|START Conference Manager|
With the continuing development of space-based instruments, new opportunities for earth observation at better spatial and temporal resolution have been provided. One of the recent examples is the launch of Sentinel satellite series with a large number of potential applications. In particular, the utilization of Sentinel-2A and 2B for mapping land cover/use features is yet to be discovered as higher spatial resolution maps at 10m compared to CORINE (100m) can be generated. In this study, the Sentinel 2 data from 2017 is used to map land cover features across Copenhagen, Denmark according on the CORINE nomenclature. Several machine learning-based supervised and unsupervised classification methods including random forest, support vector machines, decision tree, and maximum likelihood are applied to generate high quality land cover data. The models are also cross-compared and their performance is discussed. By having collected ground truth data, the quality assessment of the generated land cover maps from different methods shows accuracies above 80%. This proves a promising perspective for the integration of Sentinel data in land monitoring efforts. Our conclusions draw attentions to the use of Sentinel data to generate fine-scale land cover maps across other landscapes as well.
Topic Area: [1.2] Environmental monitoring and assessments Abstract Type: Oral Presentation and paper in IJSDIR
START Conference Manager (V2.61.0 - Rev. 4840)