Land Information System

Welcome to the LandIS Soil Portal

LandIS Case Studies

National Parks

Automatic classification of vegetation communities

A co-operation between the University of Leeds and the Yorkshire Dales National Park Authority

Yorkshire DalesVegetation maps are a valuable resource for a range of applications, such as the monitoring of vegetation in a changing climate, the study of species – habitat relationships, for conservation management or land use planning. Creating vegetation maps by ground surveys is prohibitively expensive for large areas. Mapping by automated classification using readily and widely available data sources could present a cost-effective alternative.

In Great Britain, the National Vegetation Classification (NVC) is commonly used to categorise vegetation as communities (detailed descriptors of plant species composition). Automated classification procedures to map vegetation communities did often not produce good results. A recent study (a co-operation between the University of Leeds and the Yorkshire Dales National Park Authority) has achieved high overall accuracies (87 – 92 %) for the automated classification of 24 NVC communities and a land cover class 'wood' (consisting of trees and bushes) at a high resolution (5 m) for an upland area in England. This study, funded by the Yorkshire Dales National Park Authority, used a combination of readily available data sources including the simplified version of the National Soil Map (NATMAPsoilscapes) together with a sophisticated mathematical algorithm.

Vegetation communities are often associated with environmental conditions, such as soil type, elevation, slope or aspect. These variables can influence plant species distribution through for example soil fertility, soil pH, solar irradiation or temperature. However, vegetation communities at locations with similar environmental conditions may differ due to different conditions in the surroundings. The vegetation community typical of the conditions at a particular location may be altered by other communities in the surroundings associated with different environmental conditions. Biotic factors such as dispersal or competition may prevent the community typical of the conditions at the location from being present. Therefore, the environmental variables at the location but also the average conditions within the surroundings were incorporated into the automated classification. Environmental variables were extracted from the NATMAPsoilscapes and a digital terrain model.

The associations between vegetation communities and environmental conditions can be weak or even absent, for example when vegetation is at different stages of succession (such as after fires or other disturbance events) or due to anthropogenic activities (such as when trees have been planted over a wide range of environmental gradients). Information from aerial images (describing spectral reflectance and texture of vegetation) can partly overcome this limitation and was also included in the automated classification.

If good results using this procedure can be achieved in other areas and ecosystems, the cost-effective production of highly detailed and accurate vegetation community maps for large areas could become possible. The study is published in the British Ecological Society's Journal of Applied Ecology thus:
Bradter, U., Thom, T. J., Altringham, J. D., Kunin, W. E. and Benton, T. G. (2011) Prediction of National Vegetation Classification communities in the British uplands using environmental data at multiple spatial scales, aerial images and the classifier random forest. Journal of Applied Ecology, 48: no. doi: 10.1111/j.1365-2664.2011.02010.x.
The paper can be accessed here.


If you have found LandIS information useful in your work and would like to share your experiences, we'd like to hear from you!