The classification of fine spatial resolution imagery: parcel-based approaches using HRSC-A data

Large amounts of remotely sensed data are now being collected from airborne platforms carrying digital scanners and digital cameras with spatial resolutions from 2 m down to 15 cm. Spaceborne instruments are now being launched with spatial resolutions between 1 and 4 m. These systems are used to address mapping issues in locating and identifying objects or areas on the surface. The fine spatial resolution of the data now becoming available would at first appear to be a major advantage for mapping applications compared to conventional satellite systems. However, it must be remembered that similar reasoning proceeded the launch of Landsat 4 and the Systeme Probatoire de l´Observation de la Terre (SPOT-I) in the 1980s. Work comparing TIM and HRV with the established 80 m spatial resolution Landsat Multispectral Scanner System (MSS) found that finer spatial resolutions actually reduced classification accuracy for certain land cover types. The coarse spatial resolution of the MSS smoothed out spatial complexity within heterogeneous land cover types, such as urban, as scene components, such as buildings and vegetation, become lost within a pixel. At finer spatial resolutions a scale boundary is crossed where the data recorded for each pixel is related not to the character of object or area as a whole, but to components of it and this requires a re-definition of the information that can be extracted. The current move toward even finer spatial resolution data sets should raise the same question how these types of data should be analysed using (semi-)automated techniques. This paper describes a methodology for classifying fine spatial resolution data to land cover types. The problem faced was two fold; firstly, how to extract meaningful information from the pixels within the fine spatial resolution images and secondly, how to integrate this detailed information at the pixel level to useable classes and appropriate scales