so thinking more about formulating that inverse problem about developing our models and making them fit together in a way that I Can be used to solve those Inverse problems. I think that's definitely an important area Generally modularity of software the way that we go about Standing it to add new processes and The way that we exploit parallelism these are all important things that need to think about Looking particularly at how we compare with data So it's something that I think would be really useful is Software components that just provide misfit functions that just say how different are we from these Maybe very multimodal Sets of observations. So there are lots of different observation types If I make predictions from a tectonics model or from coupled tectonics and surface processes I'm gonna have lots of possible outputs that could be tested and the thing is In like no one person or no one research group has all of the expertise to go into all of those areas Find the appropriate data Figure out how to compare the predictions that we're making on the appropriate timescales I think that's something we can actually formulate into libraries and that would be really useful And so looking at coupled problems Right now the way that we tend to go about doing that is To generalize slightly surface processes people use laughably crude models of the tectonics and Tectonics people use laughably crude models of surface processes and sometimes we target problems that are like nearly overlapping But we use different methods and the way that we try to agree with data what the parameters that we're choosing to infer They may not be very physical because it's actually a coupled problem And so if we put that together we may get better solutions So we're asking about like where are the grand challenge problems where we really need coupled models And it may be these are problems that we're already working on But we're getting kind of non-physical solutions or we're having to over constrain our models We're specifying boundary conditions that aren't physical or we're specifying some parameters That aren't physical they allow us to fit some data, but you we're not really happy about and Ideally when we couple the models we would get out really physical states things that we Believe in and really justify so it maybe we don't need to look for new problems. We just need to Care more about the quality of our of our inference so there's a lot of questions in that area so I'm Really think the idea of having sort of standard components modular things to calculate Data misfits is a very useful and generalizable important step, and I think that A big piece of making that usable that I think we've sort of been dancing around but no one has really said specifically is that when you have Different types of data each Related to something you're trying to fit with your model figuring out how you combine those two things into Something that an inversion is dealing with is not necessarily trivial. So this would be different functionally defining weights and so That's sort of another piece another layer of thinking about the model data comparison and inversion that I Think would need to go along with Misfit component idea. Yeah, so one way of dealing with uncertainty there is if those Misfit functions were appropriately scaled then they could be combined Kind of without needing to do a lot of custom work It wouldn't be so application specific to combine these like multiple contributions towards a loss function But it's the uncertainty in the data that it needs to show up there and of course uncertainty From your model needs to come through as well Yeah, I guess my thought is I I think figuring out how to scale it right in a general way I would love to find out that it's easy Quick me if I'm wrong, but I think what you were getting at was Let's say we have two different constraints say we have a gravity field and we have up left field How do we wait? How do we decide which one is more important? Well, so I would say it should come from how accurate that measurement is and so if we have Really accurate Measurements, then we should try to agree very well with those Some observations can be really sensitive to the Molly running and the questions you're asking another Might get the right gravity field with any model But you'd write it might be another parameter. That's really sensitive to whether you've got the rheology, right or whatever Yeah, so actually I think when you talk about both in wasp program and also model uncertainty Which means you need to run at least a couple times of the model So I think before you're actually doing that you need to make sure you can run Couple like a hundred times the model first. I mean the computation source or or you can kind of simplify your model to do some kind of easy way to run the simulations Yeah, you definitely need the forward model to be reliable and fast in order to solve the inverse problems But there's no way around that Let me remind everybody in the room to please introduce yourselves including the panelists So that because there are people online who can't Yeah I'm Jed Brown I'm here at CU Boulder and my background is in computational science and I the second speaker Helen Wibble from Arizona State University So I'm Lee Rangoran. I'm from Ben Gorion University in Israel and So I actually I kind of so Louise told us that The discussion here should help the people writing the right paper tomorrow So in a way, I think the people that write the White paper are expecting maybe they expected this whole meeting to come up in a grand scientific challenge of questions And although we discussed them a lot. We haven't been able to name them. We didn't really identify them So maybe it's much harder than we expected to come up with these grand challenges And to stay the obvious Maybe we were not able to identify them because as long as we don't have highly detailed surface process and your dynamic models that are coupled Can all really identify all the possible coupling We are just imagining them or even we cannot even imagine them because they are unknown so maybe we first need to have Sufficient amount of such coupled models and simulations to identify these very special coupling or feedbacks and then we can find out So I would also like