Improving the accuracy of high spatial and temporal resolution forest change mapping using reliable reference data bases to support REDD

We investigate the improvement of remote sensing deforestation mapping, by considering their error sources and types, their processing schemes and their spatio-temporal properties.

The PhD thesis approaches the essential concern of remote sensing products credibility. Politicians are reluctant about considering remote sensing as a reliable method to assess forest changes. One reason is that most remote sensing studies essentially focus on the demonstration of technology rather than generating results with proven accuracy and understanding error sources. Researching a comprehensive and comparative accuracy assessment and the quantification of error sources assists steering remote sensing into a key data source for societal decisions.

Radiometric correction, data and environment properties profoundly affect the success when mapping change of tropical forests. By inverting the traditional approach of using standardized pre-processing schemes and then performing method development, we suggest to understand capabilities of one method (BFAST type models) within a framework of varying sites (Africa, South East Asia and South America) and correction schemes using high quality reference data.

Mapping errors are modelled using a set of error parameters quantifying time series data variance and availability, atmospheric contamination and forest edge effects as leading drivers of error. However research aims also towards the choice of the best deforestation indices and the applicability of spatial parameters including spatial-temporal analysis.