Parameter sensitivity of a watershed-scale flood forecasting model as a function of modelling time-step

Abstract

Listen

Translate article

Although ongoing technological advances have alleviated data restrictions and most of the computational barriers to distributed modelling, lumped, parsimonious, conceptual and rainfall-runoff models are still widely used for flood forecasting. However both optimum parameter values and the fluxes of water through individual model components change significantly with the time-step used. Thus, such models should be used with caution in applications such as hydrograph separation or water quality studies that require the fluxes through individual flow routes through the model or which try to relate parameters to physical features of the catchment. To demonstrate this time-scale limitation, a parameter sensitivity analysis was performed on the lumped conceptual Soil Moisture Accounting and Routing with Groundwater component (SMARG) model for a 182 km2 rural catchment in Ireland for a number of time-steps, flow regimes and evaluation metrics. A global sensitivity analysis method (Higher Dimensional Model Representation) showed that sensitivity indices vary greatly with time-step and evaluation metric. The sensitivity of parameters also varied for different flow regimes. Certain parameters' sensitivities remain fairly constant across both flow regimes and time-step, while others are very much regime or time-step dependent. Care should be taken in using internal information from conceptual models because of this strong dependence on time-step.