sensors

Article Development of Wind Speed Retrieval from Cross-Polarization Chinese Gaofen-3 Synthetic Aperture Radar in

Weizeng Shao 1 ID , Xinzhe Yuan 2,*, Yexin Sheng 1 ID , Jian Sun 3, Wei Zhou 4 and Qingjun Zhang 5

1 Marine Science and Technology College, Zhejiang Ocean University, Zhoushan 316022, China; [email protected] (W.S.); [email protected] (Y.S.) 2 National Satellite Ocean Application Service, State Oceanic Administration, Beijing 100081, China 3 Physical Oceanography Laboratory/CIMST, Ocean University of China and Qingdao National Laboratory for Marine Science and Technology, Qingdao 266100, China; [email protected] 4 South China Sea Institute of Oceanology, Chinese Academy of Sciences, Guangzhou 510301, China; [email protected] 5 Beijing Institute of Spacecraft System Engineering, Beijing 100076, China; [email protected] * Correspondence: [email protected]; Tel.: +86-010-6210-5661

Received: 25 December 2017; Accepted: 29 January 2018; Published: 31 January 2018

Abstract: The purpose of our work is to determine the feasibility and effectiveness of retrieving sea surface wind speeds from C-band cross-polarization (herein vertical-horizontal, VH) Chinese Gaofen-3 (GF-3) SAR images in typhoons. In this study, we have collected three GF-3 SAR images acquired in Global Observation (GLO) and Wide ScanSAR (WSC) mode during the summer of 2017 from the China Sea, which includes the typhoons Noru, Doksuri and Talim. These images were collocated with wind simulations at 0.12◦ grids from a numeric model, called the Regional Assimilation and Prediction System- model (GRAPES-TYM). Recent research shows that GRAPES-TYM has a good performance for typhoon simulation in the China Sea. Based on the dataset, the dependence of wind speed and of radar incidence angle on normalized radar cross (NRCS) of VH-polarization GF-3 SAR have been investigated, after which an empirical algorithm for wind speed retrieval from VH-polarization GF-3 SAR was tuned. An additional four VH-polarization GF-3 SAR images in three typhoons, Noru, Hato and Talim, were investigated in order to validate the proposed algorithm. SAR-derived winds were compared with measurements from Windsat winds at 0.25◦ grids with wind speeds up to 40 m/s, showing a 5.5 m/s root mean square error (RMSE) of wind speed and an improved RMSE of 5.1 m/s wind speed was achieved compared with the retrieval results validated against GRAPES-TYM winds. It is concluded that the proposed algorithm is a promising potential technique for strong wind retrieval from cross-polarization GF-3 SAR images without encountering a signal saturation problem.

Keywords: wind speed; typhoon; cross-polarization; Gaofen-3 SAR

1. Introduction Synthetic aperture radar (SAR) operates in all-weather conditions and has the capability to detect sea surface winds, especially under extreme wind conditions [1,2]. Currently, SAR data is available from C-band (5.3 GHz) Radarsat-2 (R-2) and Sentinel-1 (S-1); X-band (9.8 GHz) TerraSAR-X/TanDEM-X, and Cosmo-SkyMed; and L-band (1.2 GHz) ALOS-2 satellite. The new generation C-band Gaofen-3 (GF-3) SAR was launched by the China Academy of Space Technology (CAST) in 2016 with a wide spatial swath coverage, i.e., more than 500 km swath for Global Observation (GLO) mode and Wide ScanSAR (WSC) mode. So far, GF-3 SAR supports operations in alternative

Sensors 2018, 18, 412; doi:10.3390/s18020412 www.mdpi.com/journal/sensors Sensors 2018, 18, 412 2 of 15 dual-polarization, i.e., vertical-vertical (VV) with vertical-horizontal (VH) or horizontal-horizontal (HH) with horizontal-vertical (HV). Our institutions are the primary participants involved in the development of marine applications using GF-3 SAR data, in particular, for wind and wave monitoring. Wind retrieval from C-band co-polarization (VV and HH polarization) SAR is a now a mature technique. Geophysical Model Functions (GMFs) for co-polarization C-band SAR wind retrieval CMODs have been widely used over the last few decades, i.e., CMOD4 [3], CMOD-IFR2 [4], CMOD5 [5], CMOD5N for neutral winds [6] and the new GMF C-SARMOD [7]. The C-band GMFs describe a complicated relationship between the normalized radar cross section (NRCS) of co-polarization SAR and wind vectors at 10 m above sea surface. CMODs have been successfully applied for wind retrieval from SAR at C-band and well validated at wind speed ranges from 0–20 m/s [8–11]. In particular, wind speeds retrieved from a large number of VV and HH polarization S-1 SAR images have been recently validated against the products from the Advanced Scatterometer (ASCAT) on board satellite Metop–A and –B [12]. Results indicate that the CMOD5N and CMOD4 plus polarization ratio (PR) models proposed in [13] are the best options for wind retrieval in VV and HH polarization, respectively. However, because two unknown variables exist in these C-band GMFs, i.e., wind speed and wind direction, it is impossible to derive the wind vector by inverting the individual C-band GMF without any prior information. It was found that spatial patterns visible on SAR imagery, known as “wind streaks” [14], form parallel to the wind direction. A local gradient (LG) method can be used for retrieving wind directions automatically, as wind direction is considered as be orthogonal to the gradient direction of the local SAR intensity image [15,16]. Unfortunately, SAR-derived wind directions have a 180◦ ambiguity. In order to remove that ambiguity, an independent source is still required. In practice, the application of C-band GMF directly employs external information on wind direction at low spatial and temporal resolution, i.e., numeric prediction and observation derived from a scatterometer. However, the spatial resolution of external wind direction is coarse, which is a mismatch with SAR imagery, especially in coastal regions. The sensitivity of SAR-wind speed on NRCS of co-polarization SAR reduces, due to the saturation of the backscattering signal under strong wind conditions (probably at wind speeds greater than 25 m/s) [17,18], causing a root mean square error (RMSE) 6.2–6.5 m/s of wind speed for VV-polarization [19]. This is considered to be the one of the major weaknesses for wind retrieval from SAR when using the traditional C-band co-polarization GMF method. Interestingly, the benefits of cross-polarization (VH and HV polarization) SAR signals have been studied over the past few years. Recent research has revealed that the NRCS of cross-polarization SAR in units of dB has a strong linear relationship with wind speed, while cross-polarization NRCS is less sensitive to wind direction [20]. Several studies have focused on retrieving wind speeds from C-band cross-polarization R-2 SAR [21–25] and S-1 SAR [26], showing that the retrieved wind speeds have a good agreement with moored measurements. It also has been found that cross-polarization SAR radar backscattering from the ocean surface enhances sensitivity to signal saturation [27–32], especially under strong wind conditions (probably for winds greater than 20 m/s). A comparison with stepped frequency microwave radiometer (SFMR) wind speeds for some cases in hurricanes has shown strong winds with speeds up to around 55 m/s with a ±5 m/s deviation [21], due to the fact the cross-polarization does not saturate as easily as the co-polarization signal. These apparent advantages bring the expectation of incorporating a VH channel into the next generation of scatterometer designs [33,34]. In this study, the wind fields from the Global and Regional Assimilation and Prediction System—Typhoon model (GRAPES-TYM) at 0.12◦ grids, which is used for typhoon simulation in the China Sea [35], have been collocated with the VH-polarization GF-3 SAR images taken of typhoons in the China Sea. Through the dataset, the dependence of wind speed and radar incidence angle on the NRCS of VH-polarization GF-3 SAR was investigated. Then we tuned an empirical algorithm for wind speed retrieval from VH-polarization GF-3 SAR image in strong winds ranging from speeds up to 40 m/s. Sensors 2018, 18, 412 3 of 15

