Seasonal Differences in Forest Height Estimation From Interferometric TanDEM-X Coherence Data

This paper demonstrates the use of X-band bistatic synthetic aperture radar (SAR) interferometric coherence for retrieving tree height of coniferous and deciduous forests during leaf-off season in early spring and leaf-on period in summer. TerraSAR-X add-on for Digital Elevation Measurements (TanDEM-X) HH and VV polarization channel coherence images were studied for over 249 ha of forests in Estonia and compared against light detection and ranging (LiDAR) and forest registry field inventory data. Strong correlation was found between interferometric coherence magnitude and LiDAR measured average tree height, especially for winter period. The regression models show the strongest correlation between pine stand heights and single-polarization interferometric coherence, where the correlation coefficients ( ) range between 0.75 and 0.97. The highest correlation for mixed deciduous tree stands was found during leaf-off period with ranging from 0.87 to 0.94, whereas leaf-on period resulted in from 0.58 to 0.75. Strong correlations were also found for spruce trees with between 0.54 and 0.83. Moreover, a simple semiempirical model based on random volume over ground model framework was constructed to describe the relation between the forest height and interferometric coherence. Also, the seasonal variability of the correlation was studied. Our results demonstrate that under Northern-European conditions, seasonal changes have a significant effect for deciduous trees as standard deviations dropped from 1.34–1.78 m during leaf-off conditions to 2.22–3.16 m for leaf-on conditions. Thus, height estimat- on of deciduous stands requires leaf-off conditions for accurate coherence-based height retrieval. Correlation coefficients for pine stands were unvarying across different weather conditions and least affected by the season. The observed strong sensitivity of interferometric coherence to forest height makes it feasible for estimating canopy height for boreal and deciduous forests in both summer and winter conditions. The estimation algorithm works best for coniferous forests.