00:00:00:00 - 00:00:07:02 Unknown So pump and treat, as you know, it is under all kind of technology, but how you deal with the data is very important. 00:00:07:03 - 00:00:59:06 Unknown And our speakers today, we’re lucky to have Inci Demirkanli and she's a senior scientist and technical advisor at PNNL and with experience in analytical and numerical modeling or multiphase flow and transport application in remediation research and radionuclide mobility, and subsurface energy and storage systems such as geologic sequestration of carbon dioxide. She provides technical support to the Department of Energy’s Richland office on integrated cleanup efforts as well as programmatic and regulatory strategy development. 00:00:59:06 - 00:01:48:03 Unknown As they have emphasized, her technical work at Hanford focuses on performance-based pump-and-treat remedy optimization approaches, characterization and monitoring of contaminants in groundwater and vadose zone, in-situ source remedy, development, treatability studies and implementation, as well as implementation of adaptive approaches for cleanup. In addition, she also supports the EPA related to injection of carbon dioxide into subsurface for storage purposes, as you know carbon dioxide, 00:01:48:05 - 00:02:21:17 Unknown it's very important to know exactly how to deal with it because of climate conditions. And also so again for permitting of industrial scale project. More importantly is she had developed the aquifer injection modeling toolbox, which is very important that many people we are using. Thank you again and would like to welcome Inci. Please, you go ahead. Thank you, and thank you for having me and assuming everyone can hear me. 00:02:21:19 - 00:03:02:10 Unknown So thanks for joining the session this afternoon and having me this afternoon. What I will be talking about is basically looking at and leveraging data driven approaches when it comes to managing our pump and treat remedies, especially related to performance based management strategy development. For these remedies, I would like to acknowledge my colleagues here. This is a collaborative effort that I will be talking about, and I have colleagues from Pacific Northwest National Laboratory that I work with here, but we also work with some colleagues at FCPA as well on this project. 00:03:02:12 - 00:03:35:13 Unknown So I would say, All right, so this is the outline of this talk. We just decided that it might be better to provide that this discussion in three parts. The initial part here is just a background information about optimization of pump and treat remedies. And we wanted to wait a little bit. We want to explain what we mean by performance based optimization a little bit more because it might be slightly different than how we typically approach the pump hatred remedy optimization approaches. 00:03:35:15 - 00:04:07:22 Unknown And now we will talk a little bit about the computational approaches and what might be necessary when it comes to looking at computational optimization of these remedies. And we wanted to introduce you to a framework that we have been working on, which we call it a optimization prescreening tool framework or prescreening tool framework. And this prescreening tool framework will allow us to look at our options in terms of computationally how we can do optimization. 00:04:07:24 - 00:04:44:13 Unknown And in second part, we want to go over the demonstration of this screening tool utilizing the information that we have from Hanford 200 West Point system here, and we will exemplify how we develop some scenarios and how we do some scenario evaluations for pump and treat optimization, just for the demonstration purposes. And in the third one, third part, we want to go over some of the work that we are currently doing in order to enhance this pre screening tool with some deep learning approaches. 00:04:44:15 - 00:05:18:05 Unknown There are two specific examples that I want to go over very briefly, and one of them is about better predicting. While locations utilize deep learning models. And the second one is kind of replacing the fake address book approaches when it comes to computational optimization of pump and treat performance. So just very quickly, pump entry systems have been used for hydraulic containment or for treating contaminated groundwater very widely. 00:05:18:07 - 00:05:46:07 Unknown And usually the components of the system include a wall network for groundwater extraction. Extraction above ground in situ treatment units for the contaminants that we might have, and a disposal system for the treated water. And this could usually be an injection well network to inject a treated water back into the aquifer. When it comes to the performance of these systems, we usually there are two campaigns. 00:05:46:07 - 00:06:15:14 Unknown One is we rely on pump and treat systems and B, we the it gives us a great option to extract a contaminated groundwater, but we also see very declining efficiencies and performance with these systems. So, sometimes we approach these pump and treat systems with some idea that they will take forever to clean the system and not be necessarily very, very successful. 00:06:15:16 - 00:06:42:08 Unknown I think the majority of the issues related to pump systems that we experience or due to initial designs and how we manage these systems. And initially we typically design these systems as a bulk system because we have a problem at hand. And we want to address that problem the immediate risk immediately. So we want to be able to contain the contamination into groundwater. 00:06:42:10 - 00:07:09:23 Unknown And so our designs are very bulk and in our early focus, usually for the pump and treat system is the volumetric pumping rates that we might set up as a matrix for success. However, we was when we start these pumping systems, we see that performance diminishes due to several factors, particularly the heterogeneity that is could be subsurface originating in geology or heterogeneity. 