Remote sensing retrieval of inland water quality parameters using Sentinel-2 and multiple machine learning algorithms

Area of Science:

• Environmental Science
• Remote Sensing
• Data Science

Background:

• Remote sensing offers efficient water quality monitoring with broad coverage.
• Traditional methods struggle with non-optically active water quality parameters.
• Machine learning shows promise in retrieving water quality from satellite data.

Purpose of the Study:

• To compare four machine learning algorithms for retrieving chlorophyll-a, dissolved oxygen, and ammonia-nitrogen.
• To assess the performance of XGBoost, SVR, RF, and ANN using Sentinel-2 imagery.
• To reconstruct and analyze the spatial-temporal patterns of key water quality indicators.

Main Methods:

• Utilized Sentinel-2 satellite imagery for inland reservoir analysis.
• Applied and compared four machine learning algorithms: XGBoost, SVR, RF, and ANN.
• Focused on retrieving chlorophyll-a (Chl-a), dissolved oxygen (DO), and ammonia-nitrogen (NH3-N).

Main Results:

• XGBoost demonstrated superior performance over SVR, RF, and ANN for quantitative retrieval.
• Successfully reconstructed spatial-temporal patterns of Chl-a, DO, and NH3-N from 2018-2020.
• Identified interannual, seasonal, and spatial variation characteristics of the monitored parameters.

Conclusions:

• XGBoost provides an efficient and practical approach for monitoring both optically active and inactive water quality parameters.
• The study offers a valuable tool for regional-scale water quality management.
• Machine learning-based remote sensing enhances water quality assessment capabilities.