Perera, Kithsiri and Herath, Srikantha and Apan, Armando and Lal, Samarakoon (2010) Application of Modis 250m images without in situ observations for mapping Mekong River Basin land cover. In: ISPRS Technical Commission VIII Symposium 2010, 9-12 Aug 2010, Kyoto, Japan.
|HTML Citation||EndNote||Dublin Core||Reference Manager|
Full text not available from this archive.
Mekong River runs from Hengduan Mountains in central-west China to Vietnam covering 805,604 sq km of land by its basin. The Lower Mekong Basin (LMB), the region mapped in this study, covers nearly 3/4 of the entire basin. About 90% of the population and agricultural activities of the Mekong River basin is located in this fertile LMB which faces disastrous floods almost annually. Mapping LMB at moderate resolution gives number of advantages for studies of flood mitigation and land utilization. However, compiling a cloud free mosaic and collecting ground truth data for training samples and map validation make map production process a challenging task. This study utilized MODIS 250m image data of the region obtained in 2005 February. Dry weather in Jan-Apr makes the sky relatively free of clouds and 2005 February also had fewer disturbances coming from smoke of biomass burning. The methodology of the study substantially relied on high resolution images in Google Earth for collection of training sample for supervised classification and accuracy assessment. Arc GIS generated KMZ file of unclassified and classified maps used to overlay image and map on Google Earth for identifying training site and field information extraction for accuracy assessment. Also ground information collected by a different research projects in 2008 were combined with information gathers from Google Earth images. The classified map showed 29.2% of the LMB under forest, 36.5% under Scrubland, when combined its Highland and Lowland subcategories. Three subcategories of paddy cultivated area covered 27.9% of LMB. Accuracy assessment conducted with randomly selected 200 points against high resolution images gave an overall accuracy of 80.7% in major land cover classes. According to the 250m resolution, urban features have not clearly separated though large urban areas like Phnom Penh and Can Tao have accurately classified. The methodology of this study produced a noteworthy success in classifying land cover of large areas like LMB, without expensive data sources and difficult and costly field investigations.
Archive Staff Only: edit this record