Science Service System

Summary of Proposal MTH2422

TitleEvaluation of TerraSAR-X data for mapping deforestation and forestdegradation in Brazilian Amazon
Investigator Waske, Björn - Freie Universität Berlin, Insitut für Geographische Wissenschaften
Team Members
Prof. Dr. Hostert, Patrick - Humboldt Universität zu Berlin, Geomatics Lab
SummaryApplication ofremote sensing methods in the context of REDD+ and on subtropical study sitesoffers many challenges. This includes natural factors like seasonality as wellas anthropogenic factors like diverse land uses or small scaled processes, e.g.selective logging. While the main objective is the derivation of meaningfulland use / land cover maps (LULC), modern methods are required to address thesechallenges. Therefore, we aim to develop an expressive classification legend,and implement latest machine learning methods using multi-scaled,multi-temporal X-band and C-band data to be independent from season-dependentoptical data. Classificationwill be implemented through a combination of novel approaches such as ImportVector Machines, Markov Random Fields (Roscher et al. 2012; Waske and Braun2009; Waske and Van der Linden 2008; Waske et al. 2012). For a later fusion ofSAR and optical data, the concept of co-training is considered. This has notbeen integrated into a remote sensing context to this date. Using advancedmethodology, we will be able to perform a reliable evaluation of the truepotentials of TerraSAR-X for subtropical land cover mapping. Maps will begenerated at different spatial resolutions, using a modular and hierarchicalclassification key to depict local and regional forest-use and -cover classes.

Back to list of proposals

© DLR 2004-2016