This is an archive of the Countryside Survey 2000 website. It
is not being updated.
Click here for the current Countryside Survey website.
|
Land Cover Map 2000 Module 7 final report |
||||||||||||||
|
|
|||||||||||||||
|
|
by Contents
Note: The report is provided in full above, and has also been divided into six smaller files (regrettably, because of graphical content, some of these sections are still quite large). In some instances, the PDF files could be further compressed as WinZip files, but this was not the case for all of them. The Executive Summary is also provided separately (below), as HTML text. Note: to read the PDF files you will need Adobe® Acrobat Reader®, if you don't
have this installed on your computer, click here. For viewers with visual difficulties, information on the services provided to improve the accessibility of Acrobat documents is available from the following web site: http://access.adobe.com Note: You will need to have WinZip installed on
your computer in order to unpack the Zip files. To download an evaluation
version of WinZip, click here: Executive Summary
Land Cover Map 2000 (LCM2000)[1] provides a census of UK habitats and land cover as digital maps and databases plus a range of derived products held in a geographical information system (GIS). The map updates and upgrades the Land Cover Map of Great Britain, made in 1990-92. LCM2000 is a classification of spectral data recorded by satellites; external datasets add context to refine the classification. The process generated 72 class Variants, combined into 26 Subclasses, which in turn gave 16 Target classes and 10 Aggregate classes. Subclasses were combined appropriately to simulate 20 Broad Habitats (BHs) of the UK Biodiversity Action Plan. In the following text, LCM2000 classes are referred to in bold lettering and Broad Habitats in italics. LCM2000 used red, near infrared and middle infrared reflectance bands from summer and winter satellite images. The image data were calibrated to reduce atmospheric haze effects, masked to remove cloud, shadow and snow, corrected for differential illumination due to undulating terrain, and registered to the British and Irish National Grids. The pixels which make up the image were grouped into areas or segments broadly equivalent to land parcels (e.g. individual fields) using a process of image segmentation. The results were generalised to exclude small segments £0.5 ha. The classification of land cover types within segments avoided the use of edge-pixels with mixed spectral signatures by using the mean spectral response of pixels within the core of each segment. Ground reference data from field reconnaissance surveys were used to identify image segments of known land cover. These formed a sample of so-called ‘training areas’ used to calculate the spectral reflectance statistics for each land cover class. Classification used a maximum likelihood algorithm applied to each segment. The procedure compared the mean reflectances of unknown segments with the training set and recorded the most likely spectral subclass in statistical terms: in fact, it stored probabilities for the top five spectral subclasses, usually covering >90% of the probability distribution. For segments which were classified with low confidence or with classes out of their natural context, knowledge-based corrections (KBCs) were used to allocate an alternative class label, where more appropriate. The individual classified satellite scenes were ‘mosaicked’ together into 100 x 100 km squares. Residual cloud-holes were patched using the best available substitute images (e.g. single date classifications). Maps showing acid-sensitivity in the landscape were used to label semi-natural grasslands as ‘acid’, ‘neutral’ or ‘calcareous’. Geological maps showing peatland were used to distinguish between ‘heaths’ and ‘bogs’. Map displays and printed maps use cartographic conventions which balance the reliability of mapping and the importance and extent of a class, whilst bringing out important patterns in the landscape. However, greater detail is available in digital products at Subclass and Variant levels and any user-defined colour scheme could be applied. At national scales, regional differences across the UK are evident. At regional scales, the inter-relations of habitats within the landscape are clearly apparent. At the level of the individual county, the continuity and fragmentation of habitats becomes visible. At local levels, the full landscape structure becomes much clearer: in unenclosed landscapes, the mosaics of semi-natural habitats are recorded; cities and towns are subdivided into urban and suburban zones with open spaces; individual fields are evident; and larger linear features such as rivers and motorways can be seen. In a GIS, the full scope of the LCM2000 data becomes clear. Each segment is delineated by ‘vector boundaries’ recording the outlines in digital form. The database carries a range of attribute data for each segment. These describe its shape, size