Model initializations are frequently assessed in terms of noise statistics or long-range forecast skill and predictability limits. However, meant motion or the utility of very short-range forecasts or nowcasts has stimulated interest in the possibility of using mesoscale models for their production. Thus, very short-range forecast skill has become an important criterion for evaluating the adequacy of model initializations. For example, previously acceptable forecast products such as total-event precipitation amounts do not provide the “nowcamer” with sufficiently detailed guidance. Models must now be able to predict hourly rain amount. It is the relationship between the quality of the very short-range forecasts of hourly rainfall and the specification of the initial divergence field that is the focus of this study.
The mesoscale initialization discussed by Tarbell et. al., in which horizontal divergence is diagnosed from a diabatic omega equation, was tested on a heavy rainfall case. The procedure for diagnosing the divergent-wind component included effects of latent heating obtained from the observed rain rates. In the real-data tests, three forecast periods were used during the SESAME III (1979) study period. Six-hour rainfall predictions initialized with the diagnosed divergence were compared to observed precipitation and to rainfall forecasts based on initial conditions containing, no divergence and observed divergence obtained from the standard rawinsonde winds. The utilization of the mesoscale rainfall information in diagnosing the initial divergent component was found to be important in correctly predicting hourly rainfall patterns, especially for the first few hours. The use of the divergence field obtained from rawinsonde data was only marginally better the use of nondivergent initial conditions.
A data-simulation procedure was also used to test this initialization technique. Model-generated data, which were in dynamic balance, represented an internally consistent high-resolution dataset that was used to solve the omega equation and to define the rain rates used both for verification and as input to the diabatic term of the omega equation. This experimental setting tested the ability of the diagnosed-divergence initialization to improve the very short-range precipitation forecast under idealized conditions—when balanced, high-resolution man/momentum data and grid-box average precipitation data are available. Results from these experiments were consistent with those that used real data. Only the diagnosed-divergence initialization produced reasonable rain rates during the first 4–6 hours, and it did so only when the observed rain rates at the initial time were used to define the diabatic term in the omega equation.