Object-based image analysis for distinguishing indigenous and exotic forests in coffee production areas of Ethiopia

Abstract

Indigenous forest management and conservation has a major importance for ecosystem services in the Eastern Afromontane Biodiversity Hotspots. In the southwestern highlands of Ethiopia, indigenous forests are particularly relevant for coffee producers, given that Coffea arabica grows as understory shrub in these forests. Currently, identifying and mapping understory coffee using remote sensing is still considered a challenging task because exotic tree plantations are largely overspread among indigenous forests. In this paper, a rule set was developed for recognizing indigenous forests from high-resolution satellite imagery using object-based image analysis (OBIA). The study applies a multiscale approach, in which aerial photographs (0.5 m), SPOT-5 satellite image (2.5 m), and field observations were integrated to discriminate indigenous from exotic forests. The rule-set combined segmentation (multiresolution, spectral difference, and contrast splitting), classification algorithms, and knowledge-based threshold functions. Moreover, principal component analysis and imagery texture indexes (e.g., homogeneity) were used to feed the classification algorithms. The results show that the applied methodology could separate indigenous from exotic forests with an overall accuracy of 84.3 % based on a fourfold cross-validation. The user and producer accuracy of indigenous forest were 84.7 and 94.4 %, respectively. On the other hand, exotic forest was classified with user accuracy of 87.9 % and producer accuracy of 61.9 %. This study contributes not only to coffee and environmental researchers but also benefits local communities by allowing the identification of indigenous and exotic forest areas, and leading to better informed natural resources management and conservation.