The ESA Permanent Facility for Altimetry Calibration: Monitoring Performance of Radar Altimeters for Sentinel-3A, Sentinel-3B and Jason-3 Using Transponder and Sea-Surface Calibrations with FRM Standards

This work presents the latest calibration results for the Copernicus Sentinel-3A and -3B and the Jason-3 radar altimeters as determined by the Permanent Facility for Altimetry Calibration (PFAC) in west Crete, Greece. Radar altimeters are used to provide operational measurements for sea surface height, significant wave height and wind speed over oceans. To maintain Fiducial Reference Measurement (FRM) status, the stability and quality of altimetry products need to be continuously monitored throughout the operational phase of each altimeter. External and independent calibration and validation facilities provide an objective assessment of the altimeter’s performance by comparing satellite observations with ground-truth and in-situ measurements and infrastructures. Three independent methods are employed in the PFAC: Range calibration using a transponder, sea-surface calibration relying upon sea-surface Cal/Val sites, and crossover analysis. Procedures to determine FRM uncertainties for Cal/Val results have been demonstrated for each calibration. Biases for Sentinel-3A Passes No. 14, 278 and 335, Sentinel-3B Passes No. 14, 71 and 335, as well as for Jason-3 Passes No. 18 and No. 109 are given. Diverse calibration results by various techniques, infrastructure and settings are presented. Finally, upgrades to the PFAC in support of the Copernicus Sentinel-6 ‘Michael Freilich’, due to launch in November 2020, are summarized.

cross-comparison by diverse and redundant Cal/Val techniques at different settings (at 1050 m, 160 m altitude and at sea level, for example) is required to establish confidence in the calibration results that are derived.
The revisit cycle of Sentinel-3A and -3B is every 27 days and that of Jason-3/Sentinel-6 is every 10 days, with ascending and descending orbits crossing over PFAC Cal/Val transponder infrastructure. This allows the PFAC to support annually about 50 calibrations for each Sentinel-3A and Sentinel-3B satellite and more than 70 calibrations per year for Jason-3 (and subsequently Sentinel-6). It can be seen from Figure 1 that the PFAC calibrates three Sentinel-3A Passes (No. 14,No. 278,and No. 335), three Sentinel-3B Passes (No. 14,No. 71,and No. 335) and two Jason-3 Passes This work presents the latest calibration results for Sentinel-3A, Sentinel-3B and Jason-3 as determined at the PFAC, making use of three independent Cal/Val techniques: transponder, sea-surface and crossover analysis. A short description of the PFAC infrastructure and instrumentation is given in Section 2, together with a summary of the three calibration methods employed. The procedures to ensure the Cal/Val measurements (compliant with the FRM standards), and the uncertainty analysis for the sea-surface and transponder calibrations, are described in Section 3. Section 4 presents the absolute Cal/Val results for each satellite (i.e., Sentinel-3A, Sentinel-3B and Jason-3), along with relative calibration results of Sentinel-3A and Sentinel-3B with respect to Jason-3 at crossover locations over the sea surface. Section 5 discusses single-mission results and evaluates performance using each Cal/Val site. Inter-validation of calibration results obtained by diverse Cal/Val methods is also reported. Section 6 summarizes calibration results and the performance of Sentinel-3A, Sentinel-3B and Jason-3 obtained by the PFAC. Finally, future plans for the calibration of the upcoming Sentinel-6 are also discussed.

Infrastructure, Instrumentation and Calibration Methods
This section presents the infrastructure and instrumentation of the Permanent Facility for Altimetry Calibration in west Crete, Greece. The principle of operations for each calibration method (i.e., transponder, sea-surface, crossover analysis) employed along with the parametrization of each is also briefly described.

PFAC Infrastructure and Instrumentation
The location and setting of the Gavdos/Crete Permanent Facility for Altimetry Calibration, in the center of the eastern Mediterranean, offers a unique location for the present generation of multi-mission altimetry calibration ( Figure 1). This work presents the latest calibration results for Sentinel-3A, Sentinel-3B and Jason-3 as determined at the PFAC, making use of three independent Cal/Val techniques: transponder, sea-surface and crossover analysis. A short description of the PFAC infrastructure and instrumentation is given in Section 2, together with a summary of the three calibration methods employed. The procedures to ensure the Cal/Val measurements (compliant with the FRM standards), and the uncertainty analysis for the sea-surface and transponder calibrations, are described in Section 3. Section 4 presents the absolute Cal/Val results for each satellite (i.e., Sentinel-3A, Sentinel-3B and Jason-3), along with relative calibration results of Sentinel-3A and Sentinel-3B with respect to Jason-3 at crossover locations over the sea surface. Section 5 discusses single-mission results and evaluates performance using each Cal/Val site. Inter-validation of calibration results obtained by diverse Cal/Val methods is also reported. Section 6 summarizes calibration results and the performance of Sentinel-3A, Sentinel-3B and Jason-3 obtained by the PFAC. Finally, future plans for the calibration of the upcoming Sentinel-6 are also discussed.

Infrastructure, Instrumentation and Calibration Methods
This section presents the infrastructure and instrumentation of the Permanent Facility for Altimetry Calibration in west Crete, Greece. The principle of operations for each calibration method (i.e., transponder, sea-surface, crossover analysis) employed along with the parametrization of each is also briefly described.

PFAC Infrastructure and Instrumentation
The location and setting of the Gavdos/Crete Permanent Facility for Altimetry Calibration, in the center of the eastern Mediterranean, offers a unique location for the present generation of multi-mission altimetry calibration (Figure 1). Remote Sens. 2020, 12,2642 6 of 39 The PFAC consists of three permanent coastal Cal/Val sites (GVD8 at Gavdos island, CRS1 and RDK1 at the south and west coast of Crete, respectively, Figure 1b) and one transponder Cal/Val site on a mountainous area in the mainland of Crete (Figure 1a). A detailed presentation of these Cal/Val sites has been provided in [12]. Table 1 presents an overview of each Cal/Val site in terms of the altimeters supported, the calibration method employed and its main instrumentation. These instruments are accessed and controlled remotely. Their field observations are transmitted to, via satellite and/or GPRS communications link, and directly archived at the PFAC Operations Control Center at the Technical University of Crete, Greece. Details of the data processing techniques and analysis of raw measurements are given in Section 3.
Two more calibration sites are currently under development: A sea-surface-elevation Cal/Val site at the south coast of Crete (SUG1, Figure 1) and a second transponder Cal/Val site (GVD1) on Gavdos island to support Sentinel-6 Cal/Val activities. A summary of the new GVD1 site is provided in Section 6.

Calibration and Validation Methods
Independent calibration and validation activities are required to assess the end-to-end performance characteristics of satellite altimeter measurements. Cal/Val activities monitor the measurement uncertainty, stability and behavior, providing independent knowledge of measurement quality. External techniques that regularly support post-launch calibration of satellite altimeters include: (1) sea-surface Cal/Val, (2) transponder Cal/Val, (3) backscatter coefficient Cal/Val, (4) crossover analysis, (5) tandem phase analysis, and (6) comparison to tide gauge networks. The latter is an indirect method, which compares the altimeter-derived sea surface heights with measurements obtained by a global network of tide gauges [13]. Details on PFAC-employed Cal/Val techniques can be found in [14]. A short description of the remaining four techniques and their implementation in the PFAC is given in the following sections.

