Analysis of Temporal and Spatial Variability of Fronts on the Amery Ice Shelf Automatically Detected Using Sentinel-1 SAR Data

remote sensing

Article Analysis of Temporal and Spatial Variability of Fronts on the Amery Ice Shelf Automatically Detected Using Sentinel-1 SAR Data

Tingting Zhu 1,†, Xiangbin Cui 2,† and Yu Zhang 3,*

1 State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan 430079, China; [email protected] 2 Polar Research Institute of China, Shanghai 200136, China; [email protected] 3 Chinese Antarctic Center of Surveying and Mapping, Wuhan University, Wuhan 430079, China * Correspondence: [email protected] † These authors contributed equally to this work.

Abstract: The Amery Ice Shelf (AIS) dynamics and mass balance caused by iceberg calving and basal melting are significant in the ocean climate system. Using satellite imagery from Sentinel-1 SAR, we monitored the temporal and spatial variability of the frontal positions on the Amery Ice Shelf, Antarctica, from 2015 to 2021. In this paper, we propose an automatic algorithm based on the SO-CFAR strategy and a profile cumulative method for frontal line extraction. To improve the accuracy of the extracted frontal lines, we developed a framework combining the Constant False Alarm Rate (CFAR) and morphological image-processing strategies. A visual comparison between the proposed algorithm and state-of-the-art algorithm shows that our algorithm is effective in these cases including rifts, icebergs, and crevasses as well as ice-shelf surface structures. We present a Citation: Zhu, T.; Cui, X.; Zhang, Y. detailed analysis of the temporal and spatial variability of fronts on AIS that we find, an advance Analysis of Temporal and Spatial of the AIS frontal line before the D28 calving event, and a continuous advance after the event. The Variability of Fronts on the Amery Ice study reveals that the AIS extent has been advanced at the rate of 1015 m/year. Studies have shown Shelf Automatically Detected Using that the frontal location of AIS has continuously expanded. From March 2015 to May 2021, the frontal Sentinel-1 SAR Data. Remote Sens. location of AIS expanded by 6.5 km; while the length of the AIS frontal line is relatively different after 2021, 13, 3528. https://doi.org/ the D28 event, the length of the frontal line increased by about 7.5% during 2015 and 2021 (255.03 km 10.3390/rs13173528 increased to 273.5 km). We found a substantial increase in summer advance rates and a decrease in winter advance rates with the seasonal characteristics. We found this variability of the AIS frontal Academic Editor: Nereida Rodriguez-Alvarez line to be in good agreement with the ice flow velocity.

Received: 7 July 2021 Keywords: frontal position; constant false alarm rate; Sentinel-1 SAR data; Amery Ice Shelf; Accepted: 1 September 2021 iceberg calving Published: 5 September 2021

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in 1. Introduction published maps and institutional affil- The fronts of ice shelves in Antarctica are critical interfaces, between the ice sheet iations. and ocean, and their geometry and variation can significantly impact the ice shelf-ocean interaction (e.g., basal melting, and iceberg calving), which further affects upstream ice sheet dynamics and sea-level rise [1]. In September 2019, the Amery Ice Shelf (AIS) calved a 1636-km2 iceberg named D28. The last major calving event from AIS was in the early 1960s. Copyright: © 2021 by the authors. After the D28 detachment, the mass loss of AIS was approximately 315 billion tonnes due Licensee MDPI, Basel, Switzerland. to the iceberg calving at the front, which, in addition to the basal melting underneath, is This article is an open access article the key mass loss process. An AIS front advance will increase stress at the rift [2]. Before distributed under the terms and the D28 detachment, five rifts were actively propagated near the AIS front among seven conditions of the Creative Commons active rifts over the 13 ice shelves in Antarctica, where most active rifts were initiated at Attribution (CC BY) license (https:// the Amery Ice Shelf fronts [3–5]. Although ice-shelf calving does not influence the global creativecommons.org/licenses/by/ sea level, the location of ice-shelf fronts is an important parameter of ice-sheet dynamics, 4.0/).

Remote Sens. 2021, 13, 3528. https://doi.org/10.3390/rs13173528 https://www.mdpi.com/journal/remotesensing Remote Sens. 2021, 13, 3528 2 of 15

which contributes to sea-level rise [6–11]. Moreover, the tabular iceberg (D28) has to be monitored and tracked since the Chinese Antarctic Zhongshan Station, Russian Progress Station and Indian Bharati Station are near the AIS where the D28 broke off may cause shipping hazards. Sentinel-10s fine spatial resolution with 40 m and high repeat acquisitions have allowed for monthly D28 events or daily analysis of the rifts to change the detection and ice fronts dynamic in AIS, the third largest ice shelf in the east of Antarctica and one of the largest glacier drainage basins. Although attempts have been made to survey the front of AIS through airborne in- struments [12], remote satellite sensors can provide continuous measurements for the early detection of ice-shelf calving and fronts dynamics in remote locations as well as can monitor extreme climate conditions due to its remote location, snow, and ice-covered surface, and extremely cold and windy weather conditions [13–15]. The advance or retreat of ice-shelf fronts can be delineated by synthetic aperture radar (SAR) based on its all-weather and frequent repeat observation capabilities, because the microwave can penetrate thick clouds and darkness in polar regions, indicating that SAR may be the most advanced method for monitoring large-scale dynamics in ice-shelf fronts [8,16]. The current generation of SAR satellites including RADARSAT-2 and Sentinel-1 can provide high spatial resolution images with less than 40 m and high temporal availability within one day. High spatial resolution SAR images are easily acquired as open-source data since Sentinel-1 was launched in early 2014. Free Sentinel-1 SAR data access facilitates the possibility of providing long-term satellite observations for the location of ice-shelf fronts and glaciers. Perennial sea ice or fast ice (sea ice “fastened” to land and motionless with tides) might appear quite similar to shelf ice in remote sensing imagery. Separation of them is a challenging task due to changing sea ice conditions and is our focus in this study. The main goal in this study is to automatically detect the calving frontal dynamics on Sentinel-1 SAR imagery because the manual delineation of calving fronts from remote sensing imagery is extremely time-consuming. The Radarsat-1 Antarctica Mapping Project (RAMP) Antarctic Mapping Mission 1 (AMM) was used to classify the coastline of Antarctica in terms of surface structure patterns is close to the calving front. Mapping and classification for calving fronts of Antarctica based on the automated edge detection method were complemented by manual control [17]. The semi-automated detection method for glacier and ice shelf fronts from Sentinel-1 imagery was proposed based on deep learning technology, which requires enormous training samples [18]. The unsupervised method mostly attempts to focus on the edge detection method for Greenland using MODIS data, such as Sobel, by selecting the peak backscattering value along the direction of ice flow as the ice frontal points [19,20]. Limited by MODIS due to illumination or contamination by clouds, coastline products are incomplete [14]. Continuous observations can be achieved by SAR technology. Using Radarsat-1 SAR data, the coastline is extracted based on a locally adaptive threshold algorithm [21–23], which significantly removes the burden of conventional manual delineation methods. Out- crop rock, surface melting, crevasses, and rifts may cause large computation costs and inaccuracies. A thorough investigation of AIS is required as all of the aforementioned algorithms focus on the continental scale. Our work was inspired by the profile method proposed in [24,25], where the standard deviation (STD) and five-maximum value extraction algorithms were proposed. SAR data with speckle noise, high interclass, and high intraclass backscatter variability degrade the image details and decrease the separability among ocean, coastal rock outcrops, sea ice, and ice shelf. Considering these characteristics, we proposed a framework combining the Constant False Alarm Rate (CFAR) and profile method to automatically detect the ice-shelf fronts in AIS by Sentinel-1 SAR data. To deal with the problem caused by surface structures as well as rifts on the ice shelf and floating ice, the maximum cumulative value-based approach is used to find the frontal points of each profile. A morphological filter is then used to remove the icebergs. Finally, the time series of the frontal position displacement, as well as the regional dynamic of the ice shelf, is investigated. Remote Sens. 2021, 13, x FOR PEER REVIEW 3 of 16

