Research on Surface Water State for Rivers in Western Ukraine Using Time Series Forecasting Methods

This study presents a data-driven forecasting framework for surface water state trends using time-series modelling based on hydrochemical monitoring data from the Ikva River (Ukraine). The monitoring campaign, conducted between 2021 and 2023, involved monthly sampling of 19 hydrochemical indicators...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Water (Basel) Jg. 17; H. 21; S. 3148
Hauptverfasser: Bytsyura, Leonid, Szczepanik-Scislo, Nina, Desyatnyuk, Oksana, Shakhovska, Natalya, Scislo, Lukasz, Sachenko, Anatoliy, Lototska, Olena, Shevchuk, Ihor, Sofinska, Oksana
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Basel MDPI AG 02.11.2025
Schlagworte:
Algorithms
Artificial intelligence
Climate change
Environmental aspects
Environmental monitoring
Forecasts and trends
Groundwater
Infrastructure
Management
Methods
Nitrogen
Pollutants
Risk assessment
River networks
Rivers
Sewer systems
Surface water
Water
Water pollution
Water quality
Water resources
Water supply
Ukraine
ISSN:2073-4441, 2073-4441
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:This study presents a data-driven forecasting framework for surface water state trends using time-series modelling based on hydrochemical monitoring data from the Ikva River (Ukraine). The monitoring campaign, conducted between 2021 and 2023, involved monthly sampling of 19 hydrochemical indicators at two sites. We applied the Prophet time series forecasting algorithm, a decomposable additive model, to predict key indicators, including water hardness and bicarbonate concentration. The approach provides a transparent and adaptable method for forecasting water state in data-limited contexts. Key contributions include the integration of high-resolution hydrochemical monitoring with an explainable machine learning model, enabling early warning insights in under-monitored river basins. The case study of best-performing models for hydrocarbonate and hardness confirmed that Prophet offered well-calibrated prediction intervals with rapid deployment, high interpretability, and dependable uncertainty estimation, though its forecasts were comparatively less accurate. Analysis of computational performance shows that Prophet enables faster implementation and quick insights, while ARIMA and LSTM achieve higher predictive accuracy at the cost of longer execution times. Results demonstrate strong predictive skill: for hardness, MAE = 1.64 and RMSE = 1.73; for bicarbonate, MAE = 54.82 and RMSE = 62.00. Coverage accuracy of 95% prediction intervals exceeded 91% for both indicators. The proposed approach provides a practical foundation for implementing early-warning systems and supporting evidence-based water resource management in regions lacking real-time monitoring infrastructure.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:2073-4441
2073-4441
DOI:10.3390/w17213148