00:07:10:03 - 00:07:50:13 Unknown It's outside it with the plume, gas distribution and usually the scale of the plumes is another factor for the diminishing return from these systems. If they have large and dispersed plumes, which might require multiple volume of upper volume of flushes, that might show as a declining performance for these systems as the time goes and the presence of solutions or diffusion limited mass transfer in different zones in the subsurface might also cause this diminishing return, as well as the recalcitrant contaminants that we might be dealing with. 00:07:50:15 - 00:08:20:11 Unknown So this diminishing return that we see with the pumping systems might not necessarily related to the most effective way of managing these systems. These systems might require more, more dynamic and iterative, more performance based management strategies. And this is why we call this management strategy development utilizing some optimization algorithms as the performance based optimization for pump and treat remedies. 00:08:20:13 - 00:09:00:11 Unknown So our objective at this performance based optimization approaches is that we would like to maintain or if necessary, increase the contaminant movement, removal, effectiveness and efficiency as much as possible throughout the remedy lifetime. And this requires some effective well work and treatment capacity management, understanding that how our decisions would impact the end state of this remedy. So for the performance based optimization approaches, we rely on continuous performance monitoring to gather the data that we would rely on for decision making and frequent updates to the system. 00:09:00:12 - 00:09:36:22 Unknown Conceptual site models. We can use data in. Another important factor for the performance based optimization strategy. Development is the periodic evaluations of the effectiveness of the performance and to remedy lifetime. And this is where we utilize the data driven approaches and the computational approaches and more specifically, computational optimization evaluations can inform the capacity needs or the well network effectiveness or evaluation needs that we might have for achieving a specific end state. 00:09:36:24 - 00:10:17:13 Unknown So on the right hand side, what you see is a workflow that we came up with. So this basically captures this iterative performance based optimization idea highlighting the design and initial campaign treat remedy, starting operating and the and the optimization approaches of this remedy. Optimization evaluations of this remedy, relying on the data that we gather from the performance assessments, whether this performance assessment is related to plume or source containment treatment processes and such that meet the evolving site conditions and the system updates. 00:10:17:15 - 00:10:59:13 Unknown We look at optimizing pump and treat remedies for direct structural mobile network efficiency or treatment process efficiency or injection network effectiveness and such. And we can also evaluate the monitoring network and strategy, develop strategies to increase their effectiveness as we go. So in the in the last year, Interstate Technology Regulatory counsel, ICRC published a performance based campaign Treat optimization guidance document, and this was published in 2023, and we worked supporting the development of this guidance document as well. 00:10:59:15 - 00:11:29:12 Unknown And on the right, what you see is that with the performance based optimization perceptually looking at it, the benefits of a performance based optimization could be great for a pump and treat remedy in the tub. What you see is the initial capital cost of developing the pump and treat remedy implementing it, and then the on and off cost pretty much constant throughout the lifetime of that pump and treat remedy until the end of pumping. 00:11:29:14 - 00:11:54:19 Unknown And this represents a total cost for this remedy to be implemented and operated until we reach an end state at the bottom. What we are depicting conceptually is that if we were to more dynamically and iteratively manage our pumping treat systems, we actually would see fluctuations year to year or every other year fluctuations in our own AMP costs. 00:11:54:21 - 00:12:44:14 Unknown In some years our air costs would actually increase that compared to the baseline one amp cost because we might need to have a significant change to the, let's say, extraction well network for that year to maintain our effectiveness master mobile effectiveness. But at the end if our optimization and computational approaches are able to assess the end state how we achieve the end state, then we might be able to actually bring the pumping time frame to a shorter period of time and we might be able to gain significant cost reduction from the lifetime cost by reducing the plumping time frame for remedies. 00:12:44:16 - 00:13:25:20 Unknown So with this idea and the concept in mind, what we developed was a pre screening tool framework or pre screening tool optimization and prescreening tool. This is a computational framework and we rely on formal optimization evaluations with this tool coupled to the fate and transport models initially. So we create a reduced complexity, fate and transport model of a remedy pump and treat remedy, and couple it to the formal optimization algorithms in order to be able to provide a compare rate of assessment of scenarios. 