Sea-Surface Cal/Val
Satellite altimeters are designed to measure the sea (or water) surface height (SSH) above a reference ellipsoid. The sea-surface calibration method compares the sea surface height, as measured by a satellite Remote Sens. 2020, 12, 2642 7 of 39 altimeter at a certain location and time in open water, to sea surface heights measured by reference ground measurements. A number of corrections need to be applied to the measured satellite range to account for propagation errors caused by the atmosphere (ionosphere, wet and dry troposphere), the sea state bias due to surface roughness impacting the radar measurement, and geophysical corrections including solid earth tide, ocean tides, etc. Ideally, the bias of an altimeter-derived sea-surface height is determined as the difference between the satellite and the ground-truth observations implemented at the same location and time.
Ground-truth observations are provided by a combination of water level sensors, absolute positioning receivers, meteorological and oceanographic sensors, and models of the geoid, tides, mean dynamic topography, etc. At the PFAC, the land mass of Gavdos and Crete contaminates the altimeter's signal, leading to invalid sea-surface measurements close to the coast (about 7.5 km). To overcome this, the Cal/Val site observations have to be transferred to the open sea at about 20 km, with about 3000 m depth, far from the coast where valid satellite measurements exist. This introduces an uncertainty in the approach compared to the ideal case where truly contemporaneous and co-located measurements are available.

Transponder Cal/Val
The sea-surface calibration method compares a derived geophysical parameter (sea-surface height or sea-surface anomaly) at a close-by location with ocean truth measurements regularly made on the coast. Nonetheless, the primary measurement of radar altimeter is the propagation time (converted to range if multiplied by speed of light) of a signal pulse from the altimeter antenna to the Earth's surface and back to the satellite. A transponder provides absolute calibration of these direct and fundamental range observations made by the altimeter if the location of the transponder and its internal path delay are precisely and accurately known.
An active microwave transponder receives the altimeter's signal when a satellite flies over it (which takes about 4 s), amplifies it with minimum distortion and sends it back to the satellite where it is recorded. At first, the received signal is distinctly separated from other surrounding reflections (clutter) of the ground surface as it had been powerfully amplified by the transponder. Then, at explicitly the same time, the geocentric positions of the altimeter in orbit and the transponder on the ground are determined. The theoretical distance between these two instruments in space is subsequently computed. Finally, after extensive processing, the observed range by the altimeter is compared against the theoretical range of the transmitting satellite point in orbit and the center of the transponder. During the overpass period, a time series of the differences between the observed range and the theoretical range provides an estimate of the altimeter range, if short time scale atmospheric components of the troposphere are sufficiently known.
The PFAC has been providing transponder calibration services for Jason-2, Jason-3, CryoSat-2, Sentinel-3A, and Sentinel-3B from September 2015 as part of the ESA FRM4ALT project. After each calibration, the echo of the transponder response is recorded by the altimeter and two recent such examples are shown in Figure 2. Transponder calibrations presuppose that the Agency operating the satellite has programmed the altimeter to observe within the measuring window of the CDN1 Cal/Val site on the mountains at the elevation of 1050 m. scale atmospheric components of the troposphere are sufficiently known.
The PFAC has been providing transponder calibration services for Jason-2, Jason-3, CryoSat-2, Sentinel-3A, and Sentinel-3B from September 2015 as part of the ESA FRM4ALT project. After each calibration, the echo of the transponder response is recorded by the altimeter and two recent such examples are shown in Figure 2. Transponder calibrations presuppose that the Agency operating the satellite has programmed the altimeter to observe within the measuring window of the CDN1 Cal/Val site on the mountains at the elevation of 1050 m.

Backscatter Coefficient Cal/Val
In addition to the signal propagation time, a radar altimeter measures the power level of these radar signals [15]. The relation between the amplitude of the transmitted and received echo signals provides information about the characteristics of the imaged surface on the ground. This is expressed as the back-scatter coefficient (sigma-naught, or sigma-0). The sigma0 coefficient is used to retrieve wind speed, and empirical models for the sea state bias required to determine sea surface height.
Calibration of the sigma-naught coefficient can be obtained using either active or passive targets with known and stable radar cross-section, as they are easily detected within the altimeter backscatter records. Previously, a sigma-naught transponder was developed by the European Space Agency and used to calibrate the ENVISAT RA-2 altimeter [16]. This old transponder has been recently (2019) refurbished and is expected to support Sentinel-3 sigma-naught calibrations in 2020 [17].
In support of Copernicus Sentinel-6, a new range and sigma-naught transponder is currently under development and is expected to be installed at the PFAC. The new processing technique of fully focused SAR altimetry [18] may also provide sigma-naught calibration with passive corner reflectors [17] and this will also be investigated at the PFAC. Details on these future plans are given in Section 6.

Crossover Analysis
This technique aims at providing relative calibration between two or more contemporaneous satellite altimeters [19]. The analysis usually takes place on a global scale and at locations where the ascending and descending ground-tracks of altimeters intersect. A crossing event occurs when the time difference of the intersecting passes of two (or more) missions is commonly less than two days [20]. Under the tenuous assumption that the variability and signature of sea level has not changed significantly in the two-day lag between measurements, the sea surface height at the crossover location, as determined by the two missions, can be compared. A significant advantage of this relative performance approach is that it does not require ground measurements and can be performed throughout the missions' lifetime and on a global scale [19].
At the PFAC, the crossover analysis has been modified and applied to retrieve the relative bias of Sentinel-3A Pass No. 14 and Jason-3 Pass No. 109 about 20 km south of Gavdos ( Figure 3). [20]. Under the tenuous assumption that the variability and signature of sea level has not changed significantly in the two-day lag between measurements, the sea surface height at the crossover location, as determined by the two missions, can be compared. A significant advantage of this relative performance approach is that it does not require ground measurements and can be performed throughout the missions' lifetime and on a global scale [19].
At the PFAC, the crossover analysis has been modified and applied to retrieve the relative bias of Sentinel-3A Pass No.14 and Jason-3 Pass No. 109 about 20 km south of Gavdos ( Figure 3).

Tandem Phase Cal/Val
The ideal scenario to characterize the relative bias between two altimeters is to make them observe the same location of the Earth at the same time or a few seconds apart [21]. This is accomplished during a tandem phase, which many new satellite missions incorporate into their regular operations. During this phase, the newly launched satellite is placed on the same orbital track as a reference orbit. The two satellites fly in unison on the same orbit but separated only some seconds apart. For example, Sentinel-3B flew in tandem [22] with Sentinel-3A for four months (6-June-2018 to 16-October-2018) before being placed on its nominal orbit. Similarly, Jason-3 was flying in tandem with Jason-2 with a temporal distance of 80 s for six months (12-Febuary-2016 to 2-October-2016). In [12], the PFAC-derived calibration results of the Jason-2/Jason-3 tandem phase have been presented. The present plan for the future Sentinel-6 will implement a 12-month tandem phase with Jason-3, where the two satellites will fly 30 s apart [23].
This work (see Section 4) presents the relative calibration results from the Sentinel-3 A/B tandem phase. The following Section presents the PFAC's in-situ measurements, the models employed, and the processing strategies followed for the implementation of the aforementioned calibration methodologies for Sentinel-3A, Sentinel-3B and Jason-3 altimeters for absolute and relative calibration.

