This study presents a detailed analysis of Iterative Self
Organizing Data Analysis (ISODATA) clustering for
multispectral data classification. ISODATA is an unsupervised
classification method which assumes that each class obeys a
multivariate normal distribution, hence requires the class means
and covariance matrices for each class. In this study, we use
ISODATA to classify a diverse tropical land covers recorded
from Landsat 5 TM satellite. The classification is carefully
examined using visual analysis, classification accuracy, band
correlation and decision boundary. The results show that
ISODATA is able to detect eight classes from the study area with
93% agreement with the reference map. The behavior of mean
and standard deviation of the classes in the decision space is
believed to be one of the main factors that enable ISODATA to
classify the land covers with relatively good accuracy.