Saemangeum is reclaimed land created by constructing a sea dike across an estuarine tidal flat on the coast of the Yellow Sea in South Korea (completed on April, 2010). The land is specified as highly saline and of spatial heterogeneity in vegetation distribution mainly by soil conditions. There are 3 typical vegetation type; halophyte vegetation located in high saline area, grass vegetation in mid or low saline area, and dense reed/small-reed vegetation. Although it is important to figure out the distribution of vegetation types, it is not easy even in high resolution satellite images because areas are dominated by grass/herbaceous plants and show similar spectral response. We compared three methods to map vegetation communities within a study area (ca. 350ha); supervised classification using maximum likelihood, indicator kriging and regression-kriging with RapidEye image. The last technique showed better predictive capability. Geostatistics, which formulates spatial structure of target quantities, can help improving accuracy of prediction and even more with auxiliary information combined. The framework of combining geostatistics and satellite image information will be presented.