A Advanced Model Output Statistics Guidance System

Three methods of improving the accuracy of Model Output Statistics (MOS) weather forecasts are investigated. One is a model consensus approach in which MOS forecasts based on two independent numerical weather prediction (NWP) models are combined. The second is a nonparametric regression technique that automatically estimates the appropriate functional (curvative) relationship for each predictor term in an additive model. The third technique involves the incorporation of a network of surface observations surrounding the forecast location as an additional source of predictive information. Each technique is subjected to an extensive multi -year verification on independent data consisting of more than a million cases for a variety of forecast variables, locations, initial times, and lead times. Results show that each technique is able to improve accuracy over traditional MOS procedures by an average of 3%-5%. It is indicated that these 3%-5% improvements in accuracy are on par with a half-decade of scientific advancement and a 6-8 h lead time gain for a 24-h forecast. Moreover, it is shown that these improvements result in MOS forecasts that are as accurate, or slightly more accurate, than subjective forecasts from the National Weather Service.