Sensitivity of short range numerical weather prediction to data availability during NAME

Integrating vast amounts of data into a Numerical Weather Prediction model requires extensive computational resources and time. When using a NWP model for short range prediction purposes, time constraints and computation resources may be considered. Therefore data reduction techniques can be applied to fast and limited computing applications such as incident forecasting.

In this research we investigate the impact of reduced data integration into NWP models using the Fifth-Generation NCAR / Penn State Mesoscale Model (MM5). The North American Monsoon Experiment (NAME) during August 2004 provided an extensive data set that was used in conducting several MM5 simulations.

The set of MM5 simulations start with an optimal first guess field and observational data set. These optimal fields serve as a control for the other model runs. We then apply various degradation techniques to the first guess fields and observations to reduce their effective resolution and density. Determination of the impact of data sensitivity on NWP models was done using 2D and 3D analyses of these simulations.

In our simulation with degraded vertical resolution and observations, the use of one radiosonde within a domain versus none produces little difference in the placement, intensity and timing of convection. In addition, the degraded simulation predicted more convective precipitation than the operational forecast simulation during an MCS event. Differences also arise in the model forecasts of U and V wind components. Under the same conditions the simulation with vertically degraded initial conditions develops substantial differences in U and V in the upper layers of the model that persist throughout the simulation.