The Characteristics of GFS MOS Temperature Forecast Guidance Errors for the Past Decade
The spatial-temporal features of operational Model Output Statistics (MOS) forecast guidance errors for 2-m temperature, 2-m dewpoint, daytime maximum temperature and nighttime minimum temperature for projections of 6 to 192 hours are analyzed for the past 10 years. During this period, the National Centers for Environmental Prediction (NCEP) Global Forecast System (GFS) and the GFS-based MOS have undergone frequent updates. The main focuses here are on the temporal variations of MOS and GFS model errors, their spatial distributions and major impacts on MOS forecast guidance due to major GFS upgrades. Efforts are currently underway to recalibrate the GFS MOS with limited training samples from the latest version of the GFS model. Results of this effort will be shown.
Further analysis presents some striking features of these MOS forecast errors: they are dominated by relatively large-scale spatial structures with low-frequency temporal variations, such as annual and semi-annual cycles, ENSO resemble mode etc., namely climate biases. Most of these MOS forecast errors originate from GFS. That is, these MOS forecast errors are inherited from errors in the GFS predictors. Although MOS can greatly reduce GFS forecast errors, they remain either because the training samples used to build MOS forecast equations are not long enough or these biases cannot be removed by traditional MOS technique. However, it is very encouraging that a large portion of these forecast errors can be removed by using simple bias correction methods.