Probabilistic forecasting from ensemble prediction systems: Improving upon the best‐member method by using a different weight and dressing kernel for each member

Abstract

SUMMARYEnsembles of meteorological forecasts can both provide more accurate long-term forecasts and help assess the uncertainty of these forecasts. No single method has however emerged to obtain large numbers of equiprobable scenarios from such ensembles. A simple resampling scheme, the 'best member' method, has recently been proposed to this effect: individual members of an ensemble are 'dressed' with error patterns drawn from a database of past errors made by the 'best' member of the ensemble at each time step. It has been shown that the best-member method can lead to both underdispersive and overdispersive ensembles. The error patterns can be rescaled so as to obtain ensembles which display the desired variance. However, this approach fails in cases where the undressed ensemble members are already overdispersive. Furthermore, we show in this paper that it can also lead to an overestimation of the probability of extreme events. We propose to overcome both difficulties by dressing and weighting each member differently, using a different error distribution for each order statistic of the ensemble. We show on a synthetic example and using an operational ensemble prediction system that this new method leads to improved probabilistic forecasts, when the undressed ensemble members are both underdispersive and overdispersive.

An evaluation of the Canadian global meteorological ensemble prediction system for short-term hydrological forecasting

“…Statistical procedures exist to calibrate unreliable probabilistic forecasts (e.g. Raftery et al, 2005;Fortin et al, 2006;Stensrud and Yussouf, 2007).…”

Improving the calibration of the best member method using quantile regression to forecast extreme temperatures

“…The BMM was first proposed by Roulston and Smith (2002a) and subsequently improved by Wang and Bishop (2005), Fortin et al (2006) and in Gogonel-Cucu et al (2011a,b). The last three references use the rank of a member in the sorted ensemble.…”

“…The best member method is the simplest amongst these three methods, and Gogonel-Cucu et al (2011a) show that it gives better results on ECMWF ensemble forecasts than Bayesian model averaging. Furthermore, we suspect that using ranks, as proposed in Fortin et al (2006), could be very useful to improve the representation of the tails, so it will be the basis Gogonel, Collet, Bar-Hen: Improving Calibration of Best Member Fig. 2.…”

An Improved Stochastic Weather Generator for Hydrological Impact Studies

“…The mean annual precipitation is about 1000 mm. The probability of observing a daily precipitation of more than 0.2 mm is approximately 50% (Fortin et al, 2006). Topography is very gentle in the northern agricultural and urbanised portion of the basin, while the southern portion, which reaches the Appalachian range, is mostly forested and hilly.…”