LARGE-SCALE FOREST HEIGHT MAPPING FROM TANDEM-X, ICESAT-2 AND LANDSAT 8 DATA USING A MACHINE-LEARNING METHOD

Forest height is an indispensable parameter for natural resource investigations. Spaceborne Interferometric SAR (InSAR) has the sensitivity to measure forest height, especially TanDEM-X, which provides high-quality interferometric coherence without the effects of atmospheric delay and temporal decorrelation. In this paper, we seriously considered the limited penetrability of TanDEM-X InSAR, and proposed a two-step machine learning (ML) method to estimate large-scale forest height by combining TanDEM-X InSAR data, ICESat-2 data, and Landsat 8 data. The forest scattering phase center (SPC) height of InSAR is estimated by the first ML, and on this basis, the relationship between the SPC height and forest height is established by the second ML. We validated the effectiveness of proposed method in a Spanish Mediterranean climate forest region with airborne LiDAR data. As validated against LiDAR data, the accuracy of the estimated SPC height ranges from 2.07 m to 2.49 m, and the root mean square error (RMSE) of the average forest height ranges from 1.79 m to 2.93 m, at the resolution of four-hectare forest stands.