Spaceborne SAR-Based Detection of Ships in Suez Gulf to Analyze the Maritime Traffic Jam Caused Due to the Blockage of Egypt’s Suez Canal

Abstract

With the convenience and connectedness of the oceans in recent years, there has been an increase in naval traffic, which has prompted maritime surveillance to attract special attention due to its significant application in marine operations. Ships, because of their uneven and rugged design, appear as a brighter patch, which aids in their identification by the Synthetic Aperture Radar (SAR), an active remote sensing technique. In this study, Sentinel-1 and Sentinel-2 datasets are used to detect vessels in the Gulf of Suez in order to examine the increasing maritime traffic induced by the Suez Canal blockage caused by the Ever Given ship becoming stranded in the canal on 23 March 2021 and being freed after 6 days on 29 March 2021. The usefulness of dual-pol spaceborne SAR datasets in ship detection is also determined. The analysis was performed within a time window spanning before, during, and after the blockage event. On the basis of the experimental results, Sentinel-1 images proved to be more effective compared to Sentinel-2 images for ship detection due to the all-weather capability of the Sentinel-1 dataset. Furthermore, the ship detection results obtained in dual polarization were substantially more accurate than the results obtained in a single polarization.

1. Introduction

The ocean drives global processes that keep the Earth livable for humans; it is an important aspect of our biodiversity since it is home to a variety of marine creatures, regulates the temperature, and offers food. The ocean is also important for economic development, with a huge percentage of trade traveling through marine channels, supporting millions of people. More than 50,000 seagoing ships transport more than 10 billion tons of important and desirable cargo each year, including goods, fuels, natural resources, and consumer items [1]. Though oceans have linked the world via trade and transportation, another aspect of the ocean is the various ship wreckages it brings with it. Various accidents have occurred in oceans and seas, either purposefully or unintentionally. The most common causes of marine casualties include capsizing, grounding, sinking, collision, or blockage. Oil and chemical spills, hazardous waste dumping, and bio invasion by foreign microbes are all causes of the environmental degradation of oceans.

