Random forest classifier for extracting water bodies from pansharpened image to detect surface water changes

Abstract

Change detection from time series multispectral Landsat imagery has been an active research in remote sensing for several years to monitor the ecosystem, environment, climate and so on. This study is focused on detecting the changes in surface water by the integration of fusion and image classification techniques in multi-temporal multispectral Landsat images. The panchromatic band and the multispectral band of Landsat OLI and TM images respectively, were fused using undecimated wavelet transform to get the pan-sharpened image. Then classification techniques like Maximum Likelihood, Support Vector Machine, Artificial Neural Network and Random Forest were employed for extracting the water pixels and changed pixels. The performances of these classification techniques were analyzed based on metrics such as overall error, commission error, precision, recall, overall accuracy, kappa coefficients and the results show that the application of random forest classifier on pansharpened image outperforms in extracting the water pixels and also in highlighting the changes with maximum accuracy.