"A Satellite Map Representation of the
blanket mires of Lewis, Scotland"

Project undertaken for Scottish Natural Heritage, completed May 1996. I.S. Ltd produced this report with the help of F Everingham (peatland ecologist) and E Reid (SNH). Subsequent projects for SNH have used a similar technique for mid-west Scotland and the Inner Isles (1997), Grampian region (1998) and Shetland Isles (1999).

Background

Blanket bog is a vegetation formation which probably reaches its extreme world development in western and northern Britain and in western Ireland, reflecting the cool, intensely oceanic climate (Lindsay, 1995) Under these conditions peat formation is able to occur on all but the more steeply sloping ground.

In order to identify areas of conservation importance within this extensive habitat the Nature Conservancy Council carried out site driven survey of blanket bog between 1985 and 1990. This small scale ('fire brigade') site specific inventory highlighted that extensive tracts of ground between the specific sites were being overlooked and valuable information on the resource was being under represented. Remote sensing was therefore investigated as a means of gathering information from the vast areas of blanket mire which are particularly prevalent in northern Scotland.



Aims and Objectives

The main requirement of this study was to produce a satellite map representation of the blanket mires of Lewis and Harris (Landsat Thematic Mapper scene 208/19, date June 1992) by interpretation of the imagery and ground data. The study had three broad aims:


Ground reference sites map Download a larger image

Methodology

Small areas of homogeneity had been delineated on the Landsat Thematic Mapper satellite image using an unsupervised classification approach. The boundaries of these areas were converted to vector lines and overlaid on the satellite image. The original satellite bands 5,4 and 3 were contrast enhanced and plotted showing the enhanced images and boundaries of homogeneous areas at approximately 1:10000 scale for 11 survey sites on Lewis.

A total of 139 ground reference areas (blocks) were visited and a survey data sheet completed for each area. Samples were taken from several different areas within the main extent of the blanket bog on Lewis in an attempt to capture the full range of variation within the habitat. Figure 1 shows the main areas selected for survey.

Existing SSSI vector data and the British Geological Survey (BGS) peatland areas were imported as vector layers in the same file containing the satellite data. Areas marked as proposed SSSI (3 bog and bird pSSSI's) were digitised on screen within the EASI/PACE image processing using 1:50000 Ordnance Survey raster data as a reference. A 1:50000 scale digital elevation model (DEM) for Lewis was purchased from the Ordnance Survey. The 139 ground reference blocks identified on photographs / satellite image enlargements were digitised on screen using the satellite image as a reference image.

Satellite data processing

The satellite data used were six Thematic Mapper image bands acquired by the Landsat 5 satellite on 6th June 1992. All satellite data processing was carried out using PCI's EASI/PACE image processing system on a Pentium PC running the Windows 3.1 operating system.

To aid the classification a number of masks were produced to restrict the statistical analyses to the required areas of interest (peatland). In doing so it was hoped to maximise the number of distinctive peatland classes. The six Landsat bands were transformed to three Principal Component images using statistics taken from all land areas below 200 metres. This reduced the size of the data to be processed and increased the separability of the bands to be used in the subsequent classification.

The final classification produced output for Lewis in three stages. Firstly, an initial classification was used to identify a mask of the gross peat area. The areas were classified to produce 15 'peat' classes. These spectral classes were examined and aggregated to produce 7 peat classes and 1 non-peat (rock) class. Secondly, the areas outside the peat class mask but below 200 metres were classified into eight categories which were subsequently amalgamated into three classes. Finally, the areas above 200 metres were assigned to a single class. In total the final classification contained 12 classes. A 'sieve' filter was applied to the classification to remove some of the small isolated pixels or polygons.

Quantitative Analysis

The quantitative analysis of the classification results (for the peatland classes only) involved a test of the robustness of the classification, and an assessment of the accuracy of the classification in comparison with the ground reference data.

A transformed divergence test was used to determine the statistical separability for each class. The average separability of all classes was 1.84 (maximum 2.0) indicating that the classes are well separated.

To test the accuracy of the peatland classification a ground reference image was constructed from the ground survey blocks. Class accuracies ranged from 63.3 to 89.8 % correctly classified pixels (overall accuracy 76.98%). Some confusion was evident between classes 2, 3 and 5.

Ecological assessment of spectral classes

It is not certain that the classes produced will be meaningful in terms of ground observations. Statistical analysis can determine if the classes are ecologically meaningful.

