Verification of rainfall forecast products for use in short-term streamflow forecasting

[摘要] This report presents a comparison of the performance of the alternative sources of forecast rainfall available in Australia and the streamflow forecasts produced using these rainfall forecasts as inputs to a hydrological model. Rainfall forecasts from ACCESS-G and blended rainfall forecast products, including the Bureau of Meteorology’s Poor Man’s Ensemble (PME), Operational Consensus Forecasts (OCF) and version two of the seamless rainfall product, were obtained and their performance is assessed in three medium-sized catchments for a period of 11 months in 2012-2013.Rainfall forecasts from Seamless-V2 have larger bias compared to other rainfall products and have tendency to underestimate the rainfall amounts. The bias of PME is similar to that of OCF. OCF has tendency to overestimate rainfall at most lead times. The bias of ACCESS-G is between that of Seamless-V2 and PME/OCF. The Seamless-V2 rainfall forecasts seem to be relatively reliable for higher rain rates. This study reveals that finer spatial and temporal resolution forecasting models may not necessarily produce more skilful and reliable rainfall forecasts than coarse resolution models.The performance of streamflow forecasts produced using the four forecast rainfall products is consistent with the performance of forcing rainfall. The performance of PME/OCF forced streamflow forecasts is comparatively better than that of mean of Seamless-V2and ACCESS-G. The performance of the PME and OCF forced streamflow forecasts is comparable. Likewise the performance of the ACCESS-G and Seamless-V2 forced streamflow forecasts is comparable. Streamflow forecasts forced with all these rainfall forecast show skill in relation to zero rainfall reference forecasts for all catchments. Ensemble streamflow forecasts forced with ensemble Seamless-V2 forecast rainfall are not reliable - the forecast distributions are too narrow at the short lead times and are biased (under-forecasting) at the longer lead times.While the bias for some products was consistently larger, no single rainfall and corresponding streamflow forecast product clearly outperformed the others with respect to errors measured by MAE/CRPS. Since OCF and PME forecasts were more strongly correlated with observations than the ACCESS-G and Seamless-V2, particularly at longer lead times, ensemble rainfall forecasts generated by post-processing the OCF and PME output at the catchment scale are likely to result in improved streamflow forecasts. Given that the performance of the OCF and PME is generally similar, it is recommended that only one of these be further investigated and the choice be informed by the availability of archived forecasts.

[发布日期] 2014-06-18 [发布机构] CSIRO

[学科分类] 地球科学(综合)