 My name is Julien Jenin. I'm a geophysical data scientist here at BlueWare, and today we'd like to showcase some accelerated interpretation methods using interactive deep learning on the OSDU. So we see that there's a variety of different business challenges that have surfaced within the recent years, especially within our industry. We've noted that we have a significantly vast amount of libraries of data, and they're growing at an exponential rate. BlueWare has developed some various solutions around that, namely VDS and cloud systems, which can compress data while eliminating duplicates, enabling us to do more. We also see that teams are expected to manage more assets with fewer resources, and so many evolving technologies today are enabling us to leverage deep learning to help reduce these cycle times and really maximize prospect identification. However, we also see that companies must now focus on both the digital transformation as well as the energy transition. And so utilizing the combination of VDS and deep learning, this can provide a more agile way in the methods that we work. The cloud will help drive business agility while really fueling innovation to support this energy as well as this digital transformation. As we talk about the presentation today, here is what interactive AI can help you improve your interpretation workflows. It's a tool that can allow you to not only enhance but accelerate your current interpretation methods. The next thing is that the geoscientist needs to be in control of the results. It's them that are responsible for the interpretation, and sometimes these deep learning and data science workflows developed in the R&D labs can be a little bit black boxy. And so the geoscientist does not see or interact with the actual model as it is being produced. However, the unique aspect with interactive AI is that now the geoscientist is in full control. They are sort of put in the driver's seat and can condition the network in real time to identify their features of interest. And finally, this is a tool around providing a huge amount of efficiency. It will streamline the data preparation process and really reduce the tedious interpretation, and it will give you back time to reduce risk and create a higher detailed interpretation in a shorter amount of time. This has direct value back to the different assets as well as the business units. Ultimately, this is a tool that will help you. Interactive AI is really positioned as a new interpretation tool, and it's really meant to integrate into the existing workflows, right? As geoscientists work in tandem with it, we want to be able to directly consume those outputs into our interpretations. Finally, it's meant to accelerate the interpretation process. Usually, 80% of our time as geoscientists is spent interpreting data, combing through it, etc., and the 20% is on critical evaluation. Now, utilizing the cloud as well as deep learning solutions, we can flip that. So now it's 20% interpretation, and now the remaining 80% can be shifted towards critical evaluation and understanding the data. And finally, you can quickly analyze fast amounts of data at all sorts of scales for your prospecting workflows. This can be something like a development block, say one kilometer by one kilometer, or several thousands of kilometers. It's really a tool that enables you to work with all sorts of data sizes. And many of the operators today are a part of the OSDU, and they're using interactive AI commercially, and we'll share some of those quotes later. But first, let's dive into some of the different things that we were able to achieve using this new deep learning solution. In machine learning, a label is what you teach the machine. The inference is what the machine comes back with, and so this is kind of where the teaching starts. What we have here is a prediction on a non-labeled slide. So I've not labeled this. This is merely after a few iterations of training. This is the prediction made by the neural network. And what is so unique about interactive AI is that it doesn't use any pre-trained networks. And so because of this, it offers a new level of flexibility where now geosciences can interpret many different geological features and does not have to be restricted to the pre-trained models. This way it's kind of an out-of-the-box, run-and-ready solution so that you can continue to interpret those, especially since the subsurface is very ambiguous. So in this case, we've gone and interpreted the different faults. So we see the predictions for those in yellow. We also see the source-rock interval, which is highlighted in green. We see the red here, which is the reservoir. And then finally in blue is the regional top seal for the Scarborough gas field offshore, Australia. And so what we're able to do is quickly characterize the petroleum system for this area and better understand what are the different relationships between the stratigraphy as well as the structure and how that plays into our area and what new opportunities we can identify. Furthermore, moving from the realm of probabilities where it provides a nice kind of de-risking idea of various features, you can immediately take those probability cubes and you can generate industry-ready formats such as in charisma faults or in tessellated geobodies and readily, again, incorporate those into the existing workflows, not meant as a roadblock, but rather as an additional toolkit for the geoscientists. Here, from the probability, I was able to extract the source-rock interval, which is shown in green, and immediately we can see that there are different expressions of structure, which is present in the area, and we can also begin to see some finer-scale details on the surface of the interval, potentially revealing other faulting directions, which may have not been previously considered, or perhaps these could be indicative of channels, as we know within the geology in that area. Furthermore, again, you can take these probability cubes, and this is sort of where Bluer has completely rewritten the game for exporting faults. We have had great feedback from our clients on grouping and the quality of the faults, and we've developed in tandem with our different clients this fault-export algorithm to really better unlock that higher level of detail within the subsurface, and this applies to other geometries as well. As we move into the reservoir, we can begin to see how the basement faults are beginning to interact with the reservoir, suggesting potential areas for compartmentalization. Furthermore, we can pull in the polygonal faults, and really, if you look at the density of these faults, it's very rare that we do this. However, using the interactive method that geoscientists working in tandem with the network, they're able to better characterize really at a higher level of detail all of these different faults present throughout the survey. We can then bring in the geobody for the regional top seal, and we can see that those same basement faults, which extend through the reservoir, are also extending through the seal, and so now we understand that there's some potential for a leak off, and there's a risk of containment. Further, if we look into the seismic section, we notice that at the shallow surface, as well as the sea floor, we see expression of shallow gas anomalies, which are also characterized by pock marks, and so we can already export those different geobodies and superimpose them with well data, and so we can really calibrate this interpretation and validate it in real time. Is this the geology that we're seeing? And again, that higher level detail that we're talking about. No geoscientists would be able to pick this density of faults, perhaps if they had several