3.1. Ocean Surface Wind (OSW)
OSW is an essential parameter for various applications, such as marine disaster monitoring, climate change modeling, water mass formations, and Numerical Weather Prediction (NWP) [
76,
77,
78,
79]. Considering the limitations of the traditional methods for OSW estimation (e.g., anemometers and buoys) [
76,
80], RS observations have emerged as cost-effective techniques [
81]. Remotely sensed OSW information mainly relies on the relationship between the OSW and the sea surface roughness, which represents emissive and reflective properties of the ocean surface [
79]. Five RS systems have been frequently applied to measure OSW: microwave radiometer, GNSS-R, SAR, scatterometer, and HF radar. The advantages and disadvantages of each system, summarized in
Table 1, are discussed in more detail in the following subsections.
3.1.1. Microwave Radiometer
Microwave radiometers can estimate OSW based on the spectrum of the microwave radiation emitted by the ocean surface [
79,
82]. OSW measurements from microwave radiometers mainly rely on the physical RT models, which estimate the microwave emission from the ocean surface and the emission and absorption by the Earth’s atmosphere [
83]. As discussed, the nonpolarimetric microwave radiometers (e.g., Advanced Microwave Scanning Radiometers (AMSR)) can only estimate OSW speed. However, the WindSat polarimetric microwave radiometer can provide both OSW speed and direction [
79]. Polarimetric microwave radiometers contain ±45° polarized and right/left circularly polarized channels along with vertical and horizontal polarizations. Polarimetric microwave radiometers retrieve OSW speed and direction by measuring the complex correlation between horizontally and vertically polarized microwave radiations [
32]. In fact, the difference between horizontal and vertical polarizations is related to OSW speed, and the complete state of polarization is related to OSW direction [
82]. The OSW direction from polarimetric microwave radiometer systems becomes less precise when wind speed is less than 8 m/s [
81,
84]. This is because the passive OSW direction signal is small in all polarizations at low wind speeds, causing noisy and inaccurate measurement of the OSW direction [
79,
85].
Generally, microwave radiometers provide better results in high wind speeds than other low-resolution RS systems, such as scatterometers. For example, it has been reported that in spaceborne L-band radiometers, received signals remained sensitive to increasing wind speeds up to 70 m/s [
86,
87]. Accordingly, some of the L-band radiometers (e.g., Soil Moisture and Ocean Salinity (SMOS) and Soil Moisture Active Passive (SMAP)) can provide a reasonable estimation of extreme OSW speeds at a coarse spatial resolution [
87]. The quality of the OSW measurements from microwave radiometers is negatively affected by rain [
83]. However, L-band radiometers (e.g., SMOS) are almost insensitive to the presence of rain or frozen precipitation [
88]. Various OSW products have been so far generated using the data collected by different microwave radiometers. For example, Yin et al. [
89] implemented a novel method to estimate OSW vector in severe weather conditions (i.e., the high precipitation environment of tropical cyclones) using WindSat data. This study used HWind data [
90] as the reference dataset and evaluated the proposed methodology in 17 tropical cyclones from 2003 to 2009. The results showed that the mean difference between the retrieved OSW speed and HWind data for 17 tropical cyclones was 0.2 m/s. The Root Mean Square Error (RMSE) was also 24.2°. The authors reported that the observed RMSE value was mainly related to a time or location mismatch between the HWind analysis and the WindSat data.
3.1.2. GNSS-R
As discussed, GNSS-R is a passive RS satellite utilizing the GNSS signals to estimate the OSW speed using the L-band (frequency and wavelength around 1.5 GHz and 20 cm) [
34]. GNNS-R measures multiple points over the ocean to estimate the OSW speed, and its geometry is not like a swath, individual point, or limited to the nadir direction [
91]. In contrast to traditional OSW speed retrieval methods, GNSS-R has better spatiotemporal sampling due to the global accessibility of the GNNS signal, which offers greater coverage over the ocean [
35]. Another advantage of GNSS-R is its lower sensitivity to small atmospheric particles and small-scale capillary waves since it utilizes a longer wavelength compared to scatterometers (i.e., wavelengths = 2–5 cm) [
35]. Additionally, lower cost and lower power needed for the GNSS-R receiver are the other benefits of these systems for OSW estimation [
92].
The first GNNS-R data were collected by the UK-DMC satellite between 2004 and 2005 [
37]. Later, in July 2014, the second spaceborne polar-orbiting satellite (TDS-1) was equipped with the Space GNSS Receiver Remote Sensing Instrument (SGR-ReSI) to collect the reflected GNNS signals [
92]. The first OSW speed estimation from the TDS-1 mission was discussed in [
92]. This study suggested an OSW speed retrieval method based on the Signal-to-Noise Ratio (SNR) and GNSS-R Biostatic Radar Equation. The Meteorological Operational (MetOp) Advanced SCATterometer (ASCAT) satellite measurements were used to validate the results. It was observed that when SNR was greater than 3 dB, OSW speed could be estimated with an accuracy of approximately 2.2 m/s for wind speed values between 3 and 18 m/s [
92]. In late 2016, an advanced GNSS-R receiver was installed on the CYGNSS to provide higher-quality OSW speed products [
93].
Fusing measurements from multiple GNSS-R systems can improve the spatiotemporal resolution of the estimated OSW speed [
94]. Retrieval of OSW speed from GNNS-R data is usually based on quantities known as observables, which are extracted from the GNSS delay-Doppler maps peak [
7,
13]. A series of algorithms, including the delay-Doppler map variance, delay-Doppler map average, trailing edge slope, leading edge slope, and Allan delay-Doppler map variance, have been so far introduced in [
8] for OSW speed estimation from GNSS-R data. The authors also suggested a new OSW speed estimation algorithm, known as minimum variance, based on the integration of OSW data from each single observation [
35]. Moreover, a new parametric Geophysical Model Function (GMF), extracted from the CYGNSS measurements, was proposed by [
95] to relate the OSW speed to the two observables of the slope of the leading edge of the radar return pulse scattered by the ocean surface and the normalized bistatic radar cross-section of the ocean surface. In this study, an overall RMSE of 1.4 m/s was obtained when the results were compared with in situ data [
95].
3.1.3. SAR
SAR sensors measure the Normalized Radar Cross Section (NRCS) variation from the wind-roughened ocean surface as a function of both OSW speed and direction. Accordingly, GMFs are developed to establish the relationship between OSW vector information, calibrated NRCS, and sensor viewing angles for different SAR bands and polarizations [
96]. However, since SAR sensors operate with a single azimuth view, it is required to infer whether backscatter variations are related to OSW speed (along-view OSW vector component) or OSW direction (across-view OSW vector component) [
81]. To resolve this issue, several different methods have been implemented, a comprehensive list of which are discussed in [
97]. It is also worth noting that rain can negatively affect the OSW estimation accuracy from SAR data, and rain contamination is more noticeable at shorter wavelengths [
81].
SAR systems are the only satellite-based system that can provide OSW information with a subkilometer spatial resolution (between a few meters and 100 m) [
96]. However, OSW retrieval using SAR data is generally implemented at spatial resolutions of 0.5–1 km for removing speckle noise as well as filtering ocean waves and other non-wind-induced features [
81]. SAR systems are efficient at both low and high winds with subkilometer spatial resolutions [
98]. The cross-polarized SAR data provide a very weak incidence and azimuth angle dependency, with no indication of saturation for the strongest wind speeds (up to 75 m/s). Accordingly, these capacities have opened new perspectives for OSW speed estimation in different settings, even in tropical cyclones [
98,
99,
100]. In this regard, Li et al. [
100] analyzed 83 SAR images, captured by Radarsat-1 and Envisat from 2001 to 2010, to investigate the morphology of the tropical cyclone eye in terms of shape and size distribution. The detailed atmospheric phenomena generated within tropical cyclones (e.g., rainbands, boundary layer rolls, and arc clouds) were also explored. This study demonstrated the application of SAR data for investigating the characteristics of windstorms with relatively higher resolution. In fact, by analyzing 30 typhoons and 43 hurricanes, the authors reported that the response of ocean surface to the storm-forced winds could be effectively explored by SAR data.
3.1.4. Scatterometer
Scatterometers have proven to be efficient tools for both OSW speed and direction measurements [
101,
102]. A scatterometer transmits a radar pulse and then measures the returned pulse ratio from the ocean surface [
103]. The differential response and the intensity of the reflected pulses determine the OSW direction and speed, respectively [
82]. The main operating radar frequencies in scatterometers are Ku-band (frequency = 13.4 GHz and wavelength = 5 cm) and C-band (frequency = 5.2 GHz and wavelength = 2 cm). Scatterometers provide global OSW vector information with spatial resolutions between 25 and 50 km, which is almost similar to that of microwave radiometers [
79]. Unlike SAR systems, the low spatial resolution of these two instruments cannot effectively satisfy several local applications, such as tropical cyclone analysis and coastal OSW mapping. For low OSW speeds, scatterometers provide more reliable information compared to microwave radiometers. However, scatterometers lose sensitivity, and the signal begins to saturate, at high wind speeds (above 35 m/s) [
104]. Like SAR systems, the rain contamination in scatterometers’ OSW measurements is more serious in the higher frequencies.
