Can three-dimensional nitrate structure be reconstructed from surface information with artificial intelligence? - A...

Area of Science:

• Oceanography
• Marine Ecology
• Artificial Intelligence

Background:

• Nitrate is a crucial ocean variable controlling upper ocean ecosystems.
• Understanding the 3D nitrate structure is vital for ecosystem dynamics.
• Reconstructing 3D nitrate from surface data remains underexplored.

Purpose of the Study:

• To reconstruct the 3D nitrate concentration in the Indian Ocean using surface data.
• To compare the performance of U-net and Earthformer AI networks for this task.
• To identify key environmental drivers influencing nitrate distribution.

Main Methods:

• Utilized U-net and Earthformer AI networks trained on ecosystem model simulations.
• Input data included wind vectors, sea surface temperature, chlorophyll-a, and precipitation.
• Input features were decomposed into climatology and anomaly components for analysis.

Main Results:

• Earthformer achieved optimal nitrate reconstruction (NRMSE = 0.1591), outperforming U-net (NRMSE = 0.2007).
• Decomposing input features into climatology and anomaly components improved reconstruction accuracy.
• Earthformer effectively identified interannual nitrate anomalies and highlighted wind as a key driver.

Conclusions:

• AI, particularly Earthformer, can successfully reconstruct 3D ocean nitrate from surface data.
• Dynamic factors like wind significantly influence seasonal nitrate variations.
• The AI approach offers a novel method for studying marine ecosystems and environmental variables.