A heuristic approach for precipitation data assimilation using the information contained in the ensemble forecast
听
Student Seminar Series
Department of Atmospheric & Oceanic Sciences
presents
a talk by
Andres Perez Hortal
PhD student
A heuristic approach for precipitation data assimilation using the information contained in the ensemble forecast
We present a simple data assimilation (DA) technique named "Localized Ensemble Mosaic Assimilation" (LEMA) for the assimilation of radar-derived precipitation observations. Our objective is to modify the trajectory of the model toward a new trajectory that is closer to observed reality for a long forecast time. The method constructs an analysis mosaic by assigning to each model grid point the information from the ensemble member that is locally closest to the precipitation observations. The new ensemble forecast is created by relaxing all the background members are relaxed towards the analysis mosaic.
Since LEMA relies only on the information contained in the ensemble forecast, under the ideal conditions were the spread of the ensemble forecast captures the actual forecast uncertainties, LEMA performs as desired. However, real DA experiments using StageIV precipitation observations show that when LEMA uses only the background members to construct the analysis, the quality of the precipitation forecast shows small or no improvements. To overcome this limitation, we expand the spread of the ensemble used to construct the analysis mosaic by considering states at different times and states from forecasts initialized at different times (lagged forecasts). The expansion of the ensemble spread provides a better representation of the actual forecast uncertainties, resulting in a better performance of LEMA.
Finally, in LEMA, all the background ensemble members are relaxed towards the single analysis, reducing the ensemble spread and produces an underdispersive forecast. This limits the time intervals between successive applications of LEMA to ~9h, at which time the ensemble spread in the forecast is restored. To shorten the cycling period, we add to LEMA another heuristic step: nudging the ensemble towards the analysis mosaic while maintaining the original background's spread. Using real DA experiments, we show that successive DA using LEMA improves the forecast quality, but it produces a slight decrease in the ensemble spread.
Wednesday Jan 22/ 2.30 PM/ Room 934 Burnside Hall
听
听
听