Remote Sens. 2021, 13, 3528 series of the frontal position displacement, as well as the regional dynamic of the ice shelf,3 of 15 is investigated. In this work, following an overview of the scientific motivations and the Sentinel-1 SAR data used on AIS in Section 1, Section 2 comprises the methodology for the determi- nationIn of this glacier work, front following location an fields, overview with ofa focus the scientific on the proposed motivations detection and the framework. Sentinel-1 Subsequently,SAR data used the on results AIS in Sectionof the AIS1, Section front location2 comprises and analysis the methodology of more than for theseven-years- determi- timenation series of glacier (2015–2021) front location for their fields, dynamics with aar focuse presented on the proposedin Section detection3. Section framework. 3 also dis- cussesSubsequently, the AIS theD28 results calving of event. the AIS Co frontnclusions location are andgiven analysis in Section of more 4. than seven-years- time series (2015–2021) for their dynamics are presented in Section3. Section3 also 2.discusses Methods the AIS D28 calving event. Conclusions are given in Section4. 2. MethodsThe frontal line extraction algorithm consists of three steps: SO (Smallest of)-CFAR for binaryThe frontal classification line extraction [26–29], algorithm morphological consists image of three processing steps: SO [30,31], (Smallest andof)-CFAR maximal cumulativefor binary classification based frontal [ 26point–29], extraction morphological [32]. CFAR image detectors processing are [ 30adaptable,31], and threshold maximal detectorscumulative that based use frontalvarious point statistical extraction models [32 ].to CFARdetect detectorstarget returns are adaptable from the threshold ice shelf againstdetectors the that background use various clutter, statistical such as models sea ice toand detect ocean. target The threshold returns from for every the ice detect- shelf ingagainst cell the(sliding background window) clutter, is adaptive such as to sea maintain ice and ocean. a constant Thethreshold probability for of every a false detecting alarm (PFA)cell (sliding according window) to an isassumed adaptive background to maintain probability a constant probabilitydensity function of a false f ()x alarm. The (PFA) slid- ingaccording window to strategies an assumed include background cell averaging, probability greatest density and smallest function off (CFARx). The named sliding as CA-CFAR,window strategies GO-CFAR, include and cellSO-CFAR. averaging, We greatest use the and SO-CFAR smallest strategy of CFAR combined named as with CA- WeibullCFAR, GO-CFAR, statistical distribution and SO-CFAR. since We this use strategy the SO-CFAR can maximize strategy the combined f ()x on with SAR Weibull image statistical distribution since this strategy can maximize the f (x) on SAR image in this in this paper. Afterward, SO-CFAR is used to obtain the binary image. To remove the paper. Afterward, SO-CFAR is used to obtain the binary image. To remove the icebergs icebergs on the binary image after CFAR, we use morphological image processing such as on the binary image after CFAR, we use morphological image processing such as opening opening and closing to remove small fragments. The maximal cumulative value-based and closing to remove small fragments. The maximal cumulative value-based method is method is a strategy designed for frontal point extraction. In this step, we firstly constrain a strategy designed for frontal point extraction. In this step, we firstly constrain the ice the ice shelf in the defined borders region. Between two borders of the ice shelf, the lon- shelf in the defined borders region. Between two borders of the ice shelf, the longitudinal gitudinalprofiles are profiles generated are generated by an equal by divide. an equal Then, divide. the frontal Then, pointsthe frontal are determined points are bydeter- the minedcumulative by the maximal cumulative values. maximal Finally, values. the frontal Finally, lines the can frontal be automatically lines can be delineated. automatically The delineated.detailed flowchart The detailed is illustrated flowchart in Figureis illustrated1. in Figure 1.

Ice shelf detection Lines delineation

Sentinel-1 GRD Morphological Opening and HH SAR image data processing closing

Statistical distribution Ice shelf lateral Weibull distribution Borders constraints modeling boundary

Parameters MoLC estimation Longitudinal Equal divide estimation method profiles

False alarm SO-CFAR threshold Frontal points Cumulative maxima

Automatic Binary image Frontal lines delineation

Figure 1. Flowchart of thethe frontalfrontal lineslines detectiondetection algorithm. algorithm. The The left left panel panel is is for for the the ice ice shelf shelf detection detection and and the the right right panel panel is is for delineating frontal lines. for delineating frontal lines.

2.1. Ice shelf Detection Using CFAR Method The ice-shelf areas in Sentinel-1 SAR imagery can be detected using the CFAR method [26]. For CFAR modeling, we first model the distribution of the SAR backscatter using Weibull distribution since it is suitable for ocean and ice clutter modeling [27]. Then, the CFAR method Remote Sens. 2021, 13, x FOR PEER REVIEW 4 of 16

2.1. Ice shelf Detection Using CFAR Method The ice-shelf areas in Sentinel-1 SAR imagery can be detected using the CFAR Remote Sens. 2021, 13, 3528 method [26]. For CFAR modeling, we first model the distribution of the SAR backscatter4 of 15 using Weibull distribution since it is suitable for ocean and ice clutter modeling [27]. Then, the CFAR method is carried out based on a constant false alarm rate to identify the SAR ispixel carried as the out ice based shelf on label a constant. To classify false the alarm SAR rate image to identify into the the ice SAR shelf pixel label as and the the ice shelfback- label.ground To classifylabel, a theSO-CFAR SAR image based into detection the ice shelf appr labeloach and is used the background [33,34]. For label, a given a SO-CFAR pixel in basedSAR imagery, detection we approach first select is used a clutter [33,34 region]. For awith given a window pixel in SARsize of imagery, (2n + 1). we We first define select the a clutterguard regionarea with with sizes a window of (2m size + 1) of as (2n the + test 1). Wecell. define Except the for guard the guard area with area, sizes theof remaining (2m + 1) aspixels the test are cell.equally Except divided for the into guard two area, parts: the P1 remaining and P2 with pixels theare size equally of (n-m). divided The SO-CFAR into two parts:strategy P1 selects and P2 the with minimum the sizeof value (n-m). P of The P1SO-CFAR and P2, where strategy P1 and selects P2 theare minimumthe mean values value Pof of the P1 sum and P2,intensity where (mean P1 and power) P2 are theof the mean loca valuesl clutter ofthe region. sum intensityThen, pixels (mean in power)the clutter of theregions local are clutter compared region. with Then, P, pixels and pixels in the larger clutter than regions P are are selected compared for Weibull with P, and modeling. pixels largerIn this than paper, P are the selected Weibull for distribution-based Weibull modeling. InSO-CFAR this paper, method the Weibull is used distribution-based to detect the ice SO-CFARshelf pixel method in the isSAR used imagery. to detect Then, the ice the shelf adap pixeltive in threshold the SAR imagery. for the Then,pixel theis calculated adaptive thresholdusing the for SO-CFAR the pixel strategy. is calculated Finally, using if the the SO-CFAR intensity strategy. of the given Finally, pixel if the is intensitylarger than of thethe giventhreshold pixel isT larger, it will than be labeled the threshold as an iceT, itshelf will with be labeled the value as an 1; iceotherwise, shelf with it is the labeled value 1;as otherwise,the background it is labeled (clutter) as thewith background a value of -1. (clutter) A detailed with aflowchart value of −for1. the A detailedSO-CFAR flowchart method foris illustrated the SO-CFAR in Figure method 2. is illustrated in Figure2.

