Freeze Detection Over Agricultural Areas Using Sentinel-1 Data (FAS)

Résumé 0

The freeze maps were carried out over 500 m x 500 m grids (maps could be provided at a plot scale). For a given site, all acquired Sentinel-1 radar images were processed (about 15 images per month). An approach that relies on change detection in the high-resolution Sentinel-1 C-band SAR backscattered coefficients was applied to determine surface states as either frozen or unfrozen. Freeze detection is carried out for two distinct land cover classes: cereals and grasslands (LC1), vineyards and orchards (LC2). Products are in vector format (shapefile): (FREEZDETECT_{S2 Tile}_{Date}T{hour}{minute}{second}). ID_PARCEL field: plot Id. CODE_CULTU field: culture code according to the RPG 2019. CODE_GROUP field: group culture code according to the RPG 2019. PARC_TYPE field: land cover classes according to our detection algorithm (LC1 and LC2). MREFSIGMA field: reference backscattering coefficient in dB (unfrozen conditions). MEANSIGMA field: mean backscattering coefficient for the given plot at the S1 acquisition date in dB. MEANTEMP field: temperature in ⁰C from ERA5-LAND. FROZ_TYPE field: the freeze state: 0 = No detected freezing; 1 = Mild-to-Moderate freezing; 2 = Severe freezing. For LC1, if Delta defined by the difference (MREFSIGMA – MEANSIGMA) is < 3.5 dB -> unfrozen, if 3.5 dB < Delta < 4.5 dB -> mild-to-moderately frozen, if Delta >4.5 dB -> severely frozen. For LC2, if Delta is < 2.5 dB -> unfrozen, if 2.5 dB < Delta < 3.5 dB -> mild-to-moderately frozen, if Delta >3.5 dB -> severely frozen.