Neural network spectral relationship to improve an inherent optical properties data processing system for residual...

Area of Science:

• Oceanography
• Remote Sensing
• Data Science

Background:

• Residual error in satellite remote sensing reflectance (Rrs) impacts ocean color data quality and application suitability.
• Existing Inherent Optical Data Processing System (IDAS) algorithms rely on invariant spectral relationships, which may not fully correct for residual errors.

Purpose of the Study:

• To develop and evaluate a novel neural network-based IDAS algorithm (IDASnn) for more effective residual error correction in Rrs data.
• To improve the accuracy and spatiotemporal consistency of retrieved inherent optical properties (IOPs) and ocean color products.

Main Methods:

• Expressed residual error spectrum as an exponential plus linear function.
• Developed neural network models to derive spectral slope coefficients from satellite Rrs data.
• Compared the performance of the IDASnn algorithm against an invariant spectral relationship-based IDAS algorithm (IDAScw) using synthetic, field, and COCTS data.

Main Results:

• The IDASnn algorithm demonstrated superior effectiveness in reducing residual error effects on IOPs retrieval compared to IDAScw.
• IDASnn produced more accurate and smoother spatiotemporal ocean color products, especially in open ocean environments.
• The algorithm enabled data quality monitoring, identifying large residual errors in COCTS images with low effective coverage.

Conclusions:

• An accurate spectral relationship of residual errors is critical for effective error correction in ocean color data processing.
• The IDASnn algorithm offers a significant advancement in improving the reliability and usability of satellite-derived ocean color products.
• Rigorous control of product effective coverage or accurate residual error correction is essential for temporal and spatial analysis of COCTS data.