Estimation of mean Estimate the The clutter in the power of the local parameters of sliding window clutter Weibull distribution Guard area window Guard P1(n-m) P2(n-m) size=2*m+1,m=5 area Parameters estimation including b and c c Clutter region window SO-CFAR cxc−1  x fx()=− ( ) exp size=2*n+1,eg.n=15 X>P=min(p1,p2) bb b

Compute the CFAR Give the constant threshold false alarm Binary image X>T, Target P =10-12 X

TheThe followingfollowing isis thethe formulaformula forfor CFARCFAR threshold: threshold:

T −=Z T 1(),Pfxdxfa (1) 1 − Pf a = 0 f (x)dx, (1) 0 where f ()x is the probability density function and we use Weibull distribution to mod- where f (x) is the probability density function and we use Weibull distribution to modify x P theify distributionthe distribution of SAR of SAR clutter. clutter.x is the intensity is the intensity value of value SAR imagery,of SAR imagery, and Pf a isand the givenfa is falsethe given alarm false rate. alarm We assume rate. We the assume clutter followingthe clutter the following Weibull the model Weibull taking model the probability taking the densityprobability functions density (PDF) functions form: (PDF) form: c x c−1 h  x cci f (x) = =−( cx) c−exp1 −  x (2) fx()b b ( ) expb (2) bb b where b and c are the scale and shape parameters, respectively, and are estimated by the methodwhere b of and logarithmic c are the scale cumulants and shape (MoLC) parameters, parameter respectively, estimation and approach are estimated [35]. T can by bethe obtainedmethod of after logarithmicb and c calculation cumulants using (MoLC) the following parameter formula: estimation approach [35]. T can be obtained after b and c calculation using the following formula: 1/c T = b(− log Pf a) , (3)

−12 In this study, the false alarm rate Pf a is set to 10 and the threshold is calculated automatically within the given sliding window. Pixels in the sliding window are compared with the threshold T, and those larger than the threshold are treated as the ice shelf; Remote Sens. 2021, 13, 3528 5 of 15

otherwise, they are treated as the background. Then, the binary image is automatically obtained by the adaptive threshold from the SO-CFAR method for the following processing. As glacier and ice-shelf fronts are dynamic due to calving events, there are a large number of disintegrating icebergs in the front of the ice shelf, as can be seen on the binary imagery from SO-CFAR results. The presence of rifts, pressure ridges, crevasses, and depressions on ice-shelf surfaces may be mistakenly detected as the background on SO- CFAR binary imagery. The heterogeneous characteristics of SAR backscatter on ice-shelf surfaces may cause significant uncertainties for the following profile processing. Therefore, a post-processing step is required for an SO-CFAR binary image to remove some floating icebergs and to fill in gaps resulting from heterogeneous ice surfaces, since it may affect the accuracy of ice-shelf front point extraction. To solve these problems, a morphological filter sequence including opening and closing is exploited [36]. The morphological opening is used to dilate an eroded image to remove isolate bright structures such as the floating icebergs. Morphological closing is used to erode a dilated image to suppress dark holes such as the heterogeneous ice surface structures. In this paper, a combination of opening and closing filters with a window size of 5 × 5 pixels can obtain the optimal result using SO-CFAR for ice shelf point extraction.

2.2. Profile Analysis Based Frontal Point Extraction The accuracy of ice-shelf extraction is affected by the ice spatial distribution in the frontal area, where there is some floating sea ice in front of the ice shelf. The fraction of floating iceberg also decreases the accuracy of the traditional algorithm for frontal position detection. For the ice shelf frontal point extraction, the real condition of the ice-shelf frontal area can be treated as the following three situations shown in Figure3. Two red lines were manually labeled as shown in Figure3, which define the border area for the AIS. The three purple lines L1-L3 show the different ice conditions in the frontal area, where the most common case L3 means that there is no obstacle in the frontal area. From 22 to 25 September 2019, a huge iceberg named D28 was breaking off from AIS, and there was a large amount of floating iceberg in the frontal position area of L1 and L2. L2 means that the profile line only crosses a small piece of iceberg. L1 shows that the profile line penetrates the main body of the D28 iceberg even though the morphological operation is used after CFAR detection, and L2 means that the profile line only crosses a small piece of the iceberg. To give a brief description of the frontal points extraction in this study, the distribution of ice-shelf points on Sentinel-1 SAR imagery and the corresponding cumulative value for the three conditions are illustrated in Figure3. The cumulative value beginning from the original to k position along the defined profile is value(K). Di f fBu f f er(k, n) is the minus of cumulative values at k and n. For L2 and L3 in Figure3d,e, the frontal point detection is treated as extracting the position along each profile, where the cumulative value value(K) at position K (Equation (4)) along the current profile reaches the peak, and L(i) is 1 for the ice shelf and −1 for the background. For the L3 profile, the peak value appears on the end area of the floating ice along the profile. To detect the real location of the ice-shelf frontal point, we first define a buffer area with the length of n along the given profile. If the difference of the cumulative value Di f fBu f f er(k, n) is –n, then the position in the buffer area is the final frontal point of the ice shelf. In this paper, the length of buffer area n is set to 5, which equals 200 m with the pixel spacing of 40 m for Sentinel-1 SAR imagery. The new filter is also suitable for the L1 and L2 cases, and it will be used as the criterion for frontal point extraction in all profiles.

K value(K) = ∑ L(i) (4) k=1

Di f fBu f f er(k, n) = value(k) − value(k − n) (5) Remote Sens. 2021, 13, 3528 6 of 15

The Sentinel-1 imagery was preprocessed using the Python toolbox and the extrac- Remote Sens. 2021, 13, x FOR PEER REVIEW 6 of 16 tion of the regions was performed by the ArcGIS software package. The SO-CFAR and morphological algorithms were executed on the MATLAB 2019b platform.

Figure 3. Frontal points extraction criterion for different ice conditions. In (a,b), the redred lineline isis thethe boundary we defineddefined inin this study and the magenta is the defineddefined profileprofile line. Figures (c–e)) illustrate the cumulative values for threethree differentdifferent situationssituations forfor fontalfontal pointspoints extraction.extraction. ((c)) itit meansmeans profilesprofiles cross the main body ofof floatingfloating ice.ice. (d)) it means profilesprofiles crosscross aa smallsmall partpart ofof floatingfloating ice.ice. ((ee)) itit meansmeans no floating ice along the profile. no floating ice along the profile.

3. ResultsThe cumulative and Discussion value beginning from the original to k position along the defined value() K k profileThe is Global Monitoring. DiffBuffer (,) for k n Environment is the minus and of cumulative Security (GMES) values programmeat and n . ledFor byL2 theand EuropeanL3 in Figure Space 3d,e, Agency the frontal (ESA), point which detectio hasn operated is treated the as Sentinel-1extracting the mission position with along free and open data access, is suitable for ice monitoring. For this purpose, Sentinel-1A/B each profile, where the cumulative value value() K at position K (Equation (4)) along the was launched in April 2014 and August 2016, respectively, providing a near-polar, sun- 𝐿𝑖 synchronouscurrent profile orbit reaches with the a 6-daypeak, and repeat time. is 1 Thefor the C-band ice shelf Sentinel-1 and -1 for SAR the image background. with a highFor the spatial L3 profile, resolution the ofpeak 40 mvalue contains appears much on more the detailedend area information of the floating with ice four along modes: the stripmapprofile. To (SM), detect wave the real (WV), location interferometric of the ice-shelf wide frontal swath (IW),point, and we first extra-wide define a swath buffer (EW). area Duewith tothe the length all-day of andn along all-weather the given capabilities profile. ofIf Sentinel-1,the difference it has of beenthe cumulative widely used value and Diff(,) k n is –n, then the position in the buffer area is the final frontal point of the ice has demonstratedBuffer its potential in cryosphere observation [37]. We delineated the defined boundaryshelf. In this of AISpaper, as thethe studylength area of buffer with thearea red n lineis set and to the5, which profile(green equals 200 line) m shown with the in Figurepixel spacing4. In Figure of 404 ,m we for first Sentinel-1 provide SAR details imagery. on the studyThe new sites filter and is dataset also suitable selected for for the the L1 temporaland L2 cases, and and spatial it will variability be used ofas fronts the criterion on the AIS.for frontal point extraction in all profiles.

