Rejoinder.

This article presents the authors' response to the comments of Claudia Tebaldi, Doug Nychka, William Briggs, and Mark S. Roulston on their article 'Calibrated Probabilistic Mesoscale Weather Field Forecasting: The Geostatistical Output Perturbation Period.' All three discussions compared the geostatistical output perturbation (GOP) method with dynamical ensemble methods, and suggested that a combination of the two approaches would be fruitful. We strongly agree. Dynamical ensemble methods generate an ensemble of initial conditions and run the numerical weather prediction model forward from each of them in turn, whereas the GOP method instead perturbs the model output rather than its input. Dynamical ensembles have the advantage, pointed out by Tebaldi and Nychka, that they can capture nonlinear aspects of forecast uncertainty, but they typically require considerable resources in terms of data, data assimilation software, and computing power. The GOP method, on the other hand, is much faster and does not require any data beyond the deterministic forecast once it has been trained using historical data. We, therefore, endorse Roulston's suggestion that the GOP method be used as a benchmark for other ensemble methods.