The PFAC Measurements and the FRM Uncertainty Analysis
In-situ measurements used for external calibration and validation of Earth observation have to follow and be traceable to internationally agreed reference standards. In addition, the uncertainty analysis of these measurements has to accompany the respective Cal/Val results. This uncertainty should be available to allow altimetry operators to evaluate the credibility and suitability of the Cal/Val results for efficient monitoring of the altimeter's performance. A strategy for assuring a uniform and absolute standardization for calibrating satellite altimeters is given in [24]. The main uncertainty constituents for sea-surface and transponder Cal/Val techniques have been identified in [9] and quantified in [14].
Accurate determination of the absolute geodetic coordinates for a Cal/Val site constitutes one of the key uncertainty components in both sea-surface and transponder calibration. Particularly, the height component is of utmost importance as it controls both the sea-surface height at the sea-surface Cal/Val site and the distance between the transponder and altimeter phase center. The following section presents the strategy followed at the PFAC to determine the absolute coordinates of the Cal/Val sites and their uncertainty. Later, the remaining constituents of uncertainty contributing to each calibration method are examined separately.

Absolute Coordinates Constituent
In the PFAC, a network of continuously operating GNSS stations distributed in western Crete and Gavdos ( Figure 4) provides absolute geodetic coordinates and velocities for reference points at each Cal/Val site.  According to FRM requirements, redundancy of the scientific equipment has to be ensured. This implies that at least two instruments should operate side by side to measure the same parameter to ensure that any potential errors and drifts in the operations are identified. For example, Table 2 presents the coordinates and velocities for the two GNSS receivers operating at the CDN1 transponder Cal/Val site (CDN0 and CDN2) and the two operating at the Gavdos sea-surface facility (GVD7 and GVD8). All sites have been processed using the GAMIT scientific software, which applies relative positioning techniques [25]. The reference epoch of these coordinates and velocities is 2013.5 and the geodetic frame used is the ITRF 2014.  According to FRM requirements, redundancy of the scientific equipment has to be ensured. This implies that at least two instruments should operate side by side to measure the same parameter to ensure that any potential errors and drifts in the operations are identified. For example, Table 2 presents the coordinates and velocities for the two GNSS receivers operating at the CDN1 transponder Cal/Val site (CDN0 and CDN2) and the two operating at the Gavdos sea-surface facility (GVD7 and GVD8). All sites have been processed using the GAMIT scientific software, which applies relative positioning techniques [25]. The reference epoch of these coordinates and velocities is 2013.5 and the geodetic frame used is the ITRF 2014.
In order to assess the values for the ellipsoidal height for each pair of the GNSS receivers (i.e., the CDN0 with the CDN2 and the GVD7 with the GVD8) their antenna reference points have to be geodetically tied together but also connected to other reference control points in the vicinity. This task is achieved through precise leveling that provides height differences with uncertainty of the order of ±1 mm. These levelling surveys are carried out on a semi-annual basis as well as during a major event (i.e., GNSS antenna replacement, earthquake, etc.). The derived uncertainty of GNSS processing is given in Table 2. The final value for uncertainties in the determination of absolute coordinates for the reference benchmarks on each Cal/Val site are used as input in the processing chain to derive the FRM constituent uncertainty.

GNSS Processing Constituent
The positioning results presented in Table 2 are those derived by relative positioning processing. Using the same GNSS observations, the alternative technique of Precise Point Positioning (i.e., GIPSY scientific software [26]) is also applied to derive solutions for the coordinates and velocities of the same GNSS stations. Processing of the same GNSS observations with different processing strategies ensures validity of the derived absolute positioning values, and follows the FRM4ALT study recommendation to use multiple approaches to arrive at the same result [24]. The results of different processing strategies are shown in Figure 5 for the GVD8 Cal/Val site in Gavdos.
Remote Sens. 2020, 12, x FOR PEER REVIEW 11 of 39 2019.99] ±1.8 mm ±2.2 mm mm In order to assess the values for the ellipsoidal height for each pair of the GNSS receivers (i.e., the CDN0 with the CDN2 and the GVD7 with the GVD8) their antenna reference points have to be geodetically tied together but also connected to other reference control points in the vicinity. This task is achieved through precise leveling that provides height differences with uncertainty of the order of ±1 mm. These levelling surveys are carried out on a semi-annual basis as well as during a major event (i.e., GNSS antenna replacement, earthquake, etc.). The derived uncertainty of GNSS processing is given in Table 2. The final value for uncertainties in the determination of absolute coordinates for the reference benchmarks on each Cal/Val site are used as input in the processing chain to derive the FRM constituent uncertainty.

GNSS Processing Constituent
The positioning results presented in Table 2 are those derived by relative positioning processing. Using the same GNSS observations, the alternative technique of Precise Point Positioning (i.e., GIPSY scientific software [26]) is also applied to derive solutions for the coordinates and velocities of the same GNSS stations. Processing of the same GNSS observations with different processing strategies ensures validity of the derived absolute positioning values, and follows the FRM4ALT study recommendation to use multiple approaches to arrive at the same result [24]. The results of different processing strategies are shown in Figure 5 for the GVD8 Cal/Val site in Gavdos.  The coordinates and velocities of each GNSS station as determined by diverse GNSS processing techniques have been found to agree at the mm-level and sub-mm level, respectively (Table 3). Table 3. GNSS processing results (Cartesian coordinates and velocities) for the GVD8 station as derived by the relative differential and precise point positioning techniques.

GVD8 Station
Relative In the following subsection, the uncertainty for the sea-surface and transponder Cal/Val, as implemented at the PFAC, is given along with the verification that in-situ measurements are in compliance with the FRM4ALT standards.

Uncertainty Budget for Sea-Surface Calibration
Apart from the uncertainty constituent in the determination of the absolute coordinates for the Cal/Val site, presented in Section 3.1, other major contributing sources of uncertainty in sea-surface calibration are: (1)