Many studies have so far investigated the potential of different scatterometers for OSW estimation. For example, Guo et al. [
103] examined the spatial variability of global OSW resources at heights of 10 m and 100 m above Sea Level (SL) using multiple scatterometers, including Quick SCATterometer (QuikSCAT), ASCAT, and WindSat (see
Figure 2). The authors also compared the results with mean OSW speeds and wind power densities collected by buoys during 1995–2015. Based on the results, although OSW information retrieved from each of these three scatterometers was fairly comparable, a combination of these instruments provided better results. Furthermore, other types of sensors, including terrestrial laser scanners, SONAR, and Autonomous Underwater Vehicles (AUV), have been utilized to collect data about the position, geometry, and morphology of icebergs [
26,
27]. Although the abovementioned approaches provide accurate information, they are resource-intensive and logistically arduous in oceans, especially in remote locations of polar regions [
28]. Consequently, it is efficient to employ other RS systems, such as satellites, which can provide broad observations about the icebergs through space and time. Various RS systems have been so far applied to identify and track icebergs [
10,
29,
31]. Optical, SAR, scatterometer, altimeter, and HF radar systems have been widely used for iceberg studies.
3.1.5. HF Radar
As discussed, HF radars measure ocean surface parameters, such as OSW speed, in a timely fashion over relatively larger areas and at a reasonable cost compared to in situ measurements. OSW speed can be extracted from either the first-order radar backscatter or the second-order sea echo. Stewart and Barnum [
105] revealed that the 10-dB width of the first-order resonant scattering could be used for the estimation of the local OSW speed. However, this method was not a robust solution since it depended on different ocean parameters, such as OSC [
106]. The second-order scattering is also employed to calculate the OSW speed because the amplitude of the second-order radar backscatter and its location in Range-Doppler is greatly dependent on the local OSW speed [
107]. In other words, by increasing the wind speed, the second-order peaks increase and move to the first-order peaks. Moreover, Barrick et al. [
108] postulated that the ratio of the second-order peak amplitudes to the first-order ones could be used for measuring the local OSW. Ahearn et al. [
109] also suggested that the ratio of the second-order continuum closer to zero Doppler to the amplitude of the first-order resonant peak can be employed for OSW speed measurement because short waves represented in this region reflect the OSW speed. Additionally, Dexter and Theodoridis [
110] proposed a method to calculate OSW speed from the significant OWH and period. Subsequently, Huang et al. [
111] applied this method to an HF radar installed in the Eastern China Sea.
3.1.6. Summary and Future Direction
OSW is among the key components of ocean/atmosphere circulation. Accordingly, accurate knowledge of OSW is of substantial importance to understand and quantify different ocean-related characteristics. Despite the tremendous progress in the utilization of RS technology for quantitative and qualitative assessment of OSW, there are still different issues and challenges that need to be addressed in future research. These include but are not limited to improving the measurement accuracy and consistency of OSW products by integrating different data sources (e.g., buoys, satellite, and NWP), proposing novel automatic and/or semiautomatic methods for concurrently retrieving OSW and other related parameters (e.g., ocean currents and precipitation), developing new RS sensors to enhance the temporal resolution and spatial extent of global OSW products, integrating active and passive sensors to provide OSW products in all-weather conditions with the highest accuracy, and employing downscaling methodologies to enhance the temporal and spatial resolutions of OSW products.
3.2. Ocean Surface Current (OSC)
OSC is the continuous and directional movements of the mass of the seawater, transferring nutrients, energy, heat, pollutants, and chemical substances around the world [
112,
113]. Ocean currents affect the global climate, the ocean’s ecosystems, and fishing productivity. More importantly, they play a key role in reducing shipping costs, fuel consumption, as well as developing policies for preventing natural disasters [
112,
114]. OSC can originate from a wide range of factors, such as wind, Coriolis effect, water density variation, Ocean Tide (OT), as well as SST and OS differences [
112,
113,
115,
116]. The seafloor and shoreline topography can also affect OSC and hinder or boost the mixing and passageway of water from different areas [
117]. Ocean currents can be generally divided into five categories: (1) geostrophic ocean current, which is balanced under pressure gradient force by the Coriolis effect; (2) tidal ocean current, which is created by the gravitational force of the moon, sun, and Earth; (3) wind-driven Ekman ocean current, which is created by the steady ocean wind; (4) wave-induced Stokes drift, which is characterized by the difference between the average Lagrangian flow velocity of a fluid parcel and the average Eulerian flow velocity of the fluid at a fixed position; and (5) small-scale ocean current, which is created by the small features such as eddies, fronts, and filaments [
118]. Ocean currents are also separated into two groups based on their temperature: warm and cold ocean currents (see
Figure 3) [
3]. For instance, the Gulf stream, Kuroshio, and the Agulhas are warm currents that transport heat from the tropics poleward and significantly affect the global climate [
112,
119,
120]. The Humboldt, Benguela, and California are cold currents that preserve highly upwelling waters and carry cold water toward the equator [
112,
120]. For example, the Labrador Current has a cooling effect with a low OS and is known for transporting icebergs from Greenland’s glaciers into shipping lanes in the North Atlantic [
120,
121].
Ocean currents can also be categorized into two groups according to their depth: surface and deep (subsurface) [
112,
118,
122]. The surface currents are horizontal water streams that occur on local to global scales, and their effects are primarily restricted to the top 400 m of ocean water [
112,
123]. Along the coasts and offshore regions, there are local surface currents, which are typically small and short-lived (e.g., hourly/seasonal), generated by OT, waves, buoyant river plumes, and local-scale winds [
112,
123]. These currents control the local flooding, algal bloom, marine pollution, sediment transport, and ship navigation [
112,
123]. The global surface currents (e.g., the Gulf Stream) are typically controlled by dominant global winds (e.g., trade winds and the westerlies) together with Coriolis force and the restriction of flow by continental deflections [
112,
114,
123]. These currents travel over long distances in the same direction as the wind and at a speed of approximately 3 to 4% of winds’ speed [
112,
123]. However, the Coriolis force deflects these currents from the equator to the right direction in the Northern Hemisphere and the left in the Southern Hemisphere, which creates the clockwise and counterclockwise circular patterns or gyres, respectively [
112,
114,
115,
123,
124]. In contrast, the deep ocean currents are vertical streams under the influence of the thermohaline circulation generated by water density differences and depend on temperature and OS [
112,
125]. Deep ocean currents are formed with upwelling and downwelling directions below 400 m of the surface water [
126].
Depending on the scale of the ocean currents, they are measured by different methods.
Figure 4 illustrates various in situ and RS methods for ocean current estimation. It should be noted that the focus of this section is on the OSC using offshore, shipborne, and spaceborne platforms.
Table 2 also summarizes these systems along with their advantages and limitations for OSC studies. More details about the applications of each system are also provided in the following subsections.
3.2.1. Optical
Optical satellites provide high/medium spatial resolution and multitemporal and multispectral images for oceanographic studies over large areas [
112,
127]. These images are typically captured in the visible (Red/Green/Blue) (0.4–0.7 µm) and Near-, Mid-, and Far-Infrared (0.7–3.5 µm) spectral ranges, and can be directly/indirectly used to investigate the spatiotemporal characteristics of OSC [
112,
127]. For example, OC products of optical satellites can be indirectly applied to estimate the OSC [
112,
116]. For instance, Sun et al. [
128] employed two matching algorithms, including area-based cross-correlation and feature-based optical flow approaches, to estimate OSC from OC image pairs. The results demonstrated a high similarity between the derived OSC and the Ocean Surface Current Analysis Real-time (OSCAR) products. It was also observed that the robust optical flow method outperformed the Maximum Cross-Correlation (MCC) algorithm in terms of accuracy. Furthermore, optical RS images can be directly used to retrieve surface wave directional properties and OSC characteristics. For instance, Yurovskaya et al. [
127] developed a new technique based on cross-spectral phase estimation to directly retrieve OSC and the characteristics of propagating ocean surface waves from Sentinel-2 images. The results showed that the retrieved OSC was well-matched with medium-resolution OSC models as well as the derived velocities from altimeter observations in deep ocean regions. The results over shallow water areas demonstrated that the retrieved wave propagation characteristics corresponded well with the sea-depth variations. These results confirmed the high potential of optical images in direct estimation of OSC velocity.
Although the optical satellite images are rich in content and are easy to use for OSC mapping, they are sensitive to cloud cover and are not suitable for nighttime. Moreover, the long revisit time and inappropriate latency of some optical datasets have made some of them unsuitable for real-time OSC measurements.
3.2.2. TIR Radiometer
In order to retrieve SST, Low Earth Orbit (LEO) and geostationary satellites acquire TIR data in the 7–14 µm region of the spectrum [
129]. The derivations of SST from TIR satellites are typically utilized to estimate the OSC velocity by tracking the motion of natural surface features (e.g., oil and algae) in cloud-free images [
112,
129]. This tracking scheme is mainly carried out by analyzing the sequential TIR images in both human interactive and automatic methods [
129]. In this regard, automated feature-tracking methods are more interested in retrieving OSC velocity due to the costly and time-consuming procedure of the human interactive approaches [
130]. To this end, the feature tracking methods based on the MCC matching algorithm have provided promising results for NRT estimation of OSC velocity fields from consecutive TIR images [
112,
129,
130,
131]. For example, Heuzé et al. [
130] used the MCC method to track features from 224 pairs of sequential TIR images acquired by Advanced Very High-Resolution Radiometer (AVHRR) between January and December 2015 in the western Mediterranean Sea. The results indicated that the satellite-based OSC resulted from tracking a small pattern with low speed and concurred well with in situ measurements, especially in summer times. Like optical satellite images, TIR radiometers are restricted by the cloud cover and undesirable viewing conditions, which reduce the spatial and temporal coverage of the resulting OSC velocity products [
130].
