Research Abstracts 1998-1999: Report No. 99-01

Barbour, P.L. and S. N. Walker, “Site Wind Forecasts from an Operational Numerical Weather Prediction Model," OSU Wind Research Cooperative, WRC Report No. 99-01, July 30, 1999, 30 pp., $22.50.

The increase in resolution provided by modern, regional, mesoscale models has improved the prospects for the development and use of new objective forecast products to predict conditions at specific locations. The traditional means of producing objective forecasts has been the use of Model Output Statistics (MOS; Glahn and Lowry, 1972). MOS forecasts typically use output from one of a number of large-scale numerical prediction models along with site observations to develop forecast regression equations. The strength of MOS forecasts is that the procedure accounts for many of the systematic biases present in any model and, with a sufficient sampling period, can be used to produce accurate and reasonable forecasts. One of the biggest problems with the MOS approach, however, is that a relatively long sampling time is required before reasonable results can be obtained. A large number of cases must be sampled before conditions of significance can be sampled and represented by the regression equations. This can be particularly troublesome when major changes are made to the model. Changes generally require re-calculation of the basic MOS relationships, a process that can be extremely time and resource intensive. In the case of regional models, changes are made quite often and it would be difficult to maintain an adequate consistency to MOS forecasts. As an alternative to the traditional MOS forecasts, a modified grid-point adjustment method is presented and examined here. This method represents the simplest approach possible above direct grid-point predictions and is viewed as a first step in developing more complete and detailed adjustment schemes. The basis of this approach lies in the belief that although most of the important physical and dynamical processes of the atmosphere can be represented by current models, systematic biases will still be present. To account for the biases, a simple linear regression equation is computed each day using only a limited sampling period consisting of the near-recent history at a site. This method was applied to model grid-point wind forecasts at a number of wind sites in the Pacific Northwest for two study periods. Results show that this method can provide improved forecasts, especially for sites where a significant topographic influence on the winds exist. At several sites, improvements on the order of 30% to 40% were obtained over using direct model data. Additional improvements may be possible through the use of longer regression periods and through improvements in the resolution of the basic weather prediction model.