We organize the remainder of this paper as follows: Section2 introduces the datasets used in our study, i.e., GF-3 SAR images in VH-polarization, Windsat polarimetric radiometer at 0.25◦ grids and wind simulation from numeric GRAPES-TYM. Then an empirical wind retrieval algorithm for VH-polarizationSensors 2018, 18, x GF-3 FOR PEER SAR REVIEW is tuned in Section3. The validation of retrieval wind speeds3 from of 14 the VH-polarization GF-3 SAR images is presented in Section4 and the discussion is presented in Section5. Sectionand6 giveswind simulation the conclusions from numeric and summary. GRAPES-TYM. Then an empirical wind retrieval algorithm for VH-polarization GF-3 SAR is tuned in Section 3. The validation of retrieval wind speeds from the 2. DescriptionVH-polarization of Dataset GF-3 SAR images is presented in Section 4 and the discussion is presented in Section 5. Section 6 gives the conclusions and summary. In total, seven VH-polarization GF-3 SAR images, which were acquired in GLO and WSC mode, were2. collected Description during of Dataset the summer of 2017 in the China Sea. These VH-polarization GF-3 SAR images were processedIn total, as seven Level-1B VH-polarization (L-1B) products GF-3 SAR and images, were taken which under were acquired typhoon in conditions. GLO and WSC They mode, have the standardwere pixelcollected size during of about the 100~500summer of m 2017 for in range the China and azimuthSea. These direction. VH-polarization A GF-3 GF-3 SAR SAR image images in GLO modewere and processed WSC mode as isLevel-1B comprised (L-1B) of products several singleand were radar taken beams. under The typhoon details conditions. of the six They collected have GF-3 SAR imagesthe standard and informationpixel size of aboutabout 100~500 the corresponding m for range typhoons and azimuth is listed direction. in Table A GF-3 A1 of SAR the image Appendix in A, in whichGLO the mode typhoon and WSC information mode is comprised has been of bilinearly several single interpolated radar beams. to the The SAR details imaging of the time.six collected We use the followingGF-3 equationSAR images for and calculating information the about NRCS the of cross-polarizationcorresponding typhoons GF-3 is SAR listed acquired in Table in A1 L-1B of mode:the Appendix, in which the typhoon information has been bilinearly interpolated to the SAR imaging time. We use the following equation for calculat Ming the2 NRCS of cross-polarization GF-3 SAR σ0 = DN2 − N (1) acquired in L-1B mode: 65535 2 2 M 0 σ0 = DN − N (1) where in σ is the NRCS in units of dB, DN is SAR-measured65535 intensity, M is the external calibration factorwhere and Nin isσ0the is the offset NRCS constant in units stored of dB, inDN the is SAR-measured annotation file. intensity, M is the external calibration Infactor the and data N collection, is the offset there constant are threestored VH-polarization in the annotation GF-3file. SAR images, which were taken during the periodsIn the of typhoonsdata collection, Noru, there Hato are and three Doksuri. VH-polarization Figure1 showsGF-3 SAR three images, quick-look which were images taken of this. Due toduring the lowerthe periods signal-to-noise of typhoons ratioNoru, of Hato dual-polarization, and Doksuri. Figure GF-3 1 shows SAR suffersthree quick-look from crosstalk images amongof the differentthis. Due polarization to the lower channels. signal-to-noise The effect ratio ofof instrumentaldual-polarization, noise GF-3 causes SAR the suffers apparently from crosstalk lighter near the edgeamong of differentthe different signature polarization beams channels. of GF-3 The SAR, effect as of aligned instrumental vertically noise shown causes inthe Figure apparently1. This is assumedlighter to near reduce the edge the instrumental of different signature noise for beams wind of retrieval.GF-3 SAR, Aas techniquealigned vertically is proposed shown in Figure [23] of 1. using This is assumed to reduce the instrumental noise for wind retrieval. A technique is proposed in [23] the information of referred noise, called the ‘de-noised’ procedure. Unfortunately, the instrumental of using the information of referred noise, called the ‘de-noised’ procedure. Unfortunately, the noise information for GF-3 SAR has yet to be stored in the annotation file. Therefore, it is difficult to instrumental noise information for GF-3 SAR has yet to be stored in the annotation file. Therefore, it effectivelyis difficult deal to with effectively the instrumental deal with thenoise instrumental in theVH-polarization noise in the VH-polarization channel in channel this study. in this study.

FigureFigure 1. VH-polarization 1. VH-polarization GF-3 GF-3 SAR SAR images images taken taken during during three three typhoons. typhoons. (a) The(a) The image image from from typhoon typhoon Noru acquired in Global Observation (GLO) mode on 2 August 2017 at 21:09 UTC. (b) The Noru acquired in Global Observation (GLO) mode on 2 August 2017 at 21:09 UTC. (b) The image image from acquired in Wide ScanSAR (WSC) mode on 13 September 2017 at 22:14 from typhoon Doksuri acquired in Wide ScanSAR (WSC) mode on 13 September 2017 at 22:14 UTC. UTC. (c) The image from typhoon Talim acquired in GLO mode on 14 September 2017 at 21:29 UTC. (c) The image from typhoon Talim acquired in GLO mode on 14 September 2017 at 21:29 UTC. The Windsat polarimetric radiometer was developed by the Naval Research Laboratory (NRL) and has the capability to measure all-weather winds from space with swath more than 350 km. A

Sensors 2018, 18, 412 4 of 15

The Windsat polarimetric radiometer was developed by the Naval Research Laboratory (NRL) andSensors has 2018 the, 18 capability, x FOR PEER to REVIEW measure all-weather winds from space with swath more than 3504 of 14 km. A comparison between Windsat winds at 10 m height above sea surface and SFMR observations showscomparison that Windsat between winds Windsat do winds not encounter at 10 m height saturation above sea problems surface and in strongSFMR observations winds ranging shows from 20that to 40 Windsat m/s [ 36winds]. In do this not study, encounter we collectedsaturation measurementsproblems in strong from winds Windsat ranging at from 0.25 ◦20grids. to 40 m/s However, [36]. weIn found this study, there we were collected only measurements a few Windsat from grids Wind insat the at 0.25° coverage grids. of However, the three we images found there in Figure were 1. only a few Windsat grids in the coverage of the three images in Figure 1. Therefore, GRAPES-TYM Therefore, GRAPES-TYM wind fields at 0.12◦ grids were used here and were collocated with those wind fields at 0.12° grids were used here and were collocated with those three GF-3 SAR images. The three GF-3 SAR images. The GRAPES-TYM wind maps are shown in Figure2, in which the rectangles GRAPES-TYM wind maps are shown in Figure 2, in which the rectangles represent the spatial represent the spatial coverage of the three VH-polarization GF-3 SAR images in Figure1. The time coverage of the three VH-polarization GF-3 SAR images in Figure 1. The time difference between difference between them is within 30 min. It was proved in [35] that GRAPES-TYM wind fields from all them is within 30 min. It was proved in [35] that GRAPES-TYM wind fields from all the typhoon the typhoon cases in the northwest Pacific Ocean and South China Sea in 2013 had good performances cases in the northwest Pacific Ocean and South China Sea in 2013 had good performances with withobservations. observations. Moreover, Moreover, the GRAPES-TYM the GRAPES-TYM winds winds map mapinformation, information, i.e., locations i.e., locations of typhoon of typhoon (TE) eye (TE)and and maximum maximum wind wind speeds, speeds, is consistent is consistent with withthe data the datalisted listed in Table in TableA1. The A1 matchup. The matchup dataset dataset has hasa large a large number number of samples of samples with with wind wind speeds speeds up to up 40 tom/s, 40 which m/s, whichis suitable is suitable for developing for developing a high- a high-windwind retrieval retrieval algorithm. algorithm.

