A study on water quality parameters estimation for urban rivers based on ground hyperspectral remote sensing...

Area of Science:

• Environmental Science
• Remote Sensing
• Analytical Chemistry

Background:

• Traditional water quality monitoring relies on time-consuming laboratory analyses.
• Urban river water quality assessment is crucial for public health and ecosystem management.
• Hyperspectral monitoring offers potential for rapid, non-invasive water quality assessment.

Purpose of the Study:

• To develop and evaluate a better inversion algorithm for water quality monitoring using hyperspectral technology.
• To explore the feasibility of hyperspectral monitoring as an alternative to laboratory testing for urban rivers.
• To provide convenient, economical, and extensive monitoring methods for urban internal river water quality.

Main Methods:

• Collected water samples from Fuyang River and obtained hyperspectral data using ASD FieldSpec 4.
• Preprocessed spectral data using Savitzky-Golay (SG) smoothing and mathematical transformations.
• Developed and evaluated water quality parameter models (Turbidity, SS, COD, NH4-N, TN, TP) using Partial Least Squares (PLS), Random Forest (RF), and Lasso algorithms.

Main Results:

• The first derivative of reciprocal logarithm spectral data (after SG smoothing) effectively modeled Turbidity, COD, NH4-N, and TP.
• The first derivative of smoothed spectral data effectively modeled SS and TN.
• Machine learning models (RF and Lasso) demonstrated superior prediction accuracy (R² > 0.8 for key parameters) and generalization ability compared to PLS.
• RF and Lasso models showed complementary strengths in applicability and prediction accuracy.

Conclusions:

• Hyperspectral monitoring, particularly with machine learning algorithms, is a viable and effective alternative for urban river water quality assessment.
• The developed inversion algorithms and models offer significant advantages in terms of convenience, cost-effectiveness, and extensiveness.
• Machine learning approaches provide a more suitable framework for classified inversion prediction of urban river water quality parameters.