Mapping National Park landscape from ground, air and space

Abstract

Aerial photograph interpretation and field mapping were used in a series of experiments to evaluate the use of Landsat and Système Probatoire pour l'Observation de la Terre (SPOT) satellite imagery for landscape mapping. The 'Monitoring Landscape Change in the National Parks' (MLCNP) Project mapped landscape in each of the National Parks of England and Wales in terms of 38 land cover classes with significant visual impact. The main source of data was aerial photography but satellite imagery for selected areas was also analysed. It was found that single-date multi-spectral imagery could be classified to an acceptable level of agreement with ground data only if the 38 sub-classes of the interpretation scheme were grouped into the seven main class headings. Visual interpretation of SPOT panchromatic imagery at the 38 sub-class level proved comparable with aerial photograph interpretation for an area of the North York Moors. This paper describes the approaches taken in data analysis and presents the main results obtained. The use of confusion matrices allowed measurements of agreement to be made between the three sources of data. A significant problem in mapping landscape was to arrive at unambiguous class definitions when many of the categories had no clear boundaries on the ground. Confusion matrix analysis, together with the use of a hierarchical classification scheme, allowed links to be made between data collected from ground, air and space. Some classification problems were attributable to all sources of data due to inherent difficulties with the classification system.