Challenges in classifying diverse and highly structured landscapes with satellite images to infer land cover and land use change

Valeska Scharsich1, Dennis Otieno Ochuodho2, Christina Bogner3
1 Ecological Modelling and Soil Physics, University of Bayreuth
2 Department of Plant Ecology, University of Bayreuth; Department of Botany, Jaramogi Oginga Odinga University of Science and Technology, KE
3 Ecological Modelling, University of Bayreuth

O 2.1 in Understanding Ecosystem Services: Geoecological processes/functions with value for the society

12.10.2017, 11:00-11:15, H36, NW III

Introduction

Changes of land use and land cover (lulc) are often driven by the growth of human population. In the Lambwe valley, Kenya, the most important reason for growth was the control of the tsetse fly, the biological vector of trypanosomes. Due to the increasing population the cultivated area expanded and the conflict between the competing ecosystem services food production and erosion prevention deepens. In particular, more and more areas at higher elevations are cultivated and land degradation increases. Here, we investigate possible effects of the farming pressure on the lulc in the Lambwe valley and its Ruma National Park.

Material and Methods

We analysed the surface reflectance of three Landsat images of the Lambwe valley from 1984, 2002 and 2014. We inferred ground data for past lulc from recent observations combining WorldView-1 images and change detection. By supervised classification with Random Forests, we identified five lulc types: “dense forest” and “light forest”, prominent on the hills; “dense shrub” and “savanna”, found in the National Park; and “agriculture”, embracing crop and fallow land. Subsequently, we compared the three classifications and identified changes of lulc classes that occurred between 1984 and 2014.

Results

We observed an increase of “agriculture” in the whole Lambwe valley by about 12%, while “dense forest” dropped by 6% and “savanna” by 9%. The National Park itself shows a quite diverse but rather stable composition. Some areas were misclassified as “agriculture” which is probably due to the fire management in the park. Burned area could be classified as “agriculture” because of absence of vegetation.

Conclusions

Our results show that agriculture expands on the upper hills and consequently on steeper slopes. Therefore, the risk of erosion is increasing, whereas the formerly “dense forest” has almost disappeared. Nevertheless, the protection of the National Park is quite strong so that no abrupt changes could be observed.



Keywords: land use and land cover remote sensing change detection Ruma National Park Lambwe valley

Export as iCal: Export iCal