to maybe offer an imaginable such a feedback and this comes from a Pietro Sturney who's a postdoc in the University of Geneva. He actually asked me to tell you about that So so the Geodynamic Community has recently been able to identify an interesting feedback between climatic oscillations and magma production So I'm referring here to the glacial interglacial cycles that possibly control the volume of magma produced at mid-ocean ridges and Controls there for the topography of the oceans So maybe there are such interesting coupling that involves surface processes For example, if you have very intensive erosion then you can depressurize and Produce more magma and that can affect your climate and that can affect the efficiency of your surface processes So that might be a classical problem to address the highly detailed both surface process model and and tectonic model and Maybe and Louise also asked us to maybe suggest some Some goals can be New future or far future So I think that the the last session that we just had might point out that maybe a very near future goal could be to define an interface such that any Karen to future model from both communities Would like to fit this interface in terms of the input and output that it supplies and then When maybe you're slightly more mature More easily couple such models that were not intended initially Yeah, so I guess in a sense I agree. Oh, yes, my name is Sarah and I am a postdoc in Melbourne University So I agree that that may be That's why we haven't been able to formulate what the big challenges are because maybe we just have to I don't know Make a list of the 10 grand questions that we actually want address And then based on the questions, then we can figure out which models we need to come up with And I guess, you know, in a sense I would be kind of invoking the scientific method of like which questions do you have in a set of making the model This is Nicole Gasparini from Tulane University. So I'm gonna try and be half glass half full, which is very much not my normal personality But I feel like in this meeting We have had people who are really really super experts in Geodynamic modeling and surface process modeling and the most interesting thing For that surface process person is to like go to the things that we don't understand Distributions of floods and try to study it and like poke all the holes and everything that we know and try to find all these details Because that's super Exciting for us as a scientist, right? And so we've been maybe Motivated so much by our own individual kind of questions that we've focused on all of the unknowns in this meeting and Put forth sort of like here's all these warts that we have, right and all of our models are so bad And we need to know all these other things But on the flip side we get a lot of things right Right and so one question I have is like how perfect do we need to be before We start doing this coupling right because we're never gonna be perfect And so I sort of loved what you said which was Some of these questions might come forward when we start doing it because a lot of us have just been sitting in our own box Not doing it So we are not even focusing on those cool questions that are going to come out when we actually start the coupling So I think that's a really great way to think about it Sorry Germany was not fake. Sorry Now I just want to say a few things regarding the big scientific question Trying to address here maybe just To remind people that in many cases other models have actually Help answering or trying to answer the question that First instance, they're actually used to find the question I think it's something we shouldn't really not saying that shouldn't have a big question But you shouldn't make this pendant on big questions and I see also that a lot of Answers to be question will be question. We have discovered that the modeling I've come because we have coupled erosion models So, you know, my recommendation something we could do as two communities is Really work on the interface and and Through this give the opportunity To any member of one of the two communities to use the tools of the other community without having to Either reinvent the wheel or worry about how to And I really like Louie Mores's talk, which is a great Example of what this can be done This ends away Guys, you need MPI here is these are the guys have never done any MPI There's a way we can do and that it's broad enough It's not really specific one But I will also say that something really that I agree with telling them that one is to work We need also co-educate You know, I'm sure many of the geodynamic model is Really no ID different place your erosion and through your erosion and which Circumstances One would be more suitable than the other Down to using the simple So, I think that's anything I would like to see in the white paper But I think that's really essential Not just define questions It means to find I'm gonna go for it. All right, this is Eric Middlestead from University of Idaho. Sorry about the cold if I start coughing So So there's been a lot of things covered already. I'll do my best to add something new one of the themes that keeps coming up and has been Mentioned several times in this panel It was just brought up by Nicole in a different way is that The meeting of these two communities has caused a really interesting phenomenon Where the communities have looked at themselves and said, oh, you want to use what we use? Oh, well, we got to fix all this before I'll let you use it and I think that's a really important step And actually I think that's one of the most productive things to come out of this meeting is for us to reassess our own communities and and even down to our fundamental assumptions, so You know as Arthur pointed out in the talk earlier today is the viscous formulation Starting from the viscous formulation for long-term tectonics. Is that the right paradigm? Is the stream power model even the right way to go or should we do fully dynamic physics? You know these types of fundamental questions are really important But I also like what Nicole is bringing up and that when when are we good enough? When are we gonna start coupling and I think we should start Thinking about that and working together and and I think one of the