Forest Type Classification Improvement Using Spatial Predictive Distribution Models

The last experiences showed that spectral data have not sufficient to classify forest types in the mountainous area. In order to clear abilities of spatial models to classify forest types and improve results, an investigation was planned in a case study in the northern forests of Iran by ETM+ data. The spatial models based on aspect, elevation, incorporated aspect-height and homogenous units constructed for each type individually. Probability occurrence rates of types were extracted in the each class. Classification was accomplished with the best spectral data sets by maximum likelihood classifier using only spectral data and with spatial models separately. The accuracy of results was assessed with a sample ground truth map. The results showed that spatial models could improve considerably results in compare with only spectral data (14%). This study exposed that spatial models based on the homogenous units in compare to other models could better improve classification.