Remote sensing based ecosystem state assessment in the Sandveld Region, South Africa

Abstract

We present a remote sensing based approach for assessing ecosystem state or intactness to inform land use management and conservation planning. Using segmented multispectral medium resolution satellite imagery, parameters related to the image objects’ spectral brightness and heterogeneity, and compactness are used to derive a scoring system of 0 to 10 for the ecosystem intactness, with 0 being completely degraded and 10 being pristine. Linked to the remote sensing approach we suggest a field validation approach that focuses on 10 ecosystem-relevant visually assessed parameters which, when combined, produce a score out of 10 as well. The approach was tested in the South African Sandveld region using a SPOT 5 image from 2009 and a Landsat 7 ETM+ image from 2011. Field assessments took place in 2011. Both image data sets returned consistent results suggesting an inter-sensor transferability of the approach. Inconsistencies between satellite and field scores occurred mainly on sites where crops were currently being grown and on fields where various stages of succession were underway, following abandonment. Masking out of those sites which are of little interest from an ecosystem state perspective would improve overall accuracies. For regions with vegetation types that differ significantly in cover and structure, a stratified approach is suggested to optimise the results per vegetation type. Outputs suggest that the approach with its standardised and robust results and its repeatability provides a suitable tool for long term monitoring of large regions with a degree of detail sufficiently high to allow for fine scale planning.