abstract fractal blue background pattern
Research for Industry

Carbon Capture and Storage

Actively remove carbon from the atmosphere

Global warming is one of the most important problems humanity needs to solve for. To reach the goal of limiting the global rise in temperatures to 1.5 degrees Celsius we need to take major steps towards mitigating anthropogenic CO2 emissions. The general scientific consensus (opens in new tab) is that to reach 1.5 degrees, carbon neutrality must already be reached by 2050 – a giant technological challenge. While decarbonization and adoption of fossil fuel alternatives are accelerating across many industries, researchers project that to reach carbon neutrality in the next 30 years, steps need to be taken to actively remove carbon from the atmosphere.

Carbon capture and storage (CCS) is among the most promising technologies that paves the way towards CO2 neutrality and has seen a huge growth over recent years, with many CCS facilities being in active development worldwide. The core idea of CCS is to capture CO2 at industrial plants or directly from the air (direct air capture, or short DAC), compress and then store it permanently several kilometers beneath the surface in CO2 storage sites. Several pilot projects such as Sleipner (opens in new tab) have demonstrated the feasibility of CCS and have shaped our understanding of the risks and mitigation techniques.

In accordance with Microsoft’s pledge of reaching carbon negativity by 2030 and along with several promising avenues such as Energy 2.0 and agricultural carbon sequestration, we are actively contributing research towards enabling CCS as a viable and cost-efficient technique for decarbonization. As a technology partner, Microsoft has joined the Northern Lights partnership (opens in new tab), one of the flagship CCS projects and a collaboration between the Norwegian government and energy companies Equinor, Shell and Total.

One of the main challenges of CCS is being able to scale this technology to a level that is required to reach current climate goals. Current ongoing and planned projects (including Northern Lights) are projected to sequester a combined amount of approximately 40 mega tonnes of CO2 per year. However, to reach 1.5 degrees, the storage capacity of CCS projects needs to increase by a hundredfold (opens in new tab). Our vision at Research for Industry (RFI), is to bring together the world’s leading experts from our industry partners, academic collaborators, and internal research labs to accelerate the adoption of CCS through digitalization, cloud and AI technologies. Our areas of research include AI-driven simulations to speed up numerical computations, open-source software to facilitate running HPC workloads in the cloud, as well as the automation of CO2 monitoring and storage site exploration.

KarbonVision — Understanding subsurface geology and fault seal is critical for assessing the containment risks and storage capacity of subsurface CO2 storage sites, as across-fault and along-fault CO2 migration/leakage poses a significant threat to CO2 containment. Traditional fault mapping relies on manual picking and is time-consuming (weeks or months for a 3D seismic image). We have developed a computer-vision based approach for automatically mapping geological faults from seismic data to detect potential leakage pathways of CO2, reducing the processing time to hours or days. It also enhances faults that are hard to detect by human eyes, and therefore reduces the potential risk for CO2 containment analysis.

Q-FNOs for 3D flow — Reservoir modeling is an essential component of carbon capture and storage to simulate CO2 movement in the subsurface and to mitigate risk and uncertainty. CCS projects require a large number of fluid flow simulations to model different storage scenarios or possible flow pathways of the CO2. Traditionally, CO2 flow and reservoir simulations are carried out by solving a set of coupled partial differential equations, which are associated with high computational cost and mathematical complexity. In this project, we are developing data-driven simulations that can scale to industry-relevant 3D problem sizes as encountered in CO2 storage.

Redwood: Towards clusterless supercomputing on Azure — As high-performance computing (HPC) is entering the mainstream of scientific computing, cloud platforms need to facilitate the process of running increasingly large-scale parallel applications on distributed infrastructure. In project Redwood, we argue for the case of clusterless supercomputing as inspired by the rise of serverless function frameworks, in which the responsibility of resource management is shifted from the user to the cloud platform. With Redwood, we take a first step towards this goal in the form of a distributed programming framework that built on top of existing Azure HPC services and turns Azure into a language extension with which users can executed distributed parallel workloads without having to manage the underlying HPC infrastructure.

Hyperwavve: Elastic and resilient HPC for modeling and inverse problems — In our Hyperwavve project, we explore its potential for hyperscale scientific modeling and inversion workloads on Azure. We demonstrate our cloud-native fault-tolerant Hyperwavve framework for hyperscale 3D seismic imaging, where Docker, Kubernetes and Dask are used to run parallel containerized seismic workloads such as RTM/FWI at scale on Azure. With this end-to-end solution, we are able to perform large-scale 3D FWI with the full 3D overthrust velocity model (20x20x5 km3) using 1000 VMs on Azure (extensible to 6000+ VMs). The Hyperwavve platform would help expedite the seismic imaging/CCS research and development on Azure for users.

Outreach

  • Georgia Institute of Technology: Seismic monitoring for CO2 storage
  • Imperial College London: Automated finite-differences for seismic modeling & inversion
  • Stanford: Data-driven fluid flow modeling

Highlights