Homogenizing pheno-cam data to understand Sentinel-1 backscatter dynamics

Plant phenology controls many ecological processes, and its observation can provide valuable insights about the status of vegetation (e.g., water stress). In recent years there has been a growing interest in capturing seasonal variations using time-lapse digital cameras (phenocams), an inexpensive alternative to field surveys. In fact, pheno-cam images have become a popular reference for phenology studies based on Earth Observation. However, the analysis of long archives acquired under difficult conditions may pose some challenges such as degraded (blur, corruption), misaligned or mis-colored images (e.g. inconsistencies due white balance).

In this study we analyzed the images acquired from the tower of the Zöbelboden LTER site, describing the process we followed to ensure the archive was consistent. Once it was homogenized, we compared the chromatic coordinates with Sentinel-1 backscatter to understand how it is linked with leaf phenology of different tree species.

Archive preparation relied on machine vision techniques. Blur estimation was used to detect degraded images. Registration relied on a group of well aligned images that were used to create a robust synthetic reference based on Sobel edge detector and principal component analysis. The rest of the images were compared with this reference, completing the alignment using Keypoint matching and enhanced cross-correlation. The impact of white balancing was reduced using vicarious calibration, matching the data distributions of stable areas (tree trunks) to the reference from well calibrated images.

When we examined the correlation between the chromatic coordinates and Sentinel-1 terrain-flattened backscatter (gamma) the absolute coefficient often exceeded 0.4 when comparing green and blue with the cross-polarized backscatter or the cross-ratio. Green is tied to photosynthetic activity, whereas the proportion of blue remains lower where leaves are still present (green, active; yellow, senescing; red/brown, dry leaves). These hints Sentinel-1  can be used to track leaf phenology, which would be a powerful asset thanks to its cloud-penetrating capabilities.