Prediction of Water Quality Parameters in the Paraopeba River Basin Using Remote Sensing Products and Machine...

Area of Science:

• Environmental Science
• Remote Sensing
• Water Quality Monitoring

Background:

• Traditional water quality monitoring is limited by cost, frequency, and spatial coverage.
• Extreme events necessitate enhanced temporal and spatial water quality assessment.
• Satellite imagery offers a complementary solution but faces resolution challenges.

Purpose of the Study:

• To develop and evaluate a methodology for predicting water quality parameters using remote-sensing data and machine learning.
• To compare different satellite data sources (Sentinel-2, raw PlanetScope, normalized PlanetScope) and machine learning algorithms.
• To assess the suitability of the proposed methodology for lotic and lentic environments.

Main Methods:

• Utilized Sentinel-2 and PlanetScope (PS) imagery, including normalized PS data.
• Collected water quality data from 24 monitoring stations (2016-2023).
• Applied Random Forest, k-Nearest Neighbors, Support Vector Machines, and Cubist algorithms for prediction.

Main Results:

• Models using normalized PS data demonstrated the best performance in estimating water quality parameters.
• Decision-tree-based algorithms (Random Forest and Cubist) exhibited superior generalization capabilities.
• Evaluated model performance using RMSE, MAE, CCC, and R-squared metrics.

Conclusions:

• The proposed methodology is effective for predicting water quality parameters using remote sensing and machine learning.
• Normalized PlanetScope data and decision-tree algorithms are highly recommended for water quality monitoring.
• The approach provides valuable insights for diverse environmental monitoring applications.