Machine learning reveals biological activities as the dominant factor in controlling deoxygenation in the South Yellow Sea

Liu, Qingyi and Liu, Chunli and Meng, Qicheng and Su, Bei and Ye, Haijun and Chen, Bingzhang and Li, Wei and Cao, Xinyu and Nie, Wenlong and Ma, Nina (2024) Machine learning reveals biological activities as the dominant factor in controlling deoxygenation in the South Yellow Sea. Continental Shelf Research, 283. 105348. ISSN 0278-4343 (https://doi.org/10.1016/j.csr.2024.105348)

[thumbnail of Liu-etal-Machine-learning-reveals-biological-activities-as-the-dominant-factor-in-controlling-deoxygenation-in-the-South-Yellow-Sea]
Preview
 Text. Filename: Liu-etal-Machine-learning-reveals-biological-activities-as-the-dominant-factor-in-controlling-deoxygenation-in-the-South-Yellow-Sea.pdf
Accepted Author Manuscript
License: Creative Commons Attribution 4.0 logo
Download (1MB)| Preview

Abstract

Dissolved oxygen (DO) is a crucial element for both biotic and abiotic processes in marine ecosystems, but has declined globally in recent decades. Therefore, there is an urgent need for solid large-scale and continuous estimation of DO concentration in vital ecosystems, such as coastal areas. A random forest (RF) model for DO in South Yellow Sea (SYS) was developed by integrating satellite data and simulation data during 2011–2019. The root mean squared error (RMSE) for the training and test sets were 0.514 mg/L and 0.732 mg/L, respectively. Spatiotemporal distributions of DO of multiple layers in the study area during 2011–2019 were very well reproduced by the RF model and showed a slight decline trend in most SYS areas, while more intense decline occurred in the deep central SYS. The analysis of the mechanisms of DO decline in the South Yellow Sea cold water mass (SYSCWM), located in the deep central SYS, indicates that the deoxygenation here is largely due to biological activities. This finding may have implications for studies on drivers of deoxygenation in coastal areas. Furthermore, integrating satellite data with machine learning models can offer a powerful approach to capturing the continuous spatiotemporal characteristics of ocean parameters over large spatial scales.

ORCID iDs

Liu, Qingyi, Liu, Chunli, Meng, Qicheng, Su, Bei, Ye, Haijun, Chen, Bingzhang ORCID logoORCID: https://orcid.org/0000-0002-1573-7473, Li, Wei, Cao, Xinyu, Nie, Wenlong and Ma, Nina;
CORE (COnnecting REpositories)