Land Cover Change in Southern and Western Africa - A combined Change Detection Approach

Kurzfassung

Savanna ecosystems exist at the transition zone between tropical forests and deserts. Due to human land use activities, inter-annual climate variations, and global climate change these ecosystems are threatened by vegetation cover loss and hence desertification. This change has to be seen as severe impact due to the loss of biodiversity [1, 2], and natural living resources, especially in terms of pasture land [3]. The detection and monitoring of these processes is the first step of prevention and the development of sustainable land use strategies. Within the international integrative research project “BIOTA-Africa” (http://www.biota-africa.org), a remote sensing based method was developed, to detect land cover changes in southern and western Africa, with special emphasize on land degradation processes. The technique uses historical and recent LANDSAT-TM and ETM imagery (up to the year 2003). It is a combined bitemporal change detection approach, based on spectral information, as well as on structural information of the altered areas. The comparison of georectified and atmospherically corrected Landsat images is performed by a modified selective Principal Component Analysis (sPCA, [4, 5]) embedded in the “ChangeIndex-Tool”, developed by Weiers et al. [6, 7]. The tool was also used to detect changed areas in an agricultural landscape in Schleswig-Holstein, Northern Germany by Kleinod [8]. C. Schultz (see Abstract … handed in for this Conference as well) used it to find non-altered pixels as base for temporal stable training areas for a retrospective Maximum Likelihood Classification of lichen fields in the Central Namib, Namibia. The Change Index-result provides information of the relative strength of change in the examined area. Additionally, an image differencing approach was applied, which is well known to deliver reasonable change values [9, 10]. The resulting difference image is then segmented with the software eCognition to group pixels with similar change values [11]. Using this threefold information base, (a) sPCA-derivate, (b) image differencing result, and (c) segmentation properties of the differencing result, a non-parametric classification scheme was established to identify detected changes. The mean classes are separated into “severe” and “slight loss” or “increase” of vegetation cover, taking into account the former or latter fraction of bare ground signal. The technique was developed on a LANDSAT-Scene in central Namibia, where mixed thorn bush savanna is the predominant vegetation type (yearly mean precipitation about 350 mm). We discovered that in this region from 1984 to 2003 desertification processes are of minor importance. Here, the vegetation degradation in form of bush encroachment affects even larger areas and leads to a loss of biodiversity due to the explosive spreading of few bush species like Acacia mellifera and Dichrostachys cinerea [12]. In many cases the degradation leads to a total abandon of the affected farmlands. We were able to apply the technique without modifications successfully as well in other ecosystems in South and West Africa: In the Kavango Region in Northern Namibia (approx. 600 mm mean precipitation) the method showed good results for a stable classification of recently logged and as well abandoned fields in the dry forest regions. Here the anthropogenic induced loss of vegetation (i.e. logging) could be distinguished from natural changes (denudation of ephemeral riverbeds) due to the nearly rectangular form of the fields, which could be extracted with the segmentation properties. In the dwarf shrub savanna of southern Namibia (ca. 150 mm prec.) the erosive effect of severe precipitation events, that led to the denudation of ephemeral riverbeds in overgrazed areas but not in sustainable used areas have been shown. In West Africa the technique described above was applied on tropical lowland rain forest in Ivory Coast. The Taï National Park close to the Liberian Border is the largest remaining fraction of protected primary rain forest, yet still undergoes a high deforestation and fragmentation rate. Mapping areas of increased use of the National Park for agroforestry, logging and changes induced by microclimate alteration was targeted. This method made it feasible to distinguish anthropogenic and natural changes (e.g. tree fall gaps and plantations) in the Taï National Park and its tendency over time. The technique has potential to detect land cover change, especially the differentiation between anthropogenic induced and natural modification of the landscape.