Error analysis and correction for extracting the forest height from airborne C-band interferometric SAR and national elevation datasets

Abstract

The Interferometric Synthetic Aperture Radar (InSAR) signal is returned from the canopy of the obscuring trees instead of bare ground when land is covered by forests. Therefore, the difference between an InSAR elevation and a bare earth model might contain information on forest height. The objective of this paper was to investigate if the difference between an airborne C-band InSAR from the National Aeronautics and Space Administration (NASA)/Jet Propulsion Laboratory (JPL) and a bare earth model of 1/3 arcsecond National Elevation Datasets can be used for regional forest height estimation. The error sources of vertical offset, uncompensated roll angle, residual vertical bias, and scattering phase centre height conversion were analysed and corrected in this estimation. The results were validated by the leastsquare linear regression analysis between Light Detection and Ranging (LiDAR) and the estimated height at different forest stand sizes within different slope categories. In areas with slopes less than 5 , the correlation coefficients increased when forest stand sizes increased. In the area of slope ranging from 5 to 10 , a similar trend of increasing correlation coefficients with increasing stand size could also be observed, but with smaller corresponding correlation coefficients than those of slope 0-5 . In areas with slopes larger than 10 , the correlation coefficients were very poor. These results indicate the difference between airborne C-band InSAR and the accurate bare earth model has the potential for regional forest height estimation in flat areas with a minimum unit of 3750-5000 m 2 . However, to accurately estimate forest height in a mountainous terrain a solution must be found to correct the significant amount of noise caused by the terrain in these areas.