Assessing the performance of postprocessing method for ensemble rainfall forecasts

[摘要] This report describes a method for post-processing ensemble rainfall forecasts produced by numerical weather prediction models. The ensemble post-processing is based on the method developed for post-processing deterministic rainfall forecasts (Robertson et al., 2013a; Shrestha et al., 2015) which uses the Bayesian Joint Probability model and the Schaake shuffle to remove bias and reliably quantify forecast uncertainty. We evaluate the ensemble post-processing method at a spatial scale relevant to hydrological modelling for 23 catchments of varying sizes that cover a wide range of climatic conditions. The performance of post-processed ensemble ACCESS-GE rainfall forecasts is compared to the performance of the post-processed deterministic ACCESS-G and PME rainfall forecasts.Rainfall forecasts from ACCESS-G, PME and ACCESS-GE NWP models are found to be biased. In most of the catchments, the bias in the forecasts from PME is the smaller than that from the ACCESS-G and ACCESS-GE. The error in the rainfall forecasts from ACCESS-GE is consistently smaller than that in the ACCESS-G and PME in all catchments. The ensemble post-processing method described here proved highly effective. The post-processing significantly reduces the forecast bias from all raw forecasts to below 20% in most catchments. The improvement is more pronounced in grassland catchments where the bias reduces from about 150 % to less than 30%. The biases in the calibrated QPFs from ACCESS-G, PME and ACCESS-GE are very similar. The error of the calibrated QPFs from ACCESS-G and PME are significantly lower than those of the raw ACCESS-G and PME, respectively for all lead times. However the error of the calibrated QPFs from ACCESS-GE is similar with that of raw ACCESS-GE in most of catchments. The forecast distributions from calibrated ACCESS-GE are more reliable than raw ACCESS-G. The reliability of the calibrated QPFs from ACCESS-G, PME and ACCESS-GE are reasonably good and similar. The ensemble spread of the raw ACCESS-GE is under-dispersed, while the spread of the calibrated ACCESS-GE, ACCESS-G and PME matches reasonably well with the ensemble error.To make informed decisions about which rainfall forecasts produce more skilful streamflow forecasts, it is recommended to evaluate the performance of the streamflow forecasts produced by forcing hydrological models with the post-processed ACCESS-GE rainfall forecasts.

[发布日期] 2016-01-04 [发布机构] CSIRO

[学科分类] 地球科学(综合)