Investigation of the capability of IRS-P6-LISS IV data for density mapping of pistachio forests (Case study: Khaje kalat forest in Khorasan)

The aim of this study was to investigate the capability of IRS-P6-LISS IV data for forest density mapping in the Pistachio forests. So LISS IV images (5 m), dated 2007 from Khaje Kalat (500ha) in Khorasan Razavi were analyzed. The geometric correction of images was implemented using nine control points extracted from an orthorectified image of the study area. The RMSE was less than one pixel. In addition to original bands, different synthetic bands from principal component analysis and transformation methods were created and used. The suitable bands set were selected by training areas and divergence indices. In order to assess the accuracy of classification results, a ground truth map covering 7% of the total area was prepared through fieldwork using 34 sample areas and canopy percent was estimated.
Satellite data were classified by supervised classification methods including minimum distance to mean (MD), maximum likelihood (ML) and fuzzy. There were spectral interferences between medium density classes (5-10%, 10-15% and 15-20%). Therefore these classes were merged. In hard supervised classification method, the highest overall accuracy and kappa coefficient, 67% and 0.40, respectively, were obtained by ML classifier with three classes (0-5%, 5-20% and > 20%). Using mode filter with a 7×7 pixel increased the accuracy up to 3%. The results of Fuzzy algorithm showed higher accuracy and kappa coefficient, 70% and 0.44, respectively. In both methods, second density class (5-20%) represented highest kappa coefficient. It could be concluded that the result of classification was not desirable regarding to low kappa, even if reaching to pretty good overall accuracy. To obtain a better result, it is suggested to use higher spectral resolution data and preparing fieldwork in smaller sample area and determining canopy percentage quantitatively.