Relying solely on the VH agricultural damage data, the evaluation is
performed for the whole of Germany. In the case of the BI data, which
are available only for the States of Baden-Württemberg, Hesse, and Thuringia
(see Fig. 4a) in sufficient density, the evaluation is restricted to
these areas. As the two insurance data sets are available for 5-digit postal
code zones only, we considered a larger area of 5×5 km 2 around the
radar grid points where hail damage was taken into account. This extension
is also necessary as falling hailstones may drift several kilometers
with the horizontal wind (Schuster et al., 2006). In order to ensure suf-
ficient sizes of the samples, we did not distinguish further among regions,
years, or seasonality. In the case of the BI data, we excluded
unsettled areas in the evaluation process by additionally implementing
CORINE (Coordination of Information on the Environment) Land Cover
2000 (CLC2000) data. Using this data set, we differentiated only between
settled (urban fabric and commercial units, CLC-codes 111, 112,
and 121) and unsettled areas (all other CLC-codes). Neither the settlement
density nor temporal changes were considered. As shown by
Kunz and Kugel (2015), considering CLC2000 in the evaluation substantially
increases the skill scores obtained. To eliminate errors in the building
insurance data (see Section 2.3), a hail damage day is classified only
Fig. 2. Example with maximum ΔH (a) without advection correction, (b) tracks of convective cells obtained from TRACE3D, (c) vector field with duplicated stormtracks, and (d) maximum
ΔH after the advection correction on 27 May 2005.
M. Puskeiler et al. / Atmospheric Research 178–179 (2016) 459–470 463
if a damage frequency of 0.1% is obtained or exceeded within a postal
code zone. For the agricultural insurance data, such a threshold cannot
be considered since only information about damage (yes or no) is
available.