Estimation of Forest Biomass and Faraday Rotation using Ultra High Frequency Synthetic Aperture Radar

• Main Author: Sandberg, Gustaf
• Language: unknown
• Published: 2013
• Subjects: Remote Sensing; Other Earth and Related Environmental Sciences; Forest Science; Forest Biomass; Ultra High Frequency; Faraday Rotation; Synthetic Aperture Radar; Faraday; Northern Sweden
• Online Access: https://research.chalmers.se/en/publication/185265

Synthetic Aperture Radar (SAR) data in the Ultra High Frequency (UHF; 300 MHz – 3 GHz)) band have been shown to be strongly dependent of forest biomass, which is a poorly estimated variable in the global carbon cycle. In this thesis UHF-band SAR data from the fairly flat hemiboreal test site Remningstorp in southern Sweden were analysed. The data were collected on several occasions with different moisture conditions during the spring of 2007. Regression models for biomass estimation on stand level (0.5-9 ha) were developed for each date on which SAR data were acquired. For L-band (centre frequency 1.3 GHz) the best estimation model was based on HV-polarized backscatter, giving a root mean squared error (rmse) between 31% and 46% of the mean biomass. For P-band (centre frequency 340 MHz), regression models including HH, HV or HH and HV backscatter gave an rmse between 18% and 27%. Little or no saturation effects were observed up to 290 t/ha for P-band. A model based on physical-optics has been developed and was used to predict HH-polarized SAR data with frequencies from 20 MHz to 500 MHz from a set of vertical trunks standing on an undulating ground surface. The model shows that ground topography is a critical issue in SAR imaging for these frequencies. A regression model for biomass estimation which includes a correction for ground slope was developed using multi-polarized P-band SAR data from Remningstorp as well as from the boreal test site Krycklan in northern Sweden. The latter test site has pronounced topographic variability. It was shown that the model was able to partly compensate for moisture variability, and that the model gave an rmse of 22-33% when trained using data from Krycklan and evaluated using data from Remningstorp. Regression modelling based on P-band backscatter was also used to estimate biomass change using data acquired in Remningstorp during the spring 2007 and during the fall 2010. The results show that biomass change can be measured with an rmse of about 15% or 20 tons/ha. This suggests ...