Post-processing rainfall forecasts from a numerical weather prediction model in Australia

[摘要] Bias free and reliable ensemble rainfall forecasts are required to produce reliable and skilful ensemble streamflow forecasts. The rainfall forecasts that are publicly available from Australian NWP models are deterministic and often contain systematic errors. Therefore, it is necessary to remove the systematic biases and reliably quantify uncertainty in rainfall forecasts before they can be used for streamflow forecasting. This report presents the further development and application of a rainfall post-processing method to remove forecast bias and reliably quantify forecast uncertainty.We post-process rainfall forecasts from a global Australian NWP model by combining a simplified version of the Bayesian joint probability (BJP) modelling approach and the Schaake shuffle. The BJP modelling approach constructs statistical relationships between NWP forecasts and observed rainfalls. It corrects biases in the NWP model forecasts and generates an ensemble of possible forecasts that reflects the uncertainty in the NWP forecast. The BJP modelling approach is applied to individual locations and individual forecast periods to produce probabilistic rainfall forecasts. Ensemble forecasts with appropriate spatial and temporal correlations are then produced by linking samples from the forecast probability distributions using the Schaake shuffle.The post-processing method is evaluated for a range of small-medium sized Australian catchments that cover a wide range of climatic conditions and hydrological characteristics. We show that the post-processing method significantly reduces the forecast bias and error, produces forecasts that successfully discriminate between events and non-events for both small and large precipitation occurrences, and reliably quantifies the forecast uncertainty.

[发布日期] 2013-11-01 [发布机构] CSIRO

[学科分类] 地球科学(综合)