Thomas Ledauphin (ICUBE-SERTIT, France); Hervé Yésou (ICUBE-SERTIT, France)

This study investigates the relationship between soil moisture conditions and false water detections observed in SWOT's class 4 "water" products over agricultural areas. The analysis addresses spurious water classifications that occur in SWOT PIXC data, particularly over temporarily saturated agricultural parcels following precipitation events.
The methodology leverages pre-existing soil moisture products generated by INRAE (UMR TETIS) using combined Sentinel-1/2 observations (El Hajj et al., 2017; Baghdadi et al.) covering agricultural parcels and grasslands at ~0-5 cm depth. The study focuses on cross-analysis between SWOT PIXC classifications (classes 2, 3, and 4), soil moisture conditions, and meteorological data (ERA5 precipitation) across two sites in France (Alsace Plain, Lorraine Plateau) from March 2023 to January 2024.
SWOT PIXC data analysis was performed using 183 soil moisture maps (100 for Alsace Plain, 83 for Lorraine Lakes area), with statistical correlation performed between soil moisture temporal evolution and PIXC classification patterns. The study incorporated precipitation data, pedological constraints (focusing on poorly-drained soils), and land cover differentiation between grasslands and crops to understand detection sensitivity variations.
Key findings reveal that SWOT effectively detects temporarily water-saturated agricultural parcels as "water" following significant precipitation events. Critical thresholds were identified: ~11% volumetric soil moisture content and 6mm precipitation levels trigger increased false water detections (for the observed area and study period). A significant temporal shift in PIXC class distribution was observed, with class 2 (land-near water) decreasing from ~70% to ~50% and class 4 (open water) increasing from ~5% to ~25% during periods of rapid soil moisture recovery following precipitation events. The relationship between soil moisture and false water detections follows a non-linear pattern, with detection frequency increasing dramatically above critical moisture thresholds. The phenomenon affects crops more than grasslands, though this observation must be considered within the context of the study area's land cover distribution (fewer grasslands than crops in the analyzed parcels). Cross-correlation analysis demonstrates strong temporal correspondence between precipitation events, soil moisture peaks, and SWOT class 4 detection spikes, particularly on poorly-drained agricultural soils with limited vegetation cover (NDVI < 0.7, corresponding to the soil moisture detection method applicability threshold).
These results provide insights for SWOT data quality assessment and suggest that soil moisture and precipitation context could be considered when interpreting water detection products over agricultural landscapes. The study contributes to understanding SWOT's detection behavior and provides preliminary quantitative thresholds that may help identify potential false positive water detections in similar agricultural areas.

References:
El Hajj, M., Baghdadi, N., Zribi, M., & Bazzi, H. (2017). Synergic use of Sentinel-1 and Sentinel-2 images for operational soil moisture mapping at high spatial resolution over agricultural areas. Remote Sensing, 9(12), 1292.