Hyperspectral Data and Machine Learning for Estimating CDOM, Chlorophyll

Area of Science:

• Environmental Science
• Remote Sensing
• Data Science

Background:

• Inland waters are vital ecosystems supporting biodiversity, necessitating effective water quality monitoring.
• Traditional in situ measurements for water quality are limited by cost, time, and spatial coverage.
• Developing advanced methods for water quality assessment is crucial for ecological management.

Purpose of the Study:

• To propose and evaluate a novel approach using hyperspectral data and machine learning for estimating inland water quality parameters.
• To assess the performance of various machine learning models in predicting concentrations of CDOM, chlorophyll a, turbidity, diatoms, and green algae.
• To establish a data-driven framework for scalable and efficient water quality monitoring.

Main Methods:

• Acquisition of in situ hyperspectral and water quality data from the river Elbe, Germany.
• Application of a regression framework employing ten distinct machine learning models.
• Utilisation of two preprocessing techniques to optimize data for model input.

Main Results:

• Machine learning models achieved high accuracy in estimating water quality parameters, with R² values ranging from 89.9% to 94.6%.
• The data-driven approach demonstrated superior performance compared to traditional methods like band ratios.
• Specific models showed excellent predictive power for CDOM, chlorophyll a, turbidity, and algal concentrations.

Conclusions:

• Hyperspectral data coupled with machine learning presents a powerful and accurate method for inland water quality monitoring.
• The developed regression framework shows significant potential for broad application across various inland water bodies.
• This approach offers a scalable and cost-effective alternative to traditional water quality assessment techniques.