Multimodel Hindcast with CFS and EU DEMETER: ohio
Using the hindcast from EU DEMETER project and the CFS hindcast, we created this multimodel-based 6-month seasonal forecast from the Southeast US and Ohio river basin. The forecast from the 8 dynamic climate models are merged with obs climatology within a Bayesian framework to produce the posterior distribution. Non-gaussian distributions are transferred to a standard normal distribution before the merging, and the normal posterior is transferred back to with the same reference distribution. ESP predictions are created using 20 randomly selected years, while CFS and CFS+ECMWF predictions are created using 20 conditionally selected years. The selection of these years is based on the similarity between the anomaly patterns in the posterior forecast and the historical observations.