Sea-Surface Height Constituent
Measurements obtained by tide gauges are used to determine the sea surface height above a reference ellipsoid at the PFAC Cal/Val sites. According to FRM standards, at least two such instruments of diverse make and operating principle are to operate at each site. This redundant instrumentation permits efficient monitoring of each instrument's performance and inter-validation of their measurements. For example, in Gavdos Cal/Val, there are five continuously operating tide gauges: KVR3 (radar), KVR4 (acoustic), KVR5 (pressure), KVR6 (radar and operating in a stilling well) and KVR7 (radar) ( Figure 6). The overall performance of these tide gauges is monitored on a daily and/or monthly basis. A 24-hour recording of four of those tide gauges, on 14-Oct-2019, is given in Figure 7. The overall performance of these tide gauges is monitored on a daily and/or monthly basis. A 24-h recording of four of those tide gauges, on 14-Oct-2019, is given in Figure 7. Figure 6. The current set up of the "Karave" Cal/Val site in Gavdos harbor for sea-surface calibrations. Five tide gauges (KVR3-radar, KVR4-acoustic, KVR5-pressure, KVR6-radar in a stilling well, and KVR7-radar), two GNSS receivers (GVD7 and GVD8) and one meteorological station comprise the main scientific instrumentation of the Gavdos Cal/Val site.
The overall performance of these tide gauges is monitored on a daily and/or monthly basis. A 24-hour recording of four of those tide gauges, on 14-Oct-2019, is given in Figure 7. A more detailed analysis of water level observations involves, among others, the evaluation of the Van de Casteele diagrams on a daily basis. The shape of the Van de Casteele diagram provides a qualitative illustration of the type of error involved in tide gauge measurements [27]. Simultaneous water level heights of a certain tide gauge are compared against a master reference tide gauge ( Figure 8). The x-axis corresponds to the difference between the two tide gauge measurements over a full tidal cycle. The y-axis is the "true" sea-surface height as measured by the reference tide gauge. The shape of the diagram may be used to identify the source of any deviation between the measurements obtained by the reference and the tide gauge under investigation [28].
In Gavdos, the KVR3 radar tide gauge is considered to be the reference water level sensor. The Van de Casteele test is applied and the performance of the remaining tide gauges (i.e., KVR4, KVR6, KVR7) is monitored. Various sampling intervals exist for the different tide gauges. In order to apply the Van de Casteele diagram, the daily tide gauge records are averaged, when needed, to match a 6-min sampling interval. Figure 8 presents an indicative example of such a single day A more detailed analysis of water level observations involves, among others, the evaluation of the Van de Casteele diagrams on a daily basis. The shape of the Van de Casteele diagram provides a qualitative illustration of the type of error involved in tide gauge measurements [27]. Simultaneous water level heights of a certain tide gauge are compared against a master reference tide gauge ( Figure 8). The x-axis corresponds to the difference between the two tide gauge measurements over a full tidal cycle. The y-axis is the "true" sea-surface height as measured by the reference tide gauge. The shape of the diagram may be used to identify the source of any deviation between the measurements obtained by the reference and the tide gauge under investigation [28].
Remote Sens. 2020, 12, x FOR PEER REVIEW 14 of 39 (14-October-2019) monitoring for the KVR4 and the KVR7 sensors. It can be seen that, for this particular day, all tide gauge records agree with an uncertainty of the order of ±1cm. The radar KVR3 sensor is considered to be the reference against which the performance of the acoustic KVR4 (upper), and radar KVR7 (lower) tide gauges is evaluated. Each diagram represents the sea level (SSH, y axis) versus the difference between records of the two instruments (ΔH, x-axis). The sampling rate is 6min.
Monitoring of the long-term performance of all tide gauges with an external and independent system without interrupting their measurements poses a challenge in the PFAC operations. Two approaches are followed: In the first, the monthly means of each tide gauge measurement are Remote Sens. 2020, 12, x FOR PEER REVIEW 14 of 39 (14-October-2019) monitoring for the KVR4 and the KVR7 sensors. It can be seen that, for this particular day, all tide gauge records agree with an uncertainty of the order of ±1cm. The radar KVR3 sensor is considered to be the reference against which the performance of the acoustic KVR4 (upper), and radar KVR7 (lower) tide gauges is evaluated. Each diagram represents the sea level (SSH, y axis) versus the difference between records of the two instruments (ΔH, x-axis). The sampling rate is 6min.
Monitoring of the long-term performance of all tide gauges with an external and independent system without interrupting their measurements poses a challenge in the PFAC operations. Two approaches are followed: In the first, the monthly means of each tide gauge measurement are monitored to verify their overall performance. The second uses an external reference instrument (i.e., The radar KVR3 sensor is considered to be the reference against which the performance of the acoustic KVR4 (upper), and radar KVR7 (lower) tide gauges is evaluated. Each diagram represents the sea level (SSH, y axis) versus the difference between records of the two instruments (∆H, x-axis). The sampling rate is 6min.
In Gavdos, the KVR3 radar tide gauge is considered to be the reference water level sensor. The Van de Casteele test is applied and the performance of the remaining tide gauges (i.e., KVR4, KVR6, KVR7) is monitored. Various sampling intervals exist for the different tide gauges. In order to apply the Van de Casteele diagram, the daily tide gauge records are averaged, when needed, to match a 6-min sampling interval. Figure 8 presents an indicative example of such a single day (14-October-2019) monitoring for the KVR4 and the KVR7 sensors. It can be seen that, for this particular day, all tide gauge records agree with an uncertainty of the order of ±1cm.
Monitoring of the long-term performance of all tide gauges with an external and independent system without interrupting their measurements poses a challenge in the PFAC operations. Two approaches are followed: In the first, the monthly means of each tide gauge measurement are monitored to verify their overall performance. The second uses an external reference instrument (i.e., tide pole) for sea-surface height determination. A leveling survey is carried out to tie the tide pole and tide gauge reference points. The Van de Casteele diagrams are again created, but using the tide pole as reference. This procedure is carried out at least once per year and a detailed record is kept. If the difference between the tide pole and one of the tide gauges exceeds certain criteria (i.e., 1 cm), then the external validation is repeated within two months. If the difference continues to exceed the criteria set, then the tide gauge is decommissioned and sent to external facilities for calibration. This technique is also used to precisely estimate the electrical center, i.e., the measuring reference point ("zero" point) of the tide gauge records.

Reference Surface Constituent
In the PFAC, the sea-surface height has to be transferred from the Cal/Val location in the harbor to open sea where satellite measurements are made uncontaminated by land mass (about 20 km away). To accomplish this transfer, the reference geoid and the mean dynamic topography (MDT) have to be accurately known. In the past, terrestrial, marine and airborne gravity surveys have been carried out in the vicinity of the PFAC Cal/Val sites. Regional geoid and MDT models have been developed and validated by dedicated field campaigns using boats and buoys. Details on those reference surfaces are presented in [29].
In 2018, additional gravity field measurements have been carried out at the PFAC Cal/Val sites and around the Gavdos island ( Figure 9). A CG-5 microgravity sensor with a resolution of 1 µGal and standard deviation less than 5 µGals has been used. Three control points with known absolute gravity values have been chosen as reference points to conduct this gravity survey. Absolute gravity measurements have been conducted in the past by the Hellenic Military Geographical Service and international research institutes. These absolute gravity points are located at Chania Airport (AIR), and Sfakia (SFK) and Paleochora (PAL) villages ( Figure 9). Absolute gravity values for various locations at the PFAC sites have been determined by relying upon those absolute reference points (AIR, PAL, SFK), and measuring relative gravity between them and the PFAC Cal/Val sites. These new gravity observations have been incorporated in the regional gravity database to improve the geoid accuracy.

FRM Uncertainty Estimation for Sea-Surface Calibration
For sea-surface calibration, the principal sources of uncertainty arise from (1) absolute coordinates of the Cal/Val site (Section 3.1), (2) the sea-surface heights (Section 3.2.1), and (3) the reference surfaces to transfer sea-surface heights to open sea (Section 3.2.2). A complete list and analysis of various constituents of uncertainty involved in sea-surface calibration may be found in [10]. The uncertainty for each constituent is classified into two categories: Type A and Type B. Their leading difference is that Type A uncertainties are evaluated by statistical analysis of observations, whereas Type B ones are estimated by scientific judgment based on manufacturer's specs, prior experience, handbooks, etc. [30]. Table 4 quantifies the uncertainty for the sea-surface Cal/Val as implemented in the PFAC. It also updates the uncertainty budget analysis that has been previously presented in [14,24]. The FRM standardized uncertainty at a 68% confidence level for the sea-surface calibration is calculated to be ± 31.91 mm.

