Characterizing changing landscape patterns in West Africa: an object based image analysis approach

Abstract

To effectively combat malaria and other mosquito-borne diseases it is important to know the habitat of their vector and whether it is changing through human influence. That is why it is useful to determine changes in the landscape. Therefore the land cover of two transects in West Africa at two different moments (1986 and 2013) was classified. Object Based Image Analysis (OBIA) was used as it is well-suited for detecting human influence. For the image segmentation a multi-resolution segmentation (MRS) approach was used. The classification was based on the spectral values, shape, relative position and border of the objects. Both transects showed a large increase in urban area (79% and 121%). In transect 1 all urban areas increased in size by 50 to 82%. However, in transect 2 only the main urban area increased in size, while the small urban areas present decreased in size by 7 to 74%. Transect 1 showed a large increase in dense vegetation, while transect 2 showed a radical decrease in dense vegetation. Combined with the different urban changes this points to a different development trajectory for the two transects. The most likely explanation is different economic growth in the region of the transects. Patch shape complexity for both transects decreased even when the total number of patches increased. This is an indication that human influence has increased. Developing a method to estimate human influence in an area proved not possible with the current data and the outliers present in it. However, there was a strong correlation (r = 0.77) between the total number of patches in the landscape (a measure of fragmentation) and the percentage of the landscape covered by urban area + bare soil, roads and grassy fields. The classification method developed for transect 1 was quite robust as few changes were needed to apply it to transect 2. To make full use of OBIA higher resolution images are required, because the 30m resolution of Landsat images was not enough to use shape in the segmentation process likely due to the small irregular fields that made up a large part of the landscape in the transects.