Mapping overused slopelands from SPOT data using support vector machines and artificial neural networks

Overuse of slopelands has become a major concern for land resources managers in Taiwan. Monitoring the overuse of slopelands is an important activity to mitigate environmental impacts. An approach for monitoring overused slopelands in central Taiwan using SPOT-5 data in 2008 is developed. Data processing consisted of four main steps: (1) data preprocessing, (2) image classification by support vector machines (SVM) and artificial neural networks (ANN), (3) accuracy assessment of the classification results using ground reference data, and (4) investigation of overused slopeland areas. The results revealed that SVM gave slightly better classification results, when compared with ANN. However, the comparison results produced by the Z-test indicated that there was no statistical difference between the two classification methods. The overall accuracy and Kappa coefficient achieved by SVM were 93.7% and 0.88%, respectively, while those achieved by ANN were 93.1% and 0.86%, respectively. The classification map produced by SVM was compared with the forestland suitability map to examine the overuse of slopeland areas. The results showed that approximately 15.6% of the slopeland area suitable for forests was identified as agricultural areas. These overused areas were verified by visual interpretation of aerial photos and field survey data. © 2013 Society of Photo-Optical Instrumentation Engineers (SPIE).