GOBAI: Gridded Ocean Biogeochemistry from Artificial Intelligence

About GOBAI

Jonathan D. Sharp, Andrea J. Fassbender, Brendan R. Carter, Gregory C. Johnson, John P. Dunne, and Cristina Schultz

GOBAI-O2 is a gridded data product that provides three-dimensional monthly fields of dissolved oxygen in the global ocean. It was constructed by training machine learning algorithms with observations of oxygen concentration ([O2]) from discrete shipboard measurements and autonomous sensors on biogeochemical Argo floats, then applying those machine learning algorithms to three-dimensional monthly gridded fields of temperature and salinity. The algorithms used to produce GOBAI-O2 have been validated using real observations and synthetic data from model output, and the data product itself has been compared against the World Ocean Atlas and selected discrete measurements. Results of these validation and comparison exercises are detailed in Sharp et al. (2023).

GOBAI HR — the high-resolution product

GOBAI HR-v1.0 is the high-resolution member of the GOBAI family, and it is what this site publishes on the cloud. It carries dissolved oxygen and nitrate on a near-global ¼-degree grid over 58 mean-pressure levels at weekly resolution, 1993-01-01 to 2025-12-05 — 1719 weekly fields per variable. It is built from Argo float data and GLODAP shipboard data applied to the temperature and salinity fields of RFROM, and shares RFROM’s grid exactly.

GOBAI HR-v1.0 is preliminary. The files carry comment = "preliminary" and cite a manuscript in preparation (Sharp et al., GOBAI High Resolution Data Products, in prep.). Expect the fields to change before the product is final.

The version string used throughout the cloud data is v202606, which is what the files themselves are stamped with, in both the file names and each file’s title attribute. The ERDDAP dataset titles say HR-v1.0 for the same data.

Acknowledging use

Please include the data product citation shown below when using GOBAI-O2. In addition, the GOBAI-O2 manuscript published in Earth System Science Data may also be cited. GOBAI HR-v1.0 has no citable release yet; cite the manuscript in preparation until one exists.

Data product citation

Sharp, J. D., Fassbender, A. J., Carter, B. R., Johnson, G. C., Schultz, C., Dunne, J. P. (2022). GOBAI-O2: A Global Gridded Monthly Dataset of Ocean Interior Dissolved Oxygen Concentrations Based on Shipboard and Autonomous Observations (NCEI Accession 0259304). NOAA National Centers for Environmental Information. Dataset. https://doi.org/10.25921/z72m-yz67.

Manuscript citation

Sharp, J. D., Fassbender, A. J., Carter, B. R., Johnson, G. C., Schultz, C., Dunne, J. P. (2023). GOBAI-O2: temporally and spatially resolved fields of ocean interior dissolved oxygen over nearly 2 decades. Earth System Science Data, 15 (10), 4481 - 4518,. https://doi.org/10.5194/essd-15-4481-2023.

Contact: jonathan.sharp@noaa.gov | oar.pmel.gobop@noaa.gov

View in a browser

Explore GOBAI HR-v1.0 on an interactive globe, with no install and no account. The viewer is gridlook, and it reads the Icechunk store directly from this bucket.

FieldViewer
Dissolved oxygenOpen o2 in the viewer
NitrateOpen no3 in the viewer

The first map can take a while to appear: each view downloads about 130 MB, because the data are stored in 100-week blocks. Moving between weeks in the same block is quick. Crossing into another block downloads again.

Data Access — ERDDAP

ERDDAP’s griddap form lets you subset by time, pressure and region in the browser and download just that subset, in netCDF, CSV or several other formats. Start here if you want part of the data rather than whole files.

GOBAI HR-v1.0 (v202606) — current

Weekly ¼° oxygen and nitrate on 58 mean-pressure levels, 1993-01-01 to 2025-12-05. Preliminary.

FieldERDDAP dataset
Dissolved oxygengobai_o2_hr_v10
Nitrategobai_no3_hr_v10

Data Access — Download

Whole files, from two places. ERDDAP serves the monthly source files through a browsable directory listing. NOAA Open Data Dissemination (NODD) holds the same data repackaged into larger, cloud-optimised blocks in the public bucket gs://noaa-oar-gobai — open data, no account or credentials.

