OXYGENERATOR: Reconstructing Global Ocean Deoxygenation Over a Century with Deep Learning

TL;DR

This work addresses the challenge of reconstructing global ocean deoxygenation over a century under sparse observations. It introduces OxyGenerator, a deep-learning framework that combines a 4D graph, a zoning-varying graph hypernetwork, and chemistry-informed regularization to learn spatio-temporal DO dynamics and chemical couplings with nitrogen and phosphorus. The approach significantly outperforms CMIP6 simulations, reducing MAPE by about 38.8%, and provides adaptive, region-specific zoning and uncertainty-calibrated reconstructions. The study demonstrates the potential of data-driven Earth-system methods to illuminate long-term ocean health trends and informs future oceanographic and climate research, while acknowledging data sparsity and the need for continued validation. Overall, the paper presents a data-driven, physics-aware pathway to understanding the evolving footprint of deoxygenation in the global ocean.

Abstract

Accurately reconstructing the global ocean deoxygenation over a century is crucial for assessing and protecting marine ecosystem. Existing expert-dominated numerical simulations fail to catch up with the dynamic variation caused by global warming and human activities. Besides, due to the high-cost data collection, the historical observations are severely sparse, leading to big challenge for precise reconstruction. In this work, we propose OxyGenerator, the first deep learning based model, to reconstruct the global ocean deoxygenation from 1920 to 2023. Specifically, to address the heterogeneity across large temporal and spatial scales, we propose zoning-varying graph message-passing to capture the complex oceanographic correlations between missing values and sparse observations. Additionally, to further calibrate the uncertainty, we incorporate inductive bias from dissolved oxygen (DO) variations and chemical effects. Compared with in-situ DO observations, OxyGenerator significantly outperforms CMIP6 numerical simulations, reducing MAPE by 38.77%, demonstrating a promising potential to understand the "breathless ocean" in data-driven manner.

Figures (16)

• Figure 1: Global ocean deoxygenation reconstruction via OxyGenerator from 1920 to 2023 based on sparse observation. Above: The proportion of ocean data observed in four-dimensional coordinates $\rho(\text{\#.Obs})$. Overall, dissolved oxygen (DO) observation are very sparse in each interval, and many areas have no observations. Below: Minimum DO reconstructed by OxyGenerator. The yellow line envelopes the oxygen minimum zone (OMZ) where $\text{DO}_{\text{min}} \leq 30 \mu \text{mol/kg}$. $\rho(\text{OMZ}_{30})$ indicates the proportion of OMZ30 regions to global oceans, which clearly shows a significant increase over a century.
• Figure 2: The framework of ocean deoxygenation reconstruction via our proposed OxyGenerator.
• Figure 3: Framework of zoning-varying message-passing.
• Figure 4: The spatial distribution of MAPE in global ocean deoxygenation reconstruction using different methods (The darker the color, the greater the error). The reconstruction error of OxyGenerator in open sea is significantly reduced.
• Figure 5: Performance Comparison of MAPE over time.