Real Time Spatialized Bias Correction of the French operational 5'-1km² Quantitative Precipitation Estimation Products using past radar and rain gauge hourly accumulations

The Weather Radar Centre of Météo France is responsible for operating the French weather radar network and producing in real time the best 5'-1km² national Quantitative Precipitation Estimation (QPE). Every 5 minutes, all PPIs of all operational radars (26 in the metropolitan France area) are processed by a complex, multi-module QPE treatment chain (including ground-clutter identification, partial beam blocking correction, vertical profile of reflectivity correction, synchronization using the advection field, correction for precipitation-induced attenuation, generation of a QPE quality map, …) and mosaicked (using quality indice for each pixel). The requirement is to disseminate the national QPE mosaic before 2 minutes after each 5 minutes cycle of scanning.

Since 2006 the QPE processing includes an adjustment of the real-time 5' single-radar QPE using past hourly radar and rain gauge precipitation estimations. This adjustment is realized independently for each radar, and uniform for all the radar coverage area. The adjustment factor (the mean field bias) is estimated using a filtering technique from the measurements made by all rain gauges (and their radar counterparts) over the past hours (up to 40h) and located up to 80 km from the radar. In the filtering technique, more recent and larger hourly depth of rainfall have a greater weight in the adjustment factor estimation. The main assumptions of the method are 1) the persistence of the calibration factor from one hour to the next one, and 2) the uniformity of the correction across the entire radar coverage area. Several parameters have been defined in order to guarantee the respect of these hypotheses, limiting variations in time of the calibration factors, and forcing its value to a mean monthly fallback value when the rain is too weak.

The poster presents the results of a new study carried out recently to upgrade the method towards a spatialization of the adjustment factors. The new method estimates an adjustment factor value for each pixel, and relies on an iterative three-step downscaling of the adjustment procedure centered on each pixel (using local areas of 128x128 km², 64x64 km² and 32x32 km²). This new method still includes a temporal filtering as well as a forcing to a fall-back value when the estimation is not possible. The length of the temporal filter decreases with the size of the local area considered, reflecting the correlation between the spatial and temporal variability of the adjustment factor. Several variants of the new method have been tested and evaluated in two steps: a statistical analysis using a complete set of 2 years of data from 24 radars and more than 1100 rain gauges ; a case study for 10 rain events representative of different situations. The method and the best set of parameters, as well as some validation results, are presented and explained in the poster. This new method will be used in the next version of the operational adjustment of the French national radar QPE.