FigureFigure 2. 2.Global Global and and Regional Regional AssimilationAssimilation and and Prediction Prediction System—Typhoon System—Typhoon model model (GRAPES-TYM) (GRAPES-TYM) wind maps, in which rectangles represent the coverage of the corresponding three VH-polarization wind maps, in which rectangles represent the coverage of the corresponding three VH-polarization GF-3 SAR images in Figure 1. (a) GRAPES-TYM wind maps from typhoon Noru on 2 August 2017 at GF-3 SAR images in Figure1.( a) GRAPES-TYM wind maps from typhoon Noru on 2 August 2017 at 21:00 UTC. (b) GRAPES-TYM wind maps from typhoon Doksuri on 13 September 2017 at 22:00 UTC. 21:00 UTC. (b) GRAPES-TYM wind maps from typhoon Doksuri on 13 September 2017 at 22:00 UTC. (c) GRAPES-TYM winds maps from typhoon Talim on 14 September 2017 at 21:00 UTC. (c) GRAPES-TYM winds maps from typhoon Talim on 14 September 2017 at 21:00 UTC. As well as the three cases, the proposed algorithm has been implemented for the other four VH- polarizationAs well asGF-3 the SAR three images cases, during the the proposed periods of algorithm typhoons Noru, has been Hato implemented and Talim. The for quick-look the other fourimages VH-polarization of these four GF-3GF-3 SAR SAR images images and during the corresponding the periods of Windsat typhoons wind Noru, maps Hato are shown and Talim. in TheFigures quick-look 3 and images4, respectively. of these It four can GF-3be clearly SAR seen images thatand there the are corresponding some measurements Windsat in windthe coverage maps are shownof these in Figuresfour images3 and with4, respectively. wind speeds It up can to be40 clearlym/s, noting seen that, that the there time are difference some measurements between Windsat in the coveragewinds and of these the collected four images four with SAR wind imaging speeds times up is to within 40 m/s, 15 notingmin. Thus, that, the retrieval time difference wind speeds between can be directly validated against Windsat gridded wind speeds.

Sensors 2018, 18, 412 5 of 15

Windsat winds and the collected four SAR imaging times is within 15 min. Thus, the retrieval wind speedsSensors can2018 be, 18,directly x FOR PEER validated REVIEW against Windsat gridded wind speeds. 5 of 14 Sensors 2018, 18, x FOR PEER REVIEW 5 of 14

FigureFigure 3. 3. Three VH-polarization VH-polarization GF-3 GF-3 SAR SAR images images taken taken during during the period the of period three typhoons. of three typhoons.(a) The Figure 3. Three VH-polarization GF-3 SAR images taken during the period of three typhoons. (a) The (a)image The image from typhoon from typhoon Noru ac Noruquired acquired in GLO in mode GLO on mode 4 August on 4 August 2017 at 201709:12 atUTC. 09:12 (b UTC.) The image (b) The from image image from typhoon Noru acquired in GLO mode on 4 August 2017 at 09:12 UTC. (b) The image from fromtyphoon typhoon Noru Noru acquired acquired in WSC in WSC mode mode on on4 August 4 August 2017 2017 at 21:26 at 21:26 UTC. UTC. (c) (Thec) The image image from from typhoon typhoon typhoonHato acquired Noru acquiredin WSC modein WSC on mode 22 August on 4 August2017 at 201722:23 at UTC. 21:26 ( dUTC.) The ( cimage) The imagefrom typhoon from typhoon Talim Hato acquired in WSC mode on 22 August 2017 at 22:23 UTC. (d) The image from typhoon Talim Hatoacquired acquired in WSC in modeWSC modeon 16 Septemberon 22 August 2017 2017 at 09:34 at 22:23 UTC. UTC. (d) The image from typhoon Talim acquiredacquired in in WSC WSC mode mode on on 1616 SeptemberSeptember 2017 at 09:34 09:34 UTC. UTC.

Figure 4. Windsat wind maps corresponding to the four VH-polarization GF-3 SAR images in Figure 2, FigureFigurein which 4. 4.Windsat Windsatrectangleswind wind represent mapsmaps corresponding the coverage of to to the the the images. four four VH-polarization VH-polarization (a) Windsat wind GF-3 GF-3 map SAR SAR from images images typhoon in Figure in FigureNoru 2, 2, ininon which which 4 August rectangles rectangles 2017 at represent represent 09:18 UTC. the the ( coverageb coverage) Windsat of of wind the the images. images. map from ((aa)) WindsattyphoonWindsat windNoruwind maponmap 4 August fromfrom typhoontyphoon 2017 at Noru 21:18 on on 4 August 2017 at 09:18 UTC. (b) Windsat wind map from typhoon Noru on 4 August 2017 at 21:18 4 AugustUTC. (c 2017) Windsat at 09:18 wind UTC. map (b )from Windsat typhoon wind Hato map on from 22 August typhoon 2017 Noru at on22:30 4 August UTC. (d 2017) The at Windsat 21:18 UTC. UTC. (c) Windsat wind map from on 22 August 2017 at 22:30 UTC. (d) The Windsat (c)wind Windsat map wind from maptyphoon from Talim typhoon on 16 Hato September on 22 August 2017 at 201709:36 atUTC. 22:30 UTC. (d) The Windsat wind map wind map from typhoon Talim on 16 September 2017 at 09:36 UTC. from typhoon Talim on 16 September 2017 at 09:36 UTC.

Sensors 2018, 18, 412 6 of 15 Sensors 2018, 18, x FOR PEER REVIEW 6 of 14

3. Development of a Wind Retrieval Algorithm for VH-Polarization GF-3 SAR Recent research has revealed that the NRCS of cross-polarization SAR is independent of wind direction [19–22]. [19–22]. In this section, the dependence of sea surface wind speed and of radar incidence angle on NRCS of VH-polarization GF-3 SAR are investigated. Based on the findings, findings, we developed an empirical wind retrieval algorithm for VH-polarization GF-3GF-3 SAR.SAR.

3.1. Dependence on VH-Pol VH-Polarizationarization GF-3 SAR NRCS During thethe process,process, each each whole whole SAR SAR image image was was divided divided into ainto number a number of sub-scenes of sub-scenes from squared from squaredimage blocks image with blocks a spatial with coverage a spatial of coverage about 8 km of ×about8 km. 8 Thekm sub-scene,× 8 km. The covering sub-scene, the GRAPES-TYM covering the windsGRAPES-TYM fields at 0.12winds◦ grids, fields was at 0.12° chosen. grids, Therefore, was chosen. GRAPES-TYM Therefore, wind GRAPES-TYM fields at 10 wind m height fields above at 10 sea m heightsurface above were matchedsea surface up were with matched the NRCSs up andwith radar the NRCSs incidence and angles radar incidence from the three angles VH-polarization from the three VH-polarizationGF-3 SAR images GF-3 in typhoons SAR images Noru, in Doksurityphoons and Noru, Talim. Doksuri The number and Talim. of co-locations The number were of co-locations more than werefive thousand more than samples five thousand in about samples 10% of in the about total 10% extracted of the total sub-scenes. extracted The sub-scenes. collocated The dataset collocated was datasetused for was studying used thefor dependencestudying the of dependence winds and ofof radarwinds incidence and of radar angle incidence on NRCS ofangle VH-polarization on NRCS of VH-polarizationGF-3 SAR. GF-3 SAR. In order to makemake aa reasonablereasonable analysis,analysis, sub-scenessub-scenes with winds smallersmaller than 1010 m/sm/s and and non- non- homogeneous sub-scenes due to heavy rainfall, wher wheree the ratio of image variance and squared image mean value was greater than 1.05 [37,38], [37,38], were exclud excludeded in this study. Figure 55aa showsshows thethe NRCSNRCS ofof VH-polarization GF-3GF-3 SARSAR versusversus wind wind speeds speeds from from GRAPES-TYM, GRAPES-TYM, in in which which the the color color represents represents the theradar radar incidence incidence angle. angle. Figure Figure5b shows 5b shows the average the average NRCS NRCS of VH-polarization of VH-polarization Gaofen-3 Gaofen-3 SAR versus SAR windversus speeds wind betweenspeeds between 10 m/s and10 m/s 40 m/sand for40 m/s a 5 m/sfor a bin 5 m/s at various bin at various incidence incidence angle ranges. angle Inranges. general, In general,it is found it is that found the NRCSthat the of NRCS VH-polarization of VH-polarization GF-3 SAR GF-3 linearly SAR enhanceslinearly enhances with the with growth the ofgrowth wind ofspeed. wind This speed. finding This isfinding consistent is consistent with the with conclusions the conclusions of several of previous several previous studies [ 18studies,20,31 ].[18,20,31]. It is also Itshown is also that shown the slop that between the slop NRCS between of VH-polarization NRCS of VH-polarization GF-3 SAR is GF-3 almost SAR unique is almost for various unique radar for variousincidence radar angles incidence at moderate angles to strongat moderate winds. to The strong range winds. of the referenceThe range GRAPES-TYM of the reference wind GRAPES- speed is TYMup to wind 40 m/s speed and is no up indication to 40 m/s ofand the no signal indication saturation of the problem signal saturation was encountered problem inwas the encountered application inof the traditional application wind of retrieval traditional algorithms wind retrieval for co-polarization algorithms for C-band co-polarization SAR. The C-band black line SAR. in The Figure black5a linerepresents in Figure the 5a linear represents fitting the result linear between fitting result VH-polarization between VH-polar GF-3 NRCSization and GF-3 wind NRCS speed. and wind speed.