3.2.3. Microwave Radiometer
SST and OSW measurements derived from microwave radiometers can be used to study OSC [
112,
132]. Compared to TIR radiometers, the coarse spatial resolution of microwave radiometers reduces the performance of sequence image analysis and MCC techniques for OSC estimation [
133]. To resolve this issue, the Surface Quasi-Geostrophic (SQG) theory and multifractal analysis have been employed to retrieve OSC from a single SST microwave image [
133,
134]. To this end, the OSC can be reconstructed by converting the SST map to OWH using a transfer function obtained by the multifractal analysis or SQG approach [
135]. In the multifractal analysis, the fact that SST multifractal structures are closely related to OSC is considered, and reasonable stream function approximation is allowed [
133,
134]. However, this technique cannot provide information on OSC orientation. In contrast, the SQG approach can reconstruct the 3D OSC field and is more commonly used by researchers [
134]. For example, González-Haro and Isern-Fontanet [
132] assessed the OSC, reconstructed from SST, using the SQG method at the global scale. The results indicated that the retrieved OSC could be improved by providing information on the energy spectrum of altimetry data.
3.2.4. SAR
SAR images are being widely used to map OSC dynamics and their variability in any weather conditions and at any time of day/night [
116,
122,
136]. SAR-based OSC estimations are mainly obtained using two approaches: (1) feature tracking and (2) Doppler-based techniques [
136]. The feature tracking methods compare sequential SAR images by measuring a similarity measure (e.g., correlation) between those images [
137]. In the best case, the outputs of these methods have the same spatial resolution as the SAR images. On the other hand, the Doppler-based techniques can be divided into two groups based on their design, implementation, and performance: (1) the Doppler Centroid Anomaly (DCA) and (2) along-track InSAR [
136]. The DCA methods model OSC at a coarse resolution (e.g., 1–2 km) with only one SAR image and are based on the Doppler shift concept [
136,
138]. These methods estimate OSC by determining the line-of-sight rate of the scatterers based on the Doppler shift [
112,
139]. On the other hand, along-track InSAR techniques are widely utilized to generate high-resolution OSC maps with two SAR images [
136]. The spatiotemporal decorrelation and atmospheric delays strongly reduce the performance of these methods. Therefore, it is necessary to select multitemporal SAR images with appropriate temporal and spatial baselines to achieve better results from these methods. Multiple studies have investigated the potential of SAR-based methods for OSC retrieval. For instance, Fu et al. [
140] proposed a new technique to provide maps of shallow water topography and OSC using SAR images. In this method, OSW and OSC were estimated by fitting the observed SAR signals to a simulated shallow water topography radar signal in an iterative procedure. The results showed the high efficacy of the proposed methods. Moreover, Elyouncha et al. [
141] investigated the potential of Sentinel-1 SAR data for OSC mapping in the Skagerrak Sea with a focus on the Norwegian Coastal Current. They implemented several postprocessing algorithms for scalloping removal and the absolute and interbeam bias correction in the Sentinel-1 data. The results showed that the retrieved OSC velocity corresponded well with a regional ocean circulation model with values of ≈0.8 m/s.
3.2.5. Altimeter
Altimeters with Ku-band (13.8 GHz) also provide information about the large-scale dynamic topography of the ocean, particularly OWH and OSW speed, which can be directly used to estimate the global OSC and its variability at coarse spatial resolutions (e.g., 10 km) [
112,
135,
142]. Moreover, integrating altimetry data with tide-gauge data and hydrodynamic models increases the knowledge about global OT and OSC variability [
136,
142]. However, altimeters cannot measure the non-geostrophic components of the OSC due to non-geostrophic and local winds [
136]. Several efforts have been made so far to derive OSC from altimetry data. For example, the NRT Archiving, Validation, and Interpretation of Satellite Oceanographic (AVISO) products, as geostrophic OSCs, have been generated by integrating the satellite altimetry and precise satellite position data and are available at a global scale (see Hwang and Fan [
90] for more information). Moreover, the OSCAR project, developed by the National Oceanic and Atmospheric Administration (NOAA) [
112,
113,
114], and the GlobCurrent project, developed by European Space Agency (ESA) [
90], provide the complementary information for AVISO products in reliable estimating of the Ekman component of the OSCs [
118].
3.2.6. HF Radar
Over the last few decades, HF radar systems have been successfully employed to remotely measure coastal and offshore currents using the concept of Bragg-scattering [
112,
122,
143]. The local currents can be typically measured by a land-based radar antenna pair over a wide area, up to 200 km or more, and with coarse spatial resolutions (e.g., 0.5–6 km), according to the selected HF radio band, which could be in the range of 3 to 45 MHz [
122,
136,
143]. Consequently, HF radar systems, as a cost-efficient tool, cover a wider range than marine radars (i.e., X-/S-band radars) [
112]. Moreover, HF radar systems are typically preferred over marine radars to support the worldwide monitoring of marine and coastal ecosystems, although they cannot produce high-quality data such as X-band image sequences [
79]. For example, there is a global HF radar network operating throughout the world to produce continuous maps of local OSCs (see
Figure 5) [
90]. This network provides NRT measurements of OSC fields which are applicable for operational marine services, coastal marine studies, OOS detection, water quality assessment, pollution tracking, search and rescue, and numerical ocean forecasting models, especially near the coast [
79,
144]. Various studies have so far investigated HF radar data for OSC estimation. For example, Ji et al. [
145] investigated the capability of the HF Hybrid Sky–Surface Wave Radar (HFHSSWR) for OSC estimation. The results were compared with in situ measurements from a current meter, and it was observed that there was a good agreement between HFHSSWR results and in situ data, with a correlation coefficient of 0.81 and an RMSE of 0.138 m/s.
The high sensitivity of HF radar signals to sea surface topography and electromagnetic interference from different sources, as well as data acquisition failures, causes frequent gaps in HF radar data, which is the main limitation in utilizing these datasets for continuous OSC retrieval [
146]. To resolve this, Kolukula et al. [
146] presented a gap-filling algorithm for HF radar data using the complex empirical orthogonal functions and validated it using two datasets with different gap rates. The results showed that the method had high efficacy in reconstructing data. The direction pattern and magnitude of the reconstructed data were well-matched with the existing data.
3.2.7. Marine Radar
Over the last four decades, marine radars, deployed on ships or land, have been widely employed for OSC [
112,
147]. In most studies, the X-band radars were preferred to S-band systems due to their clear sea-clutter images [
148]. However, X-band radar data are more sensitive to interference from raindrops; thus, this radar frequency is mostly used by weather surveillance radar for rainfall monitoring [
148,
149]. Marine radars scan the sea surface at close ranges (i.e., limited by line-of-sight propagation to the horizon) with relatively high spatial (~9 m) and temporal resolutions [
73,
147,
148] to derive OSC. The Wave and Current Analysis and Wave Spectra (WaMoS) II, developed by the OceanWaveS GmbH company, is an example of an operational system that works based on X-band [
73,
150]. The WaMoS II system measures and displays wave (e.g., peak wavelength and direction) and OSC parameters (e.g., OSC speed and direction) [
73]. These measurements are mainly obtained by taking a sequence of X-band radar images and analyzing the subsequent waves [
73,
150].
Figure 6 demonstrates a scheme of a WaMoS II X-band radar installation and its images, along with bathymetry and OSC field.
Hessner et al. [
73] investigated ocean wave variations and OSC characteristics by the WaMoS II High Resolution Current (HRC) at Tongue Point, New Zealand, during August 2011, using three scenarios based on high, low, and slack tides. The results showed that the WaMoS II HRC data had high efficiency in detecting small-scale OSC features and could demonstrate the influences of the OT and local bathymetry interaction on the diversity of current patterns. Furthermore, Chen et al. [
72] improved the cross-spectral approach to estimate the velocity of OSC from the X-band image sequences. The results indicated that the estimated current velocity had a more acceptable accuracy compared to the current velocities measured by a current meter, with a correlation coefficient of 0.88 and a relative error of 7.79%. However, conditions such as the sampling period of one to three seconds, and radar image acquisition within 10 to 40 s, were necessary to obtain reasonable results.
3.2.8. Summary and Future Direction
In summary, HF radar has widely been applied to estimate coastal and offshore OSC, while altimeters have mostly been used to measure global OSC. However, the range of the HF radars was limited to between 200 km to 600 km, which should be further improved to obtain more accurate information from coastal OSC. Moreover, HF radar and altimeters observations have data gaps (no data) due to the presence of obstacles, such as artificial features (especially for HF radar) and atmospheric conditions and clouds (especially for radiometers). Integrating these datasets with in situ data and OSC derived from TIR/Microwave radiometers and SAR systems could help reduce the gaps and improve the accuracy of the OSC map.
