The impact of time resolution on modeling performance in runoff volume and peak discharge estimation

Abstract

Conceptual rainfall-runoff models have been experiencing many advancements in recent decades. These include improvements in model structure and wider calibration options. More complex model structure allows not only long-term modeling but also those at sub-hourly. Daily runoff volume, for instance, can be calculated at disaggregated timescales. Besides, the incorporation of sub-hourly timescales into continuous simulation models means that these models can now be used to estimate instantaneous flows. Nevertheless, the impact of time resolutions on modeling performance is not clear. Some past studies showed that the impact of timescales on model performance varied from almost no different to significant. While some studies showed higher time resolutions produced better results, the opposite has also been observed in few other studies. In relation to the calibration options, non-statistical measures have also been introduced in addition to the statistical ones. Parafield Drain (PD) is a water harvesting and reuse scheme involved in the Waterproofing Northern Adelaide (WNA) Project. In order to develop a decision support system (DSS) for PD which includes real- time rainfall-runoff modeling, it is necessary to select the best modeling timescale in terms of model performance. The present study investigates how model performance at PD modeled by WaterCress was affected by modeling time resolutions. In particular, two modeling outputs, i.e. runoff volume and peak discharge, were subjected to the assessment. Daily runoff volume assessment was based on rainfall data at two timescales, 30-min and daily, while flood peaks were estimated at 30-min and 1-h. Parameters of the former were calibrated and validated against 20 months of historic streamflow data, while the latter compared against two historical flood peaks. In relation to the model performance assessment, WaterCress provides 4 performance measures of which three of them are statistical: coefficient of determination (R 2 ), Nash-Sutcliffe Efficiency (NSE) and Standard Error of Estimate (SEE); and another non-statistical measure: Percentage Volumetric Difference (VD). In runoff modeling calibration stage, it was found that not every selected performance measures were improved. For example, VD was getting worse whenever the other measures were getting better and so forth. Assessing the model performance was therefore considered to be sufficient by considering multi-objective calibration, one from each statistical and non-statistical measures. Furthermore, a minimum of 9-month streamflow data was required to obtain calibration which was insensitive to the data length. Results also showed that higher time resolution produced slightly better prediction in runoff volume calculation and remarkably improved peak flow estimation.