Machine learning algorithms with hydro-meteorological data for monthly streamflow forecasting of Kurau River, Malaysia

Abstract

Monthly streamflow forecasting is crucial in water resources management to assess the possible future streamflow patterns. It becomes vital where streamflow of Kurau River is the primary water source to irrigate the large-scale rice scheme of Kerian, Perak, coupled with future climate change uncertainty. In this context, machine learning algorithms have received outstanding attention due to their high accuracy in forecasting through high-speed input–output data processing of self-learning from physical processes. In this study, two machine learning algorithms, support vector regression (SVR) and random forest (RF), were considered to forecast the streamflow of Kurau River in Malaysia using gauged hydro-meteorological dataset for the period from 1976 to 2005. The predictions of monthly streamflows were based on hydro-meteorological data such as rainfall, minimum and maximum temperature, relative humidity, and wind speed. A comparative study is executed to evaluate the efficiency of SVR and RF in performing the streamflow predictions of Kurau River. The results show that RF outperformed the SVR in both the training and testing phases. The results have proven that machine learning algorithms, especially the RF model, can be implemented for forecasting streamflow by using only hydro-meteorological data with high accuracy, which will improve future water resources management.