and location, source images and their dates. Thematic details and class probabilities are also recorded. The map structure, related directly to real features on the ground, can be used to help our understanding of the environment. It shows the inter-connectivity of landscape features, their immediate context and the wider neighbourhood in which environmental influences operate. The map helps us to see how ecological principles can explain patterns of biodiversity. It provides spatial data on land uses which influence hydrology. The data underpin climate models; they can also be used to predict the impacts of climate change on landscapes and ecosystems. The deposition of pollutants and their environmental consequences can be assessed in context. LCM2000 indicates land uses and values which underlie planning, environmental risk assessment and socio-economic modelling. LCM2000 offers a database structure upon which users can build. They may edit data to make corrections and refinements. They may add qualifying detail. They might update the information, recording changing land cover through time. A 25 m grid-based ‘raster dataset’ was derived from the Subclasses of the vector-based LCM2000. Generalised 1 km summary products are also available; they record class-dominance and summary percentage cover per 1 km2 at the 26 Subclass level. Subclass and Aggregate class datasets are provided in the Countryside Information System which gives data-access to non-technical users. CS2000 field survey data, covering a stratified random sample of 569 one-kilometre squares, provided information to assess the quality of LCM2000 in Great Britain (field data are not in appropriate format for similar analyses in Northern Ireland). The comparison was made for the BH classes derived by both surveys. The field data are not so-called ‘ground truth’; therefore, the process of inter-comparison was one of ‘calibration’ rather than ‘validation’. Comparisons of the field survey and LCM2000 generated ‘correspondence matrices’, one for each sample square. Results show similarities and highlight differences due to different spatial resolutions, time differences in surveys, class-definition differences, and errors in one or both surveys. Statistical procedures were developed to provide confidence limits for national and regional measures of correspondence between the two surveys. The estimated direct correspondence in Britain, at BH-level, is 54% (with the 95-percentile range estimated at 53-56%). It was known from the outset that there would be mismatches when the field survey and LCM2000 were compared at BH-level. The correspondence is higher for LCM2000 Target classes: 65% across Great Britain, 73% for England-Wales and 51% for Scotland for comparisons based on field survey land parcels. The largest differences were found in upland areas where field and satellite-based mapping were most problematic. Differences in resolution, the data-model and timing of surveys contribute to the differences between LCM2000 and field survey. The comparison as a whole suggests that LCM2000 may record Target classes with around 85% success. Individual classes fare differently. LCM2000 Broadleaved and mixed woodland is near identical in extent to the field survey coverage of the Broadleaved, mixed and yew woodland BH in the UK. However, direct agreement is lower because many woodlands and clearings are at or below the minimum mappable unit of LCM2000. Coniferous woodland, generally planted and in larger blocks, records similar coverages and a greater direct correspondence. Arable and horticultural land covers nearly a quarter of the UK according to LCM2000 and equivalent field estimates. Apparent confusions between the Arable and horticultural land and Improved grassland BHs in LCM2000 relate largely to rotation farming in squares where field and satellite survey-years differed. Improved grassland, covering more than a quarter of the UK according to LCM2000 and field survey, is the most extensive single cover class. Generally, improved swards were readily recognisable and well-classified on LCM2000. However, the distinction of ‘improved’ grassland from ‘semi-natural’ types could be both difficult and controversial. Semi-natural grasslands (and Bracken) present problems in their distinction. Neutral, Calcareous or Acid grassland BHs gave no consistent spectral characteristic by which to determine soil acidity; and external data were of limited value. Bracken was problematic because so much of the imagery used to make LCM2000 was recorded in May, when bracken on the ground was at a minimum; often such stands were recorded as the underlying Acid grassland. The coverage of the Mountain, heath and bog Aggregate class is the same by field and satellite surveys. However, at BH level, the components Dwarf shrub heath and Bogs are mapped differently. LCM2000’s distinction of Bogs using a peatland map gave a conservative estimate of the class. The Montane BH was defined by altitude criteria on LCM2000 