00:13:25:20 - 00:14:27:10 Unknown So this is purely for scenario evaluations, for developing effective strategies, management strategies for a pumping treat remedy with an end state in mind and computationally defining an end state in our optimization goals. So when it comes to data driven approaches, what we have been working on is that we developed these additional data driven approaches in order to support the framework and the faith and transport simulations and the optimization algorithm, or we are trying to replace the paid and transport model with a data driven model in order to make the prescreening tool run faster and in and broaden the user capabilities for its general use. 00:14:27:12 - 00:15:00:17 Unknown So this figure here depicts our framework. Conceptually, we have three main components to our pre screening tool. The first component is the formal optimization algorithm. That's where we that's where we rely on determine the parameters for optimization, whether it's the constraint or the goal. And then we use some evolutionary algorithm, which is the stochastic approach for solving both single objective or multiple objective problems. 00:15:00:19 - 00:15:32:03 Unknown And then we use these reduced order fate and transport model coupled to the system in order to optimize the feed and transport model based on an end state goal. And as I said, the later approaches that I will be talking about for the data driven alternatives, we are trying to replace that data and transport model component in order to expedite the of the simulation time frame for the screening tool. 00:15:32:05 - 00:15:59:01 Unknown So what I will be talking about now is the demonstration cases that we ran for this prescreening tool and we relied on our data here at Hanford in order to demonstrate how we would utilize this screening tool in our data comes from Harvard 200 West Point entry system. This is a pump and treat system that is located in the Central Plateau. 00:15:59:01 - 00:16:33:07 Unknown It hampered and I'm sure everybody's familiar with Hanford, but the historical plutonium production activities at Hanford for the Manhattan Project resulted in pretty significant groundwater contaminants and contamination and plumes in the central plateau due to waste management activities or chemical separation activities. And this remedy 200 ways pump and treat is it is a very large cap and trade system, and it is located in the Central Plateau for addressing groundwater plumes like carbon tetrachloride. 00:16:33:09 - 00:17:04:10 Unknown Take 99 uranium or chromium and be historically treated nitrate, but under an optimization study, we are currently not treating nitrate anymore and the system is pretty large. Our capacity for this pumping system is 2500 GPM right now, but it is also going to be increasing due to the new work that we are doing for the optimization study for this system, which I will be talking about in a second. 00:17:04:12 - 00:17:39:05 Unknown So the optimization study in several years ago, we have due to performance based and the performance assessments that we have been conducting for the pump entry system, which was operating since 2012, we realized that our carbon tetrachloride plume, which is the largest of the plumes at the Central Plateau, was much larger than we expected and there was way more mass that we than what we initially thought during a feasibility study supporting a broad remedy decision for this location. 00:17:39:07 - 00:18:12:21 Unknown And so with that, and we also observe some declining performance at Sunburst. So with that, and we initiated an optimization study here at Harvard and the purpose of the optimization study is to evaluate increasing the capacity of the pump and treat in order to increase the removal and treatment capacity for carbon tetrachloride. And along with that, we are also looking at potential transition to MNA for nitrate. 00:18:12:21 - 00:18:59:00 Unknown At the same time. So under the optimization study, we are collecting a lot of information and we are looking at the capacity increase. But for the demonstration and purposes for the prescreening tool that we have, we utilize this system just for running some demonstration cases to see how we might need to optimize the pump and treat system. And as I mentioned earlier, one component of the pre screening tool initially stretched to develop a reduced complexity model and we utilize the which we call it plateau to revert to our model, which is the decision making tool for hepatocyte and pump and treat operations. 00:18:59:02 - 00:19:31:00 Unknown We utilize that model to create a reduced complexity model domain for coupling, coupling it to our pre-screening tool. And what you see here is the model domain for the P2R as you can see, it captures the central plateau all the way to the river. And what you see with the red square in the middle is our reduced complexity domain that we are going to be using for the demonstration purposes with the pre screening tool. 00:19:31:02 - 00:20:12:00 Unknown And we were able to bring in the carbon tetrachloride plume so initially our initial demonstrations include only one contaminant, which is the carbon tetrachloride, although this year we are working on addressing multiple contaminants at the same time with the prescreening optimization strategy, development or optimization evaluations with this pre screening tool. So we have the plume depicted in our model and we also have the extraction injection network that we have for the pump entry 200 waste pump battery demonstrated or included in our reduced complexity model setup. 00:20:12:02 - 00:20:56:08 Unknown So in this application we are going to be looking at some optimization goals just again, just to be able to demonstrate how we are using this tool. These optimization goals could be set up as mass recovery, certain amount of mass recovery or maximizing master recovery or it could be set up as pumping time frame minimization. And so with the pre screening tool, we could avail with how these different goals, computational goals, optimization goals might actually impact the lifetime of the remedy and the management strategies that we might be able to develop for the for the pump and treat optimize pumping treat remedies. 