The initialization of a three-dimensional model with operational data for Hurricane Eloise (1975) was studied to assess the impact of using bogus storm data, surface winds, rainfall rates, and a high-resolution surface pressure analysis in the initialization of forecasts of hurricane track and intensity.
Because the track and intensity forecasts based on the unaugmented NMC analyses were unsatisfactory, various data improvement procedures were used. Boundary-layer flow was diagnosed from the surface pressure with a primitive equation PBL model, a climatological hurricane circulation was inserted into the NMC wind analysis above the boundary layer, and the three-dimensional moisture field was defined with the aid of visible-image satellite photographs. Model simulations with this improved data set were designed to test the effectiveness of dynamic initialization (DI) and the data enhancement procedures in improving the numerical hurricane forecasts. A 24 h time period, starting at 0000 GMT 21 September 1975, was considered. In procedure A, all data improvements were made and surface pressure was taken directly from a detailed analysis. Procedure B represented what might be done operationally—the only modification to the original NMC data was the insertion of a bogus storm based on composite data and the diagnosis of surface pressure from the 1000 mb heights and temperatures.
For each procedure, three model integrations were made to test the effect of DI by nudging on the forecast. Model results were evaluated in terms of track, the boundary-layer flow, surface pressure and rainfall rates. All forecasts with the improved data were much better than in the preliminary model experiments with the unmodified NMC analysis. Procedure B track predictions, which were based on initial conditions that contained the least amount of mesoscale information, were somewhat better than the others, with vector position errors of <80 km. Dynamic initialization had little effect on the path of the model storm. Intensity forecasts were best using procedure A, in which the greatest amount of hurricane scale information went into the initial conditions, and when DI was employed. However, large-scale mass-momentum adjustment and the proximity of the model storm to the lateral boundaries distorted the predictions of boundary-layer flow and rainfall rates.
A time composite of surface wind reports from land-based stations, buoys, and ships represented the type of data that might be available from future remote sensing satellites like Seasat-A. Because the data were valid at only one synoptic time, a DI could not be performed. The impact of the surface winds on the initialization could only be examined in terms of a 12 h forecast. Several methods of incorporating the surface wind observations into the initial conditions included direct insertion of the data into the NMC wind analysis and a diagnosis of surface pressure from the surface winds through a divergence equation. Although satellite winds improved the mesoscale realism of the initial boundary layer winds and the surface pressure, model forecasts were virtually unimproved. Forecast errors associated with the large-scale mass momentum adjustments, the limitations of the model physics, the data enhancement procedures, and the accuracy of the surface wind analysis, prevented our reaching any definite conclusion about the benefits of supplementary near-surface wind data.
A 12 h DI was performed in which the latent heat release due to convection was externally specified based upon satellite estimates of rainfall rate. A comparison of 12 h forecasts based on this DI and a static initialization showed that this type of DI produced forecasts of surface pressure and precipitation that were greatly improved and which were reflective of observed storm intensity. Track forecasts were not significantly changed.