Statistical interpretation of model output using MOS A Canadian perspective (Invited Presentation)

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Tuesday, 6 January 2015: 11:15 AM
211A West Building (Phoenix Convention Center - West and North Buildings)
Laurence J. Wilson, EC, Dorval, QC, Canada

Model Output Statistics (MOS) is one of those rare ideas that is simple and yet incredibly effective and beneficial to operational forecasting. As an essential link between a model's version of the weather and actual observed conditions, MOS has rightly endured for more than 40 years and still is widely used for operational forecast guidance. In Canada, we learned about MOS from Dr. Glahn and Dr. Klein, and began to build a MOS system of our own in the early 1980s. Our first MOS product became operational in 1986, not for temperature, but for wind. By 2000, we had implemented the first version of the operational updateable MOS system that still runs today (UMOS). MOS is based on the idea that model output can be corrected using (usually) fairly simple statistical methods. It has traditionally been focused on the statistical adaptation of the output of deterministic models. However, strictly speaking, all that is required for MOS is that model output be built into the statistical predictive equations; one is free to use any statistical methodology one likes. Furthermore, the output can be expressed as a deterministic forecast or as a probability. In the US and Canada (since 1982) MOS probability forecasts have been used as guidance for public probability forecasts long before ensemble-based estimates of probabilities became available. This presentation will be in three parts: First, a brief history of MOS-related developments in Canada will be presented, followed by some examples of recent MOS applications. Then, finally, the question of whether MOS methods should change in future will be addressed.