Estimating the right number of clusters in remote sensing segmentation.

HI, I'm working on Remote sensing images segmentation. the images is the SST (Sea surface temperature) of north west of Morocco. My goal is to segmente the SST images in 2 regions : Region with cold temperature (correspond to Upwelling) and region with warmer temperature (No upwelling). to achieve this i'm using *Otsu* method for labeling the sst images into different clusters, and the cluster with lower mean temperature correspond to upwelling region. My big problem is to estimate the right number of cluster to achieve the *Otsu* segmentation . I also tried to calculate the validity indices : Silhouette, Dunn, ... but our images are so big to accomplish this task.