Although the airborne RS systems provide high spatial resolution data for ocean applications compared to spaceborne systems, they have been rarely used for OSC estimation due to their high cost and lower coverage over ocean. The gap between spaceborne and airborne RS data has been recently filled by data derived from Unmanned Aerial Vehicles (UAVs). UAVs can provide affordable observations with very high spatial and temporal resolutions for OSC estimation and monitoring. Moreover, various sensors (e.g., optical, SAR, and LiDAR) can be mounted on a UAV to collect data over the ocean. Such multisource observations can drive new developments in the OSC estimation. RS CubeSats (i.e., mini satellite systems from 1 to 100 kg) have also created a paradigm shift in ocean applications by providing data with high temporal and spatial resolutions. These observations can efficiently be utilized to monitor a wide variety of ocean parameters describing the OSC.
3.3. Ocean Wave Height (OWH)
Wind blowing over the ocean surface creates ocean waves with a different range of heights depending on the wind speed, wind duration, and distance. The resulting waves can travel for hundreds or even thousands of kilometers and form swell waves of various heights. Although waves are generally caused by wind, catastrophic waves (e.g., landslide surges, tsunamis, and storm surges) [
151] and internal waves (e.g., subsurface waves at the boundary between two water layers) [
152] can also generate ocean waves. OWH information is a critical parameter for coastal construction, ship navigation, and human activities in the oceans [
153].
The datasets collected by different RS systems, such as GNSS-R, SAR, altimeter, and marine radar, have been utilized for OWH estimation. In this regard, various models have been applied to retrieve OWH from these datasets [
154]. For example, due to the complexity of physical models (e.g., RT models), empirical and semiempirical models have also been developed to estimate OWH [
155]. The simplicity of empirical and semiempirical models has also led to the development of ML algorithms [
156]. In this regard, DL algorithms, as the most advanced ML models, have received more attention due to their promising performance. For example, Shao et al. [
157] proposed a hybrid statistical and a DL model in South China to predict several ocean surface variables, including OWH. In Liu et al. [
158], a short-term memory deep network was also proposed to consider the time domain data in OWH estimation.
Table 3 summarizes the advantages and disadvantages of each of these RS systems. In the following subsections, the studies that have been conducted to measure OWH based on various RS systems are discussed in more detail.
3.3.1. GNSS-R
Various GNSS-R techniques using either single or double antennas, as well as direct or reflected signals, have been proposed for OWH estimation. The signal arrival time was also used for OWH estimation in Rius et al. [
159]. The recorded SNR ratio was first used for OWH measurement by Chen et al. [
160] in 1995. Later, Larson et al. [
161] and Santamaría-Gómez et al. [
162] successfully showed the robustness of the proposed method in [
160] for OWH estimation. Moreover, Penna et al. [
163] used a GNSS Wave Glider (GNSS WG) through a 13-day experiment to measure OWH in the North Sea. Their results showed that GNSS WG could address the challenges of coastline-based tide gauges for OWH measurement, and the temporal and spatial resolution limitations of radar data.
3.3.2. SAR
SAR is a unique system for OWH observation from space because it has a high spatial resolution (e.g., 1–10 m), broad coverage, and it is independent of cloud cover and light conditions [
164]. SAR uses the omitted backscattering signal from the ocean surface to measure OWH [
165]. Theoretical and empirical algorithms are two main techniques that have been employed for OWH estimation using SAR images [
166]. Theoretical algorithms, standing on the SAR wave mapping mechanism (i.e., nonlinear velocity bunching, tilt modulation, and hydrodynamic modulation), invert the SAR intensity spectrum into a wave spectrum [
167]. The semiparametric retrieval [
168], parameterized first-guess spectrum [
169], and partition rescaling and shift [
170] are three well-known theoretical-based algorithms for OWH estimation using SAR data. On the other hand, empirical algorithms, which do not need prior information of wind and wave, can directly estimate OWH from inputs generated from SAR images [
171]. C-band WAVE models for European Remote Sensing (ERS)-2 [
172], Envisat [
173], and Sentinel-1 [
174], as well as the X-band WAVE model [
175] for TerraSAR-X, are some examples of the empirical algorithms.
Among SAR systems, Sentinel-1, which provides open-access imagery with dual-polarization, has been widely used for OWH estimation. For instance, Shao et al. [
176] used a semiempirical algorithm along with Sentinel-1 imagery (C-band VV-polarization) for OWH retrieval. The results illustrated that the algorithm could successfully estimate OWH with an RMSE of 18.6.
3.3.3. Altimeter
Generally, altimeters measure the traveling time (
T) of the transmitted radar pulses at regular intervals defined by the Pulse Repetition Frequency (PRF) from a nadir-pointed antenna to the ocean surface and back to the receiver onboard the satellite, according to the following equation:
where
c is the speed of light and
R is defined as the range from the sensor to the ocean surface. R can be calculated using Equation (2).
R is an uncorrelated range and should be modified based on Equation (3) to take into account various ranges of atmospheric and geophysical corrections, including wet and dry tropospheric correction, ionospheric correction, and solid and pole earth tide correction [
177].
where
is related to the aforementioned range corrections. The data collected by Jason-3, Chinese HY-2A/B, Satellite with ARgos and ALtiKa (SARAL), and Sentinel-3 altimeters have been widely used in OWH estimation. These satellites offer global observations under all weather conditions [
178]. For example, Wang et al. [
178] estimated OWH based on observations from HY2B. They used the data collected by the National Data Buoy Center (NDBC) from April 2019 to April 2020 for validation purposes. In this study, numerous DL algorithms were utilized to approximate the waveform of HY2B with high accuracy. Moreover, Peng and Deng [
179] enhanced the Brown model to estimate OWH based on three years of observations from the Jason-1 data. The results were validated against observations from eight buoys and compared with the retrieved OWH by the four-parameter Maximum Likelihood Estimator (MLE) retracing method. Moreover, SARAL (i.e., the first Ka-band altimeter) observations were used to retrieve the OWH in the coastal ocean and inland water bodies. Validation of the results with the in situ measurements demonstrated a significant correlation (0.98) between field measurements and satellite measurements in the coastal ocean; the performance was remarkably stable across various coastal zones.
3.3.4. HF Radar
HF radar is another RS system to measure OWH. This parameter can be calculated based on the first- and second-order sea surface scattering mechanism. Generally, OWH depends on the integral of the second-order part of the radar-measured Doppler spectra [
180]. Beamforming is required to obtain the OWH over various ranges and azimuths. Thus, a narrow-beam system along with a large aperture is essential for retrieving a wave map using HF radars [
181]. In this regard, Tian et al. [
182] proposed a method to calculate OWH based on the second-order harmonic peaks of radar Doppler spectra and its ratio to the Bragg peak power. The comparison of the results with buoy data over a 30-day experiment depicted that the RMSE was between 0.33 and 0.77 m. It is worth mentioning that the second-order scattering is vulnerable to the external clutters and noise which may reduce the robustness of the OWH measurement [
183,
184,
185]. On the other hand, the first-order radar backscatter has stronger power compared to the second-order harmonic peaks. In this regard, Zhou and Wen [
186] proposed an empirical method to calculate OWH from the power of the first-order peaks. Subsequently, Tian et al. [
187] proposed a technique using the power ratio of the first-order peaks acquired at two radar frequencies to estimate OWH. By utilizing the first-order peaks instead of the second-order peaks to calculate the OWH, the detection range improves because of the high SNR value. Nevertheless, this method can be affected by low sea state conditions due to the saturation of the Bragg waves [
182].
3.3.5. Marine Radar
Since the 1980s, marine radars have been mainly developed to measure ocean surface parameters such as OWH. This is because of their high temporal and spatial resolutions as well as the distinguishability of ocean waves in the corresponding images [
188]. Several texture-analysis-based, spectral-analysis-based, and coherent radar algorithms have been so far developed for OWH estimation using marine radar data.
Spectral-analysis-based algorithms acquire wave spectra information from a series of radar images to generate OWH information. These methods can be generally divided into three groups, namely, the 3D discrete Fourier transform-based algorithms [
189,
190], 2D continuous wavelet transform-based algorithms [
191], and the array beamforming algorithms [
192]. Regarding texture-analysis-based algorithms, wave parameters are directly generated from image texture information based on seven different methods, including the probability of illumination [
193], statistical analysis [
194], tilt-based [
195,
196], shadowing-based [
197], support vector regression [
198], empirical orthogonal function [
199], and ensemble empirical mode decomposition [
200]. Finally, coherent radar algorithms have been developed based on the X-based coherent marine radars [
201]. The image intensity and the radial velocity of the ocean surface scatters are two main advantages of coherent marine radars. These systems enable coherent radar-based algorithms to directly estimate OWH information from the radar data without calibration [
188]. Based on this feature, several empirical [
202], semiempirical [
203], and nonempirical [
204] methods have been developed for OWH estimation.
