Integrated meteocean and seismic dataset for AI-based seawater CO2 estimation at Deception Island, Antarctica

Understanding the carbon cycle in Antarctic coastal systems is vital for evaluating the role of polar oceans in regulating atmospheric CO₂ and global climate feedbacks. However, these areas remain poorly sampled and are underrepresented in existing carbon flux models. This dataset offers high-resolution environmental observations collected in February 2025 from surface waters and inland stations in Deception Island, an active volcanic caldera in the South Shetland Islands, Antarctica. It includes measurements of surface seawater pCO₂, temperature, salinity, wind speed, air temperature, solar radiation, tidal elevation, and seismic signals (long-period and tremor events), along with related spatiotemporal metadata. To enhance direct observations, we applied a data-driven modeling approach using deep learning techniques. A Bidirectional Long Short-Term Memory (Bi-LSTM) neural network was trained on multivariate sequences to estimate seawater pCO₂, with model performance evaluated through five cross-validation folds. The final datasets contain both measured and Bi-LSTM-estimated pCO₂ values. All data and processing steps adhere to FAIR principles to support research on air-sea gas exchange, remote sensing calibration, and Antarctic carbon system dynamics.