Frontiers in Forests and Global Change (Sep 2022)
Use of TanDEM-X PolInSAR for canopy height retrieval over tropical forests in the Western Ghats, India
Abstract
Canopy height is a critical parameter in quantifying the vertical structure of forests. Polarimetric SAR Interferometry (PolInSAR) is a radar remote sensing technique that makes use of polarimetric separation of scattering phase centers obtained from interferometry to estimate height. This article discusses the potential of the X-band PolInSAR pair for forest height retrieval over tropical forests in the Western ghats. A total of 19 fully polarimetric datasets with various spatial baselines acquired from November 2015 to February 2016 in bistatic mode are utilized in this study. After compensating for all possible non-volumetric decorrelations in the data-sets, the remaining volume decorrelation is modeled using a Random Volume Over Ground (RVoG) model to invert height from PolInSAR data. A modified three-stage algorithm developed by Cloude and Papathanassiou (2003) is adopted for height inversion. PolInSAR derived heights were cross-validated against reference height data measured during a field survey conducted in March 2019. RMSE values of all TerraSAR-X/TanDEM-X PolInSAR heights with respect to field measured heights range from 3.3 to 13.8 m and the correlation coefficient r2 varies between 0.16 and 0.79. The results suggest that the use of a dataset with optimal wavenumber can improve the tree height estimation process. The best performance was achieved for the dataset acquired on 11 December 2015 with RMSE = 3.4 m and r2 = 0.79. Furthermore, the effects of parameters such as angle of incidence, precipitation, and forest biomass on height inversion accuracy are assessed. A large-scale Shimoga Forest height map was generated using multiple TanDEM-X acquisitions with the best correlation results. To improve the accuracy of the height estimation, a merged height approach is explored. The best height estimates among all PolInSAR estimates for a given field plot are chosen in this regard. The merged height approach gave rise to an improved inversion accuracy with RMSE = 1.9 m and r2 = 0.92. The primary objective of this study was to demonstrate the ability of spaceborne X-band data to estimate height with maximum accuracy over natural forests in India, in which height retrieval research has seldom been done.
Keywords
