|ASADZADEH, MASOUD - University Of Manitoba|
|LEON, LUIS - Environment Canada|
|YANG, WANHONG - University Of Guelph|
|Bosch, David - Dave|
Submitted to: Journal of Hydrology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 12/14/2015
Publication Date: 1/4/2016
Citation: Asadzadeh, M., Leon, L., Yang, W., Bosch, D.D. 2016. One-day offset in daily hydrologic modeling: An exploration of the issue in automatic model calibration. Journal of Hydrology. 534:164-177.
Interpretive Summary: For computational efficiency and simplicity, computer models of watershed models are frequently set-up to produce daily simulations. However, precipitation frequently falls over a sub-daily time period. Because of this, discrepancies are frequently created between the observed and simulated timing of the runoff hydrograph. A tool was developed to help identify these inconsistencies and differentiate between errors generated due to the lumping that occurs for daily time intervals and those that are due to parameterization errors within the model. This tool can be used to more accurately interpret the accuracy of model simulations and to create better parameterization data sets.
Technical Abstract: The literature of daily hydrologic modelling illustrates that daily simulation models are incapable of accurately representing hydrograph timing due to relationships between precipitation and watershed hydrologic response. For watersheds with a time of concentration less than 24 hrs and a late day precipitation event, the observed hydrographic response frequently occurs one day after the precipitation peak while the model simulates a same day event. The analysis of sub-daily precipitation and runoff in this study suggests that, this one-day offset is inevitable if the same time interval, e.g. the calendar day, is used to measure daily precipitation and runoff datasets, and daily simulation models will fail to emulate this one-day lag. This one-day offset issue (1dOI) results in significant residuals between simulated and measured hydrographs and adversely impacts the model performance metrics that are based on the aggregation of daily residuals. This is an error introduced through data aggregation and needs to be properly addressed before evaluating the model performance. Otherwise, model performance metrics would not represent the actual modelling quality and could mislead the automatic model calibration. In this study, an algorithm called Shifting Hydrograph In order to Fix Timing (SHIFT) is developed to scan the simulated hydrograph against the measured one, automatically detect all incidents of 1dOI and shift the simulated hydrograph of those incidents one day forward before calculating the performance metrics. SHIFT is employed in the daily calibration of the Soil and Water Assessment Tool model. Results show that SHIFT can help the automatic calibration to identify a solution that accurately estimates the magnitude of peak flow rates and the shape of rising and falling limbs of the measured hydrographs. In contrast, the same automatic calibration without SHIFT finds a solution that systematically underestimates the peak flow rates while trying to emulate the one-day lags. Furthermore, it is shown that the calibrated model performs very well with an alternative precipitation dataset that has a minimal number of one-day offsets. Therefore, it is concluded that the SHIFT algorithm successfully minimized the impact of 1dOI on the model parameter estimation.