FRM Uncertainty Estimation for Sea-Surface Calibration
For sea-surface calibration, the principal sources of uncertainty arise from (1) absolute coordinates of the Cal/Val site (Section 3.1), (2) the sea-surface heights (Section 3.2.1), and (3) the reference surfaces to transfer sea-surface heights to open sea (Section 3.2.2). A complete list and analysis of various constituents of uncertainty involved in sea-surface calibration may be found in [10]. The uncertainty for each constituent is classified into two categories: Type A and Type B. Their leading difference is that Type A uncertainties are evaluated by statistical analysis of observations, whereas Type B ones are estimated by scientific judgment based on manufacturer's specs, prior experience, handbooks, etc. [30]. Table 4 quantifies the uncertainty for the sea-surface Cal/Val as implemented in the PFAC. It also updates the uncertainty budget analysis that has been previously presented in [14,24]. The FRM standardized uncertainty at a 68% confidence level for the sea-surface calibration is calculated to be ±31.91 mm.

Uncertainty Budget for Transponder Calibration
Following a similar analysis, the FRM uncertainty for the transponder calibration has also been determined. Here, the atmospheric propagation delays of satellite signals and the transponder's internal delay dominate the overall uncertainty [9].

Atmospheric Propagation Delay Constituent
Determination of the signal delays caused by the atmosphere constitutes the most challenging task in transponder calibration. This is because the radiometer on-board the satellite does not operate properly over land. Thus, other means have to be figured out to correct signal propagation delays when the altimeter flies over a transponder land site. In the PFAC, this is carried out using GNSS-derived corrections for the atmospheric delays.
The GNSS-derived delays refer to both the ionospheric and the zenith tropospheric (wet and dry component) delays at the transponder Cal/Val site. The ionosphere delays over the PFAC are regularly stable. They do not change rapidly and can be accurately determined via GNSS processing with an uncertainty of the order of ±1 mm [24]. The dry (or hydrostatic) troposphere delay is also stable. It can also be accurately modeled using in-situ surface pressure measurements made by meteorological sensors. The wet component of the troposphere is dynamic, presents unstable variability rapidly changing over time, and cannot be easily modeled. In GNSS processing, this wet troposphere delay is estimated as the remaining part of the total delay minus the dry troposphere delay. Different GNSS processing methodologies generate diverse atmospheric propagation delays ( Figure 10).
with an uncertainty of the order of ±1 mm [24]. The dry (or hydrostatic) troposphere delay is also stable. It can also be accurately modeled using in-situ surface pressure measurements made by meteorological sensors. The wet component of the troposphere is dynamic, presents unstable variability rapidly changing over time, and cannot be easily modeled. In GNSS processing, this wet troposphere delay is estimated as the remaining part of the total delay minus the dry troposphere delay. Different GNSS processing methodologies generate diverse atmospheric propagation delays ( Figure 10). The difference observed in Figure 10 may be attributed to various models applied to determine the dry troposphere delay in their derivation, the means implemented to estimate surface pressure (i.e., models or in-situ measurements), as well as to the sampling rate used for GNSS data processing ( Figure 11). The difference observed in Figure 10 may be attributed to various models applied to determine the dry troposphere delay in their derivation, the means implemented to estimate surface pressure (i.e., models or in-situ measurements), as well as to the sampling rate used for GNSS data processing ( Figure 11). In order to be FRM-compliant, the PFAC should verify the GNSS-derived wet tropospheric delays at the time of satellite overpass using an independent technique. A ground microwave radiometer, provided on loan from the European Space Agency, has been installed in 2019 at the CDN1 transponder Cal/Val site to fulfill this FRM requirement (Figure 12). This instrument measures the brightness temperature of microwave radiation from the atmosphere and determines the respective contribution/attenuation from water vapor and liquid water. Preliminary results indicate an agreement of the wet troposphere delays from the radiometer and the GNSS-derived In order to be FRM-compliant, the PFAC should verify the GNSS-derived wet tropospheric delays at the time of satellite overpass using an independent technique. A ground microwave radiometer, provided on loan from the European Space Agency, has been installed in 2019 at the CDN1 transponder Cal/Val site to fulfill this FRM requirement ( Figure 12). This instrument measures the brightness temperature of microwave radiation from the atmosphere and determines the respective contribution/attenuation from water vapor and liquid water. Preliminary results indicate an agreement of the wet troposphere delays from the radiometer and the GNSS-derived delays. At this preliminary stage, dispersion of wet delay is in the order of ±1-2 cm (maximum) (Figure 12). Research and field measurements are ongoing. As more data are built up, more robust and reliable results on the final wet troposphere estimation will be estimated.
Remote Sens. 2020, 12, x FOR PEER REVIEW 18 of 38 In order to be FRM-compliant, the PFAC should verify the GNSS-derived wet tropospheric delays at the time of satellite overpass using an independent technique. A ground microwave radiometer, provided on loan from the European Space Agency, has been installed in 2019 at the CDN1 transponder Cal/Val site to fulfill this FRM requirement ( Figure 12). This instrument measures the brightness temperature of microwave radiation from the atmosphere and determines the respective contribution/attenuation from water vapor and liquid water. Preliminary results indicate an agreement of the wet troposphere delays from the radiometer and the GNSS-derived delays. At this preliminary stage, dispersion of wet delay is in the order of ±1-2 cm (maximum) Figure 12. The microwave radiometer operating at the CDN1 transponder Cal/Val site (upper image) and preliminary assessment of GNSS-derived wet tropospheric delays against radiometer's measurements (middle). The lower picture shows the setup of the radiometer and the two GNSS stations.
An alternative technique to determine wet tropospheric delay can be derived from measurements of the imaging spectrometer flown aboard the Sentinel-3 mission. It relies upon the Integrated Water Vapor ( ) products derived from the Ocean and Land Color Instrument (OLCI). There is a direct relation between the zenith wet delay ( ) and the , Equation (1). An alternative technique to determine wet tropospheric delay can be derived from measurements of the imaging spectrometer flown aboard the Sentinel-3 mission. It relies upon the Integrated Water Vapor (IWV) products derived from the Ocean and Land Color Instrument (OLCI). There is a direct relation between the zenith wet delay (ZWD) and the IWV, Equation (1).
where: F is a temperature-dependent factor given in the right-hand equation, IWV is the atmospheric water vapor content in (kg m −2 ), ρ is the mass density of water, R V the gas constant for water vapor in (N M K −1 gr −1 ), C 1 and C 2 are refractivity coefficients in (K 2 hPa −1 ) and (K hPa −1 ), respectively, T m is the weighted temperature of the atmosphere in ( • K) and ZWD is the zenith wet delay caused by the wet troposphere in (mm). The OLCI instrument of Sentinel-3 has a swath of 1270 km and thus permits global coverage within 2 days. OLCI is an optical instrument that captures images during daytime. The spatial resolution of these images is 300 m (at nadir). Both the SAR radar altimeter (SRAL) and OLCI are collocated but OLCI measurements are only available during periods of solar illumination. For example, Sentinel-3A SRAL measurements over the CDN1 transponder Cal/Val site are available at 20:00 UTC but no OLCI images are captured at this time of day (actually night).
On the other hand, simultaneous measurements of Sentinel-3B SRAL and OLCI instruments are made over the CDN1 Cal/Val site at 08:49:15 UTC every 27 days. Therefore, it is possible to directly compare the OLCI-derived IWV value at the 300m pixel, when the CDN1 site is observed, against the one GNSS-derived by the ground operating GNSS stations ( Figure 13).

