A comparison of Markov Chain, Sarima and multiple linear regression for forecasting discharge

Developing a hydrological forecasting model based on past records is crucial for effective hydropower reservoir management and scheduling. Numerous popular discharge forecasting models have been developed; however, real-time forecasts remain challenging. This study evaluates discharge forecasts using the Markov Chain model, Seasonal Autoregressive Integrated Moving Average (SARIMA), and Multiple Linear Regression (MLR) models for forecasting monthly discharge time series. This study compares the accuracy of the discharge forecast results produced by the Markov Chain, SARIMA, and Multiple Linear Regression using five statistical indicators. Based on the simulation results, the Markov Chain, SARIMA, and MLR have accuracy levels of probability in discharge of 63%, 66%, and 76%, respectively. In comparison to other models, the highest correlation (r) is found in the MLR model (0.76) with MAPE (0.19), followed by SARIMA and Markov Chain. Therefore, the most accurate, precise, and representative water source model alternative for forecasts is the MLR model. The Markov Chain model and the SARIMA model are time series generation models, while the MLR model is a statistical regression model. In addition, this model is to be selected as the basis for modeling in forecasting river flow or optimal management of a reservoir, as well as determining future discharge, especially in monsoon climate regions.