Each bucket path below links to the Google Cloud console, where you can browse and download individual files. The bucket is public, so no project and no permissions are needed — but the console does require you to be signed in to some Google account. With no account at all, list the files with the code in Data Access (Code), or over plain HTTPS at https://storage.googleapis.com/noaa-oar-gobai?prefix=netcdf/v202606/&delimiter=/.

GOBAI HR-v1.0 (v202606) — current

FieldNODD (36 files, 247 GB)ERDDAP source files
Dissolved oxygennetcdf/v202606/o2/ — 18 files, 121 GBmonthly
Nitratenetcdf/v202606/no3/ — 18 files, 125 GBmonthly

On NODD each variable is one continuous series split into 100-week files, so the 1719 weeks come as 17 files of 100 weeks plus a final file of 19. File names carry the weeks they cover, e.g. GOBAI-O2-HR-v202606_1993-01-01_1994-11-25.nc.

GOBAI-O2 monthly — not on NODD

The monthly GOBAI-O2 product is distributed by NOAA’s National Centers for Environmental Information, not from this bucket.

FieldWhere
Dissolved oxygen, monthlyNCEI Accession 0259304 (doi: 10.25921/z72m-yz67)

Data Access — Code

WhatWhereUse it for
Icechunk storegs://noaa-oar-gobai/icechunk/v202606 The whole HR-v1.0 record — oxygen and nitrate on one 1719-week time axis, opened as a single dataset. Python only.
netCDF filesgs://noaa-oar-gobai/netcdf/v202606/ Ordinary netCDF-4 files. Any language. Stream them or download them.

The Icechunk store covers GOBAI HR-v1.0 only.

Python — install

pip install xarray gcsfs h5netcdf     # netCDF files
pip install icechunk zarr             # ...plus the Icechunk store

Python — the whole HR-v1.0 record, via Icechunk

import icechunk as ic, xarray as xr

SRC = "gs://noaa-oar-gobai/netcdf/v202606/"
repo = ic.Repository.open(
    ic.gcs_storage(bucket="noaa-oar-gobai", prefix="icechunk/v202606", anonymous=True),
    authorize_virtual_chunk_access=ic.containers_credentials({SRC: ic.gcs_credentials(anonymous=True)}),
)
ds = xr.open_zarr(repo.readonly_session("main").store, consolidated=False, chunks={})

If reads feel slow, raise Zarr’s default concurrency of 10: import zarr; zarr.config.set({"async.concurrency": 128}).

Python — a single netCDF file

import xarray as xr

url = "gs://noaa-oar-gobai/netcdf/v202606/o2/GOBAI-O2-HR-v202606_1993-01-01_1994-11-25.nc"
ds = xr.open_dataset(url, engine="h5netcdf", storage_options={"token": "anon"}, chunks={})

To see what is published: import gcsfs; gcsfs.GCSFileSystem(token="anon").ls("noaa-oar-gobai/netcdf/v202606/o2").

R

R can read these files without downloading them. Append #mode=bytes to the HTTPS URL and netCDF reads only the bytes it needs.

install.packages("ncdf4")
library(ncdf4)

u <- paste0("https://storage.googleapis.com/noaa-oar-gobai/netcdf/v202606/o2/",
            "GOBAI-O2-HR-v202606_1993-01-01_1994-11-25.nc#mode=bytes")
nc <- nc_open(u)

# one horizontal slice: 4x4 points, first pressure level, first time step
x <- ncvar_get(nc, "o2", start = c(600, 300, 1, 1), count = c(4, 4, 1, 1))
nc_close(nc)

Dimensions are reversed relative to Python: (longitude, latitude, mean_pressure, time). RNetCDF::open.nc() works the same way. There is no Icechunk reader for R, so the store is Python-only.

Streaming needs a netCDF library built with byte-range support. Most are; to check yours, run the code above — if it fails with a message about an unknown file format or an inaccessible DAP response, yours is not, and you should download the file first:

u <- paste0("https://storage.googleapis.com/noaa-oar-gobai/netcdf/v202606/o2/",
            "GOBAI-O2-HR-v202606_1993-01-01_1994-11-25.nc")
download.file(u, "gobai.nc", mode = "wb")   # ~7.1 GB
nc <- nc_open("gobai.nc")

Files run from about 1.4 to 7.3 GB each, so downloading is worth avoiding where you can.