Short term forecasting of snowbands using Doppler radar observations and a cloud-scale model

The accuracy of extrapolation technique for forecasting snowfall decreases rapidly after 30 minutes, due to the effect of storm evolution. In order to extend the predictability of phenomena such as snowbands and freezing rain events, it is necessary to use a numerical model that simulates storm evolution. In this study we use a cloud-scale model to forecast a winter storm which produced heavy snow in the New York City area on December 10, 1997. Well-defined snowband structures were the dominating feature of the storm.

To initialize the model, Doppler radar observations are used. Radar reflectivity and radial velocity are assimilated into the model using the adjoint technique. The assimilation system, which includes a cloud-scale model and its adjoints, determines the 3D wind, thermodynamical, and microphysical fields of the storm by minimizing the difference between radar observed variables and their model predictions.

Preliminary results show that using two volume scans of radar observations separated by 6 minutes, the retrieval is able to recover the band structure of the storm. The wind fields are reasonable and fit relatively well to the observed radial velocity. One hour forecasts of this event have been performed using the retrieved fields as initial conditions. Results show that the model simulates the motion of the snowband reasonably well. Work is currently being done to improve the microphysical representation in the model to allow a better quantitative snowfall forecast. We will also use the output from a mesoscale model to provide boundary conditions for the cloud-scale model so that the forecast can be extended beyond one hour.