Figure 5.5. ((aa)) TheThe relationship relationship between between VH-polarization VH-polarization GF-3 GF-3 NRCS NRCS and and wind wind speed, speed, in which in which the black the blackline represents line represents the linear the fittinglinear fitting result. result. (b) Average (b) Average NRCS ofNRCS VH-polarization of VH-polarization Gaofen-3 Gaofen-3 SAR versus SAR windversus speed wind betweenspeed between 10 m/s 10 and m/s 40 and m/s 40 for m/s a 5 for m/s a 5 bin m/s at bin various at various incidence incidence angle angle ranges. ranges.

Figure 6 shows the NRCS of the entire VH-polarization GF-3 SAR image versus collocated radar Figure6 shows the NRCS of the entire VH-polarization GF-3 SAR image versus collocated radar incidence angle, in which the color represents the wind speed. A fluctuating relationship can be incidence angle, in which the color represents the wind speed. A fluctuating relationship can be observed between the NRCS of VH-polarization GF-3 SAR and the radar incidence angle, which is observed between the NRCS of VH-polarization GF-3 SAR and the radar incidence angle, which is similar to the analyzed results in [26]. The black lines in Figure 6 also represent the fitting results between the VH-polarization GF-3 NRCS and the radar incidence angle.

Sensors 2018, 18, 412 7 of 15 similar to the analyzed results in [26]. The black lines in Figure6 also represent the fitting results betweenSensors the 2018 VH-polarization, 18, x FOR PEER REVIEW GF-3 NRCS and the radar incidence angle. 7 of 14

FigureFigure 6. NRCS 6. NRCS ofentire of entire VH-polarization VH-polarization GF-3GF-3 SAR image image versus versus collocated collocated radar radar incidence incidence angle, angle, in which the black lines represent the fitting results between VH-polarization GF-3 NRCS and radar in which the black lines represent the fitting results between VH-polarization GF-3 NRCS and radar incidence angle. incidence angle. 3.2. Tuning the Empirical Algorithm through the Collocated Dataset 3.2. Tuning the Empirical Algorithm through the Collocated Dataset Recently, some studies have made great efforts to retrieve winds from cross-polarization R-2 Recently,SAR images some [19–25,32]. studies To have date, made several great cross-polarization efforts to retrieve wind winds retrieval from models cross-polarization have been R-2 SAR imagesexploited. [19 A– 25C-band,32]. Tocross-polarized date, several ocean cross-polarization surface wind retrieval wind retrieval model for models dual-polarization have been exploited.SAR A C-band(C-2POD) cross-polarized model has been ocean widely surface used wind for strong retrieval wind model retrieval for from dual-polarization R-2 SAR images SAR that (C-2POD)only modelinclude has been the wind widely speed used term. for strongThe C-2POD wind model retrieval takes from the R-2general SAR formulation: images that only include the wind speed term. The C-2POD model takes the generalσ0 = aU10 formulation: + b [dB] (2)

where in σ0 is the VH-polarization NRCS in units of dB, U10 is the wind speed at 10 m height above σ0 sea surface, a and b are the constants determined= aU10 + bfor [dB] different R-2 mode data, i.e., fine quad- (2) polarization [19–21] and dual-polarization ScanSAR mode [22,23,32]. 0 where inInσ fact,is the dependence VH-polarization of sea surface NRCS wind in unitsspeed ofon NRCS dB, U is10 trueis the for C-band wind speedR-2 and at GF-3 10 SAR. m height aboveHowever, sea surface, based a andon the b areanalysis the constants above, it determinedwas found that for differentVH-polarization R-2 mode NRCS data, of GF-3 i.e., fineSAR quad- polarizationfluctuate [ 19over–21 the] and radar dual-polarization incidence angle. For ScanSAR convenient mode application, [22,23,32]. we have made an attempt to Indirectly fact, theretrieve dependence winds from of sea VH-polarization surface wind speed NRCS on of NRCS GF-3 isSAR true at for various C-band radar R-2 andincidence GF-3 SAR. However,angle basedranges. on Referring the analysis to recent above, achievements it was found of wind that re VH-polarizationtrieval for cross-polarization NRCS of GF-3 S-1 SAR [26], fluctuate overan the empirical radar incidence algorithm angle.is now Forproposed, convenient including application, both terms we of wind have speed made and an radar attempt incidence to directly retrieveangle, winds and is from described VH-polarization as follows: NRCS of GF-3 SAR at various radar incidence angle ranges.

Referring to recent achievements of windσ0 =retrieval f1(1 + α‖f2 for‖) + cross-polarization β [dB] S-1 SAR [26], an empirical(3) algorithm is now proposed, including both terms of wind speed and radar incidence angle, and is in which, f1 is the function of VH-polarization NRCS taking a similar formulation to Equation (3): described as follows: 0 f = AU + B (4) σ = f1(11 + αkf102k) + β [dB] (3) f2 is the function of radar incidence angle normalized into [−1,1] for various radar incidence angle in which, f is the function of VH-polarization NRCS taking a similar formulation to Equation (3): ranges1 as GF-3 SAR has a different performance of instrumental noise for each radar beam:

2 f2f 1= =C1θAU + 10C2θ+ +B C3 (5) (4) and the coefficients α, β, A, B and matrix C are the tuned constants, as shown in Table 1. The f2 is thecategories function of θ of are radar ranged incidence as [10°~20°], angle from normalized 20° to 40° for into a 5° [− bin1,1] and for [40°~50°]. various radar incidence angle ranges as GF-3 SAR has a different performance of instrumental noise for each radar beam:

2 f2 = C1θ + C2θ + C3 (5)

