The new AVN/MRF MOS development and model changes: a volatile mix?
Mary C. Erickson, NOAA/NWS, Silver Spring, MD; and J. P. Dallavalle and K. L. Carroll
On May 30, 2000, the National Weather Service (NWS) introduced an updated suite of objective Model Output Statistics (MOS) guidance. Forecast equations were derived from more recent historical samples of both the Aviation (AVN) and Medium Range Forecast (MRF) runs of the Global Spectral Model (GSM). The samples contained data from April 1997 to March 2000 and were constructed from a higher resolution model archive than used in prior developments. Data from the Reanalysis Project (years 1992 to 1996) were also included in the MRF development. Model changes, including increases in resolution, physics updates, and data assimilation modifications continued throughout the historical sample period. Although one 5-week sample of model forecasts from the summer of 1998 was eliminated due to its anomalous nature, most of these changes were deemed to not affect the statistical characteristics of the sample. With the current reality of yearly numerical model improvements, a variety of techniques were employed in the development of the MOS system to attempt to develop robust forecast equations which would accommodate the mixed development samples. In general, however, predictors sensitive to NWP enhancements were not eliminated from the equations. Evaluations of the MOS guidance on independent data demonstrated overall improvement in the new MOS system over the older guidance.
In the spring of 2001, a new suite of GSM changes was scheduled for implementation and run in a parallel version of the MRF. The new operational MOS equations were applied to this parallel run of the model, and the resulting forecasts evaluated for sensitivity to the model changes. A distinct shift in the bias characteristics was noted in the MRF low level thermal fields over North America, and this shift resulted in changes in the MOS guidance. In this paper we discuss the utility of this parallel evaluation to both the statistical and numerical weather prediction modelers involved. We also discuss a variety of techniques the Meteorological Development Laboratory (MDL) investigated to respond to this model shift. These include eliminating predictors, shifting developmental samples, blending developmental samples, and bias correcting both model fields and statistical forecasts. Finally, the broader issues of building a responsive, yet skillful, forecast system are examined.
Extended Abstract (176K)
Session 3, weather forecasting
Thursday, 17 January 2002, 8:30 AM-3:30 PM
Previous paper Next paper
Browse or search entire meeting
AMS Home Page