Estimating tree height from TanDEM-X data at the northwestern Canadian treeline

• Published in: Remote Sensing of Environment
• Main Authors: Antonova, Sofia, Thiel, Christian, Höfle, Bernhard, Anders, Katharina, Helm, Veit, Zwieback, Simon, Marx, Sabrina, Boike, Julia
• Format: Article in Journal/Newspaper
• Language: unknown
• Published: ELSEVIER SCIENCE INC 2019
• Subjects: Arctic; Canada; Northwest Territories; Tundra
• Online Access: https://epic.awi.de/id/eprint/49850/; https://hdl.handle.net/10013/epic.6478967e-aa02-4a25-ad83-2a554b58b229

The circum-Arctic transitional zone between forest and tundra, i.e. the treeline zone, is shifting northward due to current Arctic warming and, therefore, requires systematic monitoring. Up to now, radar remote sensing was hardly possible in the treeline zone due to spatial resolution and/or temporal decorrelation constraints of preceding satellite missions. The unique constellation of the TanDEM-X satellites with its bistatic mode and very high spatial resolution opens up opportunities for monitoring small (≥0.01 km2) and isolated patches of very sparse forest which are typical for the transitional zone. We focused on an area at the northern edge of the treeline zone in the Northwest Territories, Canada, and evaluated the potential of TanDEM-X bistatic data to characterize the tree height in the forest patches in this region. TanDEM-X data were acquired during the TanDEM-X Science Phase in 2015, when the perpendicular baseline was large (corresponding to the height of ambiguity of approximately 14.6 m) and kept constant. We employed TanDEM-X backscatter, bistatic coherence, and interferometric height from the stack of seven multitemporal bistatic pairs and compared them to maximum vegetation height obtained from full-waveform airborne LiDAR data. We found strong linear relationships between all TanDEM-X metrics and LiDAR vegetation height within the forest patches with r = 0.67, r = −0.69, and r = 0.78 for the backscatter, coherence, and interferometric height, respectively. Furthermore, we extracted the position of individual trees from the LiDAR data and estimated tree density as the number of trees per unit area. The linear relationships between all TanDEM-X metrics and the tree density were very weak. The relationships between all TanDEM-X metrics and tree height differentiated for three tree density classes (low, medium, and high) remained strong. Random forests regression using all three TanDEM-X metrics predicted the tree height with a mean absolute error of 0.7 m (mean forest height in the study area ...