A Comparative Study of Data-Driven Models for Discharge Forecasting: a Study Case of Siak River, Pekanbaru Water Gauge Station

The availability of long-term river discharge data covering at least 30 years is needed for proper hydrological studies, so the ability to predict river discharge is a matter of concern in the field of civil engineering. The Siak River in Pekanbaru City experiences overflowing water during the rainy season. One of the steps to prevent flooding on the Siak River is to utilize river discharge information, data-driven models utilize historical data to train or derive useful insights for predicting outputs, some data-driven models that are suitable for generating monthly historical data into new data include the Autoregressive Integrated Moving Average (ARIMA) method and the Thomas-Fiering method. The research begins by conducting the Rescaled Adjusted Partial Sums (RAPS) test to test the homogeneity of the data, then the prediction of discharge data with several schemes using the ARIMA and Thomas-Fiering methods, then the performance comparison between the two models is carried out using MAPE, RMSE, Nash-Sutcliffe, and correlation coefficient r. From the research results, it was found that the Thomas-Fiering method tends to be more accurate for predicting 1-year monthly discharge as well as long-term discharge, namely periods of 3, 5, and 7 years, with the best prediction being 1-year discharge prediction using 5 years of observed discharge with MAPE, RMSE, Nash-Sutcliffe, and correlation coefficient r values of 7.42%, 26.76 m3s-1, 0.92, and 0.96, respectively. This study could be a valuable reference for future studies in selection and further modification of data driven discharge simulation models.