Work

Features of short-term turbidity forecasting using the Amur, Angara, and Ussuri rivers as examples

Year: 2026

Type: article

Abstract: Relevance. Currently, machine learning methods and ensemble approaches are widely used. Reliable turbidity estimates based on these methods will help solve such pressing problems as planning economic activities, ensuring the reliable operation of water utilities, optimizing coagulant dosages, preventing system overload, and reducing the risk of violating sanitary standards within the water supply systems. Methods. Classic statistical methods, machine learning methods, and ensemble approaches, including ARIMA, SARIMA, ARFIMA, Holt – Winters, Prophet, Bagging, Boosting, and Stacking, were used in this study for short-term turbidity forecasting. Water turbidity observation data were collected from three observation points located near the main water treatment plants in the cities of Khabarovsk, Ussuriysk, and Angarsk in the Amur, Ussuri, and Angara rivers with a measurement frequency of once every 45 minutes. Results. Stacking and Boosting ensemble approaches are shown to outperform other models due to the use of a combination of different data sources. The effectiveness of the methods is revealed to depend on the characteristics of a specific turbidity time series: Holt – Winters performs better with strong seasonality for the Amur River, ARIMA or Stacking performs better with moderate and stable dynamics for the Angara River, and Boosting performs better with high variability for the Ussuri River. It is established that model selection requires consideration of data accuracy and availability, special features of the water body hydrological regime, and the forecast objectives. It appears necessary to use hybrid models based on physical factors (precipitation, water discharge) in future studies to improve forecast quality.

Source: Water sector of Russia problems technologies management

DOI: 10.35567/1999-4508-2026-4-28-44

Authors: Oleg M. Rozental, Tatyana N. Sintsova

Institutions: Russian Academy of Sciences, Institute of Water Problems, Mining Institute of the Ural Branch of the Russian Academy of Sciences

Language: English

Topic: Hydrological Forecasting Using AI

Subfield: Environmental Engineering

Field: Environmental Science

Domain: Physical Sciences

SDG: Clean water and sanitation

Open Access status: diamond

Funders: Ural Branch, Russian Academy of Sciences, Ministry of Science and Higher Education of the Russian Federation, Russian Academy of Sciences