GOBAI-O2: A Global Gridded Monthly Dataset of Ocean Interior Dissolved Oxygen Concentrations Based on Shipboard and Autonomous Observations (NCEI Accession 0259304)

This dataset contains a global gridded data product of observation-based ocean interior dissolved oxygen concentrations. The data product is called GOBAI-O2 for Gridded Ocean Biogeochemistry from Artificial Intelligence - Oxygen. The dissolved oxygen fields were constructed by training machine learning algorithms with observations from shipboard analyses and autonomous profiling floats, then applying those trained algorithms to global gridded fields of temperature and salinity. Those temperature and salinity fields were calculated from a long-term mean field and monthly anomaly fields constructed from the global array of Argo floats (Roemmich and Gilson, 2009), and are presented alongside GOBAI-O2 for easy analysis. Also presented are uncertainty fields for dissolved oxygen, which were calculated by combining three separate sources of uncertainty as described in Sharp et al. (2023), see the Documentation.

The scope and resolution of GOBAI-O2 are as follows: geographically, from -179.5 to 179.5 degrees longitude and -64.5 to 79.5 degrees latitude at 1-degree resolution; with respect to pressure, from 2.5 to 1975 decibars on 58 levels that become incrementally further spaced; and temporally, from January 2004 to December 2024 at monthly resolution. 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 for GOBAI-O2-v2.1 are detailed in Sharp et al. (2023).

Some updates to the methodology have been introduced for GOBAI-O2-v2.3, which will be described in an upcoming manuscript (Sharp et al., in prep):
Observational O2 data from floats is still adjusted based on a crossover comparison with bottle O2 data, however, the adjustment equation is now a linear fit of the percent difference (Argo - bottle) in oxygen saturation as a function of oxygen saturation.
Model-based experimentation has revealed some spatial and temporal discontinuities in GOBAI-O2 introduced by the Random Forest Regression models. For this reason, GOBAI-O2-v2.3 is based only on feed-forward neural networks.

Rather than basin-specific clusters for algorithm training and application (as in GOBAI-O2-v2.1 and v2.2), clusters are now developed based on unsupervised learning (Gaussian mixture modeling) with temperature, salinity, and depth data.
Algorithm-based uncertainty is now calculated from an ensemble of five model simulation experiments, rather than just one. This provides a more robust estimate of uncertainty in GOBAI-O2.
Due to limited data in the Mediterranean Sea, the neural network for the cluster covering mostly the upper water column in the Mediterranean has been trained and applied without year as a predictor variable.

Data are in netCDF, Figures are in PNG.

本数据集包含基于观测的全球网格化海洋内部溶解氧浓度数据产品。该数据产品名为GOBAI-O2，其全称为“基于人工智能的网格化海洋生物地球化学-氧”（Gridded Ocean Biogeochemistry from Artificial Intelligence - Oxygen）。溶解氧场的构建过程为：利用船载分析和自主剖面浮标的观测数据训练机器学习算法（machine learning algorithms），再将训练后的算法应用于全球网格化的温度和盐度场。上述温度和盐度场由长期平均场及基于全球Argo浮标阵列构建的月度异常场计算得出（Roemmich and Gilson, 2009），并与GOBAI-O2一同呈现，以便于分析。同时呈现的还有溶解氧的不确定性场，其通过整合Sharp等人（2023）所述的三种独立不确定性来源计算得出，详见Documentation。

GOBAI-O2的范围与分辨率如下：地理范围为经度-179.5°至179.5°、纬度-64.5°至79.5°，分辨率为1°；压力范围为2.5至1975分巴，共58层，层间距随深度递增；时间范围为2004年1月至2023年12月，分辨率为月度。用于生成GOBAI-O2的算法已通过真实观测数据和模型输出的合成数据进行验证，数据产品本身也与《世界海洋图集》（World Ocean Atlas）及选定的离散测量数据进行了对比。这些验证和对比工作的结果详见Sharp等人（2023）。

数据格式为netCDF，图表格式为PNG。

搜集汇总

背景与挑战

背景概述

GOBAI-O2是一个全球网格化的月度海洋内部溶解氧浓度数据集，基于船载和自主观测数据，通过机器学习算法生成。数据集覆盖全球范围，时间跨度为2004年1月至2024年12月，空间分辨率为1度，并提供了不确定性字段。