Countryside survey from ground and space: different perspectives, complementary results

There is a strong demand for maps and data on the state of the countryside to support regional and national studies. 'Countryside Survey 1990' is a national survey of the state of the rural environment in Britain. If combines detailed field surveys of a stratified random sample of 508 1-km squares with a generalized census, based on satellite remote sensing using Landsat Thematic Mapper, to compile a unique record of the ecology, land cover and land use in Britain. Mapping any landscape into discrete classes, though routinely practised, produces results that are both artificial and simplistic. It follows that the collection of so-called 'ground truth' data, as the single standard by which surveys can be judged, is in itself an elusive aim. This paper therefore compares and contrasts the methods and results of the two surveys, attempts to explain differences and examines the consequences for use of the data. There were clear differences between the two surveys in definitions of similarly named class es reflecting, particularly, different perceptions of land cover and land use. Landscape patterns, map scales and survey resolutions had complex inter-relationships, such that the 1:10 000 field cartography did not readily compare with the 25 m raster format of the satellite map. Spatial generalization, an artificial but inherent characteristic of conventional cartography, affected the field survey. The satellite imagery imposed an equally artificial 25 m grid on outputs. Geometric displacements were confounding factors, wrongly suggesting thematic errors. Other differences emphasized the difficulties of subdividing a continuously variable landscape. Spectral mis-classification caused most of the errors in the Land Cover Map, often due to enforced compromises over the date of image acquisition. However; field surveyors also had difficulties in classifying landscape features consistently Although evidence suggests that it is impossible to determine absolute 'accuracy: if has been possible to apportion errors to the surveys. Results suggest that the Land Cover Map is perhaps 79-84% 'accurate' with the field records about 90% 'correct'. National statistics derived from the map are based on a census, albeit incorporating these inaccuracies. Although individual field-based observations may be intrinsically more accurate, national predictions are based on statistical extrapolation that introduces errors dependent on the sample variances. Nevertheless, the accuracies achieved from both approaches are sufficient for the maps and data to have been used very widely for taking stock of environmental resources, measuring change, understanding environmental processes and predicting and managing environmental impacts