Rainfall-runoff modelling using the machine learning and conceptual hydrological models

This study compares the capability of simple data-driven and process-driven models to simulate daily discharge in a snow dominated semi-arid watershed in relation to its hydro-meteorological characteristics. M5 model tree was considered for daily discharge simulation of Taleghan watershed in north of Iran, and the results were compared with those of IHACRES and HSPF models. Results showed that with the same meteorological input data, the HSPF model performed best in predicting the daily runoff followed by the IHACRES model. M5 model overestimated the daily runoff in low flow season (May-December) as the water required to fill the watershed storage capacity was not considered by the model. Using the stream discharge of the prior day (Q(t-1)) as additional input to the M5 model resulted in much improved simulation of daily discharge (RMSE = 3.55, NSE = 0.94, KGE = 0.96).