Because of these concerns, marine security and safety are critical, and marine surveillance has risen in importance in recent years. Many techniques have been developed to make ship monitoring a simple operation, providing information on the presence and activities of ships, out of which cooperative and non-cooperative systems are two major categories. In the first system, vessels cooperate to disclose their identity and location. In the latter technology, vessel tracking systems (VMS) and automatic identification systems (AIS) are used to track maritime movement, along with providing information about the position and speed of the ship [2]. To detect objects, these systems include sensors such as cameras and radar on a variety of platforms such as ships, airplanes, and satellites. Using aerial and satellite sensors, it is now possible to monitor marine vessels from a distance, independent of ground circumstances [3]. Remote sensing data are useful in marine surveillance, and SAR imaging has made ship detection quite an easy process. SAR data are made up of the focused reflected returns of radar-frequency energy from terrain that has been brightened by a sensor’s directed pulse beam. The key parameters influencing radar returns from the topography are surface roughness, geometric structure, electrical characteristics (relative permittivity, conductivity), and sensor radar frequency. Sentinel-1 is an SAR satellite that acquires images with wide coverage and has the advantages of imaging irrespective of the weather conditions. Depending on the surface of the item, certain pixels look brighter while others appear darker. Because ships’ construction is irregular and rough, ships scatters more light and are also perpendicular to the route of propagation; therefore, more reflected light is received back by the radar, resulting in bright pixels. Smooth and calm objects, such as the water or ocean, reflect less light, so less light reaches the radar, resulting in darker patches. Because of this phenomenon alone, it is easier to see white patches against a dark backdrop, assisting in the detection of marine items distributed across a wide sea. In addition, Sentinel-2 datasets are high-resolution optical images that offer global coverage, hence making them useful for ship detection. Although the detection of ships is a major concern, it is also necessary to know other parameters of the detected ships, such as their size, length, vessel type, velocity, and so on. Various academics have used various approaches to estimate the length and direction of the vessel. Ship signature pixels are displayed as points on the Cartesian plane to determine the size and orientation of the ship. Point clouds have been obtained using the least squares line. The direction of the conforming line in relation to the range direction is used to compute the direction of the ship in relation to the range direction. Length is defined as the distance between the outermost pixels along the optimal line, and width is the distance orthogonal to the optimal line [4]. Another parameter, namely vessel categorization, is still in an early stage compared to ship detection, which has progressed due to the development of numerous approaches. The major goal of vessel classification is identifying the best qualities that may be used for categorization. The process becomes rather challenging due to the complicated structure of vessels, since signatures fluctuate as a function of the observation angle [5]. Various techniques are being proposed by various researchers for vessel classification utilizing the most recent deep learning technology. A novel convolutional network method has been created for this purpose, capable of classifying ships into several categories, such as cruise, tanker, cargo, military, and carrier. Aside from that, five pre-trained models have been employed, with Resnet-152 being named the best model and LexNet being named the poorest model with the lowest accuracy [6]. Ma et al. (2018) discuss vessel classification at the patch level using a convolutional neural network [7]. A novel architecture based on multitask neural networks was devised by Dechesne et al. (2019), consisting of a combined convolutional network linked to three task-specific networks for classification, ship identification, and length estimation [8]. For the categorization of the structure, this model has an accuracy of 97.25 percent. A great deal of research is still being conducted in this field in order to make a breakthrough. Since the well-being of millions of people and industrial sectors is dependent on the ocean, it has become clear with each passing year that the ocean’s health is deteriorating due to numerous activities that entail the usage of the ocean. The operational discharge of hydrocarbons and their accumulation in water bodies from ships, both licit and illicit, is a primary factor in the development of surface slicks, which are a source of worry regarding both the sea surface and offshore [9]. Illegal fishing is another major source of concern, as it frequently leads to resource overexploitation, along with other activities such as piracy, trafficking, smuggling, and so on. Ocean traffic is expanding rapidly as a result of the high demand for the commerce of diverse items across different countries. The congested waterways increase the risk of accidents, which has a negative impact on both the commercial and public sectors [10]. This is quite evident from the major incident that occurred in the Suez Canal, which threw the worldwide supply chain into disarray. Due to heavy winds, the 400-meter-long cargo freighter named Ever Given became trapped across the body of water on the morning of 23 March 2021, with its bow and stern wedged in the Suez Canal bank. It persisted for six days, stopping all vessels from passing through and causing major congestion on both sides of the canal. The management of the Suez Canal during this incident was accomplished using Sentinel-1 images. Blocked ships were placed in different geographical places to allow ample distance between two vessels. Finally, Ever Given was freed on 29 March 2021, and vessels began to travel across the canal, alleviating gridlock. This one catastrophe served as a wake-up call for many authorities, demonstrating how the global economy may be badly impacted by a single incident. Many research groups went on to study this incident from various perspectives. From an economic view, the canal’s closure had a significant impact on trade between Europe, Asia, and the Middle East. According to the Suez Canal Authority, the increased traffic and delays in trade supplies cost them USD 14 million in revenue in a single day [11]. A paper by Rashid et al. (2022) outlines marine traffic anomalies, such as the increased congestion caused by the Suez Canal barrier. An examination of satellite automated detection system-based ship trajectories, as well as of the Sentinel 1 and Sentinel 2 image-based ship positions, reveals that the blockage impacted maritime trade for more than three weeks [12]. Another study employed AIS data and a spatiotemporal data cube to examine the change in the number of ships, the cargo capacity obstructed near the Suez Canal and the increased waiting times for ships outside nodal ports [13]. Gast et al. (2021) developed a queueing model to study the recovery of an interrupted transport route, thereby predicting the delays induced by the disruption [14]. Another study looked at the Suez Canal barrier from an environmental standpoint. Images from the Multitemporal Sentinel 2 proved valuable in determining the impact of the canal obstruction on the Total Suspended Matter (TSM) concentration [15].

The surveillance of the ocean has become an essential task in order to safeguard it from the aforementioned unforeseen incidents. Ship detection and categorization provide crucial data for rational decisions, lowering the amount of time required to make a choice, which is critical in a real-time situation [16]. This necessary stage helps to minimize the number of marine accidents. Remote sensing imaging technology plays an important role in achieving the objective of monitoring a wide variety of sea surfaces. Because the datasets for Sentinel-1 and Sentinel-2 are not accessible for the same dates and geographical region, a direct comparison of the results is difficult. The paper’s originality is in the use of two adaptive threshold algorithms on SAR data to examine the increase in traffic, and the fact that the findings achieved are compared with the Sentinel-2 dataset.

The primary goal of the study is to investigate the maritime traffic congestion caused by the Suez Canal blockage due to Ever Given’s stranding using the Sentinel-1 dataset. We examined the fluctuations in the obstructed vessels in the Gulf of Suez before, during, and after the incident and compared the findings to the Sentinel-2 dataset in order to determine the effectiveness of both datasets for ship identification. The suitability of polarimetric channels for ship detection is also determined.

