Spectral Feature Selection Optimization for Water Quality Estimation.

Area of Science:

• Remote Sensing
• Optical Oceanography
• Water Quality Monitoring

Background:

• Bio-optical relationships in turbid inland waters are spatially heterogeneous and nonlinear, complicating chlorophyll-a (Chl-a) retrieval from satellite imagery.
• Existing Chl-a estimation models often focus on limited spectral regions (NIR to red bands), which may not be universally applicable due to varying water optical complexity.

Purpose of the Study:

• To develop and evaluate an optimized Chl-a retrieval method for turbid inland waters using multispectral satellite data.
• To address the challenges posed by spatial heterogeneity and nonlinearity in bio-optical relationships for accurate Chl-a estimation.

Main Methods:

• Proposed an optimization process involving feature candidate generation, selection, and optimization of spectral models (two-band, three-band, normalized difference chlorophyll index).
• Applied the method to remote sensing images and in situ Chl-a measurements from Lake Kasumigaura, Japan.
• Evaluated the performance of the optimized model (OptiM-3) against existing Chl-a estimation models.

Main Results:

• The optimized model (OptiM-3) demonstrated superior performance in Chl-a concentration retrieval.
• Root-mean-squared errors for Chl-a concentration decreased from 11.95 mg·m⁻³ to 6.37 mg·m⁻³.
• Pearson's correlation coefficients between predicted and in situ Chl-a improved from 0.56 to 0.89.

Conclusions:

• The proposed optimization process effectively enhances Chl-a retrieval accuracy in turbid inland waters.
• The OptiM-3 model offers a more robust and accurate solution for monitoring Chl-a concentration using satellite data.
• This approach is valuable for understanding and managing water quality in optically complex aquatic environments.