Probability of Cb and Tcu occurrence based upon radar and satellite observations

The detection of Cumulonimbi (Cb) and towering cumuli (Tcu) is relevant for aviation as
they are associated with hazardous flight conditions. Their detection is therefore a
requirement by ICAO. Since 1-8-2007 an operational algorithm, referred to as algorithm-2007, is used at the airports EHBK and EHGG to detect Cb and Tcu. It uses the radar reflection observations and lightning observations as input. The performance of the
algorithm-2007 is poor in terms of probability of detection (POD) and false alarm ratio
(FAR). At KNMI this study was initiated to develop an improved algorithm.
An automated Cb-Tcu detection algorithm based on the synergy between radar and
satellite observations is developed. The algorithm uses logistic regression to determine the
probability of Cb-Tcu occurrence. Within logistic regression a forward stepwise approach is
applied. The predictors selected by the forward stepwise regression method are related to
the highest radar contour occurring in the 15 and 30 km radii collocation area, and to the
satellite observations, reflection range of the high resolution visible channel, the cloud
temperature and its standard deviation. The latter three all in the 15 km radius collocation
area.
The obtained results show in general an improvement in performance of the developed
algorithm in comparison to the algorithm2007 results. 
The performance of the developed algorithm is dependent on season and day-night
conditions. The best performance is achieved in the Summer day category followed by the
winter day category, with the summer defined from April till October. Surprisingly the
summer night category shows the worst performance. 
Although the algorithm is developed for EHBK and EHGG no year round evaluation of the
performance of the newly developed algorithm was possible for those airports because of
the lack of sufficient Cb occurrences, which are required for a statistical analysis.
Especially for the EHBK airport data was lacking. This hampers a successful operational
application of the developed algorithm for EHBK.
Note that since there is no other observation which covers both the required spatial and
time dimensions a future assessment of the performance of the algorithm is disabled. The
METARs are the most reliable source of Cb and Tcu observation, but they are terminated
at EHGG and EHBK. At EHAM and EHRD they are still continued.
Based on the results an improved operational algorithm can be defined. The probability
threshold selection will determine the performance of the developed algorithm. For the
daytime categories a POD of 65 % and a FAR of 35 % appears feasible in the summer
and winter day categories. For the night time a POD of 55 % and a FAR of 45 % appears
achievable.
During the study it became clear that in the algorithm-2007 the evaluation area with a
radius of 30 km on June 30, 2008 is decreased to an area with a radius of 15 km. The
decrease in area leads to a lower FAR, but also to a more significant loss in POD when
compared to the METAR. It is recommendable to evaluate the effect on the results of this
radius change for the algorithm-2007 with a data set covering an area with a radius of 15
km.