Presentation

< back to presentation overview


Details of the presentation
Presentation Poster presentation
Title Object-based analysis for imagery of the farmland surrounding Kakamega Forest for socio-economic planning scenarios

PDF Download ---
Short title Object-based image analysis for socio-economic scenarios

Author(s) Lübker, T.; Schaab, G.

Presenting author Lübker, T.

Institution(s) Faculty of Geomatics, Karlsruhe University of Applied Sciences

Keywords QuickBird, land use in farmland, object-based image analysis, classification approach, spatially explicit modelling

Abstract Very high resolution (VHR) QuickBird satellite imagery has been acquired for simulating planning scenarios of rural livelihood in the agricultural matrix surrounding Kakamega Forest, western Kenya. In one of the world's most densely populated rural areas planning scenarios for the further development is a key issue in biodiversity conservation. While the impact of local population on forest disturbance is hoped to be minimized by establishing alternative income options, the propagation of hedges and trees should contribute to an enhanced biodiversity in the farmland next to the forest. Before such spatially explicit simulations based on the imagery can be conducted, it needs to be classified. As a prerequisite, a ground truthing was carried out in October 2007 in order to collect reference data and to gain a better knowledge about the structure of the farmland. For the classification of the VHR satellite imagery an objected-based approach will be followed as it can often achieve better results than a pixel-based approach. However, the determination of significant segmentation parameters and the selection of appropriate object features are very time-consuming and involve a lot of trial-and-error runs. The poster presents a methodical classification procedure employing a rather objective optimization of segmentation parameters as well as an in large parts automated selection of object features being of relevance for a classification class and for defining adequate thresholds. The methodology is to be elaborated by means of 5 areas covering 15 of the 25 farms co-operating with Biota-East. The results finally to be deduced from the complete imagery will cover structural elements like groups or lines of trees, hedges, houses, and rocks as well as information on parcels and land use, and will thus build a detailed input for the spatial modelling.

Congress Topic Land use, impact and value

Topic No. ---
Notes ---

Ref. No. 564