Abstract

The present study aims to investigate theperformance of the artificial intelligence to emulate theconventional physically-based hydrological models. Althoughthese conventional models could accurately depict theunderlying physical processes, but they require a lengthypreprocessing phase as well as a tedious calibration time.Therefore, a need to examine the potential efficient alternativefor these models is highly felt. This need becomes imperativeonce we adopt fine temporal and spatial resolutions for ourhydrological modeling, leading to a massive number ofto-be-analyzed cells. To this end, we propose a learningframework towards an accurate prediction of runoff rates usingmeteorological variables, and hence, mimicking the VariableInfiltration Capacity (VIC) by a nimble systematized predictivemodel. We also present a novel strategy to optimally select themost informative subset of data to train our predictive model,out of the pool of accessible data. This strategy would thenconsiderably enhance the performance of our prediction interms of computation time. We reported our result as thePearson correlation coefficient between the predicted andactual runoff rates. Our predictive model was able to forecastthe runoff rates with the mean correlation coefficient of 0.9007for the cells within the study basin.