Applying Procrustes shape analysis verification scheme to nowcast ensemble forecast members to determine accuracy of diagnosed convective mode

Abstract

Simple nowcast ensembles were generated by varying the reflectivity band selection in the K-Means cluster advection scheme in WDSS. This has the effect of forecasting the motion of storm cells based on the prior motion of different size clusters defined by the reflectivity threshold. As there is a correlation between cell size and reflectivity this is akin to performing an object-based spatial cascade. When performed there are clear differences between the motion and shape of the forecast storms. To objectively assess the impact of changing threshold reflectivities in the nowcast a Procrustes shape analysis verification scheme was used. This produces penalty functions based on errors in translation, rotation, dilation and shape of a forecast object and, as these penalties can be assessed separately, the mode of the error in the forecast can be determined. In some cases it was found that a significant error was observed in the shape of the forecast storm, and this could be interpreted as indicating a misdiagnosis of storm type which could lead a forecaster to infer a different mode of convection to that which is actually occurring. This, in turn, could produce an inappropriate warning of severe weather threat.