In this paper, Sentinel-1 level 1 Ground RangeK Detected (GRD) SAR scenes between 2015 and 2021 are downloadedvalue from() K the= ESA L () Scientiﬁc i Data Hub (https://scihub.(4) copernicus.eu/). We used single polarization in thek=1 HH channel with EW mode and with a spatial resolution of 40 m in the range and azimuth direction, respectively, for extracting =−− the front position of AIS. TheDiff swathBuffer (,) width k n of value the () kSAR value image ( k is n 500 ) km with the incidence(5) angle ranging from 18.3◦ to 46.8◦. We calculated the monthly frontal position extension on AISThe during Sentinel-1 the observation imagery was period preprocessed using the using proposed the Python algorithm toolbox (Figure and4). the For extrac- each month,tion of the more regions than onewas image performed is available by thewhile ArcGIS we software only select package. one scene The forSO-CFAR monitoring and themorphological AIS frontal position,algorithms as were shown execut in theed right on the panel MATLAB of Figure 2019b4. As platform. can be seen, a total of 75 Sentinel-1 image products were selected for ice fronts extraction. To show the general trend3. Results of annual and Discussion ice fronts, we use a different color for the annual AIS front position, which The Global Monitoring for Environment and Security (GMES) programme led by the European Space Agency (ESA), which has operated the Sentinel-1 mission with free and open data access, is suitable for ice monitoring. For this purpose, Sentinel-1A/B was

Remote Sens. 2021, 13, x FOR PEER REVIEW 7 of 16

launched in April 2014 and August 2016, respectively, providing a near-polar, sun-syn- Remote Sens. 2021, 13, 3528 7 of 15 chronous orbit with a 6-day repeat time. The C-band Sentinel-1 SAR image with a high spatial resolution of 40 m contains much more detailed information with four modes: stripmap (SM), wave (WV), interferometric wide swath (IW), and extra-wide swath (EW). showsDue to thethe generalall-day advanceand all-weather during capabilities the observing of Sentinel-1, period and it regional has been retreat widely due used to and the D28has demonstrated calving event. its In potential addition, in for cryosphere further analysis observation of the [37]. D28 We iceberg delineated calving the cases, defined we useboundary the acquisitions of AIS as betweenthe study 18 area September with the 2019 red andline 31and October the profile(green for a total of line) 14 Sentinel-1 shown in images.Figure 4. We In follow Figure with 4, we a discussionfirst provide of thedetails details on forthe the study AIS sites frontal and position dataset and selected the three for specialthe temporal calving and cases spatial in the variability deﬁned regions, of fronts as on shown the AIS in. Figure4.

FigureFigure 4.4.Research Research area area and and data data set set used used in in this this study. study. The The general general trend trend of iceof ice frontal frontal position position between between 2015 2015 and and 2021 2021 for Ameryfor Amery Ice Shelf.Ice Shelf. The The colors colors indicate indicate the the time time difference. difference. We We deﬁned defined three three regions regions for for dynamical dynamical and and seasonalseasonal analysis.analysis. TheThe rightright panelpanel isis thethe datadata usedused forfor thethe spatio-temporalspatio-temporal analysis.analysis. TheThe backgroundbackground isis fromfrom single-polarizedsingle-polarized Sentinel-1Sentinel-1 datadata accessedaccessed onon 2626 MarchMarch 2015.2015.

3.1. VisualIn this Performance paper, Sentinel-1 and Comparison level 1 Ground of Ice-Shelf Range Frontal Detected Point Extraction(GRD) SAR scenes between 2015To and improve 2021 are the downloaded visual performance from the ESA of frontalScientific point Data extraction, Hub (https://scihub.coperni- we proposed the strategycus.eu/). thatWe used SO-CFAR single ispolarization used for the in the binary HH ice-shelfchannel with classification. EW mode Weand extractedwith a spatial the frontalresolution position of 40 ofm AISin the from range 2015 and to azimuth 2021. The di ice-shelfrection, respectively, frontal line extraction for extracting algorithm the front is validatedposition of by AIS. comparison The swath with width the state-of-the-artof the SAR image approach is 500 onkm three with situations, the incidence including angle theranging rifts, calvingfrom 18.3° iceberg to 46.8°. as well We as calculated visible surface the monthly structures frontal (e.g., position the crevasse extension and pressure on AIS ridges).during the To showobservation the visual period performance using the prop of theosed proposed algorithm frontal (Figure line extraction4). For each method, month, threemore samplethan one Sentinel-1 image is SARavailable images while acquired we only on select 26 March one scene 2015, for 24 monitoring October 2019, the andAIS 22frontal September position, 2019 as shown are presented in the right in Figure panel5 of. AsFigure can 4. be As seen can in be Figure seen, 5aa–c, total we of show75 Sen- thetinel-1 results image from products our algorithm were selected (upper) for and ice the fronts comparison extraction. algorithm To show (bottom) the general in [27 trend] in termsof annual of rift, ice calving fronts, icebergs, we use anda different surface features,color for respectively. the annual TheAIS comparisonfront position, algorithm which mainlyshows the utilizes general the advance statistics during of backscattering the observing values period and and maximal regional values retreat to decidedue to thethe frontalD28 calving position. event. As In shown addition, in Figure for further5a, the profileanalysis line of hasthe moreD28 iceberg than one calving intersection cases, ofwe frontaluse the pointsacquisitions while between the comparison 18 September algorithm 2019 can and only 31 October provide for one a total frontal of 14 point. Sentinel- Our algorithm1 images. hasWe goodfollow performance with a discussion due to theof the strategy details of for firstly the usingAIS frontal the CFAR position algorithm and the to detectthree special the ice calving shelf. Especially cases in the in defined the calving regions, iceberg as shown case, the in profileFigure for4. the comparison algorithm is ineffective since the frontal points are on the floating iceberg, which is close to the3.1. iceVisual shelf. Performance This is because and Comparison the proposed of Ice-Shelf algorithm Frontal utilizes Point Extraction the morphological filter to removeTo theimprove isolated the icebergs. visual performance Although the of fron comparisontal point algorithmextraction, is we not proposed effective the in thesestrat- above-mentionedegy that SO-CFAR cases, is used it has for provided the binary new ice-shelf insight classification. for us to propose We aextracted new strategy the frontal based onposition this framework. of AIS from 2015 to 2021. The ice-shelf frontal line extraction algorithm is validated by comparison with the state-of-the-art approach on three situations, including the

Remote Sens. 2021, 13, x FOR PEER REVIEW 8 of 16

rifts, calving iceberg as well as visible surface structures (e.g., the crevasse and pressure ridges). To show the visual performance of the proposed frontal line extraction method, three sample Sentinel-1 SAR images acquired on 26 March 2015, 24 October 2019, and 22 September 2019 are presented in Figure 5. As can be seen in Figure 5a–c, we show the results from our algorithm (upper) and the comparison algorithm (bottom) in [27] in terms of rift, calving icebergs, and surface features, respectively. The comparison algorithm mainly utilizes the statistics of backscattering values and maximal values to decide the frontal position. As shown in Figure 5a, the profile line has more than one intersection of frontal points while the comparison algorithm can only provide one frontal point. Our algorithm has good performance due to the strategy of firstly using the CFAR algorithm to detect the ice shelf. Especially in the calving iceberg case, the profile for the comparison algorithm is ineffective since the frontal points are on the floating iceberg, which is close to the ice shelf. This is because the proposed algorithm utilizes the morphological filter to Remote Sens. 2021, 13, 3528 remove the isolated icebergs. Although the comparison algorithm is not effective in 8these of 15 above-mentioned cases, it has provided new insight for us to propose a new strategy based on this framework.

(a) (b) (c)

Figure 5. FrontalFrontal point point extraction extraction on on sample sample Sentinel-1 Sentinel-1 SAR SAR images images using using the prop the proposedosed algorithm algorithm (upper) (upper) and the and com- the comparisonparison method method (bottom). (bottom). (a) 26 (a March) 26 March 2015; 2015; (b) 24 (b )October 24 October 2019; 2019; (c) 22 (c )September 22 September 2015. 2015.

