Integrating the landscape scale supports SAR-based detection and assessment of the phenological development at the field level

Johannes Löw^1*Steven Hill²Insa Otte²Christoph Friedrich²Michael Thiel²Tobias Ullmann²Christopher Conrad¹

• ¹Martin Luther University of Halle-Wittenberg, Halle, Germany
• ²Julius Maximilian University of Würzburg, Würzburg, Bavaria, Germany

This study presents a novel framework for bridging landscape-scale vegetation dynamics with fieldlevel crop phenology using Sentinel-1 radar time series. Unlike previous approaches that focus on local algorithm optimisation or SAR feature selection, this work integrates two scales: (1) landscape patterns derived from annual distributions of time series metrics (TSMs) and (2) field-level phenology, both linked to growing degree days (GDD). TSMs were generated through breakpoint analyses over different smoothing intensities for Sentinel-1 polarisation (PolSAR) and interferometric coherence (InSAR) features, capturing crop, orbit and sensor-specific responses. The framework quantifies uncertainties inherent in both remote sensing and ground observations, and evaluates trackable progress (phenological stage detectability) and tracking range (GDD variance around stages) to assess accuracy under variable acquisition geometries, weather and smoothing parameters. Applied to the DEMMIN site (Germany), the analysis revealed consistent TSM-GDD relationships for wheat, rape, and sugar beet, with descriptors such as soil fertility and water availability explaining spatial patterns (R² ≈ 0.8). Key novelties include the identification of low tracking ranges in drought years, the demonstration of the impact of orbit-specific incidence angles on monitoring fidelity, and the highlighting of Sentinel-1's ability to resolve phenological variance across fragmented landscapes. By harmonising multi-scale SAR time series with agro-meteorological data, this approach advances transferable methods for operational crop monitoring, supporting precision agriculture and regional yield assessment beyond localised models.