FRM Uncertainty Estimation for Transponder Calibration
Several attempts have been made in [14,24] to estimate the uncertainty of the results in transponder calibration. Every time, a new and better estimate for some uncertainty constituents is presented as different elements of the uncertainty budget are improved. In fact, a primary application of a good uncertainty budget is to highlight the largest source of uncertainty in a measurement system, and then target that aspect for further improvement. In this manner, effort is given to the areas that will deliver the maximum improvement in performance, finally converging to a reliable measure of the uncertainty for transponder calibrations. In the same vein, Table 5 updates previous estimations for the uncertainty of the transponder calibration at the CDN1 transponder Cal/Val site.  In Figure 13, this comparison is made for the CDN0 GNSS station using three independent techniques: relative differential, precise point positioning and OLCI imagery. In GAMIT, the IWV is computed as a secondary product in the GNSS processing. In GIPSY, the IWV value has been estimated using Equation (1). In this present analysis, only cloud-free conditions are considered in this OLCI investigation when a Sentinel-3B transponder calibration takes place. The comparison, as shown in Figure 3, starts after the Sentinel-3B satellite reached its nominal orbit.
It seems that, at the moment, there are four different techniques for determining the wet troposphere delays during transponder calibrations of the CDN1 Cal/Val site in West Crete. As more observations are accumulated, more reliable results will be reported.

FRM Uncertainty Estimation for Transponder Calibration
Several attempts have been made in [14,24] to estimate the uncertainty of the results in transponder calibration. Every time, a new and better estimate for some uncertainty constituents is presented as different elements of the uncertainty budget are improved. In fact, a primary application of a good uncertainty budget is to highlight the largest source of uncertainty in a measurement system, and then target that aspect for further improvement. In this manner, effort is given to the areas that will deliver the maximum improvement in performance, finally converging to a reliable measure of the uncertainty for transponder calibrations. In the same vein, Table 5 updates previous estimations for the uncertainty of the transponder calibration at the CDN1 transponder Cal/Val site. In this Section, the uncertainty of the sea-surface and transponder calibration and for the procedures followed at the PFAC has been derived, following the FRM4ALT standards.

PFAC Calibration Results
This section presents the bias and uncertainty of the radar altimeters in Sentinel-3A, Sentinel-3B and Jason-3. First, absolute Cal/Val results are determined for each pass of the altimeters using the sea-surface and/or the transponder techniques. Then, relative calibration is carried out for the tandem phase of Sentinel-3A and Sentinel-3B, followed by crossover analysis for Sentinel-3A/Jason-3.

Sentinel-3A
Three Sentinel-3A passes are calibrated at the PFAC ( Figure 14): The ascending Pass No. 14 with sea-surface and transponder calibration at the Gavdos and the CDN1 Cal/Val sites, respectively, the descending Pass No. 335 also using the Gavdos Cal/Val site, and the descending Pass No. 278 with the CRS1 Cal/Val site.
In Sections 3.1 and 3.2, the calibration processing of the in-situ ground measurements and models has been described. Satellite measurements also require some kind of rectification, as the altimeter and radiometer's observations are contaminated by the land mass of Gavdos and Crete. Regions with valid satellite measurements have to be identified and selected as calibrating regions for comparison with in-situ Cal/Val measurements. This process is carried out for each satellite's pass. Using the latest processing baseline for Sentinel-3A (version 2.61) that is common for all of its  In Sections 3.1 and 3.2, the calibration processing of the in-situ ground measurements and models has been described. Satellite measurements also require some kind of rectification, as the altimeter and radiometer's observations are contaminated by the land mass of Gavdos and Crete. Regions with valid satellite measurements have to be identified and selected as calibrating regions for comparison with in-situ Cal/Val measurements. This process is carried out for each satellite's pass. Using the latest processing baseline for Sentinel-3A (version 2.61) that is common for all of its cycles, three sea-calibrating regions have been defined at the PFAC, as illustrated in Figure 14.
For a successful transponder calibration, the satellite altimeter generally needs to operate in "open-loop" mode. This is necessary to adjust its range gate to operate over the transponder's elevation and collect range measurements. The "open-loop" mode relies upon high-resolution Digital Elevation Models (DEM) stored in the altimeter or additionally with a direct commanding as in the Sentinel-6 case that additionally allows the fixing of the instrument gain given the transponder characteristics. Using this a-priori knowledge, the altimeter adjusts its internal receive window appropriately, and thus the signal amplified by the transponder can be received by the satellite altimeter. It is not the purpose of this work to explain the steps followed to process the waveforms for pulse-limited (e.g., Poseidon-3B in Jason-3) and delay-Doppler altimeters (i.e., SRAL/Sentinel-3A/B and Poseidon-4 Sentinel-6) to arrive at the value of range bias, as details can be found in [14,31].
Processing of Sentinel-3 waveforms is a complex procedure applied to about 200 radar bursts (of which about 30-40 bursts contain a useful transponder response over 1.5 to 2 s integration time) each of which contains 64 Ku-band pulses. Examples of altimeter waveforms that include the transponder's echo have been presented in Figure 2. The required parameters for transponder calibrations are obtained through Sentinel-3 telemetry and satellite calibration products [14].

Sentinel-3A Pass No. 14 Cal/Val Results
The ascending Sentinel-3A Pass No. 14 approaches the Gavdos island from the south. Calibration over sea surface is carried out using the Gavdos Cal/Val facility. About 9 s later, on its ascending pass, the satellite reaches the CDN1 transponder Cal/Val site where it is also calibrated (Figure 15). The bias in sea-surface height and range for Sentinel-3A along its Pass No. 14 are given in Figure 16 for all available passes at the time of writing.

Sentinel-3A Pass No. 14 Cal/Val Results
Τhe ascending Sentinel-3A Pass No.14 approaches the Gavdos island from the south. Calibration over sea surface is carried out using the Gavdos Cal/Val facility. About 9 seconds later, on its ascending pass, the satellite reaches the CDN1 transponder Cal/Val site where it is also calibrated (Figure 15). The bias in sea-surface height and range for Sentinel-3A along its Pass No.14 are given in Figure 16 for all available passes at the time of writing.  The sea-surface bias for Sentinel-3A Pass No.14 is calculated as −11.98 mm, whereas the range bias with the transponder is determined as +4.90 mm. These results indicate that the altimeter measures larger ranges than the "ground reference" by +4.90 mm with the present settings. The negative value of −11.98 mm in the SSH bias seems to support both Cal/Val results. The absolute difference between the two Cal/Val results is less than 1 cm, whereas the FRM uncertainty of each independent technique is of the order of ±3 cm.