ways forward and this is something that didn't come up very much except in passing Was what is this coupling? What is this coupling going to look like? What information? can long-term tectonic models give to surface process models and what information can surface process models feed back into tectonics So for example long-term tectonic models don't just give you uplift or don't just give you lateral motion They can give you information about stress and give you information about what lethologies are being exposed when They can give you information about damage So that's a lot more information than just uplift or lateral motion that can be incorporated into these surface process models I'm sure there's a lot more information that can go the other way I'd love to hear some people comment on that besides just how much material have we moved and where have we moved it but I think those kind of discussions before we go through the process of Coupling these codes are important Because we want to know what information we can use from each other's models And then that will help us hopefully answer more interesting questions So that's all I wanted to cover because a lot of other stuff has already been done So Paul Eitznerfer from the University of Pittsburgh. I'm a postdoc there and echoing Joan Browns and Comments on the grand challenge and also we are not giving you the model unless we fixed it That maybe one of the grand shares I'm wondering is can you provide the community with the tool That has been tested so that we can then Build up those research questions so that we feel conflict is that can you give it a coupled tool now? I'll test things like shall I now do a user stream power model and Let it work on the surface or any other model and then let's see maybe there is As a difference that I can actually measure. So what I would need is like I want to fix a car. I need my tool And the grand challenge can't be solved without the tool So I think one of the grand challenges is providing the community that tool Who is Carlos University of Mainz, Germany? One of the things so I Don't completely agree that there is the tool because as a matter of fact in the genomics community, we have seen many codes developed over the last I think we have made much more progress than I had thought possible Now we can do 3d multi physics problems They have learned a lot of lessons about that and the most important lesson is that we should not Should not force one code upon Let the different research groups do what they think We don't actually know the best way to to solve these equations And and that's why I think providing a framework that allows this and and takes common Features of all different codes that have been developed It's very useful thing to do and it turns out there are maybe like four or five now 3d parallel codes in the community And they all essentially are based on patsy as an underlying package And that is because patsy actually wrote we're very happy to have one of the main developers of patsy sitting in the middle here And thanks to that we have also learned that if you want to go into More complicated multi physics problems like couple magma migration with the sphere dynamics You cannot have a loose coupling on the top You cannot say okay I do one time step with the dark sea fluid flow And then I do one time step with the stokes codes and then I kind of iterate around and I hope that it will converge This doesn't work. What actually needs to be done is you need to do a tight coupling at the solver level Within the codes and within patsy there are now frameworks to do this very easily And so so one common thing is that actually patsy appears to work for big supporters The other thing is that many many people are now using more and more this kind of python frameworks and I think the best way forward in trying to provide this coupling is to to allow some kind of Framework based on patsy That allows you to kind of call one g-dynamic solver step with one particular code, which you could change and then couple it with a Surface processes framework also from within python that would work in parallel as well I think there has been a lot of progress and the development of this python tool that is very well supported And I think that make it feasible Providing one tool. I think it's the wrong But making sure that we can use and we have access to them that they explain They're talking about the interface that because some of us don't know how to code and they want to use their different parameters And maybe have a module setup, which is partly already done Here where we can use different modules and then model The way we think we can answer our question So, um, yeah, I very much agree with uh, liron and from round that It was a set of tools or one tool. I agree entirely It was more give us as many tools as possible but um, there should be some more validated and accessible in a in a way, let's say in a simple, let's say simple gui user example Or a nice manual because often receiving code and start reading a code for three months So understand the code. I think is uh, something that comes to the past and shouldn't be done in the future I think this is a Good idea. I think having these interfaces is great. And I think systems has certainly led the way here But I think one thing that we all need to keep in mind and this is something that's going to have to come from The community is if we want to do this as a joint community there's got to be some Talk with nsf or other funding agencies because we can't rely on some euro person To go out and use their research time to do this. So Um, I think this is a great idea and there's it's good to have discussion about this But at some point, you know, there's going to have to be some Some actual work into trying to get funding for this something some kind of systems type organization Maybe it could be called computational infrastructure for geodynamics or something similar to that In a way, right? I mean the these organizations these two organizations that are Here have been have a lot of experience. So actually we're in That's an optimistic statement that we can basically build on the infrastructure that we've already that we've built And that we've developed and the knowledge that we've learned about how to do it So