3.3.6. Summary and Future Direction
Most of the OWH studies using RS datasets have focused on data or algorithm-driven solutions, but the quality assessment of airborne/spaceborne-derived OWH under a variety of conditions has been investigated less. Moreover, application of advanced ML algorithms in analyzing, interpreting, and modeling RS data has received less attention from researchers, even though they can offer a great potential for accurate OWH mapping and monitoring tasks. For example, DL algorithms have rarely been used to model and estimate OWH from RS data, while they have significantly outperformed traditional methods by relying on their ability to train models with massive datasets. Indeed, DL methods can significantly facilitate understanding the complex structure of large RS data for OWH estimation/prediction.
3.4. Sea Level (SL)
SL is an important oceanographic variable that should be measured precisely for long-term trend assessments and climate studies [
205]. SL has a pivotal role in studies related to OSC, mesoscale eddies, and marine gravity fields [
206,
207,
208]. In recent decades, anthropogenic activities and global warming have mainly resulted in SL change. For instance, the Intergovernmental Panel for Climate Change (IPCC) reported a Global Mean SL (GMSL) rise of 3.6 mm/yr between 2006 and 2015 [
209,
210]. However, the relative rate of SL change is not globally identical because it depends on different spatial and temporal parameters (see
Figure 7).
Conventionally, SL estimation was based on coastal monitoring stations, tide gauges, buoys, and ship surveys [
212]. However, the high cost and sparse observations of these approaches make them inappropriate for SL measurements in most cases. Moreover, in situ measurements contain significant interannual and decadal effects and do not perfectly manifest the SL change [
213]. However, with the advancement of RS technology, satellites provide valuable datasets to study SL at different local to global scales. In addition to studies that only focused on SL measurements using RS systems, many studies related SL observations to different environmental variables. Through these analyses, it was widely argued that the main contributors to SL rise are thermal expansion of seawater [
214], Antarctic and Greenland ice sheet melting [
215], and land-water storage change due to the groundwater depletion [
216]. Consequently, SL rise has many environmental and economic impacts, including reef island destabilization [
217], wave resource alteration [
218], coastal erosion [
219], saltwater intrusion into aquifers [
220], sea turtle nesting threatening [
221], lowland and delta vulnerability [
222], coastal flooding [
223], seaport infrastructure susceptibility [
224], wetland inundation and displacement [
225], island and offshore baseline loss [
226], tidal dynamics [
227], and length-of-day changes [
228].
Among different RS systems, GNSS-R, altimeters, and gravimeters have been widely employed for SL studies [
229,
230,
231,
232,
233]. The advantages and disadvantages of these systems applied for SL mapping are provided in
Table 4. The following subsections discuss the applications of each system.
3.4.1. GNSS-R
GNSS-R systems use direct or reflected GNSS signals to measure the SL. To this end, different techniques, including floating carpet boats [
234], unmanned surface vehicles [
234], and spaceborne/airborne GNSS-R [
235], can be employed. The first two approaches carry the GPS antenna to directly compute the elevation value of the corresponding instrument, which is then converted to SL. These approaches are only suitable for relatively small regions. On the other hand, GNSS-R employs reflections of signals transmitted from navigation satellites to infer several geophysical parameters [
236,
237]. GNSS-R can provide SL data at a global coverage with high spatiotemporal resolutions [
212,
229]. The concept of GNSS-R relies on measuring the delay between direct and reflected signals reaching the receiver (e.g., aircraft, satellite) above the ocean. In particular, the computed delay waveforms (from delay Doppler maps) enable the estimation of the SL through geometric models [
238]. It was reported that the nominal precision of SL measurement from an individual 1-s GNSS-R was 5 m [
239]. Further enhancements, such as employing more observations from multiple GNSS-R data over more time ranges, could improve the precision to about tens of centimeters [
240]. For example, Qiu and Jin [
212] employed the data acquired by the eight satellites of the CYGNSS mission to estimate the GMSL. The calculated GMSL was validated by satellite altimetry and the DTU-10 sea surface model. The results showed a significant correlation of 0.97 with both datasets, demonstrating the high potential of the CYGNSS for global SL measurements. Likewise, Wang et al. [
241] implemented different ML algorithms to retrieve SL from GNSS-R data. They reported that the ensemble of ML algorithms, along with three input features (i.e., the 70% peak correlation power, peak first derivative, and leading-edge slope) from airborne delay waveform dataset, led to the best results, with an RMSE of 0.23 m concerning the DTU15 model.
3.4.2. Altimeter
The first generation of altimeters measured SL at the nadir angle along the ground track with low spatial resolution. Consequently, it was required to merge obtained data of multiple acquisitions to provide full coverage over relatively large study areas. Therefore, they had limitations in estimating SL at the mesoscale and short time intervals [
229,
237]. Moreover, the low spatial resolution of traditional altimeters and the waveform tracking effect at the sea–land boundary made these systems less applicable to near-shore regions [
48]. To overcome these shortcomings, new principles were introduced to obtain high-resolution and wide-swath altimetry data. These new principles resulted in the invention of SAR and InSAR altimeters [
49,
50].
Generally, preprocessing (e.g., orbital correction and outlier observation removal) and geophysical correction (e.g., dry/wet tropospheric correction, sea state bias correction, dynamic atmospheric correction, ionospheric correction, and tidal effect correction [
242,
243,
244,
245,
246]) steps are required to prepare altimetry data for further SL studies. For example, Ren et al. [
247] employed the Chinese Tiangong-2 Interferometric Imaging Radar Altimeter (InIRA) data to measure SL. The results were compared with one-dimensional SL data from previous altimeters of Jason-2, SARAL, and Jason-3. Then, the observed systematic and parametric biases were analyzed and removed to obtain a standard deviation of 8.1 cm in OWH estimation, indicating the promising results of the InIRA SL measurements. Additionally, Dinardo et al. [
248] investigated the potential of the CrySat-2 SAR altimetry for SL retrieval along the coasts of the German Bight and West Baltic Sea. The computed SL values were compared with in situ and regional models between 2010 and 2016. Their results showed an acceptable consistency with the regional ocean model, with a standard deviation of the differences of 24 cm. Finally, Mullick et al. [
249] investigated the coastal vulnerability of Bangladesh using the Composite Vulnerability Index (CVI). To this end, satellite altimetry data from TOPEX/Poseidon (1992–2006) and Jason 1–3 (2006–2017) were combined with other parameters (e.g., land use, mean tidal range) to map the CVI along Bangladesh’s coast. It was observed that 16% of the total coastline (87 km) was highly vulnerable. Finally, Yang, et al. [
250] developed a fusion approach based on deep belief network to integrate satellite altimetry and tide gauge data. The results revealed that the proposed method performed well when limited along-track altimetry and gauge data are available. Furthermore, it was observed that the distribution of altimetry and gauge datasets had a lower impact when the deep belief network was implemented.
3.4.3. Gravimeter
Ocean mass and steric changes cause variation in SL [
251]. The ocean mass is associated with exchanges of ice and water mass with oceans, affecting the regional and global gravity. Therefore, it is possible to measure SL changes through gravimeters [
252]. For example, the GRACE mission observes the Earth’s gravity, enabling the measurement of GMSL change associated with ocean mass alterations [
253]. Although GRACE provides global coverage of the Earth’s gravity, it is difficult to employ it for regional studies due to the existing uncertainties in low spatial resolution gravity data [
230]. For instance, Jeon et al. [
230] applied RL05 GRACE monthly data provided by the Center of Space Research (CSR) and the GeoForschungsZentrum (GFZ) in Potsdam to quantify the global SL change. The atmospheric and ocean de-aliasing model was also used to remove the contributions of atmospheric surface and ocean bottom pressures. Finally, the gravity observations revealed a GMSL rise of about 2.14 ± 0.12 mm/yr. Additionally, Elsaka et al. [
254] employed satellite gravimetry and altimetry data to evaluate the Nile Delta–Mediterranean Sea interactions. Gravity observations, acquired from GRACE, revealed an erosion rate over the eastern and western parts.
3.4.4. Summary and Future Direction
Despite the advantages of RS techniques over traditional SL mapping approaches, RS systems generally suffer from coarse spatial resolution over ocean environments. Data acquisition over an extended period is essential to obtain reliable results and to obtain comprehensive findings of SL changes. In this regard, improving the knowledge of SL either by developing more advanced RS systems with better spatial and temporal resolutions or by synergy of observations (i.e., different RS systems), consistent data acquisition, and developing advanced methodologies based on ML algorithms can assist in resolving some of the limitations of RS systems for SL studies [
232,
233]. In particular, employing ML and DL algorithms for accurate SL estimation should be considered in future studies due to their high potential for providing accurate SL data [
158]. Furthermore, integrating SL observations acquired by different sources is highly required in order to improve our understanding of the global-to-local SL dynamics and resolve the limitations of single-source observations. Finally, the continuation of using RS observations to measure SL by developing and constructing new sensors with improved capabilities is an essential prospect [
229,
230,
231].
