INDENTIFYING PATTERNS OF SATTELITE IMAGERY USING AN ARTIFICIAL NEURAL NETWORK

Abstract

An artificial neural network analysis based on the self-organizing map (SOM)  was used  to  examine  patterns  of  satellite  imagery.  This  study  used  3  ×  4  SOM  array  to  extract patterns  of  satellite-observed  chlorophyll-a  (chl-a)  along  the  southern  coast  of  the  Lesser Sunda Islands from 1998 to 2006. The analyses indicated two characteristic spatial patterns, namely the northwest and the southeast monsoon patterns. The northwest monsoon pattern was characterized by a low  chl-a concentration. In contrast, the southeast monsoon pattern was  indicated  by  a  high  chl-a  distributed  along  the  southern  coast  of  the  Lesser  Sunda Islands.  Furthermore,  this  study  demonstrated  that  the  seasonal  variations  of  those  two patterns  were  related  to  the  variations  of  winds  and  sea  surface  temperature  (SST).  The winds  were  predominantly  southeasterly  (northwesterly)  during  southeast  (northwest) monsoon, drived  offshore (onshore) Ekman transport and  produced  upwelling (downwelling) along  the  southern  coasts  of  the  Lesser  Sunda  Islands.  Consequently,  upwelling  reduce dSST  and  helped  replenish  the  surface  water  nutrients,  thus  supporting  high  chl-a concentration. Finally, this study demonstrated that the SOM method was very useful for the identifications of patterns in various satellite imageries.