Abstract:
The establishment of Trans-frontier Conservation Areas in southern Africa facilitates the roaming of wildlife across international borders. The probability of disease transfer associated with wildlife and livestock contact zones is a cause of concern for conservationists and local communities. Assessing and monitoring vegetation is an important part of ecological research and management, as vegetation characteristics are often fundamental to habitat differentiation. In the challenge to find cost-effective ways for vegetation monitoring, remotely sensed data such as satellite imagery offers a possible alternative to field based techniques. However, imagery with good spatial, spectral, radiometric and/or temporal resolution may be too expensive for frequent use. This study investigated the potential and challenges associated with the analysis of mainly savanna vegetation structure using in-situ observations and pixel-based classifications derived from multispectral SPOT 5 images in a selected subset of the Greater Limpopo Trans-frontier Park (GLTP).
The availability of cost free SPOT 5 imagery and the suitability of currently available land cover and ancillary information in the GLTP area were investigated and described. Using the acquired imagery, supervised (Maximum Likelihood) and unsupervised (ISODATA) pixel-based classification methods were examined and tested. Normalized Difference Vegetation Index (NDVI) and Second Modified Soil Adjusted Vegetation Index (MSAVI2) values were added as additional bands to the SPOT 5 image bands. Pair separation statistics and thresholds were used to evaluate and describe the potential effect of training area sizes and image-index band combinations on classification results. Classified images were assessed using qualitative (visual comparison) and quantitative (error matrix) methods. The applicability of estimated desktop and in-situ field observations as ground truth validation tools were evaluated and compared. From the various classified products, the most suitable classified image was selected and an appropriate level of generalisation was chosen based on overall accuracy and Kappa values. The potential sources of error inherent in all processes, such as field based observations, image acquisition, pre-processing, classification, generalisation and interpretation, have been acknowledged and described. Visualising techniques and guidelines aimed at the thematic presentation of a classification product along with its associated confidence levels were explored and illustrated. Furthermore, the incorporation of ancillary information to improve the applicability of the results was illustrated.
This study revealed, illustrated and discussed the influence of image resolution, classification methods, band selection, vegetation indices and training area characteristics on the suitability of remote sensing to classify vegetation characteristics in remote or inaccessible savanna areas. From the results it can be concluded that the use of medium resolution multispectral SPOT 5 imagery for pixel-based classification of vegetation structure in the study area may be limited in its application value and should be used perceptively and with caution. Overall it must be noted that although the use of satellite imagery as a whole may have reached almost unlimited potential, there are still many challenges for researchers in the various application fields of this technology.