3.5. Ocean Tide (OT)
OT refers to the regular rise and fall of the ocean water caused by the gravitational pull of the moon and sun in relationship with the geometric location of the Earth’s surface [
255]. The cyclical effects of the Earth’s and the moon’s rotations are, respectively, the primary factors of the periodic rhythm and height of OT [
256], and 24 h and 50 min is the tidal period [
256]. When the water wave slowly rises to its crest (highest level), covering much of the shore, high tide occurs. Once the water wave falls to its trough (the lowest part of the wave), it is known as low tide [
255,
256]. Thus, the tidal range is the vertical difference between the high and low OT [
256]. Ebb tide is called the flow of water from high tide to low tide [
255]. During the moon’s revolution around the Earth, the direction of its gravitational attraction is aligned with that of the Sun. High spring OTs are created when the moon, Earth, and sun are in alignment. This alignment occurs every 14–15 days during full and new moons [
256]. On the other hand, during the first and last quarters of the moon, neap OTs happen when the moon appears half-full [
255,
256]. Intertidal zones are the lands in the tidal range categorized in the splash, high-tide, mid-tide, and low-tide zones [
257,
258].
While traditionally, in situ measurements and numerical models have been used for OT studies, RS has been proposed to fill OT measurement gaps over the past four decades. RS technology has expanded our understanding of global OT [
259,
260,
261] and facilitated continuous OT monitoring and predictions over wide-spread scales. RS systems can be used to study several aspects of OT, including tidal flats, tidal channels, tidal currents, Ocean Tidal Load (OTL), and tidal wetlands. In the following, a brief description of each type of OT is provided.
The area inundated between low- and high-tide waterlines is defined as tidal flats and is a mixture of seawater and freshwater environments [
262]. Tidal flats provide essential services, such as coastal storm protection, food production, and shoreline stabilization [
262]. A recent study has discovered that tidal flats occupy approximately 130,000 km
2 of the planet (
Figure 8) [
263]. Murray et al. [
263] also reported that about 70% of the global tidal flats occurred in three continents, namely, Asia (44% of total), North America (15.5% of total), and South America (11% of total), 49.2% of which were concentrated in eight countries, namely, Indonesia, China, Australia, the United States, Canada, India, Brazil, and Myanmar. Monitoring tidal flats using field observations is limited to estimating the ebb/flood characteristics, adequate surveys for large tidal flats, and the field access. However, RS in combination with in situ measurements facilitates monitoring tidal flats in a more cost- and time-efficient approach. In fact, RS data with a high temporal resolution are necessary for tidal flats studies because there are coastal areas that fall dry during each tidal cycle [
264], and tidal flats are only exposed fully for a short period at low tides.
A tidal channel is a type of stream or a waterway that occurs during the ebb tide and flood tide in the tidal flats [
265]. Tidal channel networks are crucial aspects of the neighboring ocean and estuaries. In addition to the control of the tidal basin hydrodynamics, tidal channels connect intertidal flats to the salt marshes, which play an important role in tidal propagation [
266]. Due to various morphological characteristics from terrestrial river networks, conventional river system algorithms cannot be implemented on tidal channels. Thus, RS techniques have been effectively utilized to obtain the spatial distribution of tidal channels [
266].
The periodic movement of water created by the out-of-phase OT, the local weather patterns (radiational tides), and ocean characteristics (internal tides) is defined as tidal currents [
267]. Periodic tidal currents play an important role in the strait (a narrow waterway joining two larger water bodies) [
267]. Tidal currents’ power intensifies when flowing through the narrow and shallow channels in the islands or over the shallow ridges, although they are weak in most parts of the strait [
268]. Furthermore, tidal current power is renewable and predictable energy [
269], mostly occurring at narrow tidal channels. Tidal current generators not only cause fewer environmental concerns than barrages, but can also be installed incrementally to meet the increase in demand over time [
270]. Therefore, mapping the tidal energy density is crucial to consider the economic possibility, while tidal currents vary regionally. In this regard, a higher level of measurements, including spring and neap tidal cycles and higher temporal and spatial measurements, are required [
269].
OTL displacements are the elastic response of the Earth to the redistribution of water mass from OT, which can cause deformation gradients of several millimeters to centimeters near coastal regions [
271]. One of the most productive inland and coastal ecosystems is tidal wetlands. Tidal wetlands are important for trapping sediment and pollution, recreational purposes, and flood control. They are also natural barriers against saltwater intrusion into freshwater aquifers [
272,
273]. The tidal regime, relative SL rise, land cover change, and sedimentation are driving factors influencing tidal wetlands [
274]. Tides increase rates of relative SL rise, resulting in brackish water and a shift to becoming nontidal wetlands in freshwater tidal areas. So far, different RS systems have been successfully employed for OT studies.
Table 5 summarizes the advantages and limitations of the RS systems that have been frequently used for OT measurement.
3.5.1. Optical
Optical satellite images have been extensively exploited to study different aspects of OT mainly due to the availability of open-access imagery and proper temporal and spatial resolutions. For example, the Normalized Difference Water Index (NDWI), Modified Normalized Difference Water Index (MNDWI), Normalized Difference Vegetation Index (NDVI), Land Surface Water Index (LSWI), Automated Water Extraction Index (AWEI), Modified Soil-Adjusted Vegetation Index (MSAVI), and Enhanced Vegetation Index (EVI) generated from optical satellite images have been applied to delineate tidal flats by subtracting the classified images of the high and low tides [
263,
275,
276]. Waterlines, which are important for studying changes in tidal flats and coastlines [
277], can also be efficiently extracted using high spatial resolution optical satellite imagery. Appropriate spectral bands for waterline extraction must be selected based on various conditions. For instance, the NIR, SWIR, and TIR bands are efficient for flood tide [
278,
279].
Spatial analysis and edge detection methods in optical satellite images have also been applied to locate and map tidal channels. For instance, box-counting, a widely used fractal method, is a spatial analysis method used to identify linear features such as streams and coastlines. Moreover, a multilevel knowledge-based approach can extract channel edges using multiscale edge detection algorithms in optical images [
266]. The low-level processing uses edge detection algorithms (e.g., the Sobel and Prewitt operators) to discover channel edges and then associates neighboring nonparallel edges together to form channels [
266]. For example, Angeles et al. [
280] proposed a fractal analysis of tidal channels using box-counting and contiguity methods for Landsat-5 data to derive the degree of geomorphological control on a tidal channel network in the Bahia Blanca Estuary, Argentina.
Finally, optical satellite data have been widely used for tidal wetland classification. To this end, various spectral and textural indexes, along with advanced ML algorithms, have been employed [
272,
273,
281,
282,
283].
3.5.2. GNSS-R
GNSS-R provides range measurements, which can be used to remove vertical land motion from tide gauge records [
284]. However, many GNSS-R stations are not directly collocated to the tide gauge stations. GNNS time-series have also been applied in models to remove the influence of non-OTL from GPS coordinates [
285]. In this regard, the Precise Point Positioning (PPP) technique is used to process the GNSS-R data [
286]. OT can cause horizontal and vertical motions of more than a centimeter during a day, which is large enough to be measured with space geodetic techniques. In this regard, Zhou et al. [
286] investigated OTL displacement parameters with kinematic PPP using the harmonic analysis method from GNSS coordinate series considering the effect of mass loading.
Moreover, GNSS-R data, along with airborne RS techniques, have been utilized in a semiautomatic method to propose 3D distribution tidal channel maps including their shapes and locations [
265]. The tidal channels’ volumes and locations were verified by comparing GNNS positioning solutions and the coordinates extracted from UAV images. Moreover, Lee et al. [
265] generated 3D safety mapping of tidal channels located in the western coastal area in Korea, where tidal channels were considered the main cause of the summer drowning fatalities. Finally, GNNS technology has been extensively utilized to determine real-time tidal information using tidal level extraction algorithms, which can provide additional values to different aspects of OT studies.
3.5.3. SAR
SAR data have also been utilized for waterline extraction, which is an important step for studying tidal flats. This can be performed through measuring the roughness of the surface based on the differences in surface roughness between the smooth tidal flats and the water surface [
287]. Thus, an efficient edge detection method should be applied to delineate the border between open water and tidal flats [
288]. For instance, Heygster et al. [
288] obtained the border between tidal flats and adjacent water areas by applying a multiscale edge detection algorithm using ERS-2 images. To this end, they used the computationally efficient wavelet transform method in combination with a segmentation algorithm.
Additionally, SAR data are the most popular RS resources for studying tidal currents. Thus, different SAR methods, such as radar Line of Sight (LOS), Fourier transformed images, and interferometry, have been so far implemented for tidal current studies. The components and variations of tidal currents can be found in the radar LOS [
269]. Moreover, along-track InSAR has been widely applied to monitor surface tidal currents [
289]. To this end, ocean waves are detected using SAR and transfer functions applied to the image spectra. Furthermore, Doppler measurements by the along-track interferometry of two SAR images can retrieve surface currents. For this purpose, phase differences between coregistered pixels of two images are obtained using the signal phase [
269,
289]. The phase difference is transferred into LOS scatterer velocity, and then the average effect of the horizontal wave motion is subtracted from the current field [
269]. Moreover, tidal currents are mapped using equations including angular frequency, acceleration of gravity, two-dimension wave number, water depth, and the surface current [
270]. MCC is another method for tidal current mapping using SAR data. Since this method cannot address the large spatiotemporal variations in a typical diurnal or semidiurnal tidal current, temporal decorrelation is minimized by a virtual SAR constellation data pair [
290].
