Assessing the performance of ensemble streamflow forecasts produced by the latest generation of rainfall forecasts

[摘要] This report compares the performance of the alternative sources of quantitative rainfall forecasts (QPFs) from numerical weather prediction (NWP) models available in Australia. Deterministic rainfall forecasts from ACCESS-G and the Bureau of Meteorology’s poor man’s ensemble (PME), are assessed for 14 catchments of varying sizes that cover a wide range of climatic conditions and hydrological characteristics. We generate ensemble rainfall forecasts (calibrated QPFs) by applying the rainfall post-processor (RPP) to the deterministic ACCESS-G and PME rainfall forecasts to minimise the bias, increase accuracy and reliably quantify the uncertainty in the raw forecasts. These calibrated QPFs are then used as input to a semi-distributed conceptual rainfall-runoff modelling system to produce ensemble streamflow forecasts to lead times of 9 days. The performance of the streamflow forecasts is also evaluated and compared with reference streamflow forecasts. Deterministic PME rainfall forecasts outperformed ACCESS-G rainfall forecasts, resulting in better streamflow forecasts. For both PME and ACCESS-G, post-processing deterministic rainfall forecasts leads to lower biases in QPFs, reduced errors, and quantifies uncertainty. Streamflow forecasts generated from ensemble rainfall forecasts consistently outperformed those generated from deterministic rainfall forecasts. Post-processed PME rainfall forecasts marginally outperformed post-processed ACCESS-G rainfall forecasts. Overall, this meant that streamflow forecasts from post-processed PME QPFs slightly outperformed those generated from post-processed ACCESS-G forecasts.Our recommendations from this study are as follows:1.For deterministic streamflow forecasts, we recommend PME as the source of rainfall forecasts.2.Wherever possible, we recommend that ensemble rainfall forecasts be used to generate operational streamflow forecasts.3.We recommend that post-processed PME forecasts be used in preference to post-processed ACCESS-G QPFs to generate ensemble streamflow forecasts. We note that post-processing other ensemble rainfall QPFs (e.g. AGREPS) may ultimately produce more skilful forecasts than PME.

[发布日期] 2015-05-14 [发布机构] CSIRO

[学科分类] 地球科学(综合)