Abstract
We developed a method of discriminating land use using multi-temporal Landsat data for the area located in the humid climate region, where probability of cloud cover was very high and also complicated cropping pattern in agricultural field was presented. Due to the limitation of available cloudless scenes, we attempted to characterize seasonal variation of ground surface condition, which could be associated with land use type, by calculating the annual maximum of five indices representing surface condition of bareness of land, vegetation or water for all the data taken over several years. This procedure successfully removed the effect of cloud for the whole study site located in the western part of Java Island, Indonesia. Classification was carried out for the multi-dimensional dataset containing the maximum values of indices by ISODATA method and combination of classes. Results showed 60% of overall agreement and 79% of agreement were found if confusing classes were merged as “Upland” to “Mix Vegetation” and “Bare Land” to “Manmade” with interpreted classes from QuickBird imagery. This accuracy exhibited significant improvement compared to the case of classification using mono-temporal Landsat data. We suggested this method to be one of the most promising tool to produce pixel based land use map especially discriminating paddy and the other agricultural land use over the tropical humid climate region.