I I'm actually very pleased to hear that suggestion because it's something we have learned how to do And have some experience doing you have some experience doing the csdms. So we have Basically the ability to do it. We have the capability of doing it We don't always have the capacity to do it at the scale that is being requested, right? So that's where the resources come in to To expand the capacity to do this That's kind of Okay, I'm gonna open it up. Oh wait, was there Sean gall in color out of state university Um, so I just want to change gears a little bit. Uh, there's been a lot of chatter about this. So how many people here Model and what data sets to obvious question. I hope it's how many people here collect data That you use to constrain how many of you know of Databases they're easily accessible what sets exist You can give a handful of examples of maybe some Osmogenic nucleide data sets maybe some areas where people compile thermo geometry, but one of the data sets that I collect a lot of is Paras is other metrics of deformation Frank was all you call Heliogiotis the heliogiotic market. So those exist for gps data. They don't A lot of these classical measurements Um, and if we had sort of a community effort By all those data sets So they could easily Accessible and could be used to inform the models that might be another way forward And I see that as sort of a grand challenge to try to standardize these data sets Not just publish them in the paper, but you get them out to the public so they can utilize in a really easy way Eitan Shelef University of Eitan Shelef University of Pittsburgh in So I think to add to the point in how perfect should our models be in order for us Find them and I think this ties to the issue of scale the discussions Perhaps a way to guide the questions with us Find the scale use this as a tool to hone on the questions you can answer I The modeling tools This is katie barnhart So one thing that came up last year at the csdms meeting related to coupling human systems models and surface processes and surface dynamics models Was that oftentimes um You know in an ideal world A person in one community is going to be able to have enough time or be able to find a collaborator or something like Be fortunate enough to find a user manual that is comprehensive enough They really understand the other model that they are connecting with but for a lot of really reasonable reasons that's not always possible and and so Oftentimes the thing you really want to know is you know, what can't I interpret about this other model What shouldn't I believe is true or what isn't necessarily? a feature that is A robust result And and one of the things we came up with in in a breakout session at csdms last year is, you know, we don't really have a way to incentivize uh people being really honest about that aspect of their models and that uh, that is often one of the most important things in figuring out how to Appropriately couple and then interpret the results of a coupled model, especially if maybe you only know one half of the coupling really well So, uh, I wanted to respond to Sean's point there So if I'm not in your field and I make some predictions with my model and it's coupled to some say surface process model that Where in principle it could be related to the field observations that you have the problem is I'm going to have to understand like your Field collection techniques. I'm going to have to understand spatial and oral correlation in your model I'm going to have to understand like how do we do upscaling with that sort of data and This is hard. This is actually like it takes a lot of intellectual effort to do that and so One thing you could do is along with your data. You could provide an uncertainty model Problem is that those uncertainty models are usually really high dimensional So a different way to represent it is to have it be like a posterior distribution So a distribution that I get to sample So I say here's my prediction and you tell me how probable that is so how close to your data is that And embedded in that In that function is How close did we get to the observations that you actually made? That's in fact what we need when we're trying to To solve an inference problem. So we're trying to understand Are all the parameters in our coupled model. We really just need to know how probable is it as compared to the observations And so That's basically my suggestion in terms of loss functions is Help us out make it so that we can actually put together these components Because if we're talking about a coupled model that has all kinds of surface processes plus tectonics This isn't like two or three areas. This is like at least 10 different kinds of data sources And it just isn't feasible for any one person or group to understand all of those things Can I add something in response to it? He just said about this so and a point that's come up in a couple places has having to do with Essentially making it possible for people to use models, especially in areas where they're less familiar And so I'll just lob a couple of ideas out there and see what resonates and one of them is that You know one way to do this is to have tutorials training essentially training and by training I mean really sort of pretty intensive, you know, maybe A few days of hands-on work one of our Software development teams has done this really well Highleth which is sort of short term crustal deformation scales Which they've offered both in-person and online training that lasts a week And it has Built in what they call tanker time Basically you start and you use learn to use the software to solve certain sort of standardized problems And you go away for a few hours In the afternoon or so on with the expectation You're going to keep working on it and then you come back and by that time you have some questions You're stuck, you know, you have things that you can't figure out The tutorial continues and uh the tanker time continues and so it's kind of this way of basically building building expertise Which as you can imagine takes a lot of dedication both from the people preparing the tutorial tutorials to make and delivering them But also from the people who are taking them, but it does result in a in a cohort of people who are More expert users than they were