00:20:56:10 - 00:21:23:11 Unknown And we have to also define some constraints, optimization constraints for these type of evaluations, such as the treatment capacity can be keeping treatment capacity at certain percent capacity that we already have or do we do we need to increase the capacity or if we increase the capacity, how would that impact the lifetime of the remedy or the management approaches that we might have for the puppy treat? 00:21:23:13 - 00:22:01:17 Unknown And I will talk about this a little bit later, but we also need to set up some wall installation rules or constraints, optimization constraints for these type of evaluations. And we also in the within the evolutionary algorithm, we included some improvements to accelerate the computational process. And some of these rely on increasing the or accelerating the initial population selection during the evolutionary algorithm run. 00:22:01:19 - 00:22:47:22 Unknown So just very quickly to demonstrate, we ran two scenarios and we compared the dose scenarios to a baseline just to exemplify how this type of tool or form optimization algorithms could be used for remedy management. And indeed, what you see here is the description of our scenarios. We have one baseline which has a pumping capacity of 3400 GPM, which we expect to have at the beginning of 2024, and we ran this scenario with the 3400 GPM with no formal optimization effort. 00:22:47:24 - 00:23:19:14 Unknown So there is a static ball network, whatever the actual network was at the beginning, we just keep it static and we run the simulations until 2038, the end of 2037, which is a regulatory time frame that we have for this pump and treat remedy scenario one, we run the same simulations with the same capacity, but we apply a formal well network extraction, well network optimization. 00:23:19:14 - 00:23:52:03 Unknown So we let prescreening Tool develop a management strategy for us for the extraction mode network during this pumping timeframe and instead are scenario two, we increase our capacity and we run again a formal optimization algorithm with the prescreening framework and our goals here for we have two different goals. So each scenario is run with each one of these goals. 00:23:52:03 - 00:24:31:15 Unknown In the first goal, what we are trying to do is to minimize active cleanup timeframe. So we are telling pre screening tool to minimize pumping lifetime. And in the second goal, what we are saying is that we'll try to maximize the mass recovery for carbon tetrachloride. So we are using single objective optimization algorithms here for these scenarios. And our performance criteria is to look at the 95th percentile concentration that the baseline achieves at the end of 2037. 00:24:31:19 - 00:25:04:18 Unknown So we use that performance criteria as a comparative criteria here. So what we are looking at is with scenario one, what would we be if we do a formal optimization? Would we be able to achieve that 95th percentile concentration before 2037? So that is the question we want to be able to answer. So when we look at before we look at the results of those scenario evaluations, I want to go over how we set up some of the optimization constraints. 00:25:04:20 - 00:25:33:19 Unknown And these are related to parameterization and we call those parameterization rules. For example, just to be able to simplify some of these based on transport simulations, we do 1 to 1 while replacement. So our optimization algorithm says that yes, this well has a very diminishing performance. It needs to be replaced and we add another well and take that well off the system. 00:25:33:21 - 00:26:05:02 Unknown So we also assume that each well only has one operational period and an and a fixed pumping rate. So when a new well is added, then you will inherit the pumping rate of the wells that we are retiring. So these are some computational simplifications because running this framework and even the reduced order complexity factor and transport model, running this framework with the data and transport model is still computationally pretty burdensome. 00:26:05:04 - 00:26:40:21 Unknown But when it comes to utilizing data driven approaches, we are able to get rid of some of these rules or relax some of these rules and be able to evaluate or do scenario evaluations with a lot more capabilities. And what you see on the right hand side is some example of real realizations from the evolutionary algorithm. So the what you see on the y axis is the wells and the wells that are the that are shown with the yellow highlights. 00:26:40:23 - 00:27:10:23 Unknown Are the wells existing models. And with the wells shown with the red highlights are the new wells that the realization is testing for the for achieving the optimization goal. And on the x-axis, we have time. So we have some historical period simulation period starting from 2015 until the current time. And then from there on we start the optimization simulations and these you can see for different realizations. 00:27:10:23 - 00:27:45:01 Unknown We are testing various configurations of new wells going into the system and some old wells taken off the dust off the system for achieving the optimization goal. So we talked a little bit about initial population for the evolutionary algorithm and we also do some concentration weighted sampling to create initial populations that would allow us to accelerate the computational simulations. 