Comparison of four methods classification for Land use and Land cover mapping using RS imagery

One of the most essential information that is necessary for natural resources managers is landcover and landuse maps.providing this maps using traditional methods and aerial photo interpertion needs a lot of time and costs.satellite data because of wide and unique view,inculding most parts of electeromagnetic spectrum and update of images are suitable.this study was done for comparision 4 method of cassification and production of landcover and landuse by landsat satellite data (ETM+ sensor) and IRS (PAN and Liss III sensor) for the most of method in producing maps for arid region in Dasht-e-Qom with 183,853 hectares area.satellite data include ETM+of august 2002 and IRS-PAN data of july 2003 and IRS-Liss data of august 2003.Image were registered to digital topographic maps with scale of 1:25000 with using 35 ground control points and non-parametrical method. Image were enhaced using contrast enhancement, making false color composite images (FCC) and vegetation indices.for using panchoromatic band,multi spectral band were registered to this band and data fusion were done with septral response.for classification of image,were used vegetation indices method,one,s band supervised (density slicing),multispectral supervised(with using of maximum likelihood,minimum distance and parallelepiped classifieres) and hybrid.due to estimating of classification, a ground truth map allocating 6% of the total area was prepared by field work using random sampling and GPS. The results showed that using the regetation indices method (NDVI, SAVI, PVI), the highest overall accuracy related to PVI indice of ETM+ sensor(with 73.04% overall accuracy and 59% Kappa coefficient for landcover map and 40.8% overall accuracy and 29% Kappa coefficient for landuse map). using vegetation indices can be use for help data (specialy for agriculture, forest and rangeland classes),but that can not be one of the suitable method for classification of images. In a band supervised classification with density slicing method for landcover map,the highest overall accuracy is 62.26% and 0.47 for kappa the 2nd band of Liss III data (IRS satellite) coefficient.the highest overall accuracy for landuse is a band of ETM+ with 17.41% overall accuracy and 0.004 kappa coefficient.In a band supervised classification con not use for a perfect and indepfndent method of images classification becouse that is mixes spectral phenomena and is limited to range of classes. In the multispectral supervised classification for landcover map,the highest overall accuracy related to ETM+ data and maximum likelihood with 83.74% overall accuracy and 0. 74 kappa coefficient.Of course, accuracy results of maximum likelihood in Liss III data and fusion of Pan and Liss III data is nearest to accuracy results of ETM+ data.In landuse map, the highest overall accuracy related to ETM+ data and maximum likelihood with 71.95% overall accuracy. In the hybrid method, couldnot presented satisfactory results for classification of landcover map ( 33.14% overall accuracy for Liss III data and 28.2% for ETM+ data ) and that cannot be suitable for classification of little class ( for exampel: landcover map of Dasht-e-Qom with 3 class). The accuracy results of hybrid method for classification of maps with few class such as landuse map of Dasht-e-Qom with 7 class, presented satisfactory results. Overall accuracy this method with 76.42% is the highest accuracy and the best of 4 method classification for providing of landuse map. Finally, recommend that on the aridy region, for classification of landcover map use of the multispectral supervised classification with maximum likelihood algoritm and hybrid method for landuse map.