3.2. Spatio-Temporal Spatio-Temporal Changes in the Ice-ShelfIce-Shelf Frontal Line from 20152015 toto 20212021 InIn addition to the visual performance with the state-of-the-art algorithm, we use the manual delineation of the frontal frontal line in in three three regions regions to to evaluate evaluate the the proposed proposed algorithm, algorithm, and their accuracies reportreport isis givengiven in in Table Table1. 1. In In this this time time span, span, we we select select seven seven datasets datasets in inMarch March of eachof each year year to quantifyto quantify their their accuracies. accuracies. The The proposed proposed algorithm algorithm is superior is superior to the to thecomparison comparison method method especially especially in Region in Region 1 and 1 and Region Region 2, where 2, where the biasthe bias of the of averagedthe aver- ageddistance distance compared compared with reference with reference frontal line frontal using line the proposedusing the algorithm proposed is algorithm smaller than is smallerthe comparison than the approach,comparison especially approach, in especially Region 1 andin Region Region 1 2and area. Region This is2 area. because This the is becausefrontal line the infrontal Region line 1 andin Region Region 1 2 and is much Region more 2 is complicated much more than complicated that in Region than that 3 with in Regionregard to3 with some regard rifts, indicating to some rifts, that indicati our algorithmng that our can algorithm deal with morecan deal complex with more cases. complex cases. Table 1. Validation of frontal line extraction using different methods. Table 1. Validation of frontal line extraction using different methods. Average Distance of Frontal Position Compared with Reference Frontal Average Distance of Frontal PositionLine [m] Compared with Reference Frontal Date Line [m] Date Region 1 Region 2 Region 3 Region 1 Region 2 Region 3 Proposed Comparison Proposed Comparison Proposed Comparison Proposed Comparison Proposed Comparison Proposed Comparison 2015-03-26 26.44 156.34 38.23 145.21 12.33 33.21 2015-03-262016-03-22 26.44 28.32 156.34 143.25 38.23 44.21 145.21 165.36 12.33 9.24 33.21 32.55 2017-03-22 22.07 166.57 39.56 165.33 8.75 27.88 2018-03-15 26.33 124.32 45.12 154.06 13.65 24.35 2019-03-17 26.21 156.36 40.55 168.31 7.88 30.22 2020-03-18 23.21 137.45 35.66 144.23 10.44 25.48 2021-03-18 24.83 147.06 37.23 135.23 8.75 27.01

The time-series results of the terminus on the AIS as shown in Figure6a indicate that the front of the AIS has an advancing trend of outward expansion between 2015 and 2021. In September 2019, an area of about 1600 km2 of the iceberg D28 detached from the Amery Ice Shelf, which caused the AIS front to retreat instantaneously. The advance displacement of AIS fronts from March 2015 to September 2019 in Figure6a shows that the total displacement over these ﬁve-year periods was 4650 m, with the slope for each year meaning the monthly change rate. A signiﬁcant expanding tendency of frontal position from 2015 to 2018 is clearly seen. While in 2019, the displacement rate retreated by 280 m after the D28 calving. During the iceberg calving, the frontal line has retreated by an average of 280 m shown in Figure6b. From 2019 to 2021, the overall advance was an approximate 2.05 km extension. After the D28 event, the ice shelf continued to Remote Sens. 2021, 13, x FOR PEER REVIEW 9 of 16

2016-03-22 28.32 143.25 44.21 165.36 9.24 32.55 2017-03-22 22.07 166.57 39.56 165.33 8.75 27.88 2018-03-15 26.33 124.32 45.12 154.06 13.65 24.35 2019-03-17 26.21 156.36 40.55 168.31 7.88 30.22 2020-03-18 23.21 137.45 35.66 144.23 10.44 25.48 2021-03-18 24.83 147.06 37.23 135.23 8.75 27.01

Remote Sens. 2021, 13, 3528 to 2018 is clearly seen. While in 2019, the displacement rate retreated by 280 m after9 ofthe 15 D28 calving. During the iceberg calving, the frontal line has retreated by an average of 280 m shown in Figure 6b. From 2019 to 2021, the overall advance was an approximate 2.05 km extension. After the D28 event, the ice shelf continued to expand outward. Until May 2021,expand the outward. overall expansion Until May distance 2021, the of overall the terminus expansion of the distance AIS expanded of the terminus by approxi- of the matelyAIS expanded 6500 m compared by approximately with that 6500 in Marc m comparedh 2015. After with the that D28 in detachment, March 2015. the After AIS was the withD28 detachment,273.5 km of frontal the AIS lines was compared with 273.5 with km 255.03 of frontal km in lines March compared 2015 although with 255.03 the frontal km in pointsMarch were 2015 advanced. although the From frontal 2015 pointsto 2021, were the le advanced.ngth of fronts From is 2015increased to 2021, by 7.5%. the length During of thefronts D28 is event, increased the length by 7.5%. of Duringfrontal lines the D28 of the event, AIS thesuddenly length increased of frontal (Figure lines of 6b), the and AIS thensuddenly the length increased of the (Figure frontal6b), line and remained then the lengthunchanged. of the However, frontal line the remained frontal points unchanged. keep expandingHowever, the as can frontal be seen points in keepFigure expanding 6a. as can be seen in Figure6a.

(a) (b)

Figure 6. Time seriesseries variabilityvariability in frontalfrontal lineslines ofof thethe AISAIS monitoredmonitored duringduring MarchMarch 2015–May2015–May 2021.2021. Relative changes in frontal lines of the AISAIS duringduring thethe D28D28 calvingcalving eventevent inin SeptemberSeptember 2019.2019. The length and extension distance are coded as shown. ( a) Monthly cumulative extension distance and length of frontal line for the AIS from 2015 to 2021; (b) cumulative extension distance and length of frontal line for the AIS during the D28 calving event. extension distance and length of frontal line for the AIS during the D28 calving event.

The length of AIS fronts has also shown an increasing trend since 2015, and from the beginning of of 2018 2018 to to September September 2019, 2019, the the length length of of the the AIS AIS front front shows shows a clear a clear advancing advanc- trend.ing trend. It is Itworth is worth noting noting that thatthe rapidly the rapidly increasing increasing rate between rate between January January 2018 2018and Sep- and temberSeptember 2019 2019 is the is reason the reason for the for D28 the iceberg D28 iceberg calving; calving; the main the mainreason reason may be may due be todue the intensifiedto the intensified ice-shelf ice-shelf activity. activity. We show We the show length the lengthof the rift of thein Region rift in Region2 in Figure 2 in 7. Figure As can7. beAs seen can bein seenFigure in 7, Figure the rift7, thein Region rift in Region2 is becoming 2 is becoming wider and wider wider and. The wider. wider The rifts wider led torifts longer led to fronts. longer The fronts. AIS Thefront AIS intensifie front intensifiess its advance its advancebetween betweenMarch 2018 March and 2018 October and 2018October and 2018 increased and increased the stress the on stressthe rifts on especially the rifts especially in region in 2 regionwhich 2accelerated which accelerated the D28 the D28 calving. The advance during this period can be attributed to the ice velocity, which is discussed in the next section. Figure6b shows the time series results during the disintegration of D28. The D28 iceberg disintegration event can also be seen as having occurred from 25 September to 27 September 2019. The advance and retreat of AIS were assessed using Sentinel-1 images during 2015 and 2021. The time series of visual results for the three regions from 2015 to 2021 are shown in Figure6. The general trend for retreat or advance is shown in Figure4. In the experiment, the ice-shelf frontal line extraction algorithm is validated by visual comparison with the state-of-the-art approach, where the spatial and temporal dynamics of AIS in three defined regions is discussed. In this section, Regions 1 and 2 are used to show the dynamic change of frontal lines in detail including the D28 calving event. Region 1 is located in the west of the AIS and as can be seen in Figure8a, the profile (green line) has multiple intersections with AIS fronts. The proposed method can successfully detect all the frontal points in the profile, which can provide an accurate frontal line. In September 2019, the D28 iceberg broke up from Region 1 and Region 2, which led to a large amount of floating ice in the front of AIS. The proposed method can also suppress the influence of large floating ice in the front of the AIS. After the D28 event, we can see the obvious retreat of the frontal lines in Region 1. With respect to Region 2 located in the middle of the AIS, the retreat occurred at the rifting area while the advance is continuous in Region 3. Due to iceberg calving, Remote Sens. 2021, 13, x FOR PEER REVIEW 10 of 16 Remote Sens. 2021, 13, 3528 10 of 15

calving. The advance during this period can be attributed to the ice velocity, which is dis- cussedwe can in ﬁnd the a next signiﬁcant section. dynamic Figure 6b of shows the frontal the time position series in results Region during 1 and Regionthe disintegra- 2, while tionRegion of D28. 3 shows The D28 a stable iceberg expansion disintegration pattern. event can also be seen as having occurred from 25 September to 27 September 2019. 71°0'0"E 72°0'0"E 73°0'0"E 74°0'0"E 71°0'0"E 72°0'0"E 73°0'0"E 74°0'0"E 68°30'0"S 68°30'0"S 68°30'0"S 68°30'0"S 16646m 16903m

N N 69°0'0"S 69°0'0"S

71°0'0"E 72°0'0"E 73°0'0"E 74°0'0"E 69°0'0"S 71°0'0"E 72°0'0"E 73°0'0"E 74°0'0"E 69°0'0"S (a) 2017-10-17 (b) 2018-03-15 71°0'0"E 72°0'0"E 73°0'0"E 74°0'0"E 71°0'0"E 72°0'0"E 73°0'0"E 74°0'0"E 68°30'0"S 68°30'0"S 68°30'0"S 68°30'0"S 17796m 18272m

N N 69°0'0"S 69°0'0"S

FigureFigure 7. 7. TheThe advanced AIS fronts from MarchMarch 2018.2018. TheThe riftrift inin RegionRegion2 2 became became wider wider and and wider. wider. The The length length of of the the rift rift is ismarked marked in in the the respective respective panel. panel.

