Data fusion for reconstruction of a DTM, under a woodland canopy, from airborne L-band InSAR

This paper investigates the utility of different parameters from polarimetric interferometric synthetic aperture radar (InSAR) data for the identification of ground pixels in a woodland area to enable accurate digital terrain model (DTM) generation from the InSAR height of the selected ground hit pixels. The parameters assessed include radar backscatter, interferometric coherence, surface scattering proportion (based on Freeman–Durden decomposition), and standard deviation of the interferometric height. The method is applied to Monks Wood, a small seminatural deciduous woodland in Cambridgeshire, U.K., using airborne E-SAR data collected in June 2000. The 1428 variations of SAR-derived terrain models are validated with theodolite data and a light detection and ranging-derived DTM. The results show that increasing the amount of data used in the DTM creation does not necessarily increase the accuracy of the final DTM. The most accurate method, for the whole wood, was a fixed-window minimum-filtering algorithm, followed by a mean filter. However, for a spatial subset of the area using the $upsilon_{3}$ backscattering coefficient to identify ground pixels outperforms the minimum filtering method. The findings suggest that backscatter information may often be undervalued in estimating terrain height under forest canopies.