Using machine learning for the assessment of ecological status of unmonitored waters in Poland

Andrzej Martyszunis ¹, Małgorzata Loga ², Karol Przeździecki ³

⁰Faculty of Building Services, Hydro and Environmental Engineering, Warsaw University of Technology, Nowowiejska 20 St., 00-653, Warsaw, Poland. andrzej.martyszunis.dokt@pw.edu.pl.

Scientific Reports +

|

October 18, 2024

View abstract on PubMed

Summary

Machine learning (ML) models, including XGBoost and Random Forest, effectively assess the water status of unmonitored rivers using anthropogenic pressure data. This AI-driven approach aids in meeting European Water Framework Directive (WFD) objectives for environmental protection.

Area Of Science

Environmental Science
Computer Science
Water Resource Management

Background

The European Union's Water Framework Directive (WFD) mandates water status assessment for all European water bodies.
Intensive monitoring programs are resource-intensive, particularly for unmonitored rivers.
Artificial Intelligence (AI) offers new analytical tools for environmental management.

Purpose Of The Study

To evaluate Machine Learning (ML) techniques as a complementary method for assessing the water status of unmonitored river water bodies in Poland.
To address the challenge of absent monitoring data by utilizing anthropogenic pressures as alternative data.
To identify the most effective ML algorithms for classifying water body status according to WFD requirements.

Main Methods

Implementation of various ML algorithms: Decision Tree, Random Forest, KNN, Support Vector Machine, Multinomial Naive Bayes, and XGBoost.
Utilizing anthropogenic pressures within river catchments as input data for ML models.
Comparing model performance using Overall Accuracy (OA) and Probability of Misclassification (PoM).

Main Results

XGBoost and Random Forest algorithms demonstrated the highest efficiency in classifying unmonitored water bodies.
Achieved approximately 93% Overall Accuracy (OA) for binary classification and 72% for comprehensive classification.
Partial class accuracies and Probability of Misclassification (PoM) were also key comparison metrics.

Conclusions

AI, specifically ML, provides a practical solution for assessing unmonitored water bodies, supporting WFD objectives.
Binary classification using ML is suitable for reporting water status, while full classification aids in planning and operational uses.
Overall Accuracy (OA) and Probability of Misclassification (PoM) are effective measures for evaluating classification performance in water status assessment.

Keywords:

Ecological status assessment Machine learning Unmonitored water bodies Water quality assessment