Abstract (EN):
Earth Observation Satellite (EOS) data have a great potential for land cover mapping, which is mostly based on high resolution images. However, in tropical areas the use of these images is seriously limited due to the presence of clouds. This paper evaluates the ability of temporal-based image classification methods to produce land cover maps in tropical regions. A new approach is proposed for land cover classification and updating based exclusively on temporal series data, illustrated with a practical test using SPOT VEGETATION satellite images from 1999 to 2011 for Rondonia (Amazon), Brazil. Using the GLC2000 as reference, a Normalized Difference Vegetation Index (NDVI) time series of 15 distinct land cover classes (LCC) were created. Two classifiers were used (Euclidean Distance and Dynamic Time Warping) to produce maps of land cover changes for 1999-2011. Due to the difficulties in discriminating 15 LCC in the Amazon region, a hierarchical aggregation was performed by joining the initial classes gradually up to four broad classes. The land cover changes in the 1999-2011 period were evaluated using criteria based on the classification results for the individual years. The comparison with reference data showed consistent results, proving that this approach is able to produce accurate land cover maps using exclusively temporal series EOS data.
Language:
English
Type (Professor's evaluation):
Scientific
Contact:
dr.arlete@gmail.com
No. of pages:
13