Please login first
Daily Streamflow Modelling Using ML Based on Discharge and Rainfall Time Series in the Besós River Basin
* 1 , 2
1  Mediterranean Agronomic Institute of Zaragoza (IAMZ), International Centre for Advanced Mediterranean Agronomic Studies (CIHEAM). Av. Montañana, 1005, 50059 Zaragoza, Spain
2  Eurecat, Technology Centre of Catalonia, Unit of Applied Artificial Intelligence, 08005 Barcelona, Spain
Academic Editor: Lampros Vasiliades


The planning, design and management of water resources projects require good estimates of flow and maximum discharge at certain points within a basin. Machine Learning (ML)-based Data-driven modelling is an efficient approach to achieve such an end. This paper is mainly aimed to determine if daily streamflow can be predicted satisfactorily using ML algorithms based on open source flow discharge and rainfall historical time series. In this sense, two modelling scenarios, without and with considering the antecedent hydrologic conditions, were evaluated. Three ML algorithms—Support Vector Machines, Random Forest (RF) and Gradient Boosting (GB) —compared to Multiple Linear Regression (MLR), were implemented and applied to the Besós River basin, Spain. The prediction results were compared to observed values, based on performance metrics (root mean square error, mean absolute error, coefficient of determination and Nash-Sutcliffe coefficient of efficiency) and graphical examination (observed and predicted hydrographs). The performance comparison of the results revealed that SVR model outperformed the other suggested models. Additionally, it was deduced that taking into account the preceding hydrologic conditions clearly improves the prediction results.

Keywords: streamflow modelling; machine learning; data-driven; preceding hydrologic conditions; virtual sensor