Water quality prediction using the ARIMA time series analysis model in the Nakdong River estuary

Downstream of the Nakdong River is where water intake stations are distributed, then it is necessary to observe and predict the variation in water quality. However, it is difficult to investigate the variations due to the limit of manpower and time required for monitoring. In this study, using time-series analysis, the missing data in the Nakdong River estuary were estimated, and the short-term water quality was predicted.

The water quality data (DO, T-N, T-P, COD, and Chl-a) observed from 2013 to 2017 were collected from the Ministry of the Environment (ME). The Autoregressive integrated moving average (ARIMA) models using daily time series data were selected with the Box-Jenkins method and verified with several statistical parameters.

The models selected for the ARIMA time-series analysis can verify the changes in water quality. Moreover, the ARIMA models demanded just low autocorrelation and moving average coefficients and none of the integrated coefficient. The root mean square (RMSE), the mean absolute error (MAE), and the mean absolute percent error (MAPE) values for each models were showed that the selected models had high repeatability. As a result of the Ljung-Box verification, all models showed low significance levels (below 0.05).

The proposed ARIMA models can be estimated the missing data and predicted the short-term water quality changes. In summer, when changes in the flow rate and algal blooms occurred, the models showed relatively low reproducibility, but except for these problems, they generally showed high accuracy. It is expected that short-term variations in water quality will be predicted and eventually contribute to the management of water resources in this area using time series analysis.

The ARIMA expressions of each water quality parameter were represented as follows: