Separating grassland and shrub vegetation by multidate pixel‐adaptive spectral mixture analysis

Abstract

Monitoring and assessment of land degradation and the processes driving it require effective change indicators at appropriate scales and spatial extent. In this context, the decomposition of Mediterranean rangeland vegetation into woody and herbaceous fractions is of great significance. This study demonstrates that a stratification of vegetation into woody and herbaceous components is possible with two satellite images of moderate spatial and spectral resolution. We used a pixel‐adaptive spectral mixture analysis to derive subpixel‐level vegetation abundances from satellite imagery representing two specific phenological stages of Mediterranean rangeland vegetation. The transferability of endmember models is often a problem of multidate spectral mixture analysis because of uneven spectral dimensionality within and among datasets. In our approach, the dimensionality of the mixture model was determined automatically, based on error calculations. This method enables the transfer of the mixture model to multiple scenes and allows for quantitative comparison of vegetation abundances. The results show that the woody vegetation fraction corresponds well with field data (R ² = 0.76–0.91) and vegetation cover mapped from a very high resolution satellite image. The herbaceous vegetation fraction displays a good correlation compared to field mapped cover but still implies a moderate level of uncertainty (R ² = 0.52–0.76). The approach pursued in this research may be valuable for the characterization of rangeland plant communities and for the derivation of vegetation‐related indicators useful for the monitoring and assessment of degradation.

Acknowledgements

This work was carried out within the project GeoRange, part of the fifth framework programme of the European Union, DG Research, under contract EVK2‐CT‐2000‐00091. This support is gratefully acknowledged. We are also grateful to the Laboratory of Range Ecology (Aristotle University of Thessaloniki) for making available the field data measurements and thank F. J. García‐Haro for providing the VMESMA software package. We thank W. Mehl and S. Bärisch for support in the data processing, M. Schlerf and T. Hawbaker for helpful comments on the manuscript, and the anonymous reviewers for constructive remarks on earlier versions of the manuscript.