Techniques for wide-area mapping of forest biomass using radar data:Dissertation

• Main Author: Rauste, Yrjö
• Format: Book
• Language: English
• Published: VTT Technical Research Centre of Finland 2005
• Subjects: wide-area mapping; remote sensing; Synthetic Aperture Radar; forest biomass; SAR; Polarimetry; mosaicking; forests; backscattering; Northern Sweden
• Online Access: https://cris.vtt.fi/en/publications/b60da896-5117-483b-aecc-b929f4a152e3; https://publications.vtt.fi/pdf/publications/2005/P591.pdf

Aspects of forest biomass mapping using SAR (Synthetic Aperture Radar) data were studied in study sites in northern Sweden, Germany, and south-eastern Finland. Terrain topography - via the area of a resolution cell - accounted for 61 percent of the total variation in a Seasat (L-band) SAR scene in a hilly and mountainous study site. A methodology - based on least squares adjustment of tie point and ground control point observations in a multi-temporal SAR mosaic dataset - produced a tie point RMSE (Root Mean Square Error) of 56 m and a GCP RMSE of 240 m in the African mosaic of the GRFM (Global Rain Forest Mapping) project. The mosaic consisted of 3624 JERS SAR scenes. A calibration revision methodology - also based on least squares adjustment and points in overlap areas between scenes - removed a calibration artifact of about 1 dB. A systematic search of the highest correlation between forest stem volume and backscattering amplitude was conducted over all combinations of transmit and receive polarisations in three AIRSAR scenes in a German study site. In the P-band, a high and narrow peak around HV-polarisation was found, where the correlation coefficient was 0.75, 0.59, and 0.71 in scenes acquired in August 1989, June 1991, and July 1991, respectively. In other polarisations of P-band, the correlation coefficient was lower. In L-band, the polarisation response was more flat and correlations lower, between 0.54 and 0.70 for stands with a stem volume 100 m3/ha or less. Three summer-time JERS SAR scenes produced very similar regression models between forest stem volume and backscattering amplitude in a study site in south-eastern Finland. A model was proposed for wide area biomass mapping when biomass accuracy requirements are not high. A multi-date regression model employing three summer scenes and three winter scenes produced a multiple correlation coefficient of 0.85 and a stem volume estimation RMSE of 41.3m3/ha. JERS SAR scenes that were acquired in cold winter conditions produced very low correlations ...