Can We Calibrate a Daily Time-Step Hydrological Model Using Monthly Time-Step Discharge Data?

Abstract

Hydrological models are generally calibrated at longer time-steps (monthly, seasonal, or annual) than their computational time-step (daily), because of better calibration performance, lower computational requirements, and the lack of reliable temporally-fine observed discharge data (particularly in developing countries). The consequences of having different calibration and computation time-steps on model performance have not been extensively investigated. This study uses the Soil and Water Assessment Tool (SWAT) model to explore the correctness of calibrating a hydrological model at the monthly time-step even if the problem statement is suited to monthly modeling. Multiple SWAT models were set up for an agricultural watershed in the Indo-Gangetic basin. The models were calibrated with observed discharge data of different time-steps (daily and monthly) and were validated on data with the same or different time-steps. Intra- and inter-decadal comparisons were conducted to reinforce the results. The models calibrated on monthly data marginally outperformed the models calibrated on daily data when validated on monthly data, in terms of P- f a c t o r , R- f a c t o r , the coefficient of determination ( R 2 ), and Nash–Sutcliffe Efficiency ( N S E ). However, the monthly-calibrated models performed poorly as compared to daily-calibrated models when validated on daily discharge data. Moreover, the daily simulations from the monthly-calibrated models were unrealistic. Analysis of the calibrated parameters revealed that the daily- and monthly-calibrated models differed significantly in terms of parameters governing channel and groundwater processes. Thus, though the monthly-calibrated model captures the patterns in monthly discharge data fairly well, it fails to characterize daily rainfall-runoff processes. The results challenge the existing practice of using different calibration and computation time-steps in hydrological modeling, and suggest that the two time-steps should be the same, irrespective of the time-step required for modeling.