Jornal de Sensoriamento Remoto e GIS

Abstrato

Land Cover Land Use (LCLU) Classification Methods in Semi-Arid Botswana

Rejoice Tsheko

This paper presents Land Cover Land Use (LCLU) detected from a Landsat 8 (OLI) using two classification schemes namely Maximum Likelihood Algorithm (MLA) and Artificial Neural Networks (ANNs). Analysis was carried out using two, three and eight features (surface reflectance and indices). For all classifications, the overall accuracy and kappa statistic varied from 93.81% and 0.89 to 99.38% and 0.99, respectively. The highest classification accuracies were obtained by either using all eight features or two features (indices only) for both classification schemes. This demonstrates the importance of Normalized Difference Vegetation Index (NDVI) and Normalized Difference Buildup Index (NDBI) in LCLU mapping. The two indices are robust enough to be used to detect shrubs, trees, water, and buildup in a satellite image. Further, the ANNs classifier is also robust enough to be used for this classification. Although the MLA classifier used both the mean values and variance of the features, the ANNs classifier only used the mean values of the features. This is a demonstration of data fusion in a normalized scale -1.0 to 1.0. This work also demonstrates that acceptable classification accuracies can be achieved with fewer spectral channels