Applying rainfall ensembles to explore hydrological uncertainty

Yetemen Ö.

The 23rd International Congress on Modelling and Simulation (MODSIM 2019), Canberra, Australia, 1 - 06 December 2019, pp.1070-1076

  • Publication Type: Conference Paper / Full Text
  • Doi Number: 10.36334/modsim.2019.k14.kumari2
  • City: Canberra
  • Country: Australia
  • Page Numbers: pp.1070-1076
  • Istanbul Technical University Affiliated: Yes


The widespread presence of spatial and temporal variability in rainfall is well known. However, this variability can not be captured by point gauge measurements alone. An accurate representation of this variability is crucial for hydrological and meteorological applications. Precipitation information is an essential input for all hydrological models, but can be especially challenging in regions where no or very few rain gauge stations are established. Moreover, the uncertainties involved in these rainfall inputs are usually not considered or ignored in the hydrological simulations. Uncertainty in precipitation input arising from errors in spatial representation, measurement or estimation accuracy, can create uncertainty in the streamflow estimation. In such cases, the use of high resolution rainfall ensembles can play crucial role in modelling the rainfall-runoff relationships, particularly for high flows or flash floods. This study aims to characterise the hydrological uncertainty involved in the high flow simulations using rainfall-runoff models.

This study focuses on characterising uncertainty in rainfall-runoff model outputs through the application of ensemble precipitation estimates. We demonstrate the results for selected events in the Macleay River Basin using a simple rainfall-runoff model. The basin is situated in the New South Wales (NSW), mid-north coast of Australia and is prone to flash floods. Historically, flooding in the lower Macleay Valley occurs at every 2 or 3 years, and the largest floods have occurrence interval of 100 years. We also explored the response of basin area on this uncertainty and the cascading of this uncertainty from upstream to downstream of the basin.

The GR4H model, which is an hourly implementation of GR4J, is merged with Muskingum Routing in this study. We used GPM (Global Precipitation Mission) precipitation data at 10km x 10km spatial resolution further downscaled to 2km x 2km spatial resolution for three years (2015-2017). Further, radar data along with the GPM precipitation is used to create 50 member ensemble rainfall estimates at 2km x 2km spatial and hourly temporal resolution. In order to analyse the impact of rainfall uncertainty on streamflow we selected some of the high flows events. The three sub-basins having an area between 377-860 km2 along with the Macleay Basin (~8,000 km2 ) is used to run the simulations. Further, we compared and contrasted the runoff generated at the outlet by grid-wise simulations, basin averaged simulation, and simulations from ensemble rainfall as input with the observed streamflow.

The results show that the grid-wise streamflow generation are comparatively better in capturing the peak flow events in the Macleay Basin and sub-basins than the basin-wise streamflow output probably due to the use of the same parameter throughout the simulations, lower averaged streamflow at each sub-basins, and more amount of overall losses at the basin scale. The observed peak flow is within the range of streamflow simulated using ensemble rainfall for all the basins.

The application of interest to this study is the use of ensemble precipitation forecasts to generate ensemble streamflow forecasts. This study shows that the rainfall-runoff modelling with ensemble precipitation inputs can considerably reduce the amount of uncertainty in simulation results, particularly in data-sparce regions.