InSAR has also been used for OTL studies. Due to small changes caused by OTL, InSAR techniques are the most common RS methods for OTL studies. However, it is challenging to distinguish the OTL effects from similar signals in Differential InSAR (DInSAR) measurements. Since existing tools readily calculate the OTL effect, the OTL prediction model should be subtracted from an interferogram before tectonic analysis [
291]. The dominant component of OTL displacement is vertical, and the LOS look angle of most SAR satellites makes them sensitive to vertical-ground deformation [
291]. The proposed methods first accurately remove the orbit errors under the constraint of the relative displacements of the GPS PPP solutions in the LOS direction. Then, the crustal deformations caused by the OTL in DInSAR interferograms are analyzed. Finally, the OT model is applied to correct the OTL effect in the DInSAR interferograms [
292]. Many studies have investigated the applications of the DInSAR technique for OTL analysis. For example, DiCaprio and Simons [
291] investigated the capability of OTL correction in DInSAR using an empirical OT model to illustrate the superiority of an OT model in correcting the OTL effect in DInSAR results. The focus was on displacements from mass loading due to the OT to study processes occurring using ERS and ENVISAT data. Additionally, Peng et al. [
292] measured the OTL effect as a potential error source exploiting DInSAR data and tried to use empirical models to correct the error. They proposed a linear surface model to correct the relative displacements of the GPS PPP in LOS over the Los Angeles basin. Then, an OT model was applied to correct the OTL effect in the DInSAR interferograms. It was observed that the bilinear ramp function model was in agreement with the experimental results. Finally, multitemporal SAR data have been extensively used for tidal wetland mapping [
273,
293,
294,
295]. For instance, Wdowinski et al. [
293] presented the possibility of using high-resolution ALOS, RADARSAT-2, and TSX to characterize the extent of the OT flushing zone. They also provided quantitative constraints for detailed coastal wetland flow models using wetland InSAR techniques.
3.5.4. Altimeter
Satellite altimetry data have been utilized to study OT and improve the global OT models [
296,
297]. Furthermore, utilizing altimetry technology is one of the approaches used to detect shoreline changes over tidal flats [
298]. For two decades, global SL changes were extensively studied using satellite altimetry [
299]. For instance, Elachi et al. [
49] estimated water level over tidal flats by analysis of the waveforms. Tseng et al. [
299] also discussed the sufficiency of the coastal Digital Elevation Model (DEM) compared to current altimetry methods. The measurement of quantitative changes of tidal flats was performed by comparing the DEM produced using the waterline technique applied to SAR images and a DEM constructed from airborne altimetry data acquired in different years [
300]. As discussed, satellite altimetry also enables global mapping of significant OWH, the results of which can be used for OT studies. For example, Passaro et al. [
301] proposed a new algorithm for reprocessing altimetric waveforms to estimate significant OWH over a challenging tidal flat using Envisat, Jason-1, and Jason-2 data.
Satellite altimetry is highly effective in observing global tidal currents. The satellite altimetry sampling illustrates a broad view of the global distribution, variability, and spatial structures of currents. Considering the potential of tidal current energy at straits, Yu et al. [
302] proposed a method to provide a global map of significant OWH. Lee et al. [
303] also used TOPEX/Poseidon altimeter data as one of the inputs to model tidal currents. Additionally, Green and Pugh [
304] argued that the current magnitude is underestimated in altimetry-constrained OT databases.
3.5.5. LiDAR
LiDAR data, which provide topographic data over ocean and land, are considerably helpful for OT studies. Since elevated shorelines can be detected using multiple images over a range of tide and surge elevations, DEM data generated from LiDAR systems are useful. Therefore, DEM data have been locally generated and applied along with waterline methods to detect tidal flat changes [
18,
29,
305,
306,
307,
308,
309,
310]. However, it should be noted that the coastal DEMs which are globally generated (e.g., WorldDEMTM, SRTM, and AW3D30) do not include tidal flats because of the water-impermeable nature of existing RS approaches [
49].
As mentioned, the most useful RS data for tidal channels detection are DEM, which can be generated from LiDAR data. Different algorithms of fluvial channels extraction using DEM were provided by [
311]. For example, several studies generated a high-resolution LiDAR DEM using a multilevel knowledge-based approach with multiscale edge detection methods to map tidal channels [
266,
312]. LiDAR products, such as DEM and Digital Surface Model (DSM), are also beneficial for tidal wetland classification [
273,
282]. For example, Magolan and Halls [
273] investigated tidal wetland changes under increasing tidal range through time at two tidal creeks in south-eastern North Carolina, USA. Aerial photographs, LiDAR DEM and DSM data, Worldview-2 optical images, the National Agriculture Imagery Program (NAIP) orthophotos, and NOAA tide-gauge data were applied to classify tidal wetlands using an object-based classification algorithm. It was reported that LiDAR products were significantly helpful in tidal wetland mapping and change analysis.
3.5.6. Summary and Future Direction
The importance of areas affected by OT expands to food production, coastal storm protection, trapping pollution, and drowning prevention. Waterline extraction has widely been explored using optical, SAR, and LiDAR data to delineate tidal flats. These data have improved the monitoring of tidal flats and have facilitated shoreline stabilization programs compared to conventional techniques. Additionally, GNSS, UAV, and altimetry datasets help in determining spatial distribution of tidal channels where drowning fatalities happen. Moreover, tidal wetlands have been mapped using various types of RS data. SAR-based methods have also facilitated OTL and tidal currents studies. Finally, altimetry datasets have been employed for mapping the global distribution of tidal currents.
Although RS provides many advantages for OT studies, there are several challenges that need to be addressed in the future. For example, the temporal analysis of OTs and their effects on shorelines is limited due to the complex temporal nature of tides and the lack of suitable multitemporal RS datasets. Moreover, although a large volume of RS datasets is progressively produced, time dependency, low coverage, and complex processing steps still restrict the efficiency of RS techniques for OT monitoring. It is expected that the recent advances in ML and DL methods can fill some of these gaps. For example, DL architectures, such as CNN, Recurrent Neural Network (RNN), and Long Short-Term Memory (LSTM) can facilitate temporal analysis of OT.
3.6. Ship Detection (SD)
SD has a wide variety of environmental, civil, and military applications. For example, the benefits of automatic locating and tracking ships for the civil sector are maritime management, vessel traffic services, safety and rescue, fishery management, and illegal fishery surveillance. Moreover, the main applications in the military sector include naval warfare, battlefield environment assessment, and pirate activity surveillance [
313,
314,
315]. RS has a leading role in SD and monitoring because of its several advantages, such as the availability of open-access multitemporal datasets and large area coverage.
Although various RS datasets have been used for SD (e.g., hyperspectral [
316], TIR [
317], and UAV [
318] imagery), optical, SAR, and HF radar are the most common RS systems for SD [
313].
Table 6 summarizes the main advantages and disadvantages of each of these systems for SD.
SD methods using spaceborne RS (i.e., optical and SAR) datasets generally have three main steps (see
Figure 9): (1) ocean–land segmentation, (2) ship candidate extraction, and (3) classification of ship candidates [
319,
320,
321]. Since the objective of the corresponding studies is to detect ships in oceans, the first step is separating ocean and land regions. This is usually performed using GIS layers of coastlines. However, with the end-to-end DL SD methods, this step is not necessary anymore. Most of the SD research studies have focused on the second and third steps by developing better features for ship description and False Alarm Rate (FAR) reduction. For example, in the second step, a simple shape analysis is performed to remove obvious FAR and extract Regions of Interest (ROI) that may contain potential ship candidates. In the third step, the ship candidates are classified into ship and nonship classes.
In the following three subsections, the most commonly used approaches for SD using optical, SAR, and HF radar data are discussed. However, it should be noted that more advanced ML methods, such as DL, have been recently employed for SD with high accuracies. For instance, among many object detection DL methods, the Region-based CNN (RCNN) [
322] and its modified versions (e.g., Fast-RCNN [
323] and Faster-RCNN [
324]) are mostly used for SD. RCNN-based methods involve two major steps: (1) a CNN algorithm extracts the shared feature maps, and the region proposal network algorithm generates candidate regions, including potential ship targets; and (2) the network classifies these proposals into specified classes. DL methods can extract semantic-level features that are robust to varying ship sizes and different ocean conditions, resulting in better performance than traditional methods with human-crafted features and descriptors. However, the main limitation of DL methods is the limited accessibility to sufficient reference sample data [
325,
326].
3.6.1. Optical
With the advent of high-resolution optical systems, the volume of high spatiotemporal resolution imagery is growing, making them suitable for SD and monitoring applications [
314]. Although optical RS systems can only operate during the daytime, recently, there has been a considerable focus on SD using optical imagery due to increasing the number of Very High Resolution (VHR) optical sensors [
327]. There are a variety of methods for SD using optical images, such as threshold-based methods [
328], anomaly detection [
321], transform domain methods [
313], Bayesian decision [
329], shape and texture [
314], and visual saliency [
330]. In this study, the methods based on visual saliency and shape and texture are discussed in detail due to their relatively good performance and popularity.
