On the generation of a forest biomass map for Northeast China: SAR interferometric processing and development of classification algorithm
Maurizio Santoro(1) , Oliver Cartus(1)
, Christiane Schmullius(1)
, Urs Wegmüller(2)
, Charles Werner(2)
, Andreas Wiesmann(2)
, Yong Pang(3)
, and Zengyuan Li(3)
(2) Gamma Remote Sensing AG, Worbstrasse 225, 3073 Gümligen, Switzerland
(3) Chinese Academy of Forestry, -, Beijing, China
The ERS-1/2 interferometric SAR (InSAR) tandem coherence has been demonstrated to be the most suitable spaceborne remote sensing quantity for the estimation of forest aboveground biomass, in particular at northern latitudes. On one hand studies at test sites have showed that coherence acquired under winter stable conditions and a multi-temporal approach lead to estimates with errors of the order of 20-25%. On the other hand, using one data take, the SIBERIA Project has achieved a 1 Mio km2 large map of forest biomass levels in Central Siberia with classification accuracy greater than 90%, this currently being the best demonstration of the use of ERS-1/2 tandem coherence for large area mapping of forests using remote sensing data. The experience gained throughout more than a decade of forestry investigations based on ERS-1/2 coherence images is currently exploited to map biomass levels of a 1,5 Mio km2 region in Northeast China within the recently started ESA-MOST cooperation project Forest DRAGON.
For the generation of the biomass map all ERS-1/2 image pairs acquired during the tandem mission with baseline below 400 m are used. In total more than 250 image pairs are considered. Most of the region is covered with at least one pair, several areas being covered multi-temporally with one or more winter data takes. To achieve end products of high quality InSAR processing has to be carried out very carefully. Processing consists of co-registration at sub-pixel level, range and azimuth band filtering, slope estimation and computation of coherence using an adaptive window size. For each processing step indicators are used to assess the quality of the results. The coherence and the corresponding calibrated intensity images are geocoded to 50x50 m pixel size. The geolocational accuracy is assessed by means of internal processing parameters and a measure of overlap between neighbouring scenes. Final products consists of frames of high-resolution coherence and backscatter images as well as images of local incidence angle, pixel area normalisation factor and local topography. Data quality and geolocational accuracies are very high and coherence contrast is in most cases at optimal level for forestry applications.
For biomass classification the SIBERIA algorithm is considered. Adaptation of the algorithm following the dependence of coherence upon seasonal conditions and the strong topographic features of Northeast China are the most important tasks that need to be taken care of before proceeding to classification. At the workshop the first results on the development of the algorithm for the study region will be presented.