Seasonal and Long-Term Vegetation Effects on Sentinel-1 Range Coregistration Shifts

Abstract

In the SAR4Tectonics project, targeted at measuring ground deformation in tectonically active zones, we processed, with interferometric techniques, over 1500 stacks of Sentinel-1 SLCs, averaging 160 images per stack. During the InSAR processing, coregistration was performed between the stack reference image, located at the center of the time series, and all other images. This was achieved using patch-based incoherent cross-correlation for the range direction and the Enhanced Spectral Diversity technique for the azimuth direction. Several corrections were applied beforehand, including instrument timing correction, tropospheric and ionospheric path delays, and solid Earth tides. Tectonic plate motions were afterwards also removed from the coregistration shifts using GNSS data. The remaining shifts should primarily reflect residuals from the correction process, orbit-related errors, and random noise. Unexpectedly, residual range coregistration shifts exhibit systematic trends and yearly oscillations in forested regions, with non-vegetated areas showing the expected zero-mean, uncorrelated, and flat residuals. These signals are consistent between ascending and descending stacks, with the trends always directed towards the satellite (shortening the radar range). Over vegetated temperate regions, trends average approximately 5 cm/year, increasing to around 10 cm/year in equatorial and tropical regions. Oscillations have amplitudes ranging from 10 to 80 cm, with the largest values in tropical and temperate areas, and minimal oscillations (<5 cm) in equatorial regions. The oscillation peaks (minimum delay) occur consistently within climatic zones, around June in temperate regions, September in tropical areas, and January in the southern Hemisphere. The correlation of these signals with vegetation cover and climatic zones rules out significant correction errors, such as those linked to the ionospheric effects of the growing solar cycle or uncompensated tectonic motion. Instead, the signals seem to suggest that radar delay is influenced by seasonal vegetation dynamics and longer-term multiyear trends, possibly linked to variations in plant water content, as well as their slow growth over time. These signals contrast with results from persistent and distributed scatterers (PS/DS), which primarily capture well-known phenomena such as ground deformation from tectonic motion or subsidence due to groundwater extraction. However, PS/DS are generally absent over vegetated areas, meaning no PS/DS measurements are available to confirm the coregistration residual signals observed in these regions. To our knowledge, this phenomenon represents a novel discovery. While the exact physical mechanisms driving these trends is still unclear, these findings seem to suggest that radar delays could be used in an innovative way to monitor vegetation growth and inter-annual variability, which could also provide insights into plant health or drought stress.