Sentinel-3A Pass No. 278 and No. 335 Cal/Val Results
Two sea-surface Cal/Val sites support the calibration of the descending Pass No.335 of Sentinel-3A ( Figure 17). As the satellite descends, it meets first RDK1 Cal/Val where it is calibrated, and then about 6 seconds later it is again calibrated with the other CalVal site of Gavdos. The results obtained by these two Cal/Val sites agree at the sub-centimeter level. At this stage, the RDK1 Cal/Val results present somewhat larger dispersion compared to Gavdos Cal/Val. Larger scattering for the RDK1 Cal/Val may be caused by the employed calibrating region, covering a 35 km stretch (1000 km depth) of ocean water between Crete and Gavdos. That water stretch is shorter in length than that of The sea-surface bias for Sentinel-3A Pass No. 14 is calculated as −11.98 mm, whereas the range bias with the transponder is determined as +4.90 mm. These results indicate that the altimeter measures larger ranges than the "ground reference" by +4.90 mm with the present settings. The negative value of −11.98 mm in the SSH bias seems to support both Cal/Val results. The absolute difference between the two Cal/Val results is less than 1 cm, whereas the FRM uncertainty of each independent technique is of the order of ±3 cm.

Sentinel-3A Pass No. 278 and No. 335 Cal/Val Results
Two sea-surface Cal/Val sites support the calibration of the descending Pass No.335 of Sentinel-3A ( Figure 17). As the satellite descends, it meets first RDK1 Cal/Val where it is calibrated, and then about 6 s later it is again calibrated with the other CalVal site of Gavdos. The results obtained by these two Cal/Val sites agree at the sub-centimeter level. At this stage, the RDK1 Cal/Val results present somewhat larger dispersion compared to Gavdos Cal/Val. Larger scattering for the RDK1 Cal/Val may be caused by the employed calibrating region, covering a 35 km stretch (1000 km depth) of ocean water between Crete and Gavdos. That water stretch is shorter in length than that of the Gavdos south calibrating region (with 3000 m depth), and also fewer uncontaminated by land values are available for the satellite calibration (mountains of 2500 m height in the north of RDK1). Several updates and improvements of the marine geoid and the mean dynamic topography models in support of the RDK1 Cal/Val are currently ongoing. In the southwest tip of Crete, the CRS1 Cal/Val site provides support for the sea-surface calibration of the descending pass No. 278 of Sentinel-3A. Figure 18 presents the latest results for the Sentinel-3A. The estimated value for the bias in the sea-surface heights is found to be −3.68 mm based on this CRS1 Cal/Val site. In the southwest tip of Crete, the CRS1 Cal/Val site provides support for the sea-surface calibration of the descending pass No. 278 of Sentinel-3A. Figure 18 presents the latest results for the Sentinel-3A. The estimated value for the bias in the sea-surface heights is found to be −3.68 mm based on this CRS1 Cal/Val site. sea-surface calibration site located south of Crete site of RDK1 (upper) and the Gavdos Cal/Val sites (lower).
In the southwest tip of Crete, the CRS1 Cal/Val site provides support for the sea-surface calibration of the descending pass No. 278 of Sentinel-3A. Figure 18 presents the latest results for the Sentinel-3A. The estimated value for the bias in the sea-surface heights is found to be −3.68 mm based on this CRS1 Cal/Val site. In summary, the bias in sea-surface height for Sentinel-3A has been determined to be on average −8.04 mm (±5.5 mm, ±32 mm FRM uncertainty) along its descending pass No.335 based on Gavdos and RDK1 Cal/Val sites, while it is found to be −3.68 mm (±4 mm, ±32 mm FRM uncertainty) In summary, the bias in sea-surface height for Sentinel-3A has been determined to be on average −8.04 mm (±5.5 mm, ±32 mm FRM uncertainty) along its descending pass No. 335 based on Gavdos and RDK1 Cal/Val sites, while it is found to be −3.68 mm (±4 mm, ±32 mm FRM uncertainty) along another descending Pass of No. 278 and using the other CRS1 Cal/Val site. It appears that the bias of Sentinel-3A in sea-surface height, as monitored at the PFAC: • Displays a negative sign for all these calibrating passes; • Shows an average bias of −6.82mm (±32 mm FRM uncertainty) based on all three Cal/Val sites with descending orbits; • Has no pass directional error in SSH bias as there is not significant trend at all between ascending and descending passes (−11.98 mm for Pass No. 14 and −8.04 mm for Pass No. 335).

Sentinel-3B Cal/Val
Three Sentinel-3B passes are calibrated at the PFAC. Calibration over sea surface is performed for the ascending Passes No. 14 and No. 71 at the CRS1 and RDK1 Cal/Val sites, respectively, using the latest processing baseline for Sentinel-3B (PB 1.33). The implemented calibration regions for Sentinel-3B are shown in Figure 19. The results refer to the Operational Phase of Sentinel-3B, which is when the satellite has been on its nominal orbit. Calibration results obtained during the Sentinel-3B Tandem Phase are given separately in Section 4.3.

Sentinel-3B Cal/Val
Three Sentinel-3B passes are calibrated at the PFAC. Calibration over sea surface is performed for the ascending Passes No. 14 and No. 71 at the CRS1 and RDK1 Cal/Val sites, respectively, using the latest processing baseline for Sentinel-3B (PB 1.33). The implemented calibration regions for Sentinel-3B are shown in Figure 19. The results refer to the Operational Phase of Sentinel-3B, which is when the satellite has been on its nominal orbit. Calibration results obtained during the  The range bias of Sentinel-3B is calculated to be B(S3B) = −3.46 mm (±3.50 mmm, ±35 mm FRM uncertainty). This value displays an opposite sign with respect to the previous Cal/Val results of Sentinel-3A (Section 4.1.1, B(S3A) = +4.90 mm). In both satellites, the absolute value of their range bias appears to be about ±5 mm: This means that Sentinel-3A and Sentinel-3B seem to measure altimetric distances of about 814 km as accurately as ±5 mm based on observations of this transponder.

Sentinel-3B Cal/Val results over sea surface
The CRS1 and RDK1 sea-surface Cal/Val sites provide ocean truth measurements for the calibration of the ascending Sentinel-3B Pass No. 14 and Pass No. 71, respectively. Figure 21 presents the calibration results of Sentinel-3B over sea surface as derived by these two Cal/Val sites. The range bias of Sentinel-3B is calculated to be B(S3B) = −3.46 mm (±3.50 mmm, ±35 mm FRM uncertainty). This value displays an opposite sign with respect to the previous Cal/Val results of Sentinel-3A (Section 4.1.1, B(S3A) = +4.90 mm). In both satellites, the absolute value of their range bias appears to be about ±5 mm: This means that Sentinel-3A and Sentinel-3B seem to measure altimetric distances of about 814 km as accurately as ±5 mm based on observations of this transponder.

Sentinel-3B Cal/Val Results Over Sea Surface
The CRS1 and RDK1 sea-surface Cal/Val sites provide ocean truth measurements for the calibration of the ascending Sentinel-3B Pass No. 14 and Pass No. 71, respectively. Figure 21 presents the calibration results of Sentinel-3B over sea surface as derived by these two Cal/Val sites. The range bias of Sentinel-3B is calculated to be B(S3B) = −3.46 mm (±3.50 mmm, ±35 mm FRM uncertainty). This value displays an opposite sign with respect to the previous Cal/Val results of Sentinel-3A (Section 4.1.1, B(S3A) = +4.90 mm). In both satellites, the absolute value of their range bias appears to be about ±5 mm: This means that Sentinel-3A and Sentinel-3B seem to measure altimetric distances of about 814 km as accurately as ±5 mm based on observations of this transponder.