A common approach for SD using optical images is applying the methods based on shape and texture features. These methods are robust and provide high accuracy in SD. However, false alarms (wakes and ocean clutter) still exist. So far, several studies have used the shape and texture features extracted from VHR optical satellite images for SD. For example, Zhu et al. [
314] proposed an SD method based on shape and texture features using a combination of China–Brazil Earth Resources Satellite (CBERS) and SPOT images with a hierarchical coarse-to-fine FAR elimination process. They first carried out global and local image segmentation. Then, a simple shape analysis (e.g., region area and aspect ratio) was employed to remove obvious false candidates. Subsequently, finer shape features (e.g., compactness, rectangularity, and moment invariants), texture features (e.g., simple texture, wavelet-based, Multiscale Gaussian Differential Features (MGDFs)), and a novel operator, called local multiple pattern, proposed based on local binary pattern, were employed. Finally, they adopted a semisupervised hierarchical classification using SVM to identify ships.
Another popular approach in SD using optical satellite data is implementing methods based on visual saliency. Inspired by the human visual system, the visual saliency method tries to extract candidate regions that attract human eyes [
330]. For example, Qi et al. [
330] presented a method for extracting salient signals using a Phase spectrum of Fourier Transform (PFT). The binary saliency map was calculated through an adaptive segmentation. Then, the image was homogenized using a homogeneous filter to make it smooth. Finally, the Otsu segmentation method was applied to extract ship candidate regions. For discriminating ship targets, they developed a novel descriptor, called Ship Histogram of Oriented Gradient (S-HOG), based on the symmetrical shape of the ship. Their results on GaoFen-1 satellite images showed that the method was robust in detecting ships of various sizes in different ocean conditions.
As discussed, other than the abovementioned traditional methods, DL algorithms along with optical data have been recently used for SD. For instance, Zhang et al. [
325] proposed a method based on Faster-RCNN to detect small ships and gather ships in offshore and inland water scenes to cover shortcomings of the traditional approaches of SD. They first classified the images into water and nonwater areas and created small ROIs that might contain ships using the SVM classifier. Then, the ROIs were ingested into an RCNN-based algorithm. They also modified the network architecture of the Faster-RCNN by combining multiresolution convolutional features and performing ROI pooling on a larger feature map. Their modified Faster-RCNN showed higher precision than various state-of-the-art SD methods, including the original Faster-RCNN.
3.6.2. SAR
SAR systems have also been extensively utilized for SD due to their operational ability in all weather conditions and all times [
314]. However, compared to optical imagery, they contain a higher amount of noise. Consequently, it is difficult to detect small ships using SAR data. Moreover, there are relatively few SAR satellites compared to optical systems, causing challenges for real-time ship tracking and monitoring [
314].
Numerous algorithms have been developed for SD using SAR images, including coherent target decomposition [
331], polarimetric entropy [
332], wavelet transforms [
333], and generalized-likelihood ratio test [
334]. However, the most-used method in this regard is the Constant FAR (CFAR) [
335]. The CFAR algorithm searches for bright pixels in SAR images by comparing their grayscale value to the surrounding pixels. The most important step in the CFAR method is to design a distribution model for the background to specify an associated Probability Density Function (PDF). The Gaussian distribution, K-distribution, and Gamma PDFs have been widely used in this regard. Then, the Probability of False Alarm (PFA) for the threshold T is calculated by Equation (4).
The conventional CFAR algorithm only considers the intensity value of the pixels. However, this can be due to the SAR ambiguities and ocean clutter in complex ocean conditions [
320,
321]. Therefore, FAR increases by incorrectly detecting bright clutter pixels as ships and missing some not-that-bright ship pixels due to factors such as speckles, incident angels, and materials of targets. With the advent of high spatial resolution SAR systems, such as TerraSAR-X and Sentinel-1, researchers are encouraged to take advantage of spatial relationships alongside the intensity of pixels to improve the conventional CFAR algorithm. For example, Leng et al. [
336] proposed the bilateral CFAR algorithm, which used kernel density estimation as an analytical tool to determine the structure of the targets. Then, a combined value was generated using both intensity and spatial distributions. Finally, the standard CFAR method was applied to the SAR images with combined values to detect targets. Their results indicated that the bilateral CFAR reduced FAR compared to the conventional CFAR. Moreover, Wang et al. [
337] proposed an Intensity-Space (IS) domain CFAR SD algorithm based on two factors [
337]: (1) two neighboring pixels are more likely to be of the same kind, and (2) two pixels with similar intensity have a higher probability of belonging to the same class. In the IS-CFAR algorithm, intensity and locality were considered at the same time, as opposed to the sequential approach of the bilateral CFAR algorithm, and thus, the image was mapped into the IS domain to create a new transformed image. Then, the CFAR method was applied to extract ship targets. Finally, the results were refined using three target features of the target region area, form factor, and aspect ratio to further eliminate false alarms. The results showed that this method performed better than the bilateral CFAR only in terms of true target detection precision.
DL methods have also been widely investigated for SD using SAR data. For example, Jiao et al. [
338] proposed an end-to-end, densely connected multiscale neural network based on Faster-RCNN for SD in SAR images [
338]. They densely connected feature maps to each other from top to down, instead of using one feature map for region proposal, to achieve multiscale and multiscene (inshore and offshore) SD without ocean–land segmentation. They also proposed a training scheme to decrease the weight of easy-to-detect targets to decrease FAR in SD. Their results on multiresolution public SAR data, including imagery from RADARSAT-2, TerraSAR-X, Sentinel-1, and GaoFen-3, indicated the effectiveness of the method and its superiority over the Faster-RCNN algorithm. However, false alarms were observed, and the method had a problem when two ships were side by side (they were detected as one ship).
3.6.3. HF Radar
HF radars based on electromagnetic scattering and surface wave propagation can detect and track targets far beyond the conventional microwave radar coverage [
339,
340]. HF radars exploit the HF band (3–30 MHz) to offer a board coverage extended to more than 200 km in the range [
340]. However, the radio frequency interferences and external noise restrict the detection capability of the HF radars [
184,
185]. Regarding SD using HF radar data, the sea clutter is considered to be self-generated interference and creates the first- and second-order resonant scattering effects in the Doppler spectrum. The first-order scattering is produced by ocean waves with half of the radar wavelength, while the second-order scattering is caused by the interaction of crossing sea waves [
340]. SD using HF radar is mainly based on the CFAR algorithm following the beamforming data process. The CFAR threshold is usually calculated by the Neyman–Pearson approach and assuming a fixed PFA [
341]. With this approach, a detection decision is made for each Range-Doppler (RD) cell.
Many studies have so far utilized HF radar data for SD. For example, in Dzvonkovskaya and Rohling [
341], the adaptive threshold is obtained by applying the conventional curvilinear regression analysis [
342] along with the Doppler and range cells to detect ships. Additionally, the gating approach was used to associate each target plot with a track in polar coordinates in order to track targets such as ships. Moreover, in Gorski et al. [
343], target detection was accomplished by applying the space-time adaptive processing technique to each range cell. This algorithm was operated on a so-called data cube including complex samples, obtained from many coherent pulses for various range cells to acquire test statistics.
Although the aforementioned approaches are applicable for SD using HF radar, several issues may emerge when an HF radar observes the ocean surface and monitors moving ships. The first issue is that ships’ maneuvering during a long coherent processing interval creates Doppler blur in the Range-Doppler map. Therefore, SD may fail when the CFAR detector is applied. The second issue is that the Bragg scattering may mask a ship signal or be a considerable source of false detection since Range-Doppler cells associated with the Bragg scattering have a significant magnitude. In this regard, Roarty et al. [
344] attempted to decrease the effect of ship maneuvers by applying a median filter before CFAR detection. Additionally, the decision fusion technique can be used for improving the detection performance to solve the second issue [
70]. Park et al. [
345] proposed a method to simultaneously solve the two abovementioned issues by adding a signal-enhancement procedure before CFAR detection. Finally, Hong et al. [
346] proposed a DL method for SD using a combination of SAR and optical images. They developed a novel, robust framework based on a real-time object detection model, called YOLO, to improve the SD results compared to previous methods in the context of multisource, multiscale, and multisensor datasets. It was reported that their model could detect ships at any time with any available RS imagery with a high accuracy.
3.6.4. Summary and Future Direction
In summary, optical and SAR images have been extensively used for SD. However, they have usually been employed separately, and multisensor approaches have rarely been explored. Each of these data has its own advantages and disadvantages, and combining them for SD overcomes their limitations and provides better results. The focus of the future SD studies using SAR images should be concentrated on incorporating more precise spatial feature extraction and speeding up the algorithms of the spatial information extraction. This can improve the detection accuracy and reduce the detection time consumption [
336]. Furthermore, in case of the existence of cooperative positioning data, more accurate and better results can be achieved.
DL methods have shown promising results for a variety of SAR applications. However, some further investigations are suggested for SD. First, more annotated SAR data are needed for accurate detection because it can improve the learning network. Thus, it is expected that more high-quality SAR datasets for SD will be provided for research in the future [
338]. Additionally, it is recommended to concentrate on improving the segmentation algorithms in terms of both accuracy and computation time. In terms of optical images, it might be valuable to carry out more research on more complex CNN and RNN methods [
325]. Moreover, regarding the spaceborne RS technology, it is worthwhile to exploit the hyperspectral data to for SD.