Enabling global scale Sentinel-1 time series analysis through streaming

Abstract

Dense, high-resolution Synthetic Aperture Radar (SAR) time series from Sentinel-1 offer unique opportunities for monitoring soil moisture. The retrieval process is however challenging due to complex physical processes affecting SAR backscatter, such as vegetation and subsurface scattering effects. Furthermore, the considerable Sentinel-1 data volume introduces its own logistical and computational challenges. This study introduces a novel method for high-throughput calculation of temporal correlation, illustrating how an astute choice of algorithm can facilitate time series analysis on unfavourable data structures. Concretely, we demonstrate how a data streaming approach, with interleaved data reading and processing, can be deployed to efficiently calculate temporal Pearson correlation from a datacube structured as an image stack. Enabled by the substantially reduced computational and memory demands, global calculation of backscatter sensitivity to soil moisture dynamics at a 20 m resolution became feasible. This advancement carries potential for significantly enhancing the accuracy of soil moisture retrievals using SAR backscatter data.