Sensors 2018, 18, 412 8 of 15

α β Sensorsand the 2018 coefficients, 18, x FOR PEER, ,REVIEW A, B and matrix C are the tuned constants, as shown in Table1. The categories8 of 14 of θ are ranged as [10◦~20◦], from 20◦ to 40◦ for a 5◦ bin and [40◦~50◦]. Table 1. Coefficients in Equations (3)–(5), which are determined from the collocated data in our study. Table 1. Coefficients in Equations (3)–(5), which are determined from the collocated data in our study. A 0.152 B −28.378 Aθ 0.152 10°~20° 20°~25° B 25°~30°−28.378 30°~35° 35°~40° 40°~50° ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ θ α10 ~20−0.016 20 −~250.025 25−0.013~30 −0.00430 ~35−0.03135 ~40−0.009 40 ~50 α −0.016 −0.025 −0.013 −0.004 −0.031 −0.009 β 0.023 −0.164 0.198 0.359 −0.193 0.012 β 0.023 −0.164 0.198 0.359 −0.193 0.012 C1 C10.015 0.015 −0.001−0.001 −−0.0800.080 −0.046−0.046 0.198 0.198 0.067 0.067 − − − − C2 C2 0.400−0.400 0.360−0.360 4.264 4.264 2.972 2.972 −15.139 15.139−5.675 5.675 C3 −22.428 −16.133 −81.322 −73.816 261.555 93.644 C3 −22.428 −16.133 −81.322 −73.816 261.555 93.644

We collectedcollected many many matchups matchups from from the threethe three VH-polarization VH-polarization GF-3 SARGF-3 images SAR fromimages the typhoons.from the typhoons.As mentioned As mentioned above, the above, matchups the withmatchups wind with speeds wind smaller speeds than smaller 10 m/s than are 10 excluded m/s are in excluded the tuning in theprocess tuning in orderprocess to in give order reasonable to give reasonable fitted results fitted using results the least-squares using the least-squares method. Figure method.7 shows Figure that 7 showsthe correlation that the (COR)correlation between (COR) the between observed the NRCS obse andrved the NRCS simulated and the values simu islated 0.7, indicatingvalues is 0.7, the indicatingproposed empiricalthe proposed algorithm empirical is suitable algorithm for typhoonis suitable wind for typhoon retrieval. wind retrieval.

Figure 7. Comparison between observed NRCS of VH-polarization GF-3 SAR and simulated results using the empirical algorithm.

4. Validation Validation In addition to the three images used for tuning the algorithm, there are four VH-polarization GF-3 SAR imagesimages takentaken from from typhoons typhoons Noru, Noru, Hato Hato and and Talim. Talim. The The retrieval retrieval results results are comparedare compared with withmeasurements measurements from from the Windsat the Windsat polarimetric polarimetric radiometer radiometer at 0.25 at◦ 0.25°grids, grids, as there as there are more are more than than four fourhundred hundred matchups matchups in the in collocated the collocated dataset. dataset. Figure 88 showsshows thethe retrievedretrieved windwind mapsmaps correspondingcorresponding toto thethe fourfour GF-3GF-3 imagesimages inin FigureFigure3 .3. It It is is necessary to establish that retrieved winds with wind speeds smaller than 10 m/s are are not not reliable. reliable. Although thethe Windsat Windsat wind wind speed speed is up is toup 40 to m/s, 40 them/s, retrieved the retrieved winds dowinds not encounter do not encounter the saturation the saturationproblem as problem found inas thefound application in the application of C-band of co-polarizationC-band co-polarization GMFs. GMFs. Discontinuities Discontinuities exist in exist the inretrieved the retrieved winds mapswinds because maps thebecause GF-3 the SAR GF-3 image SAR acquired image in acquired GLO and in WSC GLO mode and isWSC comprised mode ofis comprisedseveral radar of beams.several Inradar particular, beams. theIn particular, retrieved TE the of retrieved typhoon TE Lester of typhoon shown inLester Figure shown8d is distorted,in Figure 8dbecause is distorted, of the instrumentalbecause of the noise instrumental of the radar nois beam.e of the For radar this beam. reason, For the this retrieved reason, wind the retrieved speed is windsignificantly speed is different significantly around different the edge around of each the radar edge beam of each than radar in other beam regions. than in other regions.

Sensors 2018, 18, 412 9 of 15 Sensors 2018, 18, x FOR PEER REVIEW 9 of 14

Sensors 2018, 18, x FOR PEER REVIEW 9 of 14

FigureFigure 8. 8.The The retrievedretrieved wind wind maps maps of the of four the fourS-1 images. S-1 images. (a) Lionrock (a) Lionrock acquired on acquired 29 August on at 2908:32 August UTC. at (b) Lester acquired on 31 August at 03:15 UTC. (c) Gaston acquired on 1 September at 20:30 UTC. 08:32 UTC. (b) Lester acquired on 31 August at 03:15 UTC. (c) Gaston acquired on 1 September at (dFigure) Lester 8. Theacquired retrieved on 4wind September maps of atthe 16:31 four S-1UTC. images. (a) Lionrock acquired on 29 August at 08:32 UTC. 20:30 UTC. (d) Lester acquired on 4 September at 16:31 UTC. (b) Lester acquired on 31 August at 03:15 UTC. (c) Gaston acquired on 1 September at 20:30 UTC. A( dcomparison) Lester acquired between on 4 September SAR-derived at 16:31 wind UTC. speeds and measurements from Windsat grids data is Ashown comparison in Figure between 9. It is SAR-derivedshown that there wind is speedsaround anda 5.5 measurements m/s differencefrom in wind Windsat speed grids RMSE data is shownbetweenA in comparison the Figure retrieval9. between It results is shown from SAR-derived thatVH-polarizatio there wind is around speedsn GF-3 andSAR a 5.5 measurements images m/s difference and Windsat from in Windsat grids wind wind speedgrids speeds data RMSE betweenandis shown between theretrieval in Figure10 m/s results 9.and It is40 from shown m/s VH-polarization for that a 2there m/s isbin, around GF-3in which SARa 5.5 the imagesm/s error difference and bar Windsat represents in wind grids speedthe wind standard RMSE speeds anddeviationbetween between the(STD) 10 retrieval m/s of wind and results speed 40 m/sfrom at each forVH-polarizatio a bin. 2 m/s As me bin,nntioned GF-3 in which SAR in Sectionimages the error and2, the Windsat bar “de-noised” represents grids windprocedure the speeds standard is deviationnotand applied between (STD) for of 10the windm/s VH-polarization and speed 40 atm/s each for GF-3 bin.a 2SAR m/s As image mentionedbin, in in whichthis study. in Sectionthe Weerror consider2 ,bar the represents “de-noised” this probably the procedure standard explains is notwhy applieddeviation the analyzed for (STD) the VH-polarizationof result wind isspeed worse at each GF-3than bin.the SAR As4.5 image mem/sntioned RMSE in this inof study. Section wind We speeds 2, considerthe “de-noised” in [23] this when probably procedure comparing explains is not applied for the VH-polarization GF-3 SAR image in this study. We consider this probably explains whyretrieved the analyzed wind speed result in is hurricanes worse than from the VH-polarization 4.5 m/s RMSE R-2 of SAR wind images speeds with in [Hurricane23] when Research comparing why the analyzed result is worse than the 4.5 m/s RMSE of wind speeds in [23] when comparing retrievedDivision wind (HRD) speed wind in vectors hurricanes of the from National VH-polarization Oceanic and R-2 Atmospheric SAR images Administration with Hurricane (NOAA). Research However,retrieved thewind validation speed in stillhurricanes shows thatfrom VH-polarization VH-polarization GF-3 R-2 SAR SAR images has the with capability Hurricane to derive Research high DivisionDivision (HRD) (HRD) wind wind vectors vectors of of the the National National OceanicOceanic and and Atmospheric Atmospheric Administration Administration (NOAA). (NOAA). wind speeds from typhoons, due to the fact that cross-polarization does not saturate as easily as the However,However, the the validation validation still still shows shows that that VH-polarization VH-polarization GF-3 GF-3 SAR SAR has has the the capability capability to derive to derive high high co-polarization signal. windwind speeds speeds from from typhoons, typhoons, due due to to the the fact fact thatthat cross-polarizationcross-polarization does does not not sa saturateturate as easily as easily as the as the co-polarizationco-polarization signal. signal.

