Neural network method to correct bidirectional effects in water-leaving radiance

Area of Science:

• Remote Sensing
• Oceanography
• Data Science

Background:

• Ocean color algorithms require nadir water-leaving radiances for marine constituent analysis.
• Converting measured radiances to nadir is crucial for satellite data merging and accuracy.
• Existing methods, like Morel's, require chlorophyll data and are limited to Case 1 waters.

Purpose of the Study:

• To develop a novel neural network method for converting angle-dependent water-leaving radiances to the nadir direction.
• To create a method that does not require atmospheric correction or prior knowledge of water properties.
• To provide a fast, accurate, and adaptable solution for various remote sensing instruments and water types.

Main Methods:

• A neural network was trained to directly convert remote sensing reflectance from observation to nadir directions.
• The method was validated using NuRADS measurements across diverse aquatic environments.
• Performance was compared against the standard Morel method in both Case 1 and Case 2 waters.

Main Results:

• The neural network method accurately converts radiances to nadir, independent of atmospheric correction or inherent optical properties.
• The method demonstrated suitability for both Case 1 (open ocean) and Case 2 (coastal) waters.
• Significant improvements were observed in Case 2 waters, particularly in sediment-dominated environments, compared to the standard method.

Conclusions:

• The developed neural network offers a robust and versatile tool for ocean color data processing.
• This method enhances the analysis of marine constituents in challenging coastal waters.
• It provides a valuable advancement for merging satellite data and improving global ocean monitoring.