CBaM : A calibration, bridging and merging method for post-processing GCM forecasts of meteorological variables

Thursday, 18 December 2014: 1:55 PM
Andrew Schepen1, Q.J. Wang2 and David E Robertson2, (1)CSIRO Land and Water, Dutton Park QLD 4102, Australia, (2)CSIRO Land and Water, Highett VIC 3190, Australia
A calibration, bridging and merging (CBaM) method has been developed to post-process outputs from general circulation models (GCMs) for seasonal forecasting of climate variables. An overview of the methodology and a summary of applications will be given.

Post-processing of GCM forecasts is often necessary for the outputs to be more informative. CBaM attempts to maximise the value of GCM outputs by not only post-processing the variable of interest (calibration), but also using other available outputs such as sea surface temperatures to generate forecasts (bridging). Merging forecasts from calibration and bridging models leads to the opportunity to improve forecasting skill for some regions and time periods.

In CBaM, separate calibration and bridging models are established using a Bayesian joint probability modelling approach. The models generate forecasts in the form of ensembles. Forecasts from multiple calibration and bridging models are merged using Bayesian model averaging. Ensemble time series forecasts are produced by sequencing ensemble members using the Schaake Shuffle.

Results to date are presented for a number of applications. The method is applied to produce gridded rainfall forecasts for Australia and China, using outputs from single or multiple GCMs. It is also applied to produce monthly forecasts of catchment rainfall for up to 12 months in advance. Monthly forecasts of catchment rainfall are used in a hydrological model to forecast streamflow for up to 12 months. CBaM forecasts are shown to extract skill from the atmospheric and oceanic modules of the GCM, and are also shown to be reliable.

Work is in progress to apply CBaM to forecasts of other climate variables, including temperature, and to combine forecasts from multiple GCMs, including the ECMWF System 4 and NCEP CFSv2 models.