Detecting Shifts in Agricultural Landscape Patterns of Hawassa, Ethiopia; An Assessment of Land Cover Change Between 1984 – 2014 Using Object-Based Image Analysis and Landscape Metrics

Organisator Laboratory of Geo-information Science and Remote Sensing

do 28 mei 2015 09:30 tot 10:00

Locatie Gaia, building number 101
Droevendaalsesteeg 3
6708 PB Wageningen
+31 317 48 16 00
Zaal/kamer 1

by Kirstin Abraham (Germany)


South Ethiopian landscapes have undergone rapid and mostly uncontrolled landscape change in the last few decades with high social, economic and ecological impact. Changes in cropland composition are of particular interest to explain the link between landscape and potential natural pest control. We assessed landscape change in the Hawassa area in terms of land cover and land structure in the last three decades (1984, 1998, 2014) using Landsat TM and Landsat OLI/TIRS imagery. Eleven classes were assessed utilizing a hybrid approach of OBIA and pixel-based classification. Annual and perennial crops were separated using NDMI differencing, producing overall accuracies of 77 % (1984) and 75 % (2014). The results showed high dynamics and a clear shift in cropland cultivation towards perennial crops. Largest changes in the study area were seen in rising proportions of perennial crop and built up (+204 %, +616%) and decrease of annual crop, grassland and bare soil (-77 %, -82 %, -74 %). Natural land cover was thereby replaced by cropland. A large east-west difference was observed and substantiated by using landscape metrics Simpson Diversity, Contagion, Proximity Index, Number of patches and Edge density. Western areas showed least crop diversity in 1984 with strong dominance of annual crops. The introduction of perennial crops resulted in a shift of dominance towards mixed crop classes. Eastern areas were most diverse and fragmented in 1984, but showed trends in higher aggregation as perennial crop is strongly increasing towards 2014. Rising population pressure and cash cropping can be possible explanations of the observed change. The combination of OBIA and Landscape Ecology is promising, but requires a good data choice. Landsat data is not able to detect rapid small-scale changes of the Ethiopian landscape due to the relatively large pixel size compared to small field sizes. Thus, mixed crops were created to acknowledge the presence of mixed pixels. The use of OBIA was neither effective nor feasible for the purpose of cropland classification in our study. We suggest the use of VHR data to further assess the fragmentation of the landscape and other ecologically important landscape elements such as hedgerows and tree patches. The results of this study can help understanding impacts of landscape change on biodiversity and driving forces of pest incidence in the study area.

Keywords: Land Cover Change; Landscape Assessment; Object-based Image Analysis; Landscape Ecology; Perennial Cropland; Annual Cropland; Ethiopia