The advance and retreat of AIS were assessed using Sentinel-1 images during 2015 and 2021. The time series of visual results for the three regions from 2015 to 2021 are shown in Figure 6. The general trend for retreat or advance is shown in Figure 4. In the experiment, the ice-shelf frontal line extraction algorithm is validated by visual comparison with the state-of-the-art approach, where the spatial and temporal dynamics of AIS in three defined regions is discussed. In this section, Regions 1 and 2 are used to show the dynamic change of frontal lines in detail including the D28 calving event. Region 1 is located in the west of the AIS and as can be seen in Figure 8a, the profile (green line) has multiple intersections with AIS fronts. The proposed method can successfully detect all the frontal points in the profile, which can provide an accurate frontal line. In September 2019, the D28 iceberg broke up from Region 1 and Region 2, which led to a large amount of floating ice in the front of AIS. The proposed method can also suppress the influence of large floating ice in the front of the AIS. After the D28 event, we can see the obvious retreat of the frontal lines in Region 1. With respect to Region 2 located in the middle of the AIS, the retreat occurred at the rifting area while the advance is continuous in Region 3. Due to iceberg calving, we can find a significant dynamic of the frontal position in Region 1 and Region 2, while Region 3 shows a stable expansion pattern.

RemoteRemote Sens. Sens.2021 2021, 13, 13 3528, x FOR PEER REVIEW 11 of11 16 of 15

(a) Region 1

(b) Region 2

Figure 8. The result of frontal line in the three regions from 2015 to 2021. All the background image is acquired in March of each year except for 2019 in September. For the last panel, the background image were acquired in March 2015 for Region 1, 2, and 3. (a) Region 1; (b) Region 2; (c) Region 3.

Remote Sens. 2021, 13, x FOR PEER REVIEW 12 of 16

Figure 8. The result of frontal line in the three regions from 2015 to 2021. All the background image is acquired in March Remote Sens. 2021, 13, 3528 12 of 15 of each year except for 2019 in September. For the last panel, the background image were acquired in March 2015 for Region 1, 2, and 3. (a) Region 1; (b) Region 2; (c) Region 3.

WithWith respectrespect toto RegionRegion 11 andand RegionRegion 2,2, RegionRegion 33 isis muchmuch moremore stable.stable. WeWe comparedcompared thethe seasonalseasonal dynamicsdynamics inin RegionRegion 33 usingusing thethe monthlymonthly averageaverage extensionextension distancedistance shownshown inin FigureFigure9 9,, where where it it is is clear clear that that the the frontal frontal position position of of the the AIS AIS has has a a signiﬁcant significant seasonal seasonal trend.trend. ComparedCompared with with the the expansion expansion speed speed from from October October to to March March and and April April to Septemberto Septem- inber Region in Region 3, the 3, AISthe AIS frontal frontal position position shows shows a relatively a relatively faster faster speed speed in the in australthe austral summer sum- seasonsmer seasons compared compared to the to austral the austral winter. winter During. During 2016–2020, 2016–2020, the average the average monthly monthly expansion ex- distancepansion distance from January from January to April to gradually April gradually decreases, decreases, and it starts and it to starts increase to increase from April. from TheApril. maximal The maximal average average monthly monthly expansion expansio distancen appeareddistance appeared in October in with October the expansion with the distanceexpansion larger distance than 130larger m inthan 2018. 130 During m in 2018 the. summerDuring the season, summer the speedseason, of the ice speed ﬂow on of the ice AISflow increased, on the AIS which increased, is the which main reasonis the main for the reason acceleration for the acceleration of frontal position of frontal expansion. position Forexpansion. the annual For variancethe annual from vari Aprilance from to October, April to the October, expansion the expansion speed is accelerating, speed is acceler- the expansionating, the expansion speed reaching speed the reaching top in Octoberthe top in 2018 October before 2018 the D28before event. the D28 event. of AIS frontal line [m] AIS frontal of Monthly extension distance Monthly

FigureFigure 9.9. SeasonalSeasonal advanceadvance raterate withwith thethe expansionexpansion distancedistance ofof thethe AISAIS frontalfrontal lineline inin RegionRegion 3.3. TheThe monthlymonthly averageaverage advanceadvance raterate forfor eacheach yearyear isis color-coded.color-coded.

3.3.3.3. IceIce VelocityVelocity ToTo validatevalidate the the accuracy accuracy of of the the displacement displacement retrieved retrieved using using the the proposed proposed method, method, the icethe velocity ice velocity product product from 2015–2016,from 2015–2016, provided prov byided [37, 38by], [38,39], is used. is We used. first useWe 66first matched use 66 pointsmatched to generatepoints to the generate scatter plotthe betweenscatter plot the proposedbetween the result proposed and the MEaSUREsresult and icethe velocityMEaSUREs product ice velocity in Figure product 10a; the in standard Figure 10a; deviation the standard is 165 m/yr,deviation which is 165 means m/yr, the which daily differencemeans the fordaily these difference two results for these is less two than results 0.5 m. isMoreover, less than 0.5 for m. the Moreover, relatively for stable the arearela- in the central and western parts of the AIS, we selected 38 from the 66 points to analyze tively stable area in the central and western parts of the AIS, we selected 38 from the 66 the linear relationship between these two products. The scatter plot in Figure 10e shows points to analyze the linear relationship between these two products. The scatter plot in that the standard deviation is 111 m, with a daily bias of 0.3 m. The comparison with Figure 10e shows that the standard deviation is 111 m, with a daily bias of 0.3 m. The the ice velocity dataset indicates that the displacement of the frontal position can also be comparison with the ice velocity dataset indicates that the displacement of the frontal po- used to analyze the dynamics of the AIS. Figure 10 shows the correlation analysis between sition can also be used to analyze the dynamics of the AIS. Figure 10 shows the correlation the AIS extended distance and MEaSUREs ice velocity products [33,34]. As can be seen analysis between the AIS extended distance and MEaSUREs ice velocity products [33,34]. from the ice velocity diagram in the above figure, the maximum ice velocity of the AIS As can be seen from the ice velocity diagram in the above figure, the maximum ice velocity ice shelf is mainly concentrated at the intersection of Lambert Glacier (LG), Fisher Glacier of the AIS ice shelf is mainly concentrated at the intersection of Lambert Glacier (LG), (FG), and Mellor Glacier (MG) shown in Figure 10b and at the frontal position area shown Fisher Glacier (FG), and Mellor Glacier (MG) shown in Figure 10b and at the frontal posi- in Figure 10c, the maximum ice velocity is about 4.3 m/day. In Figure 10d,e, a scatter plottion ofarea the shown expansion in Figure rate of 10c, the the glacier maximum frontal ice area velocity and its is corresponding about 4.3 m/day. ice velocity In Figure is presented.10d,e, a scatter Among plot them,of the theexpansion left picture rate containsof the glacier 66 points, frontal which area and are distributedits corresponding in the frontalice velocity area ofis thepresented. AIS. It canAmong be seen them, from the the left picture picture that contains the difference 66 points, between which theare twodis- datatributed sets ofin mostthe frontal points area is within of the 200 AIS. m/yr, It can and be that seen the from maximum the picture error that is about the difference 400 m/yr and that R2 is 0.88, indicating that there is good consistency between the ice velocity and expansion rate. The scatter diagram on the right contains 38 points, which are distributed in the front area of D28 iceberg disintegration. It can be seen from the figure that the errors of all points are within 300 m/yr, and R2 is 0.91. The ice velocity in this area is slightly Remote Sens. 2021, 13, x FOR PEER REVIEW 13 of 16

between the two data sets of most points is within 200 m/yr, and that the maximum error is about 400 m/yr and that R2 is 0.88, indicating that there is good consistency between the Remote Sens. 13 2021, , 3528 ice velocity and expansion rate. The scatter diagram on the right contains 38 points,13 which of 15 are distributed in the front area of D28 iceberg disintegration. It can be seen from the figure that the errors of all points are within 300 m/yr, and R2 is 0.91. The ice velocity in this greaterarea is slightly than the greater expansion than rate the ofexpansion the ice shelf, ratewhich of the alsoice shelf, caused which the area also tocaused be active the andarea eventuallyto be active led and to eventually the breakup led of to the the iceberg. breakup of the iceberg.

