NorESM-bcoi-v1: A bias-corrected reanalysis of ocean biogeochemistry at  >  40° N, 1980–2020, based on a global ocean model hindcast

Abstract. Accurate and gap-free historical datasets of oceanic nutrient concentrations, dissolved oxygen, and carbonate chemistry, are needed for the analysis of anthropogenic impacts and development of predictive models, including as boundary conditions for regional ocean models. We developed low-bias, quasi-optimal 4D interpolations of observed values at latitudes >40° N, excluding Mediterranean and Black Seas, for years 1980–2020 inclusive. Our approach used output from the NorESM-OCv1.2 ocean model hindcast as a basis for kernel-smoothing bias correction and optimal interpolation of anomalies, obtaining monthly reanalysis data at NorESM grid points (30–90 km spatial resolution) and all model depth levels. Based on cross-validation tests with spatially-separated training and test subset locations, the resulting products showed equal or better out-of-sample accuracy compared to standard climatological products from the World Ocean Atlas 2023 and GLODAPv2, and also compared to existing non-climatological products from machine learning and biogeochemical reanalysis/hindcast datasets from CMEMS/Copernicus. The new products showed skill in tracking smooth seasonal and interannual variations not included in the standard climatologies, and improved spatial coverage and detail resolution in northern/Arctic regions, compared to all tested alternatives. Significant uncertainties remained, however, largely due to sparse observational coverage in seasonal and permanent ice-covered regions, and limited resolution in coastal regions. The NorESM-bcoi-v1 reanalysis data, as well as the associated quality-controlled in situ observational compilations (NORBGCv1) and code used to develop the reanalysis, can be downloaded from https://doi.org/10.5281/zenodo.14525787 (Wallhead et al., 2024).