Extended Abstract

method. One can see two bulls eye rainfall maxima at two individual gauge stations,with more than 50 mm of precipitation during the given 24-hour period. Nothing of themis seen on the radar-derived precipitation field in the center panel of the figure. Theright-hand-side LAPS generated QPE has much of the same elongated rain bandstructure as the radar based distribution. However, the LAPS analysis seems to havecaptured some of the features of the northern rainfall maxima recorded by the raingauge at that site. The QPFs produced for this summer case by the three models,ECMWF, HIRLAM_RCR and MB71, presented in Figure 2b, show marked differences.As expected, the coarser models (ECMWF, RCR) have less fine-scale details than theMB71 version of HIRLAM.From the SAL statistics in this case (not shown) it is difficult to make too strongconclusions. Each of the models was verified against gauge, radar and LAPS analyses,individually, i.e. we had nine pairs of comparisons. The best S (structure) scores(i.e. smallest values) were assigned to MB71, with S values of -0.27, 0.43 and 0.40 forthe gauge/ radar/ LAPS reference, respectively. The highest S values were reported forECMWF.The amplitude (A) and location (L) components are more of a mixed bag and there is noclear signal of which model would perform best in this summer case. For the amplitude,RCR would score best and MB71 worst when using gauge comparison. Whencomparing against radar or LAPS the differences in A are very small and the absolutevalues also small, indicating good performance. The L values are generally very smallfor all models and when using whatever comparison. This is a feature we have noticedalso in our earlier studies and this SAL behavior requires more thorough investigation.The most typical way to present SAL statistics is to construct a two-dimensional, x-y plotof S (x-axis) vs. A (y-axis), where each individual plot marker indicates one S-L valuepair for each 24-hour precipitation event during a given time period. Figures 3a and 3bshow such a scatter plot for the summer period of 2010 by using radar and LAPS as theverifying “truth”, respectively. It should be noted that we had for the very first time thepossibility to utilize LAPS based QPE in verification during the summer of 2010.One can immediately see the generally very high S values. Using LAPS as the verifyingQPE produces somewhat smaller values than the radar-derived QPE. It is satisfactoryto note that the, relatively speaking, best scores (i.e. lowest S values) are gained by thefinest scale MB71 model, with average values of 0.99 and 1.29 for LAPS and radarcomparison, respectively. The large S values are probably the result of not up-scalingour QPEs to the model resolutions, i.e. the results at least partly reflect implicationsfrom comparing QPFs and QPEs of differing resolutions.The best amplitude (A) scores during summer 2010 (Figures 3a and 3b) were achieved,on average, by the meteorologists’ grid-edited forecasts, with no bias (i.e. A ~ 0) visible.-440-