2.2 Aerial photographs provide a record of land cover for the late 1940s when most of Scotland was flown by the Royal Air Force. A similar series of air photo surveys was flown by the Ordnance Survey (OS) in the early 1970s. Photography for the late 1980s also became available from an aerial survey commissioned by the Scottish Office.
2.3 The statistical methodology was based on a stratified random sample within each administrative region, as mapped in 1984. Stratification within each region was performed by means of a classification of Landsat multi-spectral scanner imagery. Typically this produced a lowland, upland and intermediate class, with an urban class where appropriate.
2.4 A 10% sample by area within each regional stratum was obtained by randomly selecting 5km-square sampling units (geometrically corrected to the National OS Grid). Where squares contained coastline, or where they fell on a regional boundary, the external area would be excluded.
2.5 In 1988 the statistical design was reassess ed in the light of experience in six regions. This allowed the sample square size to be reduced to 2.5 x 2.5km, with a minimum of 5 squares to be selected in each stratum. The overall result is a 7.5% sample coverage of Scotland's land area.
2.6 For each sample square the relevant 1:10,000 scale OS sheet served as a base map, over which a transparent film was overlaid for mapping NCMS features. Mirror stereoscopes could be used in the air-photo interpretation of lowland squares, where topographic detail was shown on the maps. Boundaries of land cover were drawn directly onto the transparency together with associated identity codes. In upland squares where land cover tended towards less distinguishable complex mosaics, and where there was an absence of reference detail on the base maps, it was necessary to use a Kern PG2 photogrammetric plotting machine to map boundaries.
2.7 In the NCMS method, the minimum mappable area is 0.1 ha and the minimum mappable length is 30m. The classification includes 31 area features and five linear features.
2.8 Interpretation difficulties arise from variable quality aerial photography, at scales ranging from 1:10,000 to 1:30,000. Field checking was carried out where there was uncertainty in the aerial photo interpretation. A more comprehensive and independent accuracy assessment has been published.
2.9 The completed land cover maps were digitised and processed on a Geographical Information System (GIS). The data were validated to check for and correct any missing linework or erroneous codes. Because of technological constraints in the early GIS configuration, vector datasets were rasterised into 10m grid cells for processing.
NCMS interpretation of land cover change
2.10 There are three 'layers' of data for each sample square, corresponding to each of the three time periods. Any two of the three matrices can be overlaid electronically to determine land cover change.
2.11 The number of 10m cells, or pixels, for each land cover type are held in a series of comparison files for each sample square. Land cover differences, represented by non-corresponding pixels, are similarly held.
2.12 After technical editing, the GIS output files for each region were input into a statistical package. Estimates of area and length were produced, together with change within, and interchange between, features. Standard errors were calculated from which confidence intervals were produced.
2.13 The project was designed to identify changes of 10%-or-more in the spatial extent of features with 95% confidence.
2.14 During the course of the study there was a migration across three GIS hardware-software platforms. The first phase of the NCMS, which compared change between the late 1940s and the mid-1970s, was undertaken with custom-built GIS software running on stand-alone desk-top personal computers with concurrent CP/M operating systems. In 1992 the project upgraded to an Intergraph workstation environment, with its improved performance partly constrained by the requirement for downward compatibility with its predecessor system. The full vector dataset is now available within an Arclnfo environment.