عنوان مقاله [English]
The present study aimed to evaluate the capability of IRS-P6-LISS III imagery to map a forest canopy density by employing the traditional hard classification and the Artificial Neural Networks (ANNs) in Marivan city, Kurdistan province. Geometric correction procedure was performed with less than 1 pixel root mean square error (RMSE). Various vegetation indices and artificially bands generated by principal component analysis (PCA) were used in the classification procedure. A ground truth map was produced based on a randomized-systematic method with a grid size of 250×400 meters and 50×50 meters sample size strata. The suitable band combinations for classification were selected through the training area using the Transformed Divergence index. Supervised classification methods i.e., parallelepiped, minimum distance to mean, maximum likelihood, and ANNs algorithms were applied to generate the canopy density map with 4 classes (very sparse, sparse, semi-dense and dense). The accuracy assessment of the generated canopy density maps was implemented using the ground truth map. Some classes were also merged because of the low spectral separation between these classes. Finally the classification was performed to produce the canopy density map with 3 classes (sparse, semi-dense and dense). The highest overall accuracy and the Kappa coefficient were achieved by maximum likelihood method with 78.47 percent and 0.66, respectively. Our results indicated the high capability of the IRS-P6 LISS III imagery compared to other satellite images, for example, Landsat and Aster data, which already tested in the previous work to map the canopy density in Zagros forests.