Initial Condition Sensitivity and Error Growth in Forecasts of the 25 January 2000 East Coast Snowstorm

Abstract

Short- and medium-range (24?96-h) forecasts of the January 2000 U.S. east coast cyclone and associated snowstorm are examined using the U.S. Navy global forecast model and adjoint system. Attention is given to errors on the synoptic scale, including forecast position and central pressure of the cyclone at the verification time of 1200 UTC 25 January 2000. There is a substantial loss of predictive skill in the 72- and 96-h forecasts, while the 24- and 48-h forecasts capture the synoptic-scale features of the cyclone development with moderate errors. Sensitivity information from the adjoint model suggests that the initial conditions for the 72-h forecast starting at 1200 UTC 22 January 2000 contained relatively small, but critical, errors in upper-air wind and temperature over a large upstream area, including part of the eastern Pacific and ?well observed? areas of western and central North America. The rapid growth of these initial errors in a highly unstable flow regime (large singular-vector growth factors) is the most likely cause of the large errors that developed in operational short- and medium-range forecasts of the snowstorm. The large extent of the upstream sensitive area in this case would appear to make ?targeting? a small set of new observations an impractical method to improve forecast skill. A diagnostic correction (derived from adjoint sensitivity information) of a part of the initial condition error in the 72-h forecast reduces the forecast error norm by 75% and improves a 1860-km error in cyclone position to a 105-km error. This demonstrates that the model is capable of making a skillful forecast starting from an initial state that is plausible and not far from the original initial conditions. It is also shown that forecast errors in this case propagate at speeds that are greater than those of the synoptic-scale trough and ridge features of the cyclone.