Urban mapping, accuracy, & image classification: A comparison of multiple approaches in Tsukuba City, Japan

The rapid growth of urban space and its environmental challenges require precise mapping techniques to represent complex earth surface features more accurately. In this study, we examined four mapping approaches (unsupervised, supervised, fuzzy supervised and GIS post-processing) using Advanced Land Observing Satellite images to predict urban land use and land cover of Tsukuba city in Japan. Intensive fieldwork was conducted to collect ground truth data. A random stratified sampling method was chosen to generate geographic reference data for each map to assess the accuracy. The accuracies of the maps were measured, producing error matrices and Kappa indices. The GIS post-processing approach proposed in this research improved the mapping results, showing the highest overall accuracy of 89.33% as compared to other approaches. The fuzzy supervised approach yielded a better accuracy (87.67%) than the supervised and unsupervised approaches. The fuzzy supervised approach effectively dealt with the heterogeneous surface features in residential areas. This paper presents the strengths of the mapping approaches and the potentials of the sensor for mapping urban areas, which may help urban planners monitor and interpret complex urban characteristics.