An evaluation of hydrological models for predicting mean-annual runoff and flood quantiles for water quality modelling

Abstract

Predicting spatial variations in runoff is important for water resource assessment and understanding spatial variation in flood quantile is valuable for assessing flooding risk and sediment loads during floods. Sediment fluxes in river networks are disproportionally sensitive to runoff events, river bank erosion and floodplain deposition. Many spatially disaggregated models of erosion and sediment transport require inputs of long term runoff volume and daily flood quantiles for each link in a stream network. This paper presents an evaluation and comparison of two different hydrological models in predicting catchment water yields and flood quantiles throughout a stream network. Using data from tropical Queensland catchments, a simple regionalization model (SedNet) and a daily time step conceptual rainfall-runoff model (SIMHYD) were calibrated using a set of gauge data and evaluated for ungauged condition using another set of gauge data. Results showed that a daily timestep rainfall-runoff model could provide better calibration to observed water yield but a regionalised model was as good as or better than the daily time step model for calibration to flood quantile. For prediction (on ungauged condition), both models produced similar results in predicting mean annual water yield and flood quantiles, with slightly better prediction of daily flow variability by the regionalised model. Based on an analysis of sensitivity of bankfull recurrence interval, we found large uncertainty in predicting bankfull flow using the SIMHYD model. Results imply that a simple regionalized model is as good as a complex daily time step model for long-term sediment budgets in water quality modelling. Using a daily time step, however, could be necessary if event modelling of flood information and sediment is required.