Image classification

Sentinel-1 and 2 data fusion for large-scale land cover mapping using object-based classification

Land cover monitoring and mapping is essential for a range of applications such as land management and planning, hydrological modelling, climate models or environmental degradation monitoring. Remote sensing and digital image processing are the most practical and cost-effective means of enabling land cover mapping and monitoring at a range of spatial, temporal and thematic scales; particularly, with the availability of free optical and SAR images from the Sentinel missions. The Department of Remote Sensing of the CTTC investigates the production of land cover maps by combining high spatial resolution optical reflectance with Synthetic Aperture Radar (SAR) backscatter measurements. Specifically, Sentinel-1 and Sentinel-2 data fusion is exploited in an attempt to identify a variety of land cover classes for different applications. BCN_S1-S2_BCN_test_poster_LR

Land cover map of Barcelona metropolitan area and adjacent territory produced using Sentinel-1 and Sentinel-2 images. The classes identified, corresponding to the Level 1 Corine land cover classification squeme, are artificial surfaces, agricultural areas, forest and semi-natural areas and water bodies. An object-based supervised classification using a Radom Trees classifier was performed. Accuracy assesment was performed using 6000 randomly selected points and the overall accuracy obtained is 80.27%.

 

References

F. C. Conesa, , A. L. Balbo, M. Madella, Use of Satellite SAR for Understanding Long-Term Human Occupation Dynamics in the Monsoonal Semi-Arid Plains of North Gujarat, India , Remote Sensing, Vol. 6, pp. 11420-11443, December 2014, http://www.mdpi.com/2072-4292/6/11/11420.