Mapping Savannah Trees in Kalahari using High Resolution Remotely Sensed Images and Object-Oriented Classification

  • J. N. Kimani The International Institute for Geoinformation Science and Earth Observation (ITC)
  • Y. A. Hussin The International Institute for Geoinformation Science and Earth Observation (ITC)
  • M. W. Lubczynski The International Institute for Geoinformation Science and Earth Observation (ITC)
  • D. Chavarro The International Institute for Geoinformation Science and Earth Observation (ITC)
  • O. T. Obakeng The International Institute for Geoinformation Science and Earth Observation (ITC)

要旨

High spatial resolution 3-band airborne images (G, R and NIR) were used to map vegetation for upscaling of transpiration measurements made over the tree-bush-shrub of Serowe area in Botswana. These images were acquired with digital TETRACAM camera mounted on a small aircraft to collect data in 30, 60 and 100 cm spatial resolutions. The airborne imaging results were also compared with one-meter pan-sharpened multi-spectral IKONOS satellite images. This paper presents the first step of the transpiration mapping study focused on vegetation mapping based on the individual species classification. Two object-oriented classification techniques corresponding to eCognition and Feature Analyst software packages were used in data processing. In the particular application of mapping of individual savannah tree species, the eCognition showed to be more accurate and reliable than Feature Analyst. The spectral characteristics of the TETRACAM images were similar to IKONOS satell ite images, while the spatial characteristics were much better in TETRACAM images than in IKONOS images. This was reflected in the substantial accuracy difference between the airborne and satellite data of the same resolution both quantitatively (e.g. in KIA values i.e. the number of tree species identified correctly) and qualitatively (i.e. their classified maps). The overall study demonstrated that the higher the spatial resolution, the higher the number of tree species properly identified with regard to the species type and canopy pattern, hence the higher the accuracy. The more fully formed individual tree crowns are, the better the distinction among their spectral signatures, especially where tree species shed and regain their leaves with season thus affecting the stable green color associated with their spectral reflectance. If these factors leading to better classification are combined, then the chances of overestimating or underestimating transpiration based on such classification are minimized.
出版済
2007-06-01
セクション
Article