Mapping Fuelwood Distribution Pattern in North-Western Zimbabwe using a Neural Network

要旨

This study sets out to model and describe the pattern of fuelwood. Population growth has put immense pressure on natural woodlands, which are dwindling in Zimbabwe and causing a threat to energy supply for rural people. Techniques to model fuelwood supply pattern in an efficient manner are therefore critical to rural planning and development. The fuelwood is mapped with remotely sensed (Landsat TM) and geographical data using an artificial neural network. Basal areas of tree species that are used for fuelwood in Zimbabwe were measured in the field and used as training samples for the neural network. The fuelwood estimates were linked with a spatial database consisting of ancillary geographic variables (rainfall, slope, soil depth, distance from settlement, aspect, cover) and Landsat TM data. Remotely sensed and ancillary variables provided data for the neural network to infer the most likely amount of fuelwood occurring at any given pixel. Several adjustments wer e made to the system parameters of the neural network to improve the mapping accuracy. To test the utility of the neural network, other algorithms (parallelepiped and minimum distance classifiers) were also applied to classify fuelwood. The map output of fuelwood by the artificial neural network had a significantly higher mapping accuracy than the maps produced by conventional classification algorithms. The result demonstrates the capability of a neural network, in combination with remotely sensed data as well as ancillary data to map fuelwood in natural woodlands.