Abstract : Regular monitoring and forecasting of water quality indicators are important for a proper management of aquatic ecosystems and for the protection of public health. This article analyzes the temporal evolution of two important indicators of nitrogen pollution: nitrite nitrogen and nitrate nitrogen. This analysis is based on data collected during 2019–2024 in the Olănești locality, on the Dniester River, Republic of Moldova.
The time series were processed using a hybrid STL-ARIMA approach. This approach combines the Seasonal-Trend decomposition using Loess (STL) with the AutoRegressive Integrated Moving Average (ARIMA) model. The STL method separated the series into trend, seasonal and residual components. It provides a clearer picture of the evolution of each indicator. After the separation of series, the ARIMA models were used. These models were applied to forecast future concentrations for the period 2025–2027. The results of the models were verified by residual analysis and by using forecast accuracy indicators.
The results indicated that both indicators will have seasonal variations, and each indicator differs. Nitrate nitrogen has more pronounced fluctuations than nitrite nitrogen. The STL-ARIMA method captured the main characteristics of the analyzed series, and the developed forecasts do not indicate a pronounced long-term increase in concentrations during the period 2025–2027. The obtained results indicate the usefulness of the proposed methodology for monitoring water quality. Forecasting is the method that can be useful to identify in advance the change of the pollution level.

Keyword : ARIMA model, Dniester River, time series