2. Algorithms and Approaches for Ship Detection

Ship detection utilizing SAR is critical in maritime security, whether in the hunt for lost ships or in commercial or military maritime control. Various methods and algorithms have been developed by researchers for this goal in order to facilitate vessel detection. One such algorithm is adaptive thresholding. Adaptive thresholding is a method used for object detection in SAR imagery. Various researchers have utilized this approach to analyze the strength of this approach in the ship detection arena. The method has been tested on TerraSAR-X stripmap SAR data and the results confirm that a high detection rate is attained, with zero missed detection [17]. There is various ship detection software that utilizes the concept of adaptive thresholding. One such tool is the Sentinel Application Platform (SNAP). The European Space Agency developed SNAP to identify ships in a clustered backdrop. This software employs an adaptive threshold method for vessel detection called the two-parameter CFAR (Constant False Alarm Rate) Detector. This software requires the preprocessing of the image, after which it can be utilized by an ocean object detection tool that helps in the easy and rapid identification of vessels. Search for Unidentified Maritime Objects (SUMO) is another ship detector developed by the fisheries control group at the Joint Research Centre (JRC) of the European community over 15 years; it is implemented as a Java 8 software package that employs the CFAR approach alongisde the K-distribution model. SUMO 1.3.5 was originally written in Interactive Data Language (IDL), but newer versions are written in JAVA, while test versions are written in MATLAB. SUMO, in its earlier versions, uses a template-matching approach; however, this was abandoned because this approach does not apply enough to all ship sizes and image resolutions [4]. Several techniques, notably the CFAR algorithm, have already been presented for vessel detection. Although working with CFAR algorithms using the software outlined above is simple, they are incapable of producing effective results in complicated contexts. Furthermore, due to the lack of a classification layer, CFAR algorithms are unable to give vessel classification. In recent times, the convolutional neural network (CNN) has joined the remote sensing arena and improved the accuracy of vessel recognition. Several experiments have been conducted utilizing various CNN networks to not only identify the ships, but also to offer the coordinates and categorize the vessels into several categories. The research conducted by Stofa et al. (2020) employs a DenseNet architecture as a CNN-based classifier, with some fine-tuning to achieve the best results. It claims to have a success rate of 99.75% percent when using the Adam optimizer [18]. The detection speed is also an important factor to consider when comparing different algorithms. Li et al. (2020) use a Faster region-based convolutional neural network (R-CNN) to extract the features from each target in order to enhance the recognition and localization task network. This study promises to deliver a detection speed that is 800% quicker [19]. Another work illustrated by Chen et al. (2021) extracts features using DarkNet-53 with residual units, and a top-down triangle design is incorporated for multiscale feature analysis with concatenation. It promises to reach 95.52% accuracy and detect speeds of up to 72 frames per second [20]. A modified Single Shot Multibox Detector (SSD) with a multi-resolution input is presented by Ma et al. (2018) in order to identify various sorts of maritime objects. According to these experiments, the output of CNN-based vessel recognition and detection is significantly superior to and more accurate than the CFAR algorithm [7]. Deep learning has the potential to significantly improve ship identification techniques. You Only Look Once version 2 (YOLOv2) is another region-based CNN. Much research has been conducted to demonstrate that this method outperforms the other current algorithms. A novel architecture with fewer layers is utilized to work around YOLOv2 to minimize the computing time and improve the detection accuracy. The results reveal that it beats Faster RCNN and SSD techniques, with an accuracy of 90.05% [21]. YOLOv2, proposed by Khan and Yunze (2018) in their study, is tested against three distinct types of datasets to see how well it performs [22]. YOLOv4 may be thought of as an improved version of the model for detecting ships [23]. Using YOLOv5, ship detection is performed by Ting et al. (2021). This study combines the feature attraction process with the GhostbottleNet approach and indicates that it not only improves accuracy, but also reduces GIoU over time [24]. This technology is still in its infancy. Many scholars are working to develop better versions than the ones that are now available. Another novel approach for ship detection is presented based on the analysis of SAR images, utilizing the discrete wavelet transform. The exposed approach employs statistical differences between vessels and the neighboring sea, reading the data with wavelet coefficients to enable better detection [25]. Another study [26] reports on the results of a wavelet treatment applied to ENVISAT SAR images. Another study conducted by Staliango et al. (2012) involves a two-stage detection strategy employing the two-dimensional Discrete Wavelet Transform (2D-DWT) and the CFAR processor [27]. There is still a lot of research being performed in this field.

