Multiscale Assimilation of Radar Reflectivity

At mesoscale, radar provides dense observation of precipitation events, and yet assimilating radar data has limited impact on the forecast quality with traditional Ensemble based Kalman Filter (EnKF). In EnKF, the localization distance controls the spread of information; this distance is selected by considering among others the ensemble size and observation density. Studies have shown that as the observation density increases the localization distance should be decreased. This constrains the spread of information to farther regions. This is important because in most regions with no echo, backgrounds have no echo either, and radar data assimilation yields no increment. To obtain a better forecast, the information from the regions with echoes should spread to regions where there is no echo as these will later influence the storm's evolution.

Here we present a method to correct the fields inside and well-outside storms by using multi-scale assimilation with radar reflectivity using Ensemble Adjustment Kalman Filter (EAKF) via DART-WRF in the context of OSSE experiments. Multi-scale assimilation is achieved by smoothing observations and background at up to 3 scales, performing an assimilation at each scale with different localization distances, and adding the increments from each of these assimilation steps. Smoothing the observations and background removes small scale errors and helps extract large-scale information with larger localization distance. The resulting analysis has both small-scale and larger-scale errors fixed.

An assimilation experiment was done at convective scales (1-km resolution with 5-min assimilation cycles) comparing traditional and multi-scale assimilation. Results show that the multi-scale assimilation improves forecasts especially at longer lead times when done once. Best results are obtained when multi-scale assimilation is not performed at every assimilation cycle.