Forest DRAGON 2: Final Results of the European Partners
Burjack, Ina1; Traut, Kerstin1; Santoro, Maurizio2; Leiterer, Reik3; Li, Zenyuan4; Ling, FL4; Schmullius, Christiana1
1University Jena, GERMANY; 2Gamma, SWITZERLAND; 3University Zurich, SWITZERLAND; 4CAF, CHINA

The Forest DRAGON Project within the DRAGON-1 and DRAGON-2 Programmes has as objectives the monitoring of the forest ecosystem of China by means of multi-source Earth Observation (EO) data. At the end of the DRAGON-2 Programme, a huge archive of satellite synthetic aperture radar (SAR) imagery spanning the last 17 years has been collected. European Remote Sensing (ERS) data acquired between 1995 and 1998 were used to map four classes of forest growing stock volume (GSV) in Northeast and South China at 50 m spatial resolution. An update of the GSV with Envisat ASAR Strip map images acquired during 2004 and 2005 was attempted but did not produce sufficiently accurate results; for this reason it was preferred to assess changes with respect to the mid-1990s baseline in form of forest/non-forest cover. The availability of extensive datasets of Envisat ASAR ScanSAR images over China since 2004 favoured the retrieval of GSV using the BIOMASAR approach. A first attempt of using the BIOMASAR algorithm for a site in Northeast China showed that the GSV estimates presented certain agreement with the MODIS Vegetation Continuous Field tree canopy cover estimates. The BIOMASAR algorithm has been applied to Northeast China to provide estimates of GSV for the years 2005 and 2010. Assessment of the ERS and Envisat ASAR ScanSAR GSV datasets over Northeast China has been pursued in the second half of the Forest DRAGON-2 project. For this, several data pools have been collected and merged into a Geoportal. Land use / land cover (LULC) products have been used to assess the thematic plausibility of the GSV estimates. The GSV plausibility has also been assessed at multiple spatial resolutions by a multi-scale cross-comparison of the GSV product from 2005 and two other forest biophysical EO products. GSV estimates of a field campaign in 2011 have been used to validate the GSV estimates of the BIOMASAR algorithm in a small area in Northeast China. EO products have been used to confirm detected changes of GSV estimates between the two epochs of the Envisat ASAR ScanSAR data used for the GSV retrieval (2005 and 2010), The ASAR-based GSV maps have then been used to provide overall statistics of forest volume changes in Northeast China between 2005 and 2010.

The European contribution to the Forest DRAGON 2 focused on the evaluation of multi-temporal, multi-sensor and multi-scale Earth Observation images and data products within the vegetation ecosystem of Northeast China. The forest growing stock volume (GSV) map produced with ERS-1/2 coherence images for 1995-1998 and two GSV maps produced from Envisat ASAR ScanSAR data for 2005 and 2010 were inter-compared with respect to several datasets (in situ, EO images and EO data products) to assess the plausibility of the GSV estimates, the contribution to land cover mapping and the dynamics over time. For this purpose, a multi-source database was set up including in situ data and EO data products. Land use / land cover (LULC) datasets identified mis-classification of GSV in the ERS dataset primarily for cropland. An a posteriori correction of the GSV resulted in an increase of overall accuracy up to 7%. LULC products can also support the fine tuning of the algorithm to estimate GSV from ASAR data particularly in transition regions between forest and shrub land. A multi-scale cross-comparison of the GSV product of 2005 and two other EO products of forest biophysical parameters was performed to assess the GSV validity at multiple spatial resolutions. Compared to the MODIS VCF product and a map of forest canopy heights, the GSV product captures the spatial variability of the forest landscape in a more reliable way compared to the other maps even at a spatial resolution of 1 km. The ASAR-based GSV estimates were consistent and highlighted areas of change. From the two ASAR maps, slight loss of volume from 2005 to 2010 was estimated.

In conclusion, it has been found that the relation between the GSV and the tree cover product is stronger than the relation between these maps and the canopy height product. Particularly, at the 1 km scale considerable structural discrepancies were found in the canopy height map compared with the other EO products. Structural similarities between the canopy height and the GSV and tree cover maps have been detected first at 10 km pixel size. Assuming that the GSV and tree cover maps are forming a reference of forest structures, the validity of the canopy height map is at 10 km pixel size higher than at 1 km pixel size. However, the tree cover product tends to saturate with increasing GSV or canopy height values in mixed forests. Only at spatial scales greater than 10 km a substantial correlation was identifiable between tree cover and GSV or canopy height values in mixed forests. Therefore, the 10 km scale seems to form the best compromise between spatial heterogeneity and map validity for the tree cover and canopy height products. However, only 20 % of the spatial variability of the 1 km scale remain at 10 km pixel size. Concerning the GSV product, it has not been determined that the 1 km scale has a lower validity than the coarser spatial scales. The DRAGON GSV map seems to capture the forest structure in a more reliable way than the tree cover or the canopy height products even at the original spatial resolution of 1 km.

Overall statistics of area and volume based on the GSV change estimates indicate that forest resources of Northeast China decreased slightly between 2005 and 2010. The statistics consider all forest and shrubland pixels according to the GLC2000 land cover product, and GSV estimates based upon at least 20 backscatter measurements (420,501 km2). For 298,844 km2, i.e., 71% of pixels, the detected change was below the uncertainty margin of a GSV estimate. Notwithstanding changes less than 25 m3/ha, the decrease of GSV was primarily in the range between 25 to 50 m3/ha. Changes corresponding to a decrease of over 100 m3/ha were rare. GSV increase was for 98.6% of the study area below 50 m3/ha (84% below 25 m3/ha). Regardless whether changes below the 25 m3/ha are considered or not, the area characterized by GSV decrease was larger than the area characterized by GSV increase. Accordingly, the loss of volume exceeded the increment over the five years. The GSV change map between 2005 and 2010 indicates a net loss of volume of 12,122 m3 (8,069 m3 for changes greater than 25 m3/ha).