Land Use Classification and Analysis Using Radar Data Mining in Ethiopia

Abstract

Land use classification in tropical areas, is hindered by frequent cloud cover which limits the availability of optical satellite data. Satellite-borne radar is a possible alternative to optical data for land use classification in tropical areas. However, radar data is affected by noise (i.e., speckle) that must be minimized before its use in land classification. Median, Lee-Sigma, and Gamma-MAP de-speckling techniques were applied to Fine Beam, Dual polarization (FBD) PALSAR radar data acquired over central Ethiopia. Each of the de-speckled images were then subjected to supervised classification using Maximum Likelihood, C4.5, Multilayer Perceptron and Stacking techniques. Validation results indicated that de-speckling techniques improved classification accuracy by up to 25%, 20% and 16% using Gamma-MAP, Median and Lee-Sigma respectively. Gamma-MAP de-speckling in combination with the Multilayer perceptron classifier achieved the best overall classification accuracy at 91.2%. This study proved the importance of radar data as an alternative source of information for land use classification in the tropics. Further research should focus on the application of radar data for forest fire detection and crop classification. The use of fully polarized radar data has the potential to further improve the proposed land use classification in tropical countries.