Figure 9. Comparison between SAR-derived wind speeds and measurements from Windsat grid winds between 10 m/s and 40 m/s for a 2 m/s bin, in which error bar represents the standard deviation FigureFigure 9. Comparison 9. Comparison between between SAR-derived SAR-derived wind wind speeds spee andds and measurements measurements from from Windsat Windsat grid grid winds (STD) of wind speed at each bin. betweenwinds 10 between m/s and 10 40m/s m/s and for 40 am/s 2 m/sfor a bin,2 m/s in bin, which in which error error bar represents bar represents the the standard standard deviation deviation (STD) (STD) of wind speed at each bin. of wind speed at each bin.

Sensors 2018, 18, 412 10 of 15 Sensors 2018, 18, x FOR PEER REVIEW 10 of 14

5.5. Discussion Discussion BecauseBecause there there are are not not many many matchups matchups at at wind wind speed speedss greater greater than 30 m/s, as as shown shown in in Figure Figure 9,9, wewe also also compared compared the SAR-derived winds with those from the 0.12 × 0.12°0.12◦ gridsgrids GRAPES-TYM GRAPES-TYM model model throughthrough the the four four GF-3 GF-3 VH-polarization VH-polarization SAR SAR images images shown shown in in Figure Figure 3 in order to further investigate thethe accuracy accuracy for for strong strong wind wind retrieval. retrieval. The The GRAPES GRAPES-TYM-TYM wind wind maps maps are are presented presented in Figure in Figure 10, 10in , whichin which the the rectangles rectangles represent represent the the spatial spatial coverage coverage of the of the four four VH-polarization VH-polarization GF-3 GF-3 SAR SAR images. images.

FigureFigure 10. 10. GRAPES-TYMGRAPES-TYM wind wind maps, maps, in in which which the the re rectanglesctangles represent represent the the coverage coverage of of the the correspondingcorresponding three three VH-polarization VH-polarization GF-3 GF-3 SAR SAR images images in in Figure Figure 3.3 (.(a) aGRAPES-TYM) GRAPES-TYM wind wind maps maps of typhoonof typhoon e on e 4 onAugust 4 August 2017 at 2017 09:00 at UTC. 09:00 ( UTC.b) GRAPES-TYM (b) GRAPES-TYM wind maps wind of mapstyphoon of Noru typhoon on 4 Noru August on 20174 August at 21:00 2017 UTC. at 21:00 (c) GRAPES-TYM UTC. (c) GRAPES-TYM wind maps windof typhoon maps ofHato typhoon on 22 HatoAugust on 2017 22 August at 22:00 2017 UTC. at (22:00d) GRAPES-TYM UTC. (d) GRAPES-TYM wind maps of wind typhoon maps Talim of typhoon on 16 TalimSeptember on 16 2017 September at 09:00 2017 UTC. at 09:00 UTC.

The comparison between SAR-derived winds and GRAPES-TYM winds is shown in Figure 11. The comparison between SAR-derived winds and GRAPES-TYM winds is shown in Figure 11. It is found that the RMSE of wind speed is 5.1 m/s, which is less than the statistical analysis exhibited It is found that the RMSE of wind speed is 5.1 m/s, which is less than the statistical analysis exhibited in Figure 8. This is not surprising as unique wind data were used for tuning and validating the in Figure8. This is not surprising as unique wind data were used for tuning and validating the proposed empirical algorithm. However, it is an obvious conclusion that the proposed algorithm proposed empirical algorithm. However, it is an obvious conclusion that the proposed algorithm works for typhoon winds retrieval from VH-polarization GF-3 SAR images without encountering the works for typhoon winds retrieval from VH-polarization GF-3 SAR images without encountering the signal saturation problem when the wind speed is up to 40 m/s. Therefore, cross-polarization GF-3 signal saturation problem when the wind speed is up to 40 m/s. Therefore, cross-polarization GF-3 SAR is an effective method for monitoring strong winds in typhoon conditions. SAR is an effective method for monitoring strong winds in typhoon conditions.

Sensors 2018, 18, 412 11 of 15 Sensors 2018, 18, x FOR PEER REVIEW 11 of 14 Sensors 2018, 18, x FOR PEER REVIEW 11 of 14

FigureFigure 11. 11.Comparison Comparison between between SAR-derived SAR-derived wind windspeeds speedsusing the using algorithm the herein algorithm and GRAPES- herein and TYM winds. GRAPES-TYMFigure 11. Comparison winds. between SAR-derived wind speeds using the algorithm herein and GRAPES- TYM winds. We also compared the SAR-derived results with GRAPES-TYM winds using the existing four Wecross-polarization also compared algorithms the SAR-derived proposed in results [20,22,23,25]. with GRAPES-TYM Most of these windshave the using same the basic existing We also compared the SAR-derived results with GRAPES-TYM winds using the existing four fourformulation cross-polarization as Equation algorithms (3) and are proposed tuned through in [20 quad-polarization,22,23,25]. Most or of dual-polarization these have the R-2 same SAR basic cross-polarization algorithms proposed in [20,22,23,25]. Most of these have the same basic formulationdata. Although as Equation the algorithm (3) and proposed are tuned in [25] through includ quad-polarizationes both term of wind orspeed dual-polarization and incidence angle, R-2 SAR formulation as Equation (3) and are tuned through quad-polarization or dual-polarization R-2 SAR the NRCS has a linear relationship of the first order with the incidence angle. Figure 12 shows that data.data. Although Although the the algorithm algorithm proposed proposed inin [[25]25] includesincludes both both term term of of wind wind speed speed and and incidence incidence angle, angle, the RMSE of wind speed is 6.4, 9.6, 6.7 and 10.9 m/s using the algorithms in [20,22,23,25], respectively. thethe NRCS NRCS has has a linear a linear relationship relationship of of the the first first order order with with the the incidence incidence angle. angle. Figure Figure 12 12 shows shows that that the RMSEThis of windanalysis speed shows is that 6.4, these 9.6, 6.7 algorithms and 10.9 all m/s perform using less the well algorithms than the results in [20 ,achieved22,23,25], using respectively. the theproposed RMSE of algorithm wind speed for isnoisy 6.4, 9.6,GF-3 6.7 SAR and data. 10.9 m/s using the algorithms in [20,22,23,25], respectively. ThisThis analysis analysis shows shows that that these these algorithms algorithms allall performperform less less well well than than the the results results achieved achieved using using the the proposedproposed algorithm algorithm for for noisy noisy GF-3 GF-3 SAR SAR data. data.

Figure 12. Comparison between SAR-derived wind speeds using the existing three algorithms and GRAPES-TYM winds. (a) using the algorithm proposed in [20]. (b) using the algorithm proposed in [22]. (c) using the algorithm proposed in [23]. (d) using the algorithm proposed in [25]. Figure 12. Comparison between SAR-derived wind speeds using the existing three algorithms and Figure 12. Comparison between SAR-derived wind speeds using the existing three algorithms and GRAPES-TYM winds. (a) using the algorithm proposed in [20]. (b) using the algorithm proposed in [22]. GRAPES-TYM winds. (a) using the algorithm proposed in [20]. (b) using the algorithm proposed (c) using the algorithm proposed in [23]. (d) using the algorithm proposed in [25]. in [22]. (c) using the algorithm proposed in [23]. (d) using the algorithm proposed in [25].

