Performance of a new ensemble streamflow forecasting system for Australia

[摘要] We describe the performance of a new ensemble continuous streamflow forecast system for Australia. The system incorporates rainfall forecasts from numerical weather prediction (NWP) models to increase streamflow forecast lead time to 9 days and uses soil moisture accounting to produce continuous streamflow forecasts. NWP model rainfall forecasts are post-processed with a Bayesian joint probability model to correct NWP model biases and quantify uncertainty. The ensemble forecasts are then ordered with the ‘Schaake shuffle’ to ensure the forecasts have realistic temporal and spatial patterns.Post-processed rainfall forecasts are then used as input to a GR4H hydrological model. Before simulating forecast streamflows, the GR4H model is run for two years with observed rainfall inputs to initialise soil moisture stores. An error correction is also applied to the streamflow forecast to ensure forecast streamflows are aligned with observations at the forecast issue time.We test the ensemble streamflow forecast system on 10 Australian catchments that span a variety of climatic conditions and hydrological characteristics. For perennially flowing streams, the ensemble streamflow forecasts outperform reference forecasts for lead times of 4 days or more. For ephemeral streams, forecasts were skilful for lead-times of 2 days or fewer.We test the performance of forecasts generated at 1-hour, 3-hour, 6-hour and 24-hour time steps. The 1-hour forecasts outperformed forecasts at other time steps largely because the streamflow error correction functioned more effectively at the 1-hour time step.Representation of forecast uncertainty is not completely reliable. We use only rainfall uncertainty to quantify uncertainty in the streamflow forecasts. Streamflow forecasts could be made more reliable with improved representation of hydrological modelling errors.

[发布日期] 2013-06-06 [发布机构] CSIRO

[学科分类] 地球科学(综合)