3. Study Area and Data

3.1. Study Area

The area of interest (AOI) of this study is the Gulf of Suez. It is connected to the Suez Canal from the north and to the Red Sea from the south. The time period of this study is from March 2021 to April 2021. Figure 1 shows the true color composite image captured using Sentinel-2’s optical multispectral capabilities, depicting the Gulf of Suez. The Suez Canal is a 152-year-old artificial sea-level waterway in Egypt with a total length of 193.30 km. The Suez Canal has a significant influence on travel from the Persian Gulf to the Northern European range, as a 21,000 km route across Africa that takes 24 days is reduced to a 12,000 km journey that takes 14 days. As a result, depending on the ship’s speed, the Suez Canal can save between 7 and 10 days of shipping time. Previously, Indian ships had to circumnavigate the Cape of Good Hope, which made the journey longer and more expensive; however, with the completion of the Suez Canal, the shipping of Indian commodities became much faster. The direct link between the Indian Ocean and the Mediterranean Sea creates a more direct maritime route between Asia and Europe, making it crucial for international trade [28]. This busy trade route was affected when the 200,000-tonne ship container, Ever Given, which is 400 m long, 59 m broad and managed by the Taiwanese shipping company Evergreen Marine, ran aground and became trapped sideways across the channel on 23 March 2021. The Ever Given was bound for the port city of Rotterdam in the Netherlands from China and was passing northwards through the canal on its way to the Mediterranean [29]. Figure 1 shows the natural color composite of Sentinel-2B MultiSpectral Instrument (MSI) data for the Suez Canal and Gulf of Suez. Two images obtained by Sentinel-2 (S2B_MSIL1C_20210329T082559_N0209_R021_T36RVT_20210329T103749 and S2B_MSIL1C_20210329T082559_N0209_R021_T36RVU_20210329T103749) that were acquired on 29 March 2021 over the Gulf of Suez were used to create a mosaic of the study area. The red circled dot in Figure 1a shows the location at which the Ever Given ship was stuck. Sentinel-2 data were acquired in 13 bands, with a 10 m spatial resolution in the bands B2 (490 nm), B3 (560 nm), B4 (665 nm), and B8 (842 nm); a 20 m spatial resolution in the bands B5 (705 nm), B6 (740 nm), B7 (783 nm), B8a (865 nm), B11 (1610 nm) and B12 (2190 nm); and a 60 m spatial resolution in the bands B1 (443 nm), B9 (940 nm) and B10 (1375 nm). Three bands corresponding to a 10 m spatial resolution in the visible range were used to generate the natural color composite image, with B2 (490 nm) represented in blue, B3 (560 nm) represented in green, and B4 (665 nm) indicated in red. Although several patches of cloud cover can also be seen in the Sentinel-2 natural color composite of the data, fortunately, the area above the Ever Given ship was not affected by cloud cover, and the visualization of the ship is possible using 10 m resolution spaceborne optical multispectral data. One of the many advantages of SAR data is that they are not affected by cloud cover, are are thus used extensively for Earth Imaging when clouds are in the atmosphere. However, the optical multispectral Sentinel-2 and Sentinel-1 SAR data acquisition dates were different, making it difficult to tell whether there was cloud cover or not on the day that the SAR data were acquired. However, if any cloud cover is present in the atmosphere during SAR acquisition, then its effect will not be visible in the data and it will look as it does in Figure 2.

Figure 2 shows the Radiometric Terrain-Corrected (RTC) product of the Sentinel 1-B data for the Suez Canal that were processed using the hyp3_gamma plugin version 6.2.2 running GAMMA release 20220630 [30]. The Sentinel 1-B data of 27 March 2021 were projected to WGS 84/UTM zone 36N. The pixel spacing of the generated product was 10 m. The false-colors RGB image was generated by keeping VV polarization in a red color, VH polarization in a green color, and VV/VH polarization in a blue color. The ration VV/VH enhances all the scatterers, which are responsible for odd-bounce or surface scattering. Blue also indicates the low backscatter from smooth surfaces in both the polarimetric channels (VV and VH) of the Sentinel-1B data. Figure 2a shows the false-color RGB view of the Suez Canal and Gulf of Suez. A large number of ships could be easily detected in the Gulf of Suez (Figure 2a). In the False Colors RGB image of Figure 2a, ships are highlighted as bright point targets, which are shown in yellow in the dotted red circle in the Gulf of Suez. The red-colored box in Figure 2b shows the SAR backscatter-based RGB view of the Ever Given ship. Since the ship contributes to a high level of backscatter, it appears as a bright object in the waters of the Suez Canal, which contributes to a low level of backscatter. Another bright object that contributes to a high level of backscatter near the Ever Given ship is obtained from the floating bridge (green-colored rounded rectangle) in the Suez Canal that connects the west and east sides of the channel.

Figure 3 is the Capella Space SAR data of 25 March 2021, taken in the X-band with spotlight mode using HH polarization, and showcasing the Ever Given as the bright pixel; this clearly depicts the blockage caused by the ship’s stranding in the canal.

The ship obstructed the way of other vessels, which were caught in lines in both directions. Dozens of vessels became mired, causing congestion, while waiting for rescue boats to free the ship, which was knocked off course by strong winds. Vessels were made to wait for the obstruction to clear in order to pass the canal. Numerous nations in the world’s East and West have faced terrible economic consequences as a result of the canal blockade. The obstruction disrupted the delivery of commodities worth nearly USD 9 billion each day, which is equivalent to USD 400 million in commerce every hour [31]. The blockade caused delays and interruptions in global supply systems, disrupting the timely delivery of commodities and potentially resulting in product shortages. Shipping, industry, and energy were among the businesses most affected by the obstruction. Shipping firms were forced to detour their vessels around the southern point of Africa, adding thousands of miles and weeks to their travels, resulting in higher fuel prices and delivery delays. As a result, firms that rely on the just-in-time delivery of raw materials and components experienced production delays and supply chain interruptions. Oil prices soared as a result of the disruption of oil tankers traveling through the canal, affecting the energy industry as well. The economic losses affected not only individual companies, but also global trade and the economy as a whole. As the ship was stranded for more than 6 days, a trade loss of nearly USB 54 billion was anticipated [32]. After the Ever Given was freed, the traffic on both sides of the canal started to decrease.

