Sentinel-1 Archive Processing for Spaceborne Synthetic Aperture Radar Derived Sea State Parameters in Scope of ESA’s Climate Change Initiative - Sea State ECV and collocation with TerraSAR-X

Kurzfassung

In the scope of ESA’s Climate Change Initiative - Sea State ECV sea state parameters are derived from Synthetic Aperture Radar (SAR) acquisitions provided by the Sentinel-1 (S1) mission (satellite S1-A launched in 2014, in operation until today; satellite S1-B launched in 2016, out of service in 2021). Within the project, ten years of archived S1 data from 2014 to 2024 are processed and analyzed in terms of sea state. The following eight integrated sea state parameters are processed from Level-1 (L1) Interferometric Wide Swath Mode (IW), Extra Wide (EW) and Wave Mode (WV) products: total significant wave height (SWH), dominant and secondary swell wave height, windsea wave heights, first and second moment wave periods, mean wave period and period of windsea. For S1 IW (coverage ca. 250×200 km, 10 m pixel spacing, acquisitions dominantly in coastal regions and shelf waters) and S1 EW (coverage ca. 450×400 km, 40 m pixel spacing, acquisitions dominantly in Arctic/Antarctic but also in North Atlantic and around Madagascar) modes, the GRD (Ground Range Detected) and for WV mode (along-orbit imagettes each 100 km ca. 20×20 km, ca. 3.5 m pixels, acquisitions only in open oceans) the SLC (Single Look Complex) ESA products are used. The scenes are processed in a spatial raster format by analysis of subscenes (ca. 2.5×2.5 km, processing raster step of 5 km for S1 IW, 10×10 km, processing raster step of 17 km for S1 EW) and result in continuous sea state fields for S1 IW and S1 EW. For S1 WV, averaged values for each along-orbit imagette are stored. The DLR ocean products include eight estimated sea state parameters, quality, uncertainty and rejection flags for each data point. For the processing, the new algorithm SAR-SeaStaR (SAR Sea State Retrieval) based on combination of linear regression (LR) and machine learning (ML) has been applied [2]. The SAR-SeaStaR was implemented in the sea state processor (SSP) software using modular architecture as one of the processing branches in the SAR AIS Integrated Toolbox (SAINT) package. Besides processing of S1 IW, EW, WV (acquired with C-band radar frequencies) the SSP modular architecture is also adapted for processing TerraSAR-X/TanDEM-X (TS-X/TD-X) StripMap (SM) modes (acquired with X-band radar frequencies). In total, ca. 3 Mio S1 IW and ca. 1 Mio S1 EW worldwide acquired scenes were processed and result into a data base of ca. 1,2 Mio and 0,5 Mio ocean scenes for IW and EW respectively. The processing of S1 WV archive for 2014-2021 with ca.150,000 products (ca. 12 Mio WV individual imagettes) has been already shown in [1]. The processing was executed on LRZ (Leibniz Supercomputing Centre, www.lrz.de). The huge amount of data (ca. 4 Mio S1 IW and EW scenes) with ca. 13 Petabyte (PB) requires the usage of such a supercomputing system enabling large scale parallel processing. For Areas of Interests (AoIs) covering the North Atlantics/Pacific and European waters including North, Baltic, Black and Mediterranean Seas [1], the processed S1 IW archive was collocated to TerraSAR-X SM archive for the years 2020 and 2021. From the data of this time period of two years (ca. 350,000 ocean S1 IW scenes and ca. 25,000 ocean TS-X SM scenes), 185 scenes are collocated within distance under 30 km and within +/-30 min time window. From those, ca. 50% of all collocations are found under 10 km distance and 5 min time difference. However, only individual scenes are directly collocated spatially and temporally under 1 min. For estimation of root mean squared error (RMSE) and BIAS from the collocated full scenes, the values of processed sea state parameters are compared for collocated subscenes extracted from the full scenes of both satellites (1.5 km processing raster for TS-X). The analysis of SWH results into RMSE≈0.45 m between S1 IW and TS-X and BIAS of ca. 0.08 m (S1 minus TS-X). These numbers correspond to the uncertainty of S1 IW compared to the model and buoys with RMSE≈42 cm [2]. The individual outliers in TS-X SM derived SWH dataset (recognizable by SWH underestimation for individual subscenes) are found being connected to large slicks-looking structures (“dark spots” with low NRCS due to e.g. oil and algae spills, wind shadowing, etc.) which are not as strongly pronounced in the S1 IW data (C-band, lower resolution). The collocation of the S1/TS-X collocated dataset to in-situ measurements (NDBC and EMODNET buoys) and to the model (MFWAM, Copernicus) results into quadruple cross-comparison and uncertainties analysis. A cross-comparison matrix of RMSE was established, where all combinations of four SWH sources are analyzed. The cross-comparison matrix represents partial RMSEs calculated only for locations where all four sources are available. These partial estimates are different from total RMSEs, estimated for each data pair worldwide (e.g. 0.40 m for S1 IW and 0.35 m for TS-X by worldwide comparison to MFWAM) where amount of collocated data pairs is significantly wider and covers larger areas. [1] Pleskachevsky, A., Tings, B., and S, Jacobsen, 2022: Multiparametric Sea State Fields from Synthetic Aperture Radar for Maritime Situational Awareness, RSE, vol. 280, 22 pp. [2] Pleskachevsky, A., Tings, B., Jacobsen, S., Wiehle, S., Schwarz, E., and D. Krause, 2024: A System for Near Real Time Monitoring of the Sea State using SAR Satellites. IEEE Transactions on Geoscience and Remote Sensing, vol. 62, 2024.