Sensors 2018, 18, 412 12 of 15

6. Conclusions and Summary To date, several co-polarization GMFs at C-band have been developed [3–7], which describe an empirical relationship between SAR NRCS and wind vector. These co-polarization GMFs have been successfully applied for various C-band SAR data, i.e., Envisat-ASAR [10], R-1/2 SAR [11,21], S-1 SAR [12] and GF-3 SAR [39,40]. However, the SAR-measured NRCS encounters a saturation problem under strong wind conditions [17,18,38], causing a limitation of the co-polarization GMF applicability. Recent achievements of R-2 signatures in cross-polarization reveal that the NRCS of cross-polarization SAR has a strong linear relationship with wind speed and it was found that the saturation boundary reaches 55 m/s [18]. The results proposed in [26] show that S-1 SAR in cross-polarization is also useful for wind monitoring. In our study, we have investigated the dependence of wind speed and of radar incidence angle on NRCS of VH-polarization GF-3 SAR and then tuned an empirical wind retrieval algorithm. Seven GF-3 SAR images in VH-polarization were collected from the China Sea, and they were acquired in four typhoons during the summer of 2017. All images were processed as L-1B products, in which discontinuities still exist near the edge of different signature beams due to the instrumental noise. Among these images, there are three images collocated with GRAPES-TYM wind fields, which had good performance for the typhoon cases in the northwest Pacific Ocean and South China Sea in 2013 [35]. Analyzed results show that the NRCS linearly increases with wind speed without encountering the signal saturation problem for wind speeds up to 40 m/s. However, it was found that the NRCS of VH-polarization GF-3 SAR is not simply linearly related with radar incidence angle, which is probably caused by the different instrumental noise of each radar beam. Therefore, based on the recent achievements for strong wind retrieval C-band cross-polarization R-2 SAR [20] and S-1 SAR [26], an empirical algorithm has been tuned for wind retrieval for VH-polarization GF-3 SAR in typhoons through the collocated dataset, which takes into account the terms of wind speed and radar incidence angle. The proposed empirical algorithm was applied for an additional four images in typhoons and the SAR-derived winds have been validated against Windsat winds at 0.25◦ grids. The comparison shows around 5.5 m/s RMSE of wind speed. We also compared the retrieved wind speed with GRAPES-TYM wind speeds, showing the RMSE of wind speed to be 5.1 m/s. Although the accuracy of wind retrieval for VH-polarization GF-3 SAR is greater than that (around 2 m/s STD) for co-polarization C-band SAR using the traditional methodology of GMF, the advantage of the proposed empirical algorithm is that it is able to operate without prior knowledge of wind direction and works under strong wind conditions without encountering the signal saturation problem. It is necessary to establish that the proposed algorithm only works for wind speeds greater than 10 m/s and homogeneous sub-scenes, as the SAR signature is weak in the presence of noise and rainfall in typhoons. In summary, we conclude that the proposed algorithm is a promising potential technique for wind retrieval from C-band VH-polarization GF-3 SAR image in typhoons. Increasingly, more GF-3 SAR images will be captured in extreme winds by our institute and the National Satellite Ocean Application Service (NSOAS). In particular, the dependence of instrumental noise on NRCS from VH-polarization GF-3 SAR acquired in GLO and WSW mode can be investigated. In future, we could further tune the empirical algorithm in order to improve the accuracy of retrieved winds from VH-polarization GF-3 SAR images.

Acknowledgments: The authors also appreciate GF-3 SAR images, which are downloaded via http://dds.nsoas. org.cn as authorized account issued by NSOAS under the contract of Specific Project of Chinese High Resolution Earth Observation System (No. 41-Y20A14-9001-15/16). Windsat winds data are kindly provided by Remote Sensing System (RSS) team and openly accessed via the sever: ftp.remss.com. The information of typhoons was provided by National Meteorological Center via http://www.typhoon.org.cn. The research is partly supported by National Key Research and Development Program of China under Grant Nos. 2017YFA0604901, 2016YFC1401605 and 2016YFC1401905 and National Natural Science Foundation of China under Grant Nos. 41776183 and 41376010. Sensors 2018, 18, 412 13 of 15

Author Contributions: Weizeng Shao, Xinzhe Yuan and Jian Sun came up the original idea and carried out the experiments. Xinzhe Yuan designed the experiments and processed the GF-3 SAR images. Wei Zhou provided great help for the winds fields in typhoons. Weizeng Shao, Yexin Sheng and Xinzhe Yuan analyzed the dataset. Qingjun Zhang provided great help for the data analysis. All authors contributed to the writing and revising of the manuscript. Conflicts of Interest: The authors declare no conflict of interest.

Appendix A

Table A1. Information of seven VH-polarization GF-3 SAR images and corresponding typhoons.

Acquisition Time Cyclone Eye Maximum Wind Radius of Maximum Central Typhoon (YYYY-MM-DD) (◦ E, ◦ N) Speed (m/s) Wind Speed (km) Pressure (hPa) Noru 2017-08-02 21:09 135.6, 26.1 45 50 950 Noru 2017-08-04 09:12 131.4, 28.4 40 100 960 Noru 2017-08-04 21:26 130.7, 29.0 40 80 960 Hato 2017-08-22 22:23 117.1, 20.8 34 80 970 Doksuri 2017-09-13 22:14 114.3, 15.8 23 180 990 Talim 2017-09-14 21:29 124.3, 27.4 50 80 940 Talim 2017-09-16 09:30 126.1, 29.4 38 40 965

References

1. Li, X.F.; Zhang, J.A.; Yang, X.F.; Pichel, W.G.; DeMaria, M.; Long, D.; Li, Z.W. morphology from spaceborne synthetic aperture radar. Bull. Am. Meteorol. Soc. 2013, 94, 215–230. [CrossRef] 2. Li, X.F. The first Sentinel-1 SAR image of a typhoon. Acta Oceanol. Sin. 2015, 34, 1–2. [CrossRef] 3. Stoffelen, A.; Anderson, D. Scatterometer data interpretation estimation and validation of the CMOD4. J. Geophys. Res. 1997, 102, 5767–5780. [CrossRef] 4. Quilfen, Y.; Bentamy, A.; Elfouhaily, T.; Katsaros, K.; Tournadre, J. Observation of tropical cyclones by high-resolution scatterometry. J. Geophys. Res. 1998, 103, 7767–7786. [CrossRef] 5. Hersbach, H.; Stoffelen, A.; de Haan, S. An improved C-band scatterometer ocean geophysical model function: CMOD5. J. Geophys. Res. 2007, 112, 225–237. [CrossRef] 6. Hersbach, H. Comparison of C-Band scatterometer CMOD5.N equivalent neutral winds with ECMWF. J. Atmos. Ocean. Technol. 2010, 27, 721–736. [CrossRef] 7. Mouche, A.A.; Chapron, B. Global C-Band Envisat, RADARSAT-2 and Sentinel-1 SAR measurements in co-polarization and cross-polarization. J. Geophys. Res. 2015, 120, 7195–7207. [CrossRef] 8. Lehner, S.; Horstmann, J.; Koch, W.; Rosenthal, W. Mesoscale wind measurements using recalibrated ERS SAR images. J. Geophys. Res. 1998, 103, 7847–7856. [CrossRef] 9. Monaldo, F.M.; Thompson, D.R.; Beal, R.C.; Pichel, W.G. Comparison of SAR-derived wind speed with model predictions and ocean buoy measurements. IEEE Trans. Geosci. Remote Sens. 2002, 39, 2587–2600. [CrossRef] 10. Yang, X.F.; Li, X.F.; Zheng, Q.A.; Gu, X.; Pichel, W.G.; Li, Z.W. Comparison of ocean-surface winds retrieved from QuikSCAT scatterometer and Radarsat-1 SAR in offshore waters of the U.S. west coast. IEEE Geosci. Remote Sens. Lett. 2011, 8, 163–167. [CrossRef] 11. Yang, X.F.; Li, X.F.; Pichel, W.G.; Li, Z.W. Comparison of ocean surface winds from ENVISAT ASAR, Metop ASCAT scatterometer, buoy Measurements, and NOGAPS model. IEEE Trans. Geosci. Remote Sens. 2011, 49, 4743–4750. [CrossRef] 12. Monaldo, F.M.; Jackson, C.; Li, X.F.; Pichel, W.G. Preliminary evaluation of Sentinel-1A wind speed retrievals. IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens. 2015, 9, 2638–2642. [CrossRef] 13. Thompson, D.R.; Elfouhaily, T.M.; Chapron, B. Polarization ratio for microwave backscattering from the ocean surface at low to moderate incidence angles. In Proceedings of the IEEE International Geoscience and Remote Sensing Symposium, Seattle, WA, USA, 6–10 July 1998; pp. 1671–1673. 14. Alpers, W.; Brummer, B. Atmospheric boundary layer rolls observed by the synthetic aperture radar aboard the ERS-1 satellite. J. Geophys. Res. 1994, 99, 12613–12621. [CrossRef] Sensors 2018, 18, 412 14 of 15