weeks, absolutely, but it doesn't sort of make sense, but when we utilize the cloud, as well as deep learning, this is kind of the agility that we're talking about, allowing the geoscientists to unlock that higher level of detail, and so in just a matter of 11 and a half hours from importing the data to training to exporting these and visualizing in different interpretation packages, we're able to see the reservoir distribution, the structural controls on stratigraphy, as well as the seal integrity, and now sort of this flexibility, right? We identified various different petroleum system elements. It's not just restricted to faults, but now we can characterize reservoirs, seals, and et cetera, and really unlock that higher level of value within the different business units. But I want to reinforce that we have these blank networks, right? And so because of that, it has this higher level of flexibility. It's not pre-trained, rather you can train it on that specific feature that you're looking for. And so you can really pick anything, and we can use these same kind of workflows and new and emerging energy sort of projects such as geothermal as well as CCS. And so taking that same sort of workflow, right? Now the method that we use for characterizing reservoirs within the petroleum realm, that can be used as the same kind of methodology for reservoir assessment, whether it be for geothermal or sequestration opportunities, right? Because you want to be aware of sort of those compartmentalizations and how it will flow in the subsurface. Furthermore, going back to the seal, right, we talked about containment risk. This is crucial within your petroleum system element modeling. However, you can also use it within the sequestration world. Is there any risk for whatever I'm injecting to percolate back up to the surface? So you can use this workflow to evaluate that. And then finally, shallow hazards, whether it be in petroleum exploration or in new energies, this allows you to really characterize the shallow hazards presence and better optimize your development plans. And so one of the features that we have been working on internally at BlueWare and I'm very excited to show you today is using interactive AI for 2D workflows. We've seen it used for 3D, but really VDS allows you to use multiple different dimensions. And it's available today as through the open group on the OSDU. And so here we'll be looking at the Pegasus basin offshore New Zealand, characterizing some 2D hydrates going back to sort of shallow hazard analysis. What I've shown you here is a regional 2D line. And we can begin to see these sort of cross crudding reflectors which are present in the seismic line. Here we can see that they're reversed polarity and they seem to follow the sea floor. And this is what we call BSRs, bottom simulating reflectors in the very characteristic of hydrate presence. However, in some areas, it may not be as immediately obvious where they connect. Here they're very well pronounced, but it begins a little more difficult to follow them through, especially in more blank seismically blank zones. So what you can use is the interactive deep learning process and it really helps you characterize the distribution of those hydrates. Where they could be present, where do they extend? And this was all done in a matter of 20 minutes, but it's not just on one line. This extends throughout the entire seismic survey. So you can readily incorporate those into the interpretation applications, understand the distribution of hydrates, right, and especially where they intersect. So it really gives you that better confidence in your interpretation, that higher level of detail. And it also extends throughout many of the different lines. What's really key about this is that now you can use this tool in just a matter of 20 minutes, right, comb through massive amounts of 2D seismic lines or 3D lines. Let's say that you're going into a data room or a bid round. You have a limited amount of time to comb through massive libraries of data. Leveraging a tool like interactive AI allows you to really hone in on those features of interest and better understand what it is that you're looking at. Quickly you can begin to evaluate potential opportunities and further characterize those, rather than spending the 80% of your time manually interpreting. Now let's leverage deep learning to accelerate that process and help deliver value faster in a more agile environment. But enough about all this interpretation stuff, right? Let's actually talk about the OSDU. So here we see that I'm launching interactive AI from the OSDU interface. We already have a data set loaded. This is going to be offshore Taranaki Basin in New Zealand. We'll be looking at deepwater channels. I mentioned earlier that IAI is a cloud-native application. And what's interesting is that it can be scaled up through Kubernetes. All data inside of OSDU can be accessed as well, right? So these different seismic surveys that we have, the label sets, etc., they're readily available. What an interpreter might be doing in AIIs is sort of what's playing in the background. There are various tools that we can use to label. But here the really powerful tool about AI is that you can use the inference from the network very quickly to continue to improve on your labels and further improve on that interpretation. It's sort of this live feedback. So every 45 seconds or so you're getting a new prediction from the network. Based off of your labels. And so because of this, as you work in real time, it's learning from your labels, understanding what a channel is, how it's expressed seismically, and then where else it could also be present in that survey. What's really interesting is that many of our clients today are actually importing their old labels from previous fields, importing them into IAI, continuing to build upon the initial label set. And once they have that new improved interpretation, they're bringing them back in and they're able to unlock other opportunities such as attic oil, missed opportunities, and really characterizing those fields at a higher level of detail. If you want to make these predictions a little more permanent and part of your interpretation, you're able to export not only the probability cubes, but also different surfaces, whether they be in fault sticks or Geobody.ts format. And so you can consume them in multiple different 3D visualization packages. Really, again, being a part of your workflow, another toolkit for the GeoScientist. And so far we've had great feedback from our customers. The tool has sort of been a revelation for them, enabling them to really improve on their different interpretations, unlocking that extra level of detail, which really, at the end of the day, actually provides more context to different things, right? These sort of subtleties are key, especially in these very risky projects. If you can get that higher level of detail, it really goes a long way to helping the risking process and the understanding of the basin. So today, we've shown how to enhance our interpretations. We've also shown how the GeoScientist is really in control, right? This back and forth, which we showed through interactive AI on the OSDU, and how it really enhances the end user workflow, and they can do so much more in this kind of environment. And really the most important thing today is that we can all do this in the cloud on OSDU today. Interactive AI is not an R&D tool. It's commercially being used today, and many of our clients have seen tremendous value in using it for characterizing their different GeoScience questions. So thank you again for your time. It's been a pleasure speaking with all of you, and I look forward to the questions. Thank you again.