Jim Zidek (UBC) Uncertainty in ensembles of deterministic models with application to probabilistic weather forecasting. This talk will provide an extension of Bayesian melding . The extension can combine measurements with outputs (simulated data) from an ensemble of deterministic models to determine model bias. Or it can combine the simulated data alone and thereby provide an integrated output with probabilistic estimates of model uncertainty. The model outputs can be on different scales. We apply the Bayesian ensemble melding model to the sea level temperature data at Pacific Northwest area. We combine the measurements with model outputs from an ensemble of five deterministic models for spatial predictions. The predictions of Bayesian ensemble melding model are compared with those of averaging model outputs and Kriging approach. We also show how this purely spatial approach can be turned into a 48 hour ahead probabilistic weather forecaster.