Please use this identifier to cite or link to this item:
|Title:||Combinig Multi-Sensor Medium Resolution Satellite Imagery for Forest Cover Change Assessment in Central Africa|
|Authors:||DESCLÉE BAUDOUIN; SIMONETTI Dario; MAYAUX Philippe; ACHARD Frederic|
|Citation:||Proceedings of the ESA Sentinel-2 Preparatory Symposium|
|Publisher:||European Space Agency|
|Type:||Articles in periodicals and books|
|Abstract:||Forest monitoring is crucial over tropical regions to assess the evolution of forest cover changes and give figures about deforestation. Based on a systematic sample of image extracts, a processing chain had been developed for producing deforestation estimates over the years 1990-2000-2005. Whereas this monitoring exercise was based on Landsat images, limitations in Landsat image availability for the year 2010 over Central Africa required alternative imagery. Given its high revisit period and characteristics close to the Landsat TM sensor, DMC images are considered in this paper to replace Landsat TM data gaps over Central Africa. However the classification module of the existing processing chain is based on Tasseled Caps (TC-based) analysis of Landsat TM imagery and needs to be adapted to such data in order to be sensor-independent. This adaptation is described in this paper. A sub-sample of the available image extracts has been used for the selection of the best object-based features through the analysis of existing Land-Cover maps for 1990 and 2000. The processing chain has been adapted for the production of Land-Cover change maps for year 2010. The resulting maps from the two methods, original TC-based classification and adapted Multi-Sensor approach, have been compared and evaluated. The overall accuracies of the 2010 Land-Cover classification results are 90% for the TC-based approach and 92% for the Multi-Sensor approach. When considering only objects for which label is changing between 2000 and 2010, the accuracies of the 2010 LC classifications are 45% and 72% for the TC-based and Multi-Sensor approaches respectively. These results show that, even with lower radiometric quality of DMC imagery the performance of the automated classification has been improved thanks to an appropriate selection of object-based features. As similar adaptation will be developed for other satellite imagery such as SPOT and Rapid-Eye in order to be sensor-independent, the future adaptation to Sentinel-2 data will be very easy using the same approach.|
|JRC Directorate:||Sustainable Resources|
Files in This Item:
There are no files associated with this item.
Items in repository are protected by copyright, with all rights reserved, unless otherwise indicated.