which recorded more of this class than the field survey estimate. The LCM2000 and field surveys recorded the Fen, marsh, and swamp BH differently, mainly through differences in the treatment of rush-pastures. Water (inland) on LCM2000 is an aggregation of the Standing open water and canals and Rivers and streams BHs. There was no attempt to distinguish standing from flowing water but the overall coverage is similar to field estimates. The Built up and gardens BH is mapped in more detail by LCM2000. The field survey treated urban land as continuous without recording open spaces in the urban zone. LCM2000 recorded open spaces >0.5 ha and distinguished Suburban and rural development from Continuous urban / industrial land. Coastal habitats of Supralittoral rock, Supralittoral sediment, Littoral rock and Littoral sediment were recognised at BH level in LCM2000 Subclasses but shown on maps as the combined classes Supralittoral rock and sediment and Littoral rock and sediment. Differences between field and LCM2000 estimates relate the tidal state at the time of survey and, in part, to LCM2000’s greater geographical offshore coverage. The Boundary and linear features BH was not targeted by LCM2000. LCM2000 only includes linear habitats which have an area >0.5 ha: to have been resolved by the images they will also have been ≥2 pixels wide. Calibration against field data Calibration of LCM2000 against field survey allows the generation of BH cover-statistics equivalent to those from sample-based field survey but benefiting from the comprehensive coverage of LCM2000. The resultant statistics are probably CS2000’s best current estimates for BH coverage at regional and national scales. The mean values compare very closely with field survey estimates, but with confidence limits which are much tighter (about half the field survey range). A weakness of the field survey has been its inability to make reliable estimates of cover at local scales. At Target class level, LCM2000 is likely, in general, to be 85% correct in its mapping, but with the potential for significant local errors. Local results are likely to be mostly reliable for Aggregate classes. Further work to provide and assess calibrated BH estimates at local levels is planned. According to LCM2000, more than half the UK is used for intensive agriculture or is developed. The remainder is largely semi-natural. Woodlands occupy a quarter of the semi-natural land, with Broadleaved woodland and Coniferous woodland about equal in extent. Mountain, heath and bog cover a third of the UK’s low intensity land; semi-natural grass swards (including rougher examples of improved swards) form over a third of all semi-natural cover. Coastal habitats and Open water, while important, are small in extent. The four countries of the UK differ markedly. Intensive uses affect nearly three-quarters of England, about two-thirds of Northern Ireland and about half of Wales; in Scotland, less than a quarter is intensively farmed or developed. The semi-natural land of England is evenly split between woodlands and grasslands. In Wales, the balance is similar within a far greater extent. In Scotland, Mountain, heath and bog make up more than half of all the semi-natural land. Northern Ireland also has reasonably extensive Mountain, heath and bog and Semi-natural grass but, at the resolution of LCM2000, is notably short of woodland cover. The measurement of changes in land cover demands high levels of precision to map real differences and to distinguish them from survey errors and generalisations. In a comprehensive national survey, the necessary precision for change detection cannot be achieved consistently by satellite-based mapping alone. The LCM2000 classification sought to remove known deficiencies in the 1990 classification and to bring field and satellite surveys into closer match. The segment-based approach of LCM2000 generated different results from the 1990 raster product. These differences preclude direct comparisons with the 1990 product. It may be possible to select intelligently, from those differences mapped, the elements which are attributable to change and those attributable to error and / or differences in the data products. This approach will be the subject of research and development, beyond the scope of the production phase. In conclusion, LCM2000 has, for the first time, mapped the land cover of the whole of the UK from satellite images. The resultant vector data record the ‘real’ structure of the landscape and thus can satisfy wide ranging user-needs. LCM2000 offers so much more scope than the conventional per-pixel products of the earlier mapping. It has a detailed spatial resolution which is far better than other vector-based maps of full UK land cover. LCM2000’s structural picture of the landscape shows the spatial inter-relations of parcels and habitats. It therefore lends itself much better to applied uses where patterns affect processes and determine their consequences. |
||||||||||||||
|
|
|||||||||||||||