15. Koch, W. Directional analysis of SAR images aiming at wind direction. IEEE Trans. Geosci. Remote Sens. 2004, 42, 702–710. [CrossRef] 16. Zhou, L.Z.; Zheng, G.; Li, X.F.; Yang, J.S.; Ren, L.; Chen, P.; Zhang, H.G.; Lou, X.L. An improved local gradient method for sea surface wind direction retrieval from SAR imagery. Remote Sens. 2017, 9, 671. [CrossRef] 17. Fois, F.; Hoogeboom, P.; Chevalier, F.L.; Stoffelen, A. Future ocean scatterometry: On the use of cross-polar scattering to observe very high winds. IEEE Trans. Geosci. Remote Sens. 2015, 53, 5009–5020. [CrossRef] 18. Hwang, P.; Fois, F. Surface roughness and breaking wave properties retrieved from polarimetric microwave radar backscattering. J. Geophys. Res. 2015, 120, 3640–3657. [CrossRef] 19. Zhang, B.; Perrie, W. Cross-polarized synthetic aperture radar: A new potential technique for hurricanes. B. Am. Meteorol. Soc. 2012, 93, 531–541. [CrossRef] 20. Vachon, P.W.; Wolfe, J. C-band cross-polarization wind speed retrieval. IEEE Geosci. Remote Sens. Lett. 2011, 3, 456–459. [CrossRef] 21. Zhang, B.; Perrie, W.; Vachon, P.W.; Li, X.F.; Pichel, W.G.; Guo, J. Ocean vector winds retrieval from C-band fully polarimetric SAR measurements. IEEE Geosci. Remote Sens. 2012, 50, 4252–4261. [CrossRef] 22. Zadelhoff, G.J.; Stoffelen, A.; Vachon, P.W.; Wolfe, J.; Horstmann, J.; Belmonte Rivas, M. Retrieving hurricane wind speeds using cross polarization C-band measurements. Atmos. Meas. Tech. 2014, 7, 437–449. [CrossRef] 23. Shen, H.; Perrie, W.; He, Y.J.; Liu, G. Wind speed retrieval from VH dual-polarization RADARSAT-2 SAR images. IEEE Trans. Geosci. Remote Sens. 2014, 52, 5820–5826. [CrossRef] 24. Hwang, P.; Stoffelen, A.; Zadelhoff, G.J.; Perrie, W.; Zhang, B.; Li, H. Cross polarization geophysical model function for C-band radar backscattering from the ocean surface and wind speed retrieval. J. Geophys. Res. 2015, 120, 893–909. [CrossRef] 25. Zhang, G.S.; Li, X.F.; Perrie, W.; Hwang, P.; Zhang, B.; Yang, X.F. A hurricane wind speed retrieval model for C-band RADARSAT-2 cross-polarization ScanSAR images. IEEE Trans. Geosci. Remote Sens. 2017, 55, 4766–4774. [CrossRef] 26. Huang, L.Q.; Liu, B.; Li, X.F.; Zhang, Z.; Yu, W. Technical evaluation of Sentinel-1 IW mode cross-pol radar backscattering from the ocean surface in moderate wind condition. Remote Sens. 2017, 9, 854. [CrossRef] 27. Hwang, P.; Zhang, B.; Perrie, W. Depolarized radar return for breaking wave measurement and hurricane wind retrieval. Geophys. Res. Lett. 2010, 37, 70–75. [CrossRef] 28. Hwang, P.; Zhang, B.; Toporkov, J.; Perrie, W. Comparison of composite Bragg theory and quad-polarization radar backscatter from Radarsat-2: With applications to wave breaking and high wind retrieval. J. Geophys. Res. 2010, 115, 246–255. [CrossRef] 29. Voronovich, A.; Zavorotny, V. Full-polarization modeling of monostatic and bistatic radar scattering from a rough sea surface. IEEE Trans. Antennas Propag. 2014, 62, 1362–1371. [CrossRef] 30. Zhang, B.; Perrie, W.; Zhang, J.A.; Uhlhorn, E.W.; He, Y.J. High-resolution hurricane vector winds from C-band dual-polarization SAR observations. J. Atmos. Ocean. Technol. 2014, 31, 272–286. [CrossRef] 31. Zhang, B.; Perrie, W. Recent progress on high wind-speed retrieval from multi-polarization SAR imagery: A review. Int. J. Remote Sens. 2014, 35, 4031–4045. [CrossRef] 32. Shen, H.; Perrie, W.; He, Y.J. Evaluation of hurricane wind speed retrieval from cross-dual-pol SAR. Int. J. Remote Sens. 2016, 37, 599–614. [CrossRef] 33. Lin, C.; Betto, M.; Belmonte Rivas, M.; Stoffelen, A.; Kloe, J.D. EPS-SG wind scatterometer concept tradeoffs and wind retrieval performance assessment. IEEE Trans. Geosci. Remote Sens. 2012, 50, 2458–2472. [CrossRef] 34. Belmonte Rivas, M.; Stoffelen, A.; Zadelhoff, G. The benefit of HH and VV polarizations in retrieving extreme wind speeds for an ASCAT-Type scatterometer. IEEE Trans. Geosci. Remote Sens. 2013, 52, 4273–4280. [CrossRef] 35. Zhang, J.; Ma, S.H.; Chen, D.H.; Huang, L.P. The improvements of GRAPES_TYM and its performance in northwest Pacific Ocean and South China Sea in 2013. J. Trop. Meteorol. 2017, 33, 64–73. 36. Meissner, T.; Wentz, F.J. The emissivity of the ocean surface between 6 and 90 GHz over a large range of wind speeds and earth incidence angles. IEEE Trans. Geosci. Remote Sens. 2012, 50, 3004–3026. [CrossRef] 37. Li, X.M.; Lehner, S.; Bruns, T. Ocean wave integral parameter measurements using Envisat ASAR wave mode data. IEEE Trans. Geosci. Remote Sens. 2011, 49, 155–174. [CrossRef] 38. Shao, W.Z.; Li, X.F.; Hwang, P.; Zhang, B.; Yang, X.F. Bridging the gap between cyclone wind and wave by C-band SAR measurements. J. Geophys. Res. 2017, 122, 6714–6724. [CrossRef] Sensors 2018, 18, 412 15 of 15

39. Shao, W.Z.; Shen, Y.X.; Sun, J. Preliminary assessment of wind and wave retrieval from Chinese Gaofen-3 SAR imagery. Sensors 2017, 17, 1705. [CrossRef][PubMed] 40. Wang, H.; Yang, J.S.; Mouche, A.A.; Shao, W.Z.; Zhu, J.H.; Ren, L.; Xie, C.H. GF-3 SAR ocean wind retrieval: The first view and preliminary assessment. Remote Sens. 2017, 9, 694. [CrossRef]

© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).