00:27:45:03 - 00:28:15:12 Unknown So in this case, high concentration locations are more likely to be chosen for installing new wells in initial realizations. So we start our evolutionary algorithm with a better start. And later on I will go through some of the deep learning approaches, examples. We also improve some of these some of the strategies to even further increase the selection of the initial populations. 00:28:15:14 - 00:29:03:08 Unknown So here I would like to go through some of the results that we have from the prescreening tool just to demonstrate how it works and this this flight represents the results that we have for the first optimization goal that we set up for our scenarios, which was minimizing an active pumping timeframe and in scenario one where we kept the pumping capacity constant as the baseline, but we let the prescreening tool to conduct a formal optimization, we were able to achieve the same end state criteria, which is the 95th percentile of concentration from the baseline simulation. 00:29:03:10 - 00:29:30:06 Unknown And with the 8% reduction in active remediation type three. And let me look at the graphic here. What you see with each dot is a realization or sets of realization groups of four realizations and scenario one which is depicted with the orange dots. That is our solution field So you see each realization tested event in the evolutionary algorithm and their results. 00:29:30:08 - 00:30:21:20 Unknown And on the y axis you see the year that this realization reaches the 95th percentile based on concentration and on the x axis, what you see is that with the optimization, how many new wells the pre screening tool is adding for each realization, each simulation, in order to achieve that end state the preference criteria goal. So the black circle indicates our baseline in that case and the red circle that the highlight highlights one of the realization, which is the most optimal realization indicates our optimization result, the most optimum solution for the scenario that we set up. 00:30:21:22 - 00:31:13:24 Unknown So what you see is that we are able to, if we do just a formal optimization with addition of seven new wells throughout the simulation period until the end of 2037, we are able to reduce the total pumping timeframe by two years, which just formal optimization without increasing our capacity. So just as a back of the envelope sort of calculation and estimate for costs and benefits, if our wells, let's say, cost us a million or two millions, but if our pump and treat O&M cost is about, let's say 28 and these are just random numbers that I'm coming up with 15 to 20 million a year, reducing the pumping creek operations two years 00:31:13:24 - 00:31:58:12 Unknown can give us pretty significant benefits with only replacing seven months. So this just demonstrates the benefit of conducting formal optimization for with the end state included in it computationally in our evaluations and state driven formal optimization for managing our pumping systems. And in the second scenario, where we are actually increasing our capacity to 4500, what we are seeing is that with addition of 17 new wells, we are able to achieve the same baseline 95th percentile concentration six years before 2037. 00:31:58:14 - 00:32:41:20 Unknown So we are gaining a six year reduction in the pumping cheat lifetime. But at the same time this requires the scenario includes increasing to pump battery capacity to 45 and that has an additional cost as well. But these type of computational evaluations and then our unit valuations with a tool like prescreening tool, which is still pretty fast to run, can allow many, many scenario evaluations and be able to develop more effective strategies for managing and managing these type of remedies. 00:32:41:22 - 00:33:14:12 Unknown And in the second set of results that I would like to demonstrate here, what we are trying to do is instead of, you know, pumping treat, pumping time frame being as the goal, now we are looking at maximizing domestic coverage and what you see here is that we are still able to we are still able to reduce our pumping lifetime pretty significantly with the with this goal optimization goal. 00:33:14:12 - 00:33:58:18 Unknown But it does not perform as well as just simply stating that our optimization goal is to reduce remedy life that the pumping like that. So we still achieve some reduction, but it is not as much as reducing the pumping time frame. Okay, so just very quickly, what I would like to show you here is the optimal solution for our scenario of the first scenario where we had two, where we had seven new wells, and this is how our code prescreening tool developed the strategy in order to achieve that two year reduction in our pumping lifetime. 00:33:58:20 - 00:34:41:20 Unknown And here are the locations. What you see at the back is the plume that we are dealing with to treat carbon tetrachloride plume and these are already year by year while locations that the prescreening tool selected based on formal optimization for the addition of those seven wells and here what you see is the addition of 17 new wells with the increased capacity and how the prescreening tool determined which wells and where they might need to go for achieving the seven year reduction or six year reduction in the pumping timeframe. 00:34:41:22 - 00:35:22:11 Unknown So this is a just an animation to show you how the plume dynamics and the while network evolution is happening with the prescreening tool, as you can see, year by year, our well network is adjusting to plume changes and our performance. And what you see on the right is that we are able to increase the mass recovery or maintain the effectiveness of mass recovery with this type of formal optimization algorithms, if we are to manage our pumping treat more dynamically and iteratively. 