Assessing Thematic Map Using Sampling Technique

The estimation of accuracy of thematic maps derived from remotely sensed data used in conjunction with a geographic information system is presented. A systematic random sampling techique is proposed to asses the accuracy. The main objective of this work is the determination of the optimal sampling intensity and the appropriate sample size used to assess the accuracy. The data set used is a Landsat-TM scene from three different regions in Mazandaran and Gilan provinces of Iran. The regions are classified using 9 different approaches and subsequently used as thematic maps. The overall accuracy and kappa coefficient of these results are compared with a complete (100%) ground-truth from these regions that are used as a comparison reference. 87 sampling networks with intensities ranging from 0.5 6% and the systematic random sample plots, each measuring between 900–32’400 square meters are prepared in raster format. These networks are overlaid with the thematic maps and the reference ground-truth to obtain the final results. These results are then compared using matrices which contain information such as the overall accuracy and the kappa coefficient. The estimated sample accuracy is then compared with the real accuracy. The results of this research indicate that through sample ground-truth with 2% intensity in a systematically random technique and appropriate sample size, considering diversity and uniform phenomena in a region, we can estimate relatively accurate (± 4-6%) the accuracy of thematic maps, prior to use in GIS-database. PC Arc/Info and IDRISI are the main data analysis packages used for this work.