Multiple-Clustering Comparison in Spatial Data for Additional Accuracy Measurement in Land-use Classification

Satellite imageries have been widely used to analyze a region by planners. Data from the satellite usually have lower accuracy than other expensive methods e.g. drone, aerial view, etc. However, the data from satellite have wide range area and sufficient enough for modeling a land use. The accuracy assessment, therefore, becomes a vital task to ensure the model from the satellite imagery meets the minimum requirement. Validating the classification result by comparing to the real location or by other higher resolution images is needed. The paper proposed additional validation by comparing the classification result by another result in different date through the cross-tabulation method. Two satellite imageries in the same year were processed before classification to get the land-use and land-cover classification. Comparing two land-use classified images gave the accuracy statistics using cross-tabulation. The kappa statistic and accuracy showed the classification performance of 0.7592 that similar to the sampling-based accuracy assessment (0.75390). Therefore, the proposed method was appropriate as the accuracy assessment.