KLIWAS

Seite  15

shown  by  Castelao  (2006)  and  by  Belkin  and  O’Reilly  (2009).  However,  gradient
algorithms  are  not  able  to  recognise  and  discard  false  fronts  caused  by  noise.  On  the
other  hand,  histogram  algorithms  are  able  to  detect  weak  fronts  in  the  presence  of
high  background  noise  as  shown  by  Cayula  and  Comillon  (1992,  1995,  1996),  Diehl
et  al.  (2002)  and,  Ullman  and  Cornillon  (1999,  2000,  2001).  In  the  present  study,  a
new  approach  was  developed  which  combines  and  modifies  the  gradient  algorithm  of
Canny  (1986)  and  the  histogram  algorithm  of  Cayula  and  Cornillon  (1992)  in  order  to
improve  the  front  detection  ability.  It  was  also  found  that  a  proper  pre-processing  of
the  data  is  important  in  order  to  have  a  good  detection  quality  for  fronts.  The  preprocessing ­
  includes  mainly  cloud  filtering  and  noise  reduction.  We  call  this  new
approach  GRADHIST.  The  flowchart  in  Fig.  2  outlines  the  overall  processing  chain
of  GRADHIST  which  includes  three  steps:  pre-processing,  the  application  of  the
modified  gradient  and  histogram  algorithms  and  the  matching  module.

“  7-'
Input  data

Fig.  2:  Flowchart  of  front  detection  algorithm

KLIWAS
Climatology
of  North  Sea
Fronts