3.2. Dataset

This study takes into account Sentinel-1 and Sentinel-2 datasets with varied acquisition dates. Since the Ever Given was mired between 23 March 2021 and 29 March 2021, in order to examine the increased traffic congestion, before, during, and after acquisition dates were selected.

Table 1. Detailed overview of Sentinel-1 dataset used in this study and its characteristics.

Characteristics	Dataset 1	Dataset 2	Dataset 3
Mission	Sentinel-1B	Sentinel-1B	Sentinel-1B
Mission datatake ID	203,453	204,876	206,295
Sensing Start (UTC)	15 March 2021, 03:44:09	27 March 2021, 03:44:10	8 April 2021, 03:44:10
Sensing Stop (UTC)	15 March 2021, 03:44:34	27 March 2021, 03:44:35	8 April 2021, 03:44:35
Mode	IW	IW	IW
Product Type	GRDH	GRDH	GRDH
Polarization	VH, VV	VH, VV	VH, VV
Pass Direction	Descending	Descending	Descending
Looking Direction	Right	Right	Left
Product Level	L1	L1	L1
Instrument Name	SAR (C-band)	SAR (C-band)	SAR (C-band)

includes a full description of the Sentinel-1 dataset and

Table 2. Detailed overview of Sentinel-2 dataset used in this study and its characteristics.

Characteristics	Dataset 1	Dataset 2	Dataset 3
Mission	Sentinel-2B	Sentinel-2B	Sentinel-2A
Sensing Start (UTC)	19 March 2021, 08:26:39	29 March 2021, 08:25:59	13 April 2021, 08:26:01
Bands	13 Bands	13 Bands	13 Bands
Pass Direction	Descending	Descending	Descending
Instrument Name	MSI	MSI	MSI

includes a full description of the Sentinel-2 dataset used in this study.

4. Methodology

The overall goal of this study was to identify vessels in the Gulf of Suez utilizing active remote sensing data obtained from Sentinel-1 to examine the maritime traffic caused by the Suez Canal blockage.

4.1. Pre-Processing of SAR Data

The process of pre-processing includes subset selection, orbit file application, calibration, and speckle filtering. When Sentinel-1 SAR data are loaded into the SNAP, the first step is to build a subset of the image that only focuses on the area of interest (AOI). This will help to reduce the processing time. The orbit information in Sentinel-1 image data has various issues. To remedy these issues, the Sentinel-1 image orbit data must be updated using accessible orbit file data. As a result, following sub-setting, the orbit file is applied. Using an orbit file offers precise satellite location and velocity data. The orbit state vectors in the product’s abstract metadata are modified based on this information. Sentinel-1 IW GRDH raw data are an amplitude image. For quantitative use, it is critical to convert the amplitude image to a calibrated product. This calibration is required for SAR images in order for the pixel values to reflect the radar backscatter of the reflector appropriately. The SAR image features speckles that look like salt and pepper, making interpretation challenging. It is critical to mitigate this impact. As a result, speckle filtering is used to reduce the impact of speckle noise on the SAR image.

4.2. Ship Detection Using Sentinel-1 Dataset

To detect the vessels using the Sentinel-1 dataset, two distinct adaptive threshold algorithms are used. The first approach used is a two-parameter CFAR detector whose overall methodology is depicted in Figure 4.

The two-parameter CFAR detector is a one-of-a-kind adaptive threshold method that is widely utilized in the Sentinel Application Platform (SNAP). To obtain results via this approach, SNAP version 9 is used.

Following preprocessing, the image is ready to be used in the ship detection procedure. The SNAP tool’s ocean object detection operator recognizes objects such as ships on the sea surface using SAR images. Object detection operations are divided into three steps: land–sea masking, pre-screening, and object discrimination.

• Land–Sea Masking: To minimize the erroneous detection of light poles, bridges, and land mass, which appear as bright pixels in SAR images, a land–sea mask must be built to mask the land area. By default, the SRTM 3-sec digital elevation model is utilized to detect and filter off regions of positive elevation. In certain ports, extensive pixel masking from the beach can be used to evade the detection of another object. Another approach might be to use vector data (.shp) to mask the region of interest and avoid false detection.
• Pre-Screening: Adaptive thresholding is a popular approach used for the prescreening process. It is applied to moving windows and, depending on each pixel under test, a new threshold is calculated based on the statistical characteristics of its local background. If the pixel under evaluation is higher than a certain threshold, it is considered an acceptable target [33]. SNAP’s CFAR algorithm for ocean object detection aids in the quick detection of ships. The user defines the parameters of the moving window. Each pixel being examined is divided into three panels. As illustrated in Figure 5, they are, from the outer to the inner layer, the background window, guard window, and target window.