when they walked in the door, obviously the next level up often involves basically bringing Users and elevating them so that they become user developers where they're contributing something back To the software and not everybody wants to do that But some people do want to do that and they kind of opens the door to understanding what's Under the hood to continue the car repair metaphor uh, and essentially um allowing people to Even contribute back in a modest way still gives them kind of more of an investment In that software and so those are two things in terms of the quality of the Of software one of the exercises that we found has helped with that is uh, what we call benchmarking and usually this it's much more than Running a bench than a benchmark it often involves a group of people getting together defining Problems that they believe they understand the solution to and then Putting them out there to the world and having people run them on their software with guidance with Oversight so that we know what's being run and then getting them back together. That is a very big Uh dedicated effort. It can take several years From start to finish to get that done but it does give you kind of confidence in the models and also knowledge about the models and It's something that this group might want to think about what sorts of Benchmarks in the future. Would you like to see in terms of these couple problems? How would you go about defining it? What would even be the process of going about defining Attractable problems that could be done in this way. It's a long-term effort But it really does increase everybody's sort of understanding of what the models do and also confidence in the models And so I would really recommend it and then we also talk about Publication of software and how to publish it and how to treat it. This is part of this reward system Right how to treat it as an intellectual product because it is in the same way that a publication is an intellectual product We're used to the sort of academic publication sphere, but we are Trying to Push and there's a much actually much much larger scientific community trying to do this and information scientists and so on to try and push recognition of this sort of effort as an intellectual product making it both Publishable and then citable and that citability means that when the model uses something It says exactly what it did what is in the model What model version was used so what capabilities that has and that really helps with replicability of science Which is of course fundamental to the scientific enterprise So those are kind of i'm just throwing those out there as things to think about as you think about the next steps Whether you like any of those ideas or don't like whether you some of you might think well, that's too much work Some of them I think that's really essential to what we do. These are kind of ideas to consider as you move forward Oh, Rolf Alto University of Exeter and just One of the things I've really picked up from this meeting is learning a lot about Geodynamics that I wasn't previously aware And I was thinking it'd be very useful one one thing I'm trying to assemble in my head is a a list of things that Are fairly well known in geodynamics that I didn't know but Suddenly but also a list of key problems In geodynamics that may not be able to help it, but at least we know they're outstanding So just just having that kind of knowledge of capabilities and the outstanding questions particularly in so far As they interface with the surface in a way that someday One of the things that I found very interesting was the problem of scale In that seeing all these excellent talks from Geodynamics Very large problems Are being increasingly sliced Into higher higher resolution using all kinds of new options for parallelism to to really decrease that resolution to a spatial scale That is something that we At the earth's surface and similarly seeing how some of the earth's surface models are scaling up to larger problems So other than that kind of list of the key outstanding questions Hopefully addressed by the other maybe just an idea of where key scales Spatially We must help each other Does anybody want to reflect on that or not? If not, I'm gonna ask or you're gonna Present a question Okay, so here's the question from online from Lorraine And she asked asks is there a straw horse grant challenge problem that will illustrate what we know What is not known And what our capabilities are Either To the front or or to anybody in the audience Well, I mean my my comment was sort of along similar lines and I I wanted to come back what you mentioned earlier That the aces is where it all started and and to me some of the some of the issues that have come up over and over are sort of about Validating and examining the cascade of natural deport, right? We sort of want to understand Does our general understanding of plate tectonics and continental dynamics allow us to predict the sediment boxes right now? That's really important for planetary evolution for the carbon cycle. We want to understand say for You know for for continent like Africa. Do we get the overall transport? Like zooming in we got major originic belts of the zenozoic You know the Andes and Tibet and then you know sort of going down in scale You know people have of course done this for centuries trying to understand transports surface transfer processes all these different scales, but I think It would be somehow useful to to just sort of rally around this concept that we need to understand these different laboratories to to check, you know, what are the scales where which algorithm applies the scales have to capture And what is sort of ringing in the back of my head is of course, how are we how are we going to get You know money for this and of course one extent it's sourced to sink, right? This is to lie about it. The all industry is interested in the same sort of stuff But that is an opportunity and then of course we have things like the integrated earth systems program Which is just about that but as a geodynamicist always feel like IES as a program expects this fully formed model Whereas we are just talking about well, how should we come and somehow when you get involved in these projects? You always get stuck never get it