FigureFigure 10.10. Correlation analysis analysis between between the the frontal frontal extension extension rate rate of ofthe the AIS AIS and and the the MEaSUREs MEaSUREs ice velocity ice velocity product. product. The relation of ice velocity and the extension rate using the scatter plot by the extension rate and MEaSUREs ice velocity The relation of ice velocity and the extension rate using the scatter plot by the extension rate and MEaSUREs ice velocity products. (a) MEaSURE ice velocity product; (b) ice velocity at the intersection of Lambert Glacier (LG), Fisher Glacier products. (a) MEaSURE ice velocity product; (b) ice velocity at the intersection of Lambert Glacier (LG), Fisher Glacier (FG), (FG), and Mrellor Glacier (MG); (c) ice velocity in the front of AIS; (d) scatter plot of the extension rate in the front of AIS andand MrellorMEaSUREs Glacier ice (MG);velocity (c )product(yellow ice velocity in theand front light of blue AIS; points); (d) scatter (e) scatter plot of plot the of extension the extended rate in distance the front at of the AIS frontal and MEaSUREsposition of icethe velocityD28 disintegration product(yellow area andand light MEaSUREs blue points); ice velocity (e) scatter product plot of (yellow the extended points). distance at the frontal position of the D28 disintegration area and MEaSUREs ice velocity product (yellow points). 4. Conclusions 4. ConclusionsThe calving and expansion of ice shelves have significant impacts on ice sheet insta- bilityThe and calving sea-level and rise. expansion As the third-largest of ice shelves ice have shelf significant in Antarctica, impacts the ondynamic ice sheet AIS insta- is of bilitygreat andimportance sea-level for rise. ice shelf As the monitoring. third-largest In this ice shelfstudy, in we Antarctica, generated the a time dynamic series AIS of the is offrontal great line importance for the AIS for using ice shelf Sentinel-1 monitoring. SAR data In this based study, on a we profile generated analysis a time method. series This of thestudy frontal for the line first for thetime AIS performed using Sentinel-1 an automated SAR data mapping based onof Antarctic a profile analysisAIS frontal method. lines Thisusing study Sentinel-1 for the SAR first imagery time performed for the period an automated 2015–2021. mapping In detail, ofwe Antarctic modified AIS an ice frontal shelf linesdetection using with Sentinel-1 SO-CFAR SAR algorithm imagery forand the profile period cumulative 2015–2021. method In detail, for frontal we modified point de- an icetection. shelf The detection main aim with during SO-CFAR algorithm algorithm devel andopment profile was cumulative the spatio-temporal method foranalysis frontal of pointthe advance detection. and Theretreat main of AIS aim frontal during lines algorithm as well developmentas to provide complementary was the spatio-temporal mapping analysisproducts. of The the advanceproposed and algorithm retreat of was AIS construct frontal linesed on as 75 well Sentinel-1 as to provide acquisitions complementary covering mapping products. The proposed algorithm was constructed on 75 Sentinel-1 acquisitions covering AIS and additional 14 Sentinel-1 acquisitions covering the D28 event period. By taking into account surface structure on the ice shelf and floating ice in the frontal area, the SO-CFAR target detection method is used to label the SAR as a binary image with the ice shelf and the background. After morphological filtering, a cumulative value-based frontal Remote Sens. 2021, 13, 3528 14 of 15

point extraction method is proposed. The time-series results show that since March 2015, the average extension distance of the front position of the AIS is 6.5 km, and the length of the front line has increased by 7.5%. Moreover, during the break-off of the D28 iceberg in September 2019, there was a signiﬁcant increase in the frontal length of the AIS, and it has been in a steady advanced state since then. While the frontal position of the ice shelf retreated during the D28 breakup, and it has been advancing since then. We examined the stable region without an iceberg calving event and found that the frontal advancing rate slowed signiﬁcantly after the austral summer. In comparison with MEaSUREs ice velocity products, we found that the front expansion rate and ice velocity have a good consistency, and an R2 is 0.88. Further analysis found that before the AIS ice shelf breaks off, the daily average expansion rate of the east and west sides of the D28 ice was greater than 2 m, which caused the rift on the west side of the central AIS image to increase and eventually lead to the breakup of the iceberg. In the future, a long-time series of ice velocity products could be used for correlation analysis. The results of this study demonstrate the high potential of using Sentinel-1 SAR data for monitoring the dynamics of the AIS frontal position.

Author Contributions: Conceptualization, T.Z., X.C. and Y.Z.; methodology, T.Z.; software, Y.Z.; validation, T.Z. and Y.Z.; formal analysis, all authors; original draft preparation, T.Z.; review and editing, all authors. All authors have read and agreed to the published version of the manuscript. Funding: This work was supported by the National Key Research and Development Program of China under Grant 2017YFA0603104; the National Natural Science Foundation of China (No. 41801266, 41901275, 41776186, and 41531069); and the China Postdoctoral Science Foundation (No. 2017M612512 and 2018M632922), and LIESMARS special research funding. Institutional Review Board Statement: Not applicable. Informed Consent Statement: Not applicable. Data Availability Statement: The Sentinel-1 SAR data can be downloaded from the ESA Scientific Data Hub (https://scihub.copernicus.eu/). The MEaSUREs ice velocity product is acquired from the National Snow and Ice Data Center (https://nsidc.org/data). Conflicts of Interest: The authors declare no conflict of interest.

References 1. Porter-Smith, R.; McKinlay, J.; Fraser, A.D.; Massom, R.A. Coastal complexity of the Antarctic continent. Earth Syst. Sci. Data 2021, 13, 3103–3114. [CrossRef] 2. Walker, C.C.; Becker, M.K.; Fricker, H.A. A High Resolution, Three-Dimensional View of the D-28 Calving Event from Amery Ice Shelf with ICESat-2 and Satellite Imagery. Geophys. Res. Lett. 2021, 48, e2020GL091200. [CrossRef] 3. Walker, C.C.; Bassis, J.N.; Fricker, H.A.; Czerwinski, R.J. Structural and environmental controls on Antarctic ice shelf rift propagation inferred from satellite monitoring. J. Geophys. Res. Earth Surf. 2013, 118, 2354–2364. [CrossRef] 4. Walker, C.C.; Bassis, J.N.; Fricker, H.A.; Czerwinski, R.J. Observations of interannual and spatial variability in rift propagation in the Amery Ice Shelf, Antarctica, 2002–2014. J. Glaciol. 2015, 61, 243–252. [CrossRef] 5. Tong, X.; Liu, S.; Li, R.; Xie, H.; Liu, S.; Qiao, G. Multi-track extraction of two-dimensional surface velocity by the combined use of di ff erential and multiple-aperture InSAR in the Amery Ice Shelf, East Antarctic. Remote Sens. Environ. 2018, 204, 122–137. [CrossRef] 6. Pittard, M.L.; Roberts, J.L.; Warner, R.C.; Galton-Fenzi, B.K.; Watson, C.S.; Coleman, R. Flow of the Amery Ice Shelf and its tributary glaciers. In Proceedings of the 18th Australasian Fluid Mechanics Conference, AFMC 2012, Launceston, Australia, 3–7 December 2012; pp. 18–21. 7. Young, N.W.; Hyland, G. Velocity and strain rates derived from InSAR analysis over the Amery Ice Shelf, East Antarctica. Ann. Glaciol. 2002, 34, 228–234. [CrossRef] 8. Köhler, A.; Nuth, C.; Kohler, J.; Berthier, E.; Weidle, C.; Schweitzer, J. A 15 year record of frontal glacier ablation rates estimated from seismic data. Geophys. Res. Lett. 2016, 43, 12155–12164. [CrossRef] 9. Fricker, H.A.; Young, N.W.; Allison, I.; Coleman, R. Iceberg calving from the Amery Ice Shelf, East Antarctica. Ann. Glaciol. 2002, 34, 241–246. [CrossRef] 10. Glasser, N.F.; Scambos, T.A. A structural glaciological analysis of the 2002 Larsen B ice-shelf collapse. J. Glaciol. 2008, 54, 3–16. [CrossRef] 11. Rignot, E.; Mouginot, J.; Scheuchl, B.; Van Den Broeke, M.; Van Wessem, M.J.; Morlighem, M. Four decades of Antarctic ice sheet mass balance from 1979–2017. Proc. Natl. Acad. Sci. USA 2019, 116, 1095–1103. [CrossRef] Remote Sens. 2021, 13, 3528 15 of 15