The target window should be around the length of the smallest vessel to be recognized, the guard window should be about the size of the largest window to be recognized, and the background window should be large enough to compute the local statistics [33]. The detection design parameter t is determined by the probability of a false alarm (PFA) [33], as shown below:

$$\mathrm{P}\mathrm{F}\mathrm{A}=\frac{1}{2}-\frac{1}{2}\mathrm{e}\mathrm{r}\mathrm{f}\left(\frac{\mathrm{t}}{\sqrt{2}}\right)$$

(1)

The valid PFA value is in the range [0,1].

3.

Discrimination: The object discrimination operator performs the discrimination process. It assists in removing false target detection based on target dimensions. First, all contiguous detected pixels are combined into one cluster. Second, the width, as well as the length information of the clusters, are retrieved, followed by the exclusion of targets that do not fall within the detection limit, which helps to reduce false detection.

After using the ocean object detection tool to generate the findings, there is a potential that the instrument recognizes a false target or perhaps misses a vessel. As a result, it is critical to visually inspect each identified target and eliminate false alarms in order to optimize the findings.

The second approach used to achieve the objective of this study involves using a Constant False Alarm Rate with a K-distribution model, whose overall methodology is depicted in Figure 6.

The method is designed to detect brighter pixels among a noisy and chaotic sea background. A vessel can be clearly distinguished from its surroundings if the radar reflection produces a high backscatter, resulting in brighter pixel values that are higher than the typical background. On the other hand, the bright pixels of the SAR data in the background can sometimes interfere with the algorithms and be detected, resulting in false alerts. As a result, maintaining an error-free threshold is critical for preventing error sources. To account for the unanticipated scattering of radar backscatter, the clutter background characteristics are modeled using the K-distribution model. The three parameters of this model are one PDF mean and two PDF width measurements. The below formula is used to calculate the CFAR threshold [3].

$${\int }_{\mathsf{\vartheta }\mathrm{F}\mathrm{A}}^{\mathrm{\infty }}(\mathrm{f}\left(\mathrm{k}\left(\mathrm{a}\right).\mathrm{d}\mathrm{a}\right)=\mathrm{P}\mathrm{F}\mathrm{A}$$

(2)

Non-overlapping windows are used to determine the local background pixel parameters. For each local window, the adaptive threshold is established. To assist in determining the local threshold, the CFAR technique is utilized. The PDF of the background clutter is analyzed to establish the threshold value, and values that exceed it are considered targets. The number of false alarms remains constant [3].

One such interface that utilizes the CFAR with the k-distribution model is Search for Unidentified Objects (SUMO). It is a Java software package used to recognize ships. It uses amplitude pictures to detect locally bright pixels in the ocean and can work in both fully automated and semi-automatic modes. It provides users with a variety of global vector shoreline files to differentiate land from sea. Various processing parameters are required to begin the ship detection process, such as the Equivalent Number of Looks (ENL) and the K-distribution model. SUMO also creates attributes for recognized targets, such as length, width, pixel number and intensity, heading location, geographic location, reliability value, and the number of identified pixels.

4.3. Ship Detection Using Sentinel-2 Dataset

For ship detection utilizing Sentinel-2 datasets, a visual examination was performed in the Normalised Difference Water Index (NDWI). The presence and quantity of water in an image are quantified using the NDWI product. It is used in satellite imagery to emphasize open-water features, enabling a body of water to stand out against soil and vegetation. The NDWI is computed utilizing Band 8 (NIR) and Band 12 (MIR) of Sentinel-2 data, which allows it to detect small changes in the water content.

When ships are present in an ocean landscape, their reflectance pattern differs from that of the water [34]. Because of their structural materials and colors, ships often reflect less near-infrared light and more visible light. This contrast in the reflectance values between ships and water can be captured by NDWI, hence making ships appear as a white patch that can be identified using visual inspection.

5. Results and Discussion

When Ever Given became grounded, the traffic on both sides of the canal surged dramatically. Various vessels waited for the blockade to be eased. In this section, two adaptive threshold methods are used to analyze the traffic caused by stranded vessels in the Gulf of Suez, and the obtained results are compared with the results obtained using the Sentinel-2 dataset.

5.1. Ship Detection Results Using Sentinel-1 Dataset

Two different adaptive threshold methods are used to obtain the results using Sentinel-1 images. The first method is the two-parameter CFAR detector. Sentinel Application Platform version 9 is utilized in this study to obtain the findings via this method. The Sentinel-1 image is opened in the software. In order to lessen the effects of SAR, preprocessing of the Sentinel-1 IW GRDH image is carried out, followed by subsetting and speckle filtering. To reduce the chances of false detection, the land mask is applied with a shoreline extension of 50 pixels. The minimum and maximum target sizes in this investigation are set to be 30 m and 600 m, respectively. The detection results obtained using this method are listed in

Table 3. Vessel detection results obtained using the two-parameter CFAR approach.

