Module 7 Land Cover Map 2000

A complete census of land cover for GB using Earth observation via satellite imagery collected in the same season as the field survey.

• Click here to access the final report for this module

Visit the LCM2000 image gallery! Click here.

Background
Land Cover Map 2000 (LCM2000) will provide a census of the countryside of the UK, in the form of digital maps and databases, plus a range of derived products, for use in a geographical information system (GIS) and statistical packages. It is funded by a consortium consisting of NERC, DEFRA, NAW, SE, EA, SNH, EHS, DARD and CCW (see Sponsors).

Land Cover Map of Great Britain 1990
The Land Cover Map of Great Britain (LCMGB) pioneered, in 1990-92, a land cover census by remote sensing. The LCMGB recorded 25 land cover types, on a 25 m grid, for all of Britain. The generalised census, as a part of the Countryside Survey 1990, complemented the sample-based detail of the field survey. The LCMGB and the field data were compared and contrasted qualitatively and quantitatively, giving correspondences which varied from 50%-90% and suggested that the map accuracy was generally 80%-85% - an exact quantification was precluded by crucial differences in the dates of survey and the choices of class definitions. The LCMGB data give complete, though generalised, national coverage and are uniquely capable of generating useful information at a field-by-field level.

Since 1992, map data have been made available, under licence, to over 436 users (Oct 2000) in Government Departments, the Agencies, utilities and commerce, and in research organisations advising these users. Incorporation of both field and satellite data sets in the Countryside Information System (CIS) has offered integrated applications to many others: since release in 1994, the CIS has sold over 384 licences (Oct 2000). The conversion of the LCMGB to the CORINE Land Cover format was completed in January 2000, contributing the British component to this European map and database.

Land Cover Map 2000
Four key refinements are incorporated into the production of the LCM2000:

• improved accuracy of classification,

• added thematic detail,

• compatibility with other systems of environmental survey and evaluation,

• closer integration between field and satellite data.

With these improvements in mind, CEH (then ITE) developed a range of new methodologies for land cover mapping. The most important development was the ability to perform parcel-based analyses of remotely sensed data rather than using conventional pixel-based approaches. Features in the landscape, such as field or stands of trees, could then be analysed and classified as a single entity in their true context. This work was undertaken within the CLEVER-Mapping - Classification of Environment with Vector- and Raster-Mapping - project and developed jointly with Cambridge University Department of Geography, Laser-Scan Limited and the Ordnance Survey under the British National Space Centre’s LINK programme (see article in CS2000 News, issue 3).

Summary of methods
The following list summarises the methods being used in the construction and validation of LCM2000:

• Dates of image acquisition are, as far as possible, matched to the date of the field survey, i.e. summer 1998; multi-temporal and multi-sensor imagery are being used, with the winter component ideally coming from winter 1997/98.

• LCM2000 production employs a parcel-based approach, which segments the UK into land parcels, the segments being based on spectral data from the satellite images.

• The parcel-based procedures uses vector land parcel boundaries, defined by land parcel segments, to extract data on the image pixels and thereby derive reflectance statistics per-parcel.

• Ground reference data identify a sample of land parcels to provide ‘training data’ which exemplify the reflectances for each target land cover class, subclass or variant, including definition of additional spectral classes where statistically necessary.

• The training data are extrapolated, with a maximum likelihood classifier operating on the reflectance statistics per-parcel, to attach land cover classes to the land parcels.

• Contextual information, such as terrain height or soil characteristics, are used as necessary and as available to enhance classification accuracy.

• The classification delivers a standard list of 24 target land cover types, the aim being that >90% of parcels will be correctly allocated. The target land cover types have been selected to allow assessments, as far as possible, of widespread Broad Habitat types. These have been defined to meet objectives of the Biodiversity Action Plan and UK obligations to protect Priority Habitats under the EU Habitats Directive.

• The main output of LCM2000 will be a vector GIS database, segmenting the UK landscape into land parcels, recording the dominant land cover and a range of other attributes for each parcel. Additional outputs will be generated from this core database, such as a raster-based GIS, recording land cover on a 25 m grid (similar to LCMGB) and summary 1 km data (% cover, dominant cover per 1 km square) from the above products, including incorporation in the CIS.

• Field survey data is being used to calibrate and validate the LCM2000 database. The calibration of the satellite classes against field survey classes may allow the removal of systematic errors in the former, using a retrospective weighting to correct the classification. It would maximise the correspondences for integrated use of the generalised satellite census and sample-based field details of CS2000.

• LCM2000 data will be evaluated against LCM1990 data for trial areas to assess the potential to measure change and to demonstrate analytical methods which later might be developed to refine distinction between error and real change.

Database Characteristics
Parcel-based approaches offer much more than conventional per-pixel classifications. The parcel-based approach will increase the power of the analyses used in production, provide a richer set of results and offer enormous opportunities for later operations and application. Put simply, each spectrally distinct land parcel in UK will be defined by LCM2000, stored in an easily accessable database and analysed as a single entity. Each land parcel will carry a wide range of attribute data, for example:

Click on image for expanded view

• Land cover type:

  • dominant class 2000 - contextually corrected maximum likelihood class;

  • alternative classes - ranked on statistical probabilities;

  • heterogeneity - cover on a per-pixel basis, summarised as a %;

  • past class - LCMGB land cover type, summarised as a %;

• Parcel area;

• Boundary length;

• Vegetation indices per summer and winter image;

• Contextual information - soil type, terrain height.