947 / 2024-09-19 21:14:06

Drivers, Trends, Predictability, and Ecosystem Implications of the Arabian Sea Oxygen Minimum Zone

Arabian Sea Oxygen Minimum Zone,HYCOM-ECOSMO,BGC-Argo,Ocean modelling,Artificial Intelligence,LSTM model

Session 15 - Ocean deoxygenation: drivers, trends, and biogeochemical-ecosystem impacts

The Arabian Sea OMZ has limited horizontal extent, but large vertical extent, rendering it the third most intense OMZ in the global oceans. This study examines the interannual variability and trends in deoxygenation processes of Arabian Sea OMZ over the period of 2000 to 2020 using the new coupled regional physical-biogeochemical model, HYbrid Coordinate Ocean Model- ECOSystem (HYCOM-ECOSMO). The model showed a high level of accuracy in simulating Dissolved Oxygen (DO) profiles with an overall model efficiency score of 0.81 and a percentage bias of 31%, when validated with the BGC-Argo derived DO profiles. The performance of the model showed consistency when validated against both BGC-Argo data and World Ocean Atlas 2018, yielding Root Mean Square Errors (RMSEs) of 21 µmol kg^-1 and 16.5 µmol kg^-1, respectively. The model was able to simulate both seasonal and interannual variations in dissolved oxygen (DO) content, but with slight overestimation of the BGC-Argo profiling float DO data. To better understand the oxygen dynamics within the oxygen minimum zone (OMZ), we utilized a supervised learning methodology incorporating a Long Short-Term Memory (LSTM) model to forecast mean oxygen concentrations in the OMZ core (ranging from 0 to 20 µmol kg^-1), using monthly data from the HYCOM-ECOSMO model. The input variables included the averages of detritus, primary production, and temperature from the surface to the euphotic depth, as well as nitrate, phosphate, and silicate. The LSTM model was trained and validated, achieving a Nash-Sutcliffe Efficiency of 0.83 for training and 0.65 for testing, with corresponding Mean Absolute Relative Errors of 0.14 and 0.22, and Kling-Gupta Efficiencies of 0.87 and 0.70, respectively. Hyperparameter tuning was then carried out to improve the performance of the model, which included adjusting dropout rates as well as the number of units in the LSTM. This highlights the capability of the model in predicting oxygen concentrations in OMZ and provides important information on the dynamics of this critical feature in the ocean that reacts to climatic changes. The findings of this study would yield better understanding of the relation between variations in OMZ and pelagic fishery of the Arabian Sea.