Work

Integrating vision transformers with multi-criteria analysis for direct air capture and CO2 storage (DACCS) siting

Year: 2026

Type: article

Abstract: Direct air capture with geological storage (DACCS) is a promising carbon dioxide removal pathway, but siting remains a multi-dimensional planning problem spanning energy supply, CO 2 transport and storage, environmental constraints, and social and regulatory feasibility. Existing DACCS siting studies either rely on region-specific GIS-MCDA workflows that are costly to rebuild and difficult to transfer across regions, or require deployment-derived labels that are scarce due to limited large-scale DACCS deployment. Here we introduce a hybrid, transparent siting framework that couples a PESTLE-grounded (Political, Economic, Social, Technological, Legal, Environmental) GIS-based multi-criteria decision analysis with an earth-observation surrogate model. Using the North German Basin and the German North Sea as a case study, we (1) operationalize PESTLE dimensions into spatially explicit exclusion masks and opportunity layers for both onshore and offshore siting and generate high-resolution reference suitability maps, and (2) train a deep regression model that predicts suitability directly from globally available Sentinel-2 imagery. The model uses a self-supervised vision transformer model (DINOv3) as feature extractor and a deep neural network head for suitability regression. The surrogate achieves strong agreement with the reference suitability map (R 2 ≈ 0.89, RMSE ≈ 0.028 on a 0–1 scale) and, through quantile-aware training, improves accuracy in the highest-suitability deciles that are most relevant for planning. A spatial-block validation further indicates moderate spatial robustness for satellite-observable suitability patterns, although not all MCDA criteria were included in the transferability model. Overall, the proposed framework provides a reproducible pathway from MCDM screening criteria to label-calibrate remote sensing suitability approximation, enabling rapid within region early-stage DACCS hotspot screening.

Source: Carbon Capture Science & Technology

DOI: 10.1016/j.ccst.2026.100647

Authors: Yifan Xu, Mrityunjay Singh, Cornelia Schmidt-Hattenberger, Marton Pal Farkas, Tomas Fernandez-Steeger

Institutions: GFZ Helmholtz Centre for Geosciences, Technische Universität Berlin, Technische Universität Darmstadt

Language: English

Topic: CO2 Sequestration and Geologic Interactions

Subfield: Environmental Engineering

Field: Environmental Science

Domain: Physical Sciences

SDGs: Climate action, Affordable and clean energy, Industry, innovation and infrastructure

Open Access status: gold

Estimated APC paid: $2,000