Field measurements of the proportion of shrub cover, herb cover, moss cover, bare peat cover and erosion had been made to characterise each ground reference block. A null hypothesis is set up to state that there is no significant difference in the proportions of cover type for a given spectral class.

The Analsysis of Variance (ANOVA) statistic tests the proportion of each cover type separately against the spectral class to see if the variation in spectral class can be explained by a corresponding variation in the proportion of a given cover type. The Multivariate ANOVA statistic tests for all of these variables together, to see if the variation in spectral class can be explained by a corresponding variation in the proportion of all the cover types as a whole.

Examination of the univariate ANOVA test reveals that whilst the Null Hypothesis can be rejected for Shrub, Herb, Bare Peat and Erosion cover, it should be accepted for Moss cover as the probability is less than the 0.05 level. From the multivariate ANOVA statistic the probability that the Null Hypothesis is true is less than 0.001. In other words the Null Hypothesis should be rejected.

It is therefore clear that the spectral classes produced by the image analysis are based upon real changes in vegetation cover type. However there is less discrimination between the classes within the % moss cover variable. It should be noted that the moss cover variable does not however attempt to distinguish between the different moss types present within a ground reference block. This is best determined by investigating the presence or absence of National Vegetation Classification (NVC) types. For each class, the proportion of each NVC type has been plotted on Figure 2 below. The Y axis shows the proportion of each NVC type per class.


NVC categories per classDownload a larger image

For classes 1,2, and 5, which were not well discriminated by the proportion of moss cover, it is clear that the proportion of M17a decreases from class 1 through to class 5. Also classes 2 and 5 both contain higher proportions of M17b vegetation categories; and class 1 contains a higher proportion ofthe M1 vegetation community. All of these classes are therefore represented in the above figure as mixtures of NVC communities.

Results

It appears that the majority of Lewis and Harris is covered by active bog. Classes 1- 5 , and 7, the peatland classes account for 66.5 % of the total land area (see table below).


Peatland Classification Results
Spectral Class Description % of area
1 M17 mixture 7.34
2 M17 microbroken 14.77
3 M17a smooth, wet 6.81
4 M19 shrub dominated 10.64
5 M17b eroded 15.97
6 Marginal peatland 3.54
7 Bare peat 11.06
8 Rocky ground 5.55
9 Grassland 5.62
10 Woodland 0.77
11 Sandy soil / dunes 1.33
12 Land above 200m 11.55
0 Unclassified 5.04


Peatland classification of an SSSIDownload a larger image

Further reading

Horsfield, D., Thompson, D.B.A., & Tidswell, R. (1996). Revised atlas of upland NVC plant communities in Scotland. Information and Advisory Note. Scottish Natural Heritage, Perth.

Joint Nature Conservation Committee. (1994). Supplement to The guidelines for the selection of biological Sites of Special Scientific Interest: bogs, Nature Conservancy Council (1989). Joint Nature Conservation Committee, Peterborough.

Macaulay Land Use Research Institute. (1993). The Landcover of Scotland 1988 Final Report. The Macaulay Land Use Research Institute, Aberdeen.

Quarmby, N.A. & Reid, E (2000). Determining the composition of the blanket bogs of Scotland using Landsat TM. In: 'From Patch to Planet'. (Ed. by A. Millington), Biogeography Research Group Conference, Chester 1995, Wiley.

Reid, E., Mortimer, G.N. , Thompson, D.B.A and Lindsay, R.A. 1994. Blanket Bogs in Great Britain:An Assessment of large-scale pattern and distribution using Remote Sensing and GIS. In Large Scale Ecology and Conservation Biology (edited by P.G. Edwards, R.M. May and N.R. Webb) pp.229-246. 35th Symposium of the British Ecological Society, Blackwell Scientific Publications, Oxford.

Rodwell, J.S., ed. 1991. British plant communities. Vol. 2: Mires and heaths. Cambridge University Press.

Tudor, G.J., Mackey, E.C. & Underwood, F.M. (1994). The National Countryside Monitoring Scheme: the changing face of Scotland, 1940s to 1970s. Main Report. Scottish Natural Heritage, Perth.


Further information:
Remote Sensing and GIS Mapping Software
Teaching materials
Satellite Image Catalogue
Mapping Peat Bogs in Lewis, Scotland
Fuelwood Assessment and Management in Chad
Conservation Areas Captured in SW England
LANDMAP project