Date	Detected Targets	Ambiguities Removed	Missed Targets	Total Targets
15 March 2021	199	47	9	161
27 March 2021	298	62	17	253
8 April 2021	211	53	6	164

.

Another approach involving adaptive thresholding is utilized as a CFAR with a K-distribution model. SUMO helps to obtain the ship detection findings via this approach. The Sentinel-1 image in the natural coordinate range and in azimuth is imported into the SUMO GUI, together with its information. A land mask with a seaward buffering of 250 m is used to avoid false alarm detection. SUMO uses the K-distribution method for detection, with threshold modifications of 1.8 for co-pol bands and 1.5 for cross-pol bands. The detection results are listed in

Table 4. Vessel detection results obtained using CFAR with K-distribution model.

Date	Detected Targets	Ambiguities Removed	Missed Targets	Total Targets
15 March 2021	274	96	1	179
27 March 2021	382	101	0	281
8 April 2021	265	94	2	173

.

5.2. Traffic Analysis Using Sentinel-2 Dataset

A visual inspection of the Sentinel-2 data was conducted in NDWI in order to detect the vessels in the Gulf of Suez. The detection results are listed in

Table 5. Vessel detection results obtained via visual inspection of Sentinel-2 dataset.

Date	Targets Detected
19 March 2021	87
29 March 2021	259
13 April 2021	76

.

5.3. Traffic Analysis in the Gulf of Suez

The ship detection results obtained via adaptive thresholding that are discussed above are analyzed to determine the traffic in the Gulf of Suez.

Some false alarms were detected as a result of the thorough manual assessment. A few minor targets were also missed. Total vessels were determined as follows:

TotalVessels = TotalDetectedTargetsbyTool − FalseAlarms + MissedTargets

(3)

The Ever Given became stranded in the Suez Canal on 23 March 2021. Prior to the event, 15 March 2021 was a day of normal operation in the Suez Canal. Therefore, the number of vessels at this stage is expected to be moderate. The findings acquired using both ship detection approaches show that the usual traffic in the Gulf of Suez is maintained at around 179 vessels. However, during the period in which the ship was stuck, the number of vessels detected on 27 March 2021 was nearly around 280 vessels. A significant surge in traffic can be seen owing to the blockade, as depicted in Figure 7. When a dredger and more than a dozen tugboats assisted in loosening the ship, and the ship was finally freed on 29 March 2021, the backlog of ships began to clear, and therefore the traffic started to reduce. The data demonstrate that on 8 April 2021, the number of vessels in the study region dropped below 175 yet again, approaching the pre-event total.

Similarly, the results obtained using the Sentinel 2 dataset via visual inspection reflect the same trend of maritime traffic even though detection was more challenging due to cloud cover causing low visibility. The 29 March 2021 datasets results reflect more than 250 ships waiting for the blockage to be cleared in the Gulf of Suez. Heavy congestion can be seen via this dataset compared to the results obtained in pre- and post-event datasets.

The results show that the findings achieved using SAR data seem more accurate than those achieved with Sentinel 2 data, as Sentinel-1 operates using microwave frequencies, allowing it to penetrate through clouds, rain, and fog. This all-weather capability ensures more easy detection. Meanwhile, optical sensors like those on Sentinel-2 cannot penetrate clouds, and therefore in the cloudy atmosphere, the possibility of visually inspecting the vessel reduces. Clearly, disasters like the one that happened in the Suez Canal are inescapable; businesses should thus consider the issue and increase their expertise in preparation for future development [35].

Figure 8 shows a visual comparison of the Optical Multispectral Sentinel-2 data and the ASF RTC product of the Sentinel-1B SAR data. Since it is a well-known fact that optical multispectral data are mainly sensitive to the reflectance of the sunlight from the top of the object’s surface, a good sensitivity towards the discrimination of color-based features is shown. A first look at the visual interpretation of the optical multispectral data in Figure 8a does not show a good result for ship detection; this is because the color of the ships is merged with the background water and it becomes very difficult to obtain any information pertaining to the ships in the true color composite image of the optical multispectral data. On the other hand, ships are clearly visible as bright objects in the SAR image (Figure 8b); this clear identification of the ships in the SAR data is due to the sensitivity of the microwaves used in the SAR imaging toward the structural and electrical properties of the targeted objects. Figure 8c shows the zoomed view of the optical multispectral image; on careful observation, the ships are visible as some dim and indistinct objects of small size on the upper surface of the water. As far as the visibility of ships in SAR data is concerned, it is clear from Figure 8d that not only ships can be detected, but their size and type can also be estimated.

Although detecting ships using optical data is a difficult task, the advantage of a large spectral resolution is its ability to generate spectral indices using spectral band combinations. One of those indices is the Normalized Difference Water Index (NDWI), which helps to highlight the water surface. Apart from water, for things in which water or moisture are not present, their NDWI value is very low. Figure 9a shows the NDWI map of the Gulf of Suez. Figure 9b shows the land-masked NDWI image of the Gulf of Suez that is generated by masking out the land using the Shuttle Radar Topography Mission Global 3 arc second product. Ships can be seen as small white objects in the zoomed NDWI product (Figure 9c).

