Statistical forecasting of water temperatures and dissolved oxygen concentrations for monitoring water quality in reservoirs: A SARIMA modeling-based application

Water temperatures (T) and dissolved oxygen concentrations (DO) have a great deal of significance for monitoring water quality is surface waters such as reservoirs since the interaction between them maintains the ecological balance Hence, applying the advanced monitoring models that require continuous field measurements of high cost and accuracy is complicated, especially for the developing countries where data is insufficient. In this paper, a statistical-based model, the seasonal autoregressive integrated moving average model (SARIMA), was employed for forecasting water temperatures and dissolved oxygen concentrations in Saddat Al-Hindiyah Reservoir, Iraq, using monthly measurements during a study period of four years (2021-2024). Results indicated to a seasonal period length of 12 months and identified two models [SARIMA(5,0,1)(1,0,0)12 and SARIMA(0,1,2)(0,1,1)12] for forecasting an entire year period of water temperatures and dissolved oxygen time series, respectively, after training the models during the years (2021-2023) and testing them during the year 2024. Consequently, the SARIMA model captured the temporal patterns in field data with good agreement. The model exhibited a satisfactory performance for water temperatures by achieving Mean Absolute Errors (MAEs) of 2.384 °C and 3.813 °C for the training and testing phase, respectively, while for dissolved oxygen MAEs were 1.026 mg/L and 0.445 mg/L for both phases, respectively. According to that an inverse relationship between the reservoir's water temperatures and dissolved oxygen concentrations revealed strongly. Furthermore, the models’ forecasting results highlighted the reservoir water temperatures range of (18.073-27.790) °C with a mean value of 23.562 °C, and the reservoir dissolved oxygen concentrations range of (5.078- 7.643) mg/L with a mean value of 6.271 mg/L above the minimum permissible limit. Thus, the study suggest that SARIMA model can be a useful tool for monitoring water quality in surface waters due to it applicability in capturing nonlinearity in time series data with low computational cost using single univariate time series.