Crown closure estimation of Argan trees in southwest Morocco from SPOT imagery

The Argan [Argania spinosa (L.) Skeels] is a species of tree endemic to the calcareous semi-desert Sous valley of southwest Morocco, covering more than 8000 km2. Argan wooded savannas, and scattered trees in rainfed cereal fields, not only act as a buffer against desertification, but are also a source of livelihood for 2 million people in rural Morocco, who depend on the trees for oil, fodder, honey, charcoal, fuel and construction wood. These agroforests suffer from continued degradation induced by intense use such as fuelwood gathering and grazing in the hilly areas, and, in the plain, from tree removal to introduce irrigated crops. It is then needed to map and monitor Argan tree density in these endangered areas.
In the present work, we are using SPOT images (panchromatic at 5m resolution and multispectral at 10m resolution) obtained in the dry season. A multiband data fusion technique (Gram-Schmidt spectral sharpening) preserving both spatial and spectral informations, is used to derive a multispectral image at 5m resolution. Then an object oriented classification (feature extraction implemented in ENVI software) is performed, in order to discriminate Argan woodlands from other land-cover classes. Finally, considering only the area of Argan woodlands, a filtering technique, based upon a set of laplacian filters is applied to the sum of the 4 channels, in order to enhance the contrast between trees and background, and obtain a thresholded binary image. This image (Argan vs background) is refined through identification of linear features and thresholding of NDVI vegetation index. Results of tree identification leads to tree density estimations, which are validated against field transects, aerial photographs, and a few Quickbird high resolution satellite images. The possibilities of generalization of tree density mapping to the whole Argan tree distribution area through linear regressions applied to ASTER or Landsat imagery is discussed.