Radar and Optical Data Sensor Integration for Land Cover Extraction

This study evaluated the advantages of combining traditional spaceborne optical data from the visible and infrared wavelengths with the longer wavelengths of radar. Several East African landscapes and one from Nepal including areas of settlements, natural vegetation and agriculture were examined. For these study sites, multisensor data sets were digitally integrated with training and truth information derived from field visits. Three general processing strategies were examined. The number of radar bands were expanded by use of texture measures to create new bands. Various spatial aspects of the data were examined including texture window size and postclassification filtering. Finally, different integration strategies including number of sensor bands, weighting of bands and principle components were explored. These results show that radar results can be improved by various manipulations such as derived texture measures and filtering and the potential of optical/radar merger for mapping basic land use/cover patterns in different environments. Given the range of landscapes examined, the results of this study should have wide applicability. The radar results can be improved by various manipulations such as derived texture measures and filtering. An important observation is that there are no consistent strategies for using radar and sensor integration that will provide the best results. As with independent optical data, the most productive processing strategies may be site and data specific.