Multi-Sensor Monitoring System for Forest Cover Change Assessment in Central Africa

Forest monitoring from Earth observation is crucial over tropical regions to assess forest resources status and provide up-to-date estimates of deforestation rates. Based on a systematic sample of 20 × 20 km size sites where medium resolution satellite imagery has been acquired, a processing chain had been developed at JRC for producing deforestation estimates between years 1990, 2000 and 2005. Whereas this monitoring exercise was based on Landsat imagery, limitations in Landsat availability over Central Africa for year 2010 required alternative imagery such as the Disaster Monitoring Constellation (DMC). The classification module of the existing JRC processing chain is based on Tasseled Caps analysis (TC-based). We adapted this module to DMC imagery. The most suitable object-based features for the classification are selected through the analysis of a sub-sample of the existing Land Cover (LC) maps for years 1990 and 2000. The processing chain is adapted for the production of 2000-2010 LC change maps and the resulting 2010 LC maps produced by the two methods, original TC-based and adapted Multi-Sensor, are assessed and compared. The overall accuracies are 90% and 92% for the TC-based and Multi-Sensor approaches respectively, with significant improvement when considering only changed objects (46% and 72% respectively). These results show that, even with DMC imagery lower radiometric quality (compared to Landsat) automated classification can provide accurate LC maps thanks to an appropriate features selection. Similar adaptation need to be developed for other satellite imagery such as SPOT and RapidEye.

DESCLÉE Baudouin;  SIMONETTI Dario;  MAYAUX Philippe;  ACHARD Frederic; 

2013-05-07

IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC

JRC71834

1939-1404,   

http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=6420968,    https://publications.jrc.ec.europa.eu/repository/handle/JRC71834,   

10.1109/JSTARS.2013.2240263,