5.4. Determining the Suitability of Polarimetric Channel

This study made use of a dual polarization Sentinel-1 image. The targets are identified in VV, VH, and the merged products of VV and VH using the CFAR K-distribution model employed by SUMO; finally, the findings are compared against each other in order to determine the suitability of polarimetric channels.

Vessel Detection Results in Different Polarization

VV polarization is a kind of radar polarization in which signals are both transmitted and received in a vertically oriented manner. The threshold modification for co-pol bands is set to be 1.8. The detection result in this band from the SUMO depicts the vessels in a blue rectangle, as shown in Figure 10.

VH polarization is a kind of radar polarization in which signals are transmitted in the vertical plane and received in a horizontally oriented manner. The detection result in this band from the SUMO depicts the vessels in a green triangle, as shown in Figure 11. The threshold modification for cross-pol bands is set to be 1.5.

In the merged product of VV and VH, the detection result from the SUMO depicts the vessels in a yellow plus sign, as shown in Figure 12. SUMO performs a union over all detection results in all polarisation channels, identifying the target as a detected ship if it is identified in any of the polarisation channels. The vessels detected in the VV and VH bands are comparatively lighter pixels and indicate tiny ships. The detection that is indicated in both polarizations seems brighter than the rest, indicating a huge object or a standard-sized vessel.

Table 6. Vessel detection results using different polarisations.

Date	VV Polarisation	VH Polarisation	Merged Products of VV and VH
15 March 2021	250	220	274
27 March 2021	305	342	382
8 April 2021	217	226	265

shows the number of vessels detected in each band. For the 15 March 2021 dataset, vessel detection using the CFAR K-distribution model shows 250 vessels being detected in VV polarization and 220 vessels identified in VH polarization;when vessels were identified in the merged product of VV and VH, the number of detected vessels was drastically increased to 274. Similarly, for the 27 March 2021 dataset, using the CFAR K-distribution model, vessels were detected in each polarization. In VV polarization, 305 vessels were detected, while in VH polarization, 342 vessels were detected. The discovered vessels in the merged output of VV and VH polarization grew dramatically to 382. The application of the CFAR K-distribution model to the 8 April 2021 dataset in order to detect vessels in different polarizations also produced similar results. In total, 217 ships were identified in the VV polarization, followed by 226 ships in the VH polarization. A drastic increase in the detection of ships for the merged output of VV and VH is observed. The ships identified in this polarization rose to 265 vessels.

It is clear from the results that some vessels are not identified in VV but are detected in VH, and vice versa. However, the majority of the vessels are visible in the merged product of VV and VH polarization, albeit in slightly different positions. Dual polarization is preferable since it aids in the efficient identification of vessels. It assists in overcoming the limitations associated with single polarization data by giving two channels of intensity and phase information. Scattering processes are distinguished using the two data channels. It also gives extra information about surface properties via the various and complimentary echoes. A VV and VH combination would be ideal for ship surveillance since the VH channel provides point target information against a very dark cluttered backdrop. At the same time, the VV polarisation would produce enough ocean surface backscatter to enable wake analyses [36]. Due to all of these benefits, dual polarisation data are sufficient.

6. Conclusions

Ocean monitoring is becoming increasingly crucial in order to maintain the maritime environment, the economic sustainability of the maritime sector, and navigation safety. With increased traffic in the Gulf of Suez and its essential connecting route to the Suez Canal, vessel surveillance is becoming increasingly important. The prime focus of the study was to utilize two methods of adaptive thresholding in order to produce ship detection results in the Gulf of Suez in order to analyze the maritime traffic caused by the stranding of the Ever Given in the Suez Canal. The experimental results show that on 15 March 2021, when the Suez Canal was operating normally, the number of ships detected was less than 179 vessels. When the Ever Given became stranded, it is clear that the traffic in the Gulf of Suez drastically increased in the six days, as is evident from the results obtained via the 27 March 2021 dataset, wherein around 280 ships were detected. As soon as it was freed, the traffic in the study area was shown to decrease gradually, as proven by the results obtained via the 8 April 2021 dataset; here, the number of vessels detected was close to the pre-event total. Other findings show that the merged product of VV and VH polarization is more suitabable for ship identification, as is evident from the results obtained in Table 6. Accurate findings and measurements are very vital for the decision-making process when potential environmental, economic, and livelihood aspects are at risk; therefore, continuous advancements in the ship detection algorithm are required to boost accuracy and precision. Since ship detection software does not guarantee full accuracy, athough the detected results have been manually inspected, detection with complete precision still remains a challenge. Improving the accuracy of the findings produced may be investigated in future research using more advanced algorithms such as deep learning, CNN, and so on.