00:35:22:13 - 00:36:04:12 Unknown So how we improve this prescreening tool utilizing deep learning approaches is another discussion that I would like to have here. So the first deep learning model that we wanted to develop is in is basically for better predicting the well about performance locations or new wild locations for better performance. So we used a deep learning model and we used multichannel 3D convolutional neural network to predict future wild performance for a for the model domain. 00:36:04:14 - 00:36:33:06 Unknown And we use this because this type of deep learning model can help us rebalance the pumping rates as you can. If you remember, one of our constraints for optimization was to keep the pumping rate constant When we included peak and transport models in our framework, this type of deep learning models can help us actually do rate optimization in year to year with the for the existing models. 00:36:33:08 - 00:37:02:05 Unknown And it can also help us reduce number of candidate wild locations for pumping triggered optimization simulations. So for current, the what typically is done is to look at the concentration plume plume concentrations and say, okay, high concentration will have a better performance. So we should put our wells in that better performance location. But the performance of the well also depends on the geology and the subsurface characteristics. 00:37:02:05 - 00:37:30:23 Unknown So what we did for developing this deep learning model is we used a plume concentration, we used the data that we have from stratigraphic units, and we used information such as hydraulic conductivity in the in the model domain and utilize the performing data for each extraction well. And we brought all that information together in order to develop this 3D performance ranking map, basically. 00:37:31:00 - 00:38:00:07 Unknown So here green, orange and red colors that you see on the right, they basically categorized the potential performance for a well, the that we might place in the future where is high, medium or low. So this type of deep learning model can improve our effectiveness or the computational or reduce our computational burden as we are running the tool. 00:38:00:09 - 00:38:25:22 Unknown The second deep learning model that we are currently working on is a deep learning surrogate model for completely replacing the plate and transport component of the prescreening tool to speed up the scenario evaluations and be able to make the tool available to a broader set of users. So we use unit algorithm and for developing the surrogate modeling. 00:38:25:24 - 00:39:07:06 Unknown And in if you look at the historical use of this algorithm, it's been widely used for inputs like hydraulic conductivity or well locations to conduct some predictions of water level predictions. In our approach for our 3D model, we are improving the well location input by replacing it with a theme like analytical solution so that we can actually provide more physically based information than just a point location that we historically did for root locations. 00:39:07:08 - 00:39:43:12 Unknown And when we apply it to the unit 3D model, and this is this is basically just to show you conceptually how this unit 3D model is going to replace the reduced order complexity model that we have in our framework. And when we run this model and this is still in development phase, but we were able to do some training results and we were able to get some pretty significant testing accuracy our cleanup level here for carbon tetrachloride is 3.5 micrograms per liter and we achieved an accuracy testing accuracy believe that. 00:39:43:14 - 00:40:23:01 Unknown But our model predictions suggest it showed that our prediction error increases with time. So we are currently working on reducing that prediction error in the future. As we you know, when we look at the pump entry, these we are looking at long term simulations and this is another way of looking at some of our preliminary results. So what you see here is that our input is that the top row was or is our input for the theme like solution for drove down for each well pumping. 00:40:23:03 - 00:40:54:00 Unknown And we also have the plume distributions which is shown in the second row and we developed a methodology where we are using the previous year's plume images as the for the following year. And the third row is the target plume distribution that we get from the data and transport model and the fourth row is our prediction from this model, a unit model that we are working on currently in very bottom row. 00:40:54:00 - 00:41:25:19 Unknown What you are seeing is our prediction error basically, which is the prediction minus the targets that we have. And if you pay attention to the scale, it's very small. You might not be able to see. But our maximum number there for the difference is 3.4, our clean up concentration. So a lot of our prediction errors that we have actually are lower than our clean up standard that we are trying to reach concentration that we are trying to reach. 00:41:25:21 - 00:42:05:06 Unknown But again, this is a work in progress and we are hoping to further test the unit model with some scenario evaluations within our prescreening optimization tool. So just as you know, as a summary campaign treat optimization prescreening tool, it allows evaluation of system behavior for multiple scenarios and multiple objectives, and it leads to proposed active management strategies to achieve certain optimization goals and to and certain end state. 00:42:05:08 - 00:42:35:18 Unknown So formal optimization of pump and treat while work can is very important in terms of better the network size of all locations and pumping strategies and it could also allow us to look at the treatment capacity considering and it can help us to develop strategies that can reduce cleanup time frame or in and also increase mass recoveries.