| dc.description.abstract | Thunderstorms can lead to flash floods, and can be accompanied by large hail, severe wind gusts and tornadoes,
possibly having a high social impact. Accurate and detailed warnings for these storms are therefore
desired. Because operational meteorologists, which monitor the weather and generate forecasts, are usually
busy in situations with thunderstorms, they will benefit from algorithms that can automatically generate
storm forecasts. These algorithms could determine the intensity of storms and their velocity, enabling it
to produce a forecast. Such a forecast could provide a quick overview of storm intensities and velocities
to operational meteorologists, or if accurate enough, might even be used to warn the public. The focus
of this project is the creation and evaluation of a storm-tracking algorithm, which automatically produces
storm motion estimates. Such a storm-tracking algorithm usually receives input from weather radars, which
provide information about the precipitation intensity. The coupling (association) of precipitation patterns
at consecutive times enables these algorithms to determine the motion. Several of these storm-tracking
algorithms have been created in the past, but most of them suffer from storm interactions, e.g. the splitting
of one storm into two, or the merging of two storms into one. This is because these algorithms perform
one-to-one association of storms at consecutive times, which is not possible in the case of storm splits and
mergers. Separate treatment of these processes is therefore required in these algorithms. The proposed new
algorithm, named CBVELOCITY, uses a method that automatically treats storm interactions, in the same
way as the usual translation of storms is treated. The key concept is the division of storm cells into subcells,
which are subsequently associated between consecutive times, instead of the storm cells themselves. The
velocity of the storm cells is then determined as the average of the velocities of the subcells in the storm.
The subcell concept is based on the approach used in the storm-tracking algorithm of Matthews and Trostel
(2010), but is optimized for determining velocity estimates instead of following thunderstorms through time.
An evaluation of the performance of CBVELOCITY shows promising results compared to SCIT, the only
existing algorithm to which it has been compared. It is further shown that the performance depends on the
storm speed, with decreasing performance for decreasing storm speed. | |
