Landcover Classification of Luambe National Park and Surrounding Game Management Areas, Zambia

Abstract

Luambe National Park (LNP) is a small, remote and relatively undeveloped national park in the Luangwa Valley, eastern Zambia. Baseline ecological data have been lacking and few publications relating to the ecology of the national park and surrounding game management area (GMA) exist. The aim of this work was to produce an accurate landcover classification that could be used as a baseline dataset for monitoring ecological health in the park. Fuzzy set theory was used to classify remotely sensed Landsat 7 ETM+ imagery with a spatial resolution of 30m. A ground survey to collect training and test data was conducted in August and September 2005. The most recent L1G Landsat dataset was obtained from the Global Land Cover Facility maintained by the University of Maryland (acquisition date 04/10/2001, path 170, row 069, cloud cover 0%). Bands one, two, three, four, five and seven were used for the classification. Erdas Imagine 8.4 (Leica Geosystems AG, Atlanta, USA) was used to perform a supervised classification using the maximum likelihood classifier with activation of the fuzzy classification function. The eight-layered output dataset was then processed to a single-layer hard classification using the fuzzy convolution facility. An error matrix was produced and producer’s and user’s accuracies calculated for each class. Nine landcover classes were identified and the overall accuracy of the classification was 71.2% (95% CI: 65.3-76.7%). The overall kappa statistic was 0.67 and the estimator of kappa (KS) for stratified random sampling was 0.74. This is the complete dataset covering both the national park and the surrounding GMA. The dataset for the national park alone can be found using the identifier http://hdl.handle.net/10672/606. Developed as part of Neil Anderson’s thesis entitled ‘An Investigation into the Ecology of Trypanosomiasis in the Luangwa Valley, Zambia’ (PhD thesis, University of Edinburgh 2009). Landsat imagery was processed using Erdas Imagine 8.4 (Leica Geosystems AG, Atlanta, USA). Funding was provided by the Royal Zoological Society of Scotland and the United Kingdom Department for International Development Animal Health Programme (DFID-AHP). Other. This dataset was first accessioned in the EDINA ShareGeo Open repository on 2014-09-25 and migrated to Edinburgh DataShare on 2017-02-22.