A series of experiments was performed to test various methods of incorporating sounding data from the visible and infrared spin-scan radiometer atmospheric sounder (VAS) into the initial conditions of the Penn State University/National Center for Atmospheric Research mesoscale model. The VAS data for this oceanic-cyclogenesis case consist of 110 irregularly distributed temperature and humidity soundings located over the North Pacific Ocean and apply at approximately 1200 UTC 10 November 1981.
The use of the VAS data produced relatively large changes to the National Meteorological Center's (NMC) analysis, which was the only source of meteorological data over the Pacific Ocean where the cyclone developed. Both static and dynamic initialization procedures were tested. When the model was statically initialized at an early stage of the cyclogenesis, the cyclone was only forecast well when VAS data were used to help define the initial conditions and when a reasonable distribution function was used for the latent heating associated with the parameterized precipitation. When the model was initialized 12 hours earlier with only large-scale data from the NMC analysis, a good cyclogenesis forecast was also produced. The use of dynamic-initialization and geostrophic correction procedures in order to provide mesoscale structure in the windfield based on the VAS-derived mass-field information, resulted in mixed success and proved to be unnecessary in this case.
In addition to showing that VAS data were a useful supplement, in this case at least, to the operational meteorological analysis over a data-sparse region of the ocean, these results illustrate two other points. First of all, data-impact studies with numerical models frequently assume that the veracity of the initial data is the factor most seriously impairing forecast quality. From an experimental design standpoint, this is convenient because it is not feasible to perform a complete predictability assessment for each case in order to determine other limitations imposed by the model/s numerics, physical parameterizations and boundary conditions. However, in this case we show that VAS data only had a positive impact when an apparently critical aspect of the precipitation parameterization was properly treated. A reasonable, but perhaps inconvenient, compromise is to perform a modest number of sensitivity tests on each case in order to identify any major “weak links” in the modeling system other than the initial data. Obviously the nature of these tests will depend on the meteorological setting.
Secondly, the need for mesoscale data in defining the model initial state for a forecast can depend on the development stage of the phenomenon (e.g., cyclone, mesoscale convective system) being forecast. In this case, VAS data were not needed in order to provide a good cyclogenesis forecast when the model was initialized in the precyclogenesis period with only a smooth analysis. This is because the model was able to provide the nonlinear interactions, response to surface fluxes, etc. that were necessary to define the precursor conditions for development. In contrast, initialization at a later stage in the development benefited from the additional information or structure provided by VAS because the model was not, in effect, used as a dynamic-initialization device. Thus, the impact of the VAS data was also dependent on the time of the initialization during the life cycle of the phenomenon.