The Canadian Updateable Model Output Statistics (UMOS) system, developed and implemented by the Meteorological Research Branch (MRB) and the Candian Meteorological Centre (CMC) of the Meteorological Service of Canada (MSC), has been supplying operational guidance for land-based stations across Canada since March of 2000. Recently the UMOS system has also been adapted, on an experimental basis, to the forecast of marine winds at the moored weather buoys deployed on the Great Lakes. Employing the winds observed at these buoys as predictands, UMOS constructs regression equations using predictors from the Global Environmental Multiscale (GEM) numerical weather prediction (NWP) model, which supports the short-term weather and environmental forecast programmes of the MSC. UMOS forecasts for the winds on the Great Lakes have been produced in real time since December of 2002, and an assessment of these forecasts is the subject of this paper.

In order to effect this assessment, the quality of the UMOS solutions will be compared against that of Great Lakes wind forecasts obtained from three sources: the direct output of the regional GEM model, the official marine forecast issued for the Great Lakes by the Ontario Region of the MSC, and the recently implemented MOS system developed by the Meteorological Development Laboratory (MDL) of the National Weather Service (NWS). Conventional summary scores, such as the mean absolute error and root-mean-square error, attest to the superiority of the UMOS solutions over the direct output from the GEM model, particularly for the forecasts of windspeed. The inherently discrete nature of the official marine forecasts suggested a distributions-oriented approach to the verification, for which the categories of wind speed and direction were dictated by the highly structured and codifed form (the MAFOR) of the marine forecast issued for the St. Lawrence Seaway. Not only does such a distributions-oriented comparison demonstrate the overall superiority of the UMOS forecasts, it refines the characterization of the performance of the forecast systems by displaying their systematic biases as a function of the speed and direction of the wind. The MDL MOS forecasts were introduced into operational service in the spring of 2003, so a comparison of this forecast system to UMOS will take place in the fall of 2003, with the accumulation of a larger sample of forecasts. Although both forecast systems are based on the MOS formulation, there are some significant differences in the implementation of this formulation, particularly in the choice of predictands, and a casual inspection of their solutions reveals some noticeable differences. Results of the comparison between the UMOS and MDL MOS solutions will be discussed in the context of these differences.

Lastly, some consideration is given to the climatological predictors used in the regression equations for the windspeed. This predictand exhibits a strong seasonal signal, whose accurate resolution in the statistical models is complicated by the incomplete monitoring programme in place on the Great Lakes. This problem will be elucidated, and a possible solution discussed.