|Paper title||Multiparametric Sea State from spaceborne Synthetic Aperture Radar and Sentinel-1 Wave Mode Archive Processing in Scope of ESA Climate Change Initiative CCI|
|Form of presentation||Poster|
The study presents a method and application for estimating series of integrated sea state parameters from satellite-borne synthetic aperture radar (SAR), allow processing of data from different satellites and modes in near real time (NRT). The developed Sea State Processor (SSP) estimates total significant wave height SWH, dominant and secondary swell and windsea wave heights, first, and second moment wave periods, mean wave period and period of wind sea. The algorithm was tuned and applied to the Sentinel-1 (S-1) C-band Interferometric Wide Swath Mode (IW), S-1 Extra Wide (EW) and S-1 Wave Mode (WM) Level-1 (L1) products and also to the X-band TerraSAR-X (TS-X) StripMap (SM) and TS-X SpotLight (SL) modes. The scenes are processed in a spatial raster format and result in continuous sea state fields. However, for S-1 WV, the averaged values for each sea state parameter are provided for each 20 km×20 km imagette acquired every 100 km along the orbit.
The developed empirical algorithm consists of two parts: CWAVE_EX (extended CWAVE) based on widely known empirical approach and additional machine learning postprocessing. A series of new data preparation steps (i.e. filtering, smoothing, etc.) and new SAR features are introduced to improve accuracy of the original CWAVE. The algorithm was tuned and validated using two independent global wave models WAVEWATCH-3 (NOAA) and CMEMS (Copernicus) and National Data Buoy Center (NDBC) buoy measurements. The reached root mean squared errors (RMSE) for CWAVE_EX for the total SWH are 0.35 m for S-1 WV and TS-X SM (pixel spacing ca. 1–4 m) and 0.60 m for low-resolution modes S-1 IW (10 m pixel spacing) and EW (40 m pixel) in comparison to CMEMS. The accuracies of the four derived wave periods are in the range of 0.45–0.90 s for all considered satellites and modes. Similarly, the dominant and secondary swell and wind sea wave height RMSEs are in the range of 0.35–0.80 m compared to CMEMS wave spectrum partitions. The postprocessing step using machine learning, i.e., the support vector machine technique (SVM), improves the accuracy of the initial results for SWH. The resulting accuracy of SWH reaches an RMSE of 0.25 m by SVM postprocessing for S-1 WV validated using CMEMS. Comparisons to 61 NDBC buoys, collocated at distances shorter than 100 km to S-1 WV worldwide imagettes, result into an RMSE of 0.31 m. All results and the presented methods are novel in terms of achieved accuracy, combining the classical approach with machine learning techniques. An automatic NRT processing of multidimensional sea state fields from L1 data with automatic switching for different satellites and modes was also implemented. The algorithms provide a wide field for applications and implementations in prediction systems.
The SSP is designed in a modular architecture for S-1 IW, EW, WV and TS-X SM/SL modes. The DLR Ground Station “Neustrelitz” applies the SSP as part of a near real-time demonstrator service that involves a fully automated daily provision of surface wind and sea state parameters estimates from S-1 IW images of the North and Baltic Sea. Due to implemented parallelization, a fine raster for scene processing can be applied: for example, S-1 IW image with large coverage of around 200 km×250 km can be processed using a raster of 1 km (~50,000 analyzed subscenes) within few minutes.
The complete archive of S-1 WV L1 Single Look Complex (SLC) products from December 2014 until February 2021 was processed to create a sea state parameter database (121,923 S-1 WV overflights with around 3,000 IDs/months, each overflight consisting of 40-180 imagettes, total around 14 Mio S-1 WV imagettes). All processed S1 WV data including derived eight state parameters, quality flag and imagette information (geo-location, time, ID, orbit number, etc.) are stored as ascii and in netcdf format for convenient use. The derived state parameters are available to the public within the scope of ESA’s climate change initiative (CCI).
The validation carried out for the whole S-1 WV archive using CMEMS sea state hindcast for latitudes of -60° < LAT < 60° to avoid ice coverage with around 13,5 Mio collocations resulted in an RMSE of 0.245/0.273 m for wv1/wv2 imagettes, respectively. The SWH accuracy for different wave height domains for wv1/wv2 is as follows: 0.28/0.34 m (SWH < 1.5 m), 0.19/0.22 m (1.5 m < SWH < 3 m), 0.30/0.33 m (3 m < SWH < 6 m) and 0.51/0.55 m (SWH > 6 m). The monthly estimated total RMSE varies form 0.22 m to 0.31 m. These RMSE fluctuations around the mean value are caused by different amounts of acquired storms in individual months. As high waves have a higher RMSE, they increase the total RMSE when their relative percentage in a month is higher: in total, SWH distribution in the worldwide acquired SAR data is SWH < 3 m for ~75% of all cases, 3 m < SWH < 6 m for around 24% and only around 1% for SWH > 6 m and even less than 0.1% for SWH > 10 m. However, SWH > 6 m can reach around 2% for individual months with quadratic impact of the SWH values on RMSE.
The cross validations carried out using CMEMS, WW3 and mixed CMEMS/WW3 ground truth show: in terms of total SWH, in comparison to NDBC data, using only CMEMS ground truth resulted into an accuracy ~3 cm better than when the model function was tuned using WW3 data. This might be consequence of the better CMEMS spatial model resolution of 1/12 degree in comparison to WW3 (1/2 degree, spatially interpolated). The SWH comparison between CMEMS, WW3 and NDBC resulted into an RMSE=0.26 cm for CMEMS/NDBC and an RMSE=0.23 cm for CMEMS/WW3 at NDBC buoy locations. Generally, in terms of SWH, the ground truth noise can be assessed to an error of ~0.25 m. As can be seen, the resulting RMSE of 25 cm for S1 WV brings the results down to the noise level of the ground truth data.