Abstract:
Monitoring tropical rain forests via remotely sensed imagery has become very useful in understand-ing land use/cover change over time for three East Africa rain forest areas: Kakamega-Nandi forests area in Kenya, Mabira and Budongo Forest areas in Uganda. Landsat imagery, provided free of charge by the U.S. Geological Survey Center for Earth Resources Observation and Science (USGS/EROS), enables consistent and timely forest cover change over time.
Analyses of the landuse/ cover changes since the early 1970s until 2003 for these three rain forests were done by processing Landsat Multispectral Scanner (MSS), Thematic Mapper (TM) and En-hanced Thematic Mapper plus (ETM+) imagery for eight or seven time steps at regular intervals by Biodiversity Monitoring Transect Analysis in Eastern Africa (BIOTA East Africa). For continuous forest change analysis, the three existing time series data are to be extended by another two time steps from more recent years (2005/2006 and 2007/2008) as part of the remote sensing activities within BIOTA East Africa sub-project E02 . Since, on 31 May 2003 Landsat ETM+ suffered the loss of its scan line corrector (SLC) which removes the “zigzag” motion of the imaging field of view produced by the combination of the along and cross track motion, there is data loss of about 22% of the total area of the scene, then there is need to get alternative solution.
This study describes a methodology for combining several SLC-off images of Budongo Forest area into a one single dataset to be used as basis for land use/cover classification. The approach of filling gaps used in this method involved techniques of adding classified images in order to come up with a meaningful classification. Two images per time step are used to come up with one meaningful classification. Additionally, the suitability of SPOT-4 multispectral image data for deriving land use/cover classifications for another two time steps of Kakamega Nandi and Mabira forests areas have been investigated to give truly comparable results to the existing Landsat-derived time series data.
Both SPOT and Landsat SLC-off data offered the chance of extending the existing times series with truly comparable classification results. The same land cover classes have been distinguished as in the previous time steps using supervised multispectral classification. The applied methodology re-sulted in high classification accuracies.