Development and testing of a multi-model rainfall-runoff streamflow forecasting application

[摘要] This document reports the progress of WIRADA research to support short-term streamflow forecasting at the Australian Bureau of Meteorology. Eight rainfall-runoff models have been implemented in a forecasting application that also includes multiple catchment- and channel routing schemes. Four of these models (abcd, the Budyko Framework, Wapaba, GR2M) aremonthly water balance models, three (AWBM, SimHyd, PDM) are daily continuous simultation models, and one (Coefficient) is useful is a naïve baseline. The application also uses the Shuffled Complex Evolution parameter optimization routine. The hydrologist can use a weighted average of any combination of six objective functions. This study employed three measures that optimize performance for large, moderate and low flows and onemeasure that minimized bias. The application was used to establish a model for the Ovens catchment in Southeast Australia. Daily spatially lumped and semi-distributed simulations of flow were performed at21 internal sub-catchments. The simplest naïve baseline (a coefficient rainfall-runoff model without catchment routing) performed very poorly although a somewhat more complex baseline (a coefficient model with catchment routing) performed significantly better. The full continuous simulation models performed better than the baselines with the AWBM model performing the best of all the models during independent model validation in 53% of cases. The PDM and SimHyd models were top performers in 24% and 18% of cases respectively.The spatially semi-distributed models performed better than lumped models for moderate and low flows, and for bias, even though the channel routing parameters were not optimized. Nash Sutcliffe scores of 0.75-0.85 were not uncommon although some problem subcatchmentswere identified. Future tasks for this research will involve the implementation of other rainfall-runoff models for comparison, such as the GR4J or GR5H models. Improved operationally-relevant forecast evaluation measures will be developed. A broader array of catchments will be tested, with a particular emphasis on performing sub daily experiments. The use of uncertain future rainfall should be evaluated, along with the use of state updating and/or error correction.

[发布日期] 2009-12-31 [发布机构] CSIRO

[学科分类] 地球科学(综合)