Precise LULC classification of rural area combining elevational and reflectance characteristics using UAV

Abstract

With the development of unmanned aerial vehicle (UAV) in the recent decade, very high-resolution aerial imagery has been used for precise land use/land cover classification (LULC). However, special structures in rural areas of developing countries such as traditional thatched houses have posed challenges for precise LULC classification due to their undistinctive appearance and confusable characteristics in both reflectance and structure. LULC mapping is essential particularly in rural areas which have high data scarcity and vulnerability to natural disasters. With high-resolution observation has been achieved by UAVs, it is important to propose high-precision LULC classification methods which can fully use the advantages of UAVs. To emphasize the differences among the common LULC types in rural areas, this study proposed an original index, the rural residence classification index (RCI). RCI was calculated as the product of the above ground height and the square of the difference between the NDVI value and one. Then, a comprehensive classification method was established by combining the RCI, the traditional threshold method and a machine learning method. As a result of the comparison with the traditional threshold method, object-based image analysis, and random forest methods, the method by this study achieved the highest overall accuracy (overall accuracy = 0.903, kappa = 0.875) and classification accuracy for detecting thatched houses (user's accuracy = 0.802, producer's accuracy = 0.920). These findings showed the possibility on identifying the confusable structures in rural areas using remote sensing data, which was found difficult by the previous studies so far. The method by this study can promote the further utility of UAVs in LULC classification in rural areas in developing countries, thereby providing precise and reliable material for hydrological, hydraulic or ecosystem modelling, which eventually contributes to more accurate natural hazard risk assessment, rural development, and natural resource management.