done, right? And so I feel like as the service process models and geodynamicist if we could somehow get Get a more coherent strategy going to have these is kind of collaborative things be driven by examining these systems And be serious about the model right to not have as much I've experienced this right. Oh, we didn't get funded. Let's ask a geodynamicist next time I want to get funded and then no one really cares, right? But I think there is an opportunity here to think about the overall transport budgets Like you said, right on the basin scale and what's happening like in this mountain with the basin scale and on a global scale All right, if you want to get an atmospheric evolution on planets things like that Think about cascades of validation exercises natural report Alison Anders, University of Illinois So one thing I was really struck by in this meeting is that unlike the climate erosion tectonics coupling that we were talking about back when I was in grad school where Apography was the only Connection, I think we are starting to think more about other ways that they might connect And I think the example that really struck me as one that might be a place where There are gains on both sides is about fracture spacing or damage or something like that, which is increasingly being thought about in the surface processes and critical zone communities as being driven by topographic stresses, but also by the tectonic history of things and then this seems like this is quite involved in strain localization and And similar problems from the geodynamic standpoint. So My eyes are really opened about how there's something else that might be a way that That the talking happens But also that maybe these some of these outstanding problems Are not actually so disparate. I mean, is this is really a problem about how The properties of the material change as they move through the system toward the surface and then In transport on the surface. It's really more of a continuum thing than Splitting everything into the surface and the subsurface So that I don't know that struck me as like a new problem. Perhaps all of you have thought of that already before but that was new to me If I could just respond to that it's actually a really nice opportunity for possible, you know mutual interest and joint benefits one of the Big unknowns this was mentioned. I think in Brian's talk, especially How does that erosion efficiency? end on material properties We need what we need a lot more data. We probably need some theoretical thinking we ought to work how how do we make that connection make the robust measurements of the trail properties translate them to erodeability essentially Because we're going to need that to know how that fracturing Exactly in that context um, it also matters these kind of interface properties like uh, putonic activity that changes the mountain range And that's something that right now. Maybe nobody really feels like they hold What strikes me is a lot of people saying I didn't know that you knew this and I didn't know that this was an Open scientific question. I think that's really very interesting so that's Already a benefit I would say Are there other thoughts that add to this like people are running out of steam? Are there other thoughts from the panel? We identify grand challenge questions or big questions or at least think about how That would that would be a major contribution i'm karah johnson University of Nevada, you know, I'm postdoc and I have a question for you. What's the we've been hearing a lot about big mental convection Plates subducting what's the small scale of the modeling that you guys do how how fine do you get how far are we from this idea of modeling fracture spacing or Stuff at the scale that a river might interact with or or surface processes Let's take that out I may not offer a complete answer, but I mean the smallest models we saw in the talks were a single fault So that's a common scale that people look at You know, it's just about scaling the size of your model if you wanted to look at Joint spacing in a hundred meter by hundred meter by 10 meter area You could set up that problem Now there's going to be comments on that, but it's the same mechanical equations are also added on the nanometer scale So as long as you can resolve the grain scale or whatever you can use the same approach to model planetary accretion you can model Behavior of crystals and acorniability as long as the physics doesn't change if you go to the earthquake cycle Things are very fast and you need to start adding Also, that's But of course, I cannot address that and at the same time model the whole Yeah, that's fair to opt in from GNS in New Zealand basically Say what Boris said, but we can definitely you choose your scale to a certain extent based on your problem and they're you know, so We can model the alpine fault of New Zealand at several different scales But it's just what processes are you looking at and what question you're trying to answer and we can take say A plate scale model take the boundary conditions from that And and scale them down to a much smaller model so that we can look at the valley scale or you know The fault zone on on sub kilometer scale Yes, so just to add to that. It's this question about, you know, the continuum approximation So the stuff we do We assume that the micro physical processes And the micro physical heterogeneities can be averaged over on some meaningful scale and some some point that's going to break Some point you're going to run into we don't want to deal with dislocation motions if we have a business people But if you really want to zoom in then at some point You're going to have to be open material ahead of genetic that you may or may not know So it's a question of averaging when this come it's come up over and over again With this informal way Organization approaches and course Or you can try to find approximate descriptions I think a lot of it hinges on the validity of these Approximations and when they break down you have to be really careful that you're not Stretching the parameters space and moving beyond the range of where those conversations Okay, so we are coming to the end of our time and that was a very interesting and lively discussion I appreciate everybody's input I'd suggest we thank the panel