Water quality monitoring for coastal hypoxia: Integration of satellite imagery and machine learning models

Abstract

Effective monitoring of water quality is essential to mitigate the development and expansion of dead zones—regions with dangerously low dissolved oxygen (DO) levels that threaten aquatic ecosystems. Traditional methods for water quality assessment are often expensive and time-intensive, highlighting the need for more efficient approaches. Advances in remote sensing technology and high-resolution satellite imagery have created new possibilities for large-scale water quality monitoring research. This study aims to estimate DO levels in Long Island Sound, New York, to detect and map dead zones using Landsat 9 satellite imagery and seven Artificial Intelligence (AI) models: Adaptive Boosting (AdaBoost), Extreme Gradient Boosting (XGBoost), Categorical Boosting (CatBoost), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest (RF), and M5 Model Tree. Using observational DO data and spectral properties from 11 Landsat 9 bands, the AI models were trained and validated for accuracy and uncertainty performance. Among them, XGBoost showed the best performance, offering the highest prediction accuracy and the lowest uncertainty. The results demonstrate the potential of integrating satellite-based remote sensing with AI models for efficient, scalable, and cost-effective monitoring of water quality and dead zones, enabling informed decision-making for environmental management and ecosystem conservation.