Marine carbon dioxide removal (mCDR) has the potential to remove atmospheric CO₂ and help mitigate anthropogenic climate change. However, its effectiveness remains highly uncertain because it depends on fine-scale ocean dynamics, including mesoscale and submesoscale processes that regulate the residence time of added materials near the ocean surface and influence air–sea CO₂ exchange. Resolving these processes requires high-resolution regional ocean models, making large ensembles of mCDR simulations computationally heavy and motivating the development of data-driven, machine learning–based emulators.

Here, we introduce a multiresolution modeling dataset for the Gulf of Mexico to support systematic assessment of mCDR performance and uncertainty. The dataset consists of coupled physical–biogeochemical simulations at horizontal resolutions of 1, 5, 10, and 25 km under multiple idealized mCDR deployment scenarios. It provides a benchmark for quantifying the effects of model resolution and deployment strategies while serving as a foundation for data-driven digital twins of the ocean targeting mCDR. The dataset is designed to enable surrogate modeling, uncertainty quantification, scenario optimization, and rapid prediction, accelerating the development and evaluation of future mCDR strategies.