Using Post-Classification Enhancement in Improving the Classification of Land Use/Cover of Arid Region (A Case Study in Pishkouh Watershed, Center of Iran)

Abstract

Classifying remote sensing imageries to obtain reliable and accurate Land<br />Use/Cover (LUC) information still remains a challenge that depends on many factors such<br />as complexity of landscape especially in arid region. The aim of this paper is to extract<br />reliable LUC information from Land sat imageries of the Pishkouh watershed of central<br />arid region, Iran. The classical Maximum Likelihood Classifier (MLC) was first applied to<br />classify Land sat image of 15 July 2007. The major LUC identified were shrubland<br />(rangeland), agricultural land, orchard, river, settlement. Applying Post-Classification<br />Correction (PCC) using ancillary data and knowledge-based logic rules the overall<br />classification accuracy was improved from about 72% to 91% for LUC map. The improved<br />overall Kappa statistics due to PCC were 0.88. The PCC maps, assessed by accuracy<br />matrix, were found to have much higher accuracy in comparison to their counterpart MLC<br />maps. The overall improvement in classification accuracy of the LUC maps is significant<br />in terms of their potential use for land change modeling of the region.