12. Cui, X.; Greenbaum, J.S.; Lang, S.; Zhao, X.; Li, L.; Guo, J.; Sun, B. The Scientific Operations of Snow Eagle 601 in Antarctica in the Past Five Austral Seasons. Remote Sens. 2020, 12, 2994. [CrossRef] 13. Kachouie, N.N.; Huybers, P.; Schwartzman, A. Localization of mountain glacier termini in Landsat multi-spectral images. Pattern Recognit. Lett. 2013, 34, 94–106. [CrossRef] 14. Bhardwaj, A.; Sam, L.; Singh, S.; Kumar, R. Automated detection and temporal monitoring of crevasses using remote sensing and their implications for glacier dynamics. Ann. Glaciol. 2016, 57, 81–91. [CrossRef] 15. Jeong, S.; Howat, I.M.; Bassis, J.N. Accelerated ice shelf rifting and retreat at Pine Island Glacier, West Antarctica. Geophys. Res. Lett. 2016, 43, 11720–11725. [CrossRef] 16. Euillades, L.D.; Euillades, P.A.; Riveros, N.C.; Masiokas, M.H.; Ruiz, L.; Pitte, P.; Elefante, S.; Casu, F.; Balbarani, S. Detection of glaciers displacement time-series using SAR. Remote Sens. Environ. 2016, 184, 188–198. [CrossRef] 17. Wesche, C.; Jansen, D.; Dierking, W. Calving Fronts of Antarctica: Mapping and Classification. Remote Sens. 2013, 5, 6305–6322. [CrossRef] 18. Baumhoer, C.A.; Dietz, A.J.; Kneisel, C.; Kuenzer, C. Automated extraction of antarctic glacier and ice shelf fronts from Sentinel-1 imagery using deep learning. Remote Sens. 2019, 11, 2529. [CrossRef] 19. Seale, A.; Christoffersen, P.; Mugford, R.I.; O’Leary, M. Ocean forcing of the Greenland Ice Sheet: Calving fronts and patterns of retreat identified by automatic satellite monitoring of eastern outlet glaciers. J. Geophys. Res. Earth Surf. 2011, 116, 1847. [CrossRef] 20. Kumar, A.; Srivastava, A.; Yadav, J.; Mohan, R. Spatio-temporal changes and prediction of Amery ice shelf, east Antarctica: A remote sensing and statistics-based approach. J. Environ. Manag. 2020, 267, 110648. [CrossRef] 21. Liu, H.; Jezek, K.C. A complete high-resolution coastline of antarctica extracted from orthorectified radarsat SAR imagery. Photogramm. Eng. Remote Sens. 2004, 70, 605–616. [CrossRef] 22. Borstad, C.; Khazendar, A.; Scheuchl, B.; Morlighem, M.; Larour, E.; Rignot, E. A constitutive framework for predicting weakening and reduced buttressing of ice shelves based on observations of the progressive deterioration of the remnant Larsen B Ice Shelf. Geophys. Res. Lett. 2016, 43, 2027–2035. [CrossRef] 23. Borstad, C.; McGrath, D.; Pope, A. Fracture propagation and stability of ice shelves governed by ice shelf heterogeneity. Geophys. Res. Lett. 2017, 44, 4186–4194. [CrossRef] 24. Rosenau, R.; Scheinert, M.; Dietrich, R. A processing system to monitor Greenland outlet glacier velocity variations at decadal and seasonal time scales utilizing the Landsat imagery. Remote Sens. Environ. 2015, 169, 1–19. [CrossRef] 25. Han, L.; Floricioiu, D.; Baessler, M.; Eineder, M. An algorithm for the detection of calving glaciers frontal position from TerraSAR-X imagery. In Proceedings of the International Geoscience and Remote Sensing Symposium (IGARSS), Beijing, China, 10–15 July 2016; pp. 6171–6174. [CrossRef] 26. Ni, W.; Yan, W.; Wu, J.; Zheng, G.; Lu, Y. Statistical analysis and modeling of TerraSAR-X images for CFAR based target detection. In Proceedings of the International Geoscience and Remote Sensing Symposium (IGARSS), Melbourne, Australia, 21–26 July 2013; pp. 1983–1986. [CrossRef] 27. Anastassopoulos, V.; Lampropoulos, G.A. Optimal CFAR Detection in Weibull Clutter. IEEE Trans. Aerosp. Electron. Syst. 1995, 31, 52–64. [CrossRef] 28. Tao, D.; Doulgeris, A.P.; Brekke, C. A Segmentation-Based CFAR Detection Algorithm Using Truncated Statistics. IEEE Trans. Geosci. Remote Sens. 2016, 54, 2887–2898. [CrossRef] 29. Hou, B.; Chen, X.; Jiao, L. Multilayer CFAR detection of ship targets in very high resolution SAR images. IEEE Geosci. Remote Sens. Lett. 2015, 12, 811–815. [CrossRef] 30. Lee, I.K.; Shamsoddini, A.; Li, X.; Trinder, J.C.; Li, Z. Extracting hurricane eye morphology from spaceborne SAR images using morphological analysis. ISPRS J. Photogramm. Remote Sens. 2016, 117, 115–125. [CrossRef] 31. Modava, M.; Akbarizadeh, G. Coastline extraction from SAR images using spatial fuzzy clustering and the active contour method. Int. J. Remote Sens. 2017, 38, 355–370. [CrossRef] 32. Perron, J.T.; Royden, L. An integral approach to bedrock river profile analysis. Earth Surf. Process. Landf. 2013, 38, 570–576. [CrossRef] 33. Gao, G.; Liu, L.; Zhao, L.; Shi, G.; Kuang, G. An adaptive and fast CFAR algorithm based on automatic censoring for target detection in high-resolution SAR images. IEEE Trans. Geosci. Remote Sens. 2009, 47, 1685–1697. [CrossRef] 34. Di Bisceglie, M.; Galdi, C. CFAR detection of extended objects in high-resolution SAR images. IEEE Trans. Geosci. Remote Sens. 2005, 43, 833–843. [CrossRef] 35. Krylov, V.A.; Moser, G.; Serpico, S.B.; Zerubia, J. On the method of logarithmic cumulants for parametric probability density function estimation. IEEE Trans. Image Process. 2013, 22, 3791–3806. [CrossRef] 36. Gil, J.Y.; Kimmel, R. Efficient dilation, erosion, opening, and closing algorithms. IEEE Trans. Pattern Anal. Mach. Intell. 2002, 24, 1606–1617. [CrossRef] 37. Mouginot, J.; Scheuch, B.; Rignot, E. Mapping of ice motion in antarctica using synthetic-aperture radar data. Remote Sens. 2012, 4, 2753–2767. [CrossRef] 38. Rignot, E.; Mouginot, J.; Scheuchl, B. Ice flow of the Antarctic ice sheet. Science 2011, 333, 1427–1430. [CrossRef]