The East China Sea (ECS) shelf plays a vital role in the marginal sea carbon cycle, significantly modulating regional air–sea carbon exchange and coastal biogeochemical dynamics. However, due to complex shelf circulation, strong spatiotemporal variability, and limited observational coverage, the development of high-resolution four-dimensional (4D) climate data records (CDRs) for ocean carbonate systems, particularly anthropogenic carbon (Cₐₙₜ), remains challenging. Here, we present a comprehensive, quality-controlled 20-year (2003–2022) 4D dataset of oceanic carbonate parameters and anthropogenic CO₂ on the ECS shelf, developed through an integrated modeling and machine learning framework. The dataset was generated by combining the Regional Ocean Modeling System (ROMS) with a Tabular Prior-data Fitted Network (TabPFN)–based bias correction approach. Multi-source satellite observations, including ocean color products, together with other environmental predictors, were integrated with ROMS outputs through the TabPFN framework to reconstruct full-depth time-series fields of total alkalinity (TA), dissolved inorganic carbon (DIC), and dissolved oxygen (O₂). These reconstructed fields were further coupled with a localized TrOCA method (Tracer combining Oxygen, dissolved inorganic Carbon and Alkalinity) and a carbonate system equilibrium constraint to resolve 4D Cₐₙₜ distributions. The reconstructed fields achieved high accuracy, with an R² of 0.95 and an RMSE of 20.54 μmol kg⁻¹ for TA (improving R² by 0.43 and reducing RMSE by 139.02 μmol kg⁻¹ compared with raw ROMS outputs), an R² of 0.95 and an RMSE of 28.66 μmol kg⁻¹ for DIC (improving R² by 0.39 and reducing RMSE by 132.68 μmol kg⁻¹), and an R² of 0.94 and an RMSE of 16.21 μmol kg⁻¹ for O₂ (improving R² by 0.32 and reducing RMSE by 28.04 μmol kg⁻¹). Furthermore, validation against Cₐₙₜ derived from in situ measurements demonstrated high fidelity of the reconstructed product (R² = 0.98, RMSE = 4.17 μmol kg⁻¹). This 4D carbonate system and anthropogenic carbon dataset provides a satellite-informed climate data record for a dynamic marginal sea, bridging surface observations and subsurface biogeochemical variability. The reconstructed product offers a valuable resource for investigating long-term changes in shelf carbon cycling, quantifying anthropogenic carbon accumulation, and advancing the development and application of ocean color–supported marine carbon CDRs.