Woody vegetation increase in Alpine areas: a proposal for a classification and validation scheme

Abstract

This paper presents a change detection analysis based on a region growing segmentation approach which combines both spectral and spatial information. The test site is a French Alpine protected area, which like many other mountain areas is characterised by a general increase of forest and woody vegetation due to the abandonment of traditional land use practices. Two Landsat images of the years 1984 and 2000 were used and a classification scheme nomenclature based on four vegetation change classes, implying a gradual modification of land cover, was adopted. The accuracy of the change map was assessed both during two visits on the field and using a bi‐temporal aerial photographic coverage. A sampling scheme specifically conceived for change detection products was adopted. Error matrices and accuracy indices to assess commission and omission errors of the change maps were generated.

The proposed change detection methodology circumvents limitations which are intrinsic to traditional classification procedures based only on spectral information. On the basis of the accuracy assessment, overall accuracy was 90.1% and the increase of woody vegetation turned out to be the vegetation change class better estimated, with user and producer accuracies of, respectively, 62.3% and 70%. However, confusion between the no change and the other vegetation change classes was noticed, due to standard problems encountered in change studies. Advantages and drawbacks of the use of multitemporal aerial photographs as the validation data set are also discussed.

Acknowledgments

This study was supported by a grant for training through research of the European Commission.

The Landsat TM image of 23/07/1984 was kindly provided by Jean Bernard‐Brunet of CEMAGREF‐Grenoble; the digital orthophotos and the DEM by the National Park of Mercantour. The Landsat ETM+ image is an IMAGE2000 (EC) product (http://image2000.jrc.it/). All the staff of the National Park of Mercantour, who contributed to the field validation, are gratefully acknowledged. The authors also wish to thank Frédéric Achard, Jacek Kozak, Michel Deshayes, and Samuel Djavidnia for their comments and suggestions.