Results of Sensitivity Testing of MOS Wind Speed and Direction Guidance Using Various Sample Sizes from the Global Ensemble Forecast System (GEFS) Reforecasts

A collaborative effort between NOAA's Oceanic and Atmospheric Research (OAR) and the Meteorological Development Laboratory (MDL) is underway to statistically calibrate model output as part of the National Blend of Global Models Project. MDL is employing Model Output Statistics (MOS) to statistical post-process model forecasts for input to the national blend. As with any MOS development, a representative training sample size is crucial for developing stable forecast equations to generate guidance. In the testing phase of this project, MDL has used NOAA/OAR's 30 year GEFS reforecast data set (Hamill et al. 2013) to determine the optimum training sample length to accurately predict both the common and extreme events for a variety of weather elements.

This paper explores the effects of calibrating MOS wind speed and direction forecasts with various sample sizes of GEFS mean, single-valued reforecasts. The results shown here suggest that while there is an MAE improvement in overall MOS wind speed forecasts with increasing sample size, especially in the extended range, the relative improvements wane for sample sizes of five years or greater. Furthermore, the MAE performance of MOS wind direction forecasts that were generated from sample sizes of five years or greater are generally insensitive to the sample length even in the extended range.