Sentinel-3B Cal/Val results over sea surface
The CRS1 and RDK1 sea-surface Cal/Val sites provide ocean truth measurements for the calibration of the ascending Sentinel-3B Pass No. 14 and Pass No. 71, respectively. Figure 21 presents the calibration results of Sentinel-3B over sea surface as derived by these two Cal/Val sites.  The bias of Sentinel-3B for the sea-surface height is +4.00 mm (±7.50 mm, ±32 mm FRM uncertainty) based on both ascending Pass No. 14 (B(S3B) = +4.55 mm, CRS1 Cal/Val) and Pass No. 71 (B(S3B) = +4.85 mm, RDK1 Cal/Val). As expected, the sign of the sea-surface height bias is opposite to the transponder range bias, while it appears that their magnitude agrees at the millimeter level.

Jason-3 Cal/Val Results
Jason-3 occupies a specific reference orbit for the multi-mission altimetry constellation. It flies on exactly the same ground-track as its predecessors TOPEX/Poseidon, Jason-1 and Jason-2, its repeat cycle is 10 days, and its measurements have provided accurate and consistent time-series for nearly three decades. In west Crete, the Gavdos Island is located under a crossover of Jason ground tracks of Pass No. 18 (descending) and Pass No. 109 (ascending) (Figure 22). The bias of Sentinel-3B for the sea-surface height is +4.00 mm (±7.50 mm, ±32 mm FRM uncertainty) based on both ascending Pass No. 14 (B(S3B) = +4.55 mm, CRS1 Cal/Val) and Pass No. 71 (B(S3B) = +4.85 mm, RDK1 Cal/Val). As expected, the sign of the sea-surface height bias is opposite to the transponder range bias, while it appears that their magnitude agrees at the millimeter level.

Jason-3 Cal/Val Results
Jason-3 occupies a specific reference orbit for the multi-mission altimetry constellation. It flies on exactly the same ground-track as its predecessors TOPEX/Poseidon, Jason-1 and Jason-2, its repeat cycle is 10 days, and its measurements have provided accurate and consistent time-series for nearly three decades. In west Crete, the Gavdos Island is located under a crossover of Jason ground tracks of Pass No.18 (descending) and Pass No. 109 (ascending) (Figure 22). Range as well as sea-surface calibrations are carried out for every cycle of Jason-3 Pass No. 18 using the CDN1 and Gavdos Cal/Val sites, respectively. First, the satellite passes over the CDN1 transponder Cal/Val site on the west Crete mountains and about 9 seconds later it reaches the Gavdos island in the south. Figure 23 presents the latest results in range and sea-surface calibration for Jason-3 over its descending Pass No.18.

Jason-3 Cal/Val Results (No. 18)
Range as well as sea-surface calibrations are carried out for every cycle of Jason-3 Pass No. 18 using the CDN1 and Gavdos Cal/Val sites, respectively. First, the satellite passes over the CDN1 transponder Cal/Val site on the west Crete mountains and about 9 s later it reaches the Gavdos island in the south. Figure 23 presents the latest results in range and sea-surface calibration for Jason-3 over its descending Pass No. 18. The sea-surface bias of Jason-3 is determined to be B(JA3) = −4.34 mm, whereas the transponder range bias is B(JA3) = +5.30 mm. As already described, the sign of the Jason-3 bias derived from the transponder and the sea-surface calibrations is opposite. Both calibrations result in the same absolute value of bias, thus supporting the reliability of the derived results.

Jason-3 Cal/Val Results (No. 109)
The ascending Pass No. 3 of Jason-3 has been calibrated over the sea surface using the Gavdos Cal/Val site. The latest calibration results for Jason-3, along the sea surface of Pass No. 109, with GDR-D data and for about 150 cycles, are given in Figure 24.  The ascending Pass No. 3 of Jason-3 has been calibrated over the sea surface using the Gavdos Cal/Val site. The latest calibration results for Jason-3, along the sea surface of Pass No. 109, with GDR-D data and for about 150 cycles, are given in Figure 24.  Table 6 summarizes the contribution of each PFAC Cal/Val site to the determination of the altimeter bias for Sentinel-3A, Sentinel-3B and Jason-3. Interpretation of these Cal/Val results per satellite follows at the end of Section 4. The Gavdos sea-surface Cal/Val site appears to produce consistent results per satellite altimeter. Two different passes in ascending and descending mode of Sentinel-3A and of Jason-3 are supported by the Gavdos Cal/Val facility. It seems that there is no significant difference for ascending and descending passes of Sentinel-3A and Jason-3. Moreover, the difference in bias results between transponder and sea-surface calibrations using the same orbit and with calibration taken place a few seconds apart, agree within ±1 cm for Sentinel-3A (Pass No. 14, ascending) and for Jason-3 (Pass No. 18, descending). This setting clearly advocates the FRM requirements for having diverse, redundant and independent methodologies to assess performance of satellite altimeters.   Over the sea surface and during tandem, the bias difference of Sentinel-3B with respect to Sentinel-3A is calculated to be +20.66 mm (average). This implies that Sentinel-3B seems to measure sea-surface height by +2 cm higher than Sentinel-3A, at least during their tandem phase. Nonetheless, this result is based only on very few cycles and thus cannot be statistically significant but only indicative. Over the sea surface and during tandem, the bias difference of Sentinel-3B with respect to Sentinel-3A is calculated to be +20.66 mm (average). This implies that Sentinel-3B seems to measure sea-surface height by +2 cm higher than Sentinel-3A, at least during their tandem phase. Nonetheless, this result is based only on very few cycles and thus cannot be statistically significant but only indicative.
During the tandem Sentinel-3A and Sentinel-3B phase, transponder calibrations were carried out at the CDN1 Cal/Val site ( Figure 27). One of them was not successful, as Sentinel-3B's altimeter was operating in Low Resolution Mode. Here also, it can be seen that Sentinel-3A seems to measure a slightly shorter range than Sentinel-3B in that short transponder Cal/Val analysis. Over the sea surface and during tandem, the bias difference of Sentinel-3B with respect to Sentinel-3A is calculated to be +20.66 mm (average). This implies that Sentinel-3B seems to measure sea-surface height by +2 cm higher than Sentinel-3A, at least during their tandem phase. Nonetheless, this result is based only on very few cycles and thus cannot be statistically significant but only indicative.
During the tandem Sentinel-3A and Sentinel-3B phase, transponder calibrations were carried out at the CDN1 Cal/Val site ( Figure 27). One of them was not successful, as Sentinel-3B's altimeter was operating in Low Resolution Mode. Here also, it can be seen that Sentinel-3A seems to measure a slightly shorter range than Sentinel-3B in that short transponder Cal/Val analysis. Both results in sea surface and transponder analysis corroborate each other during the tandem phase of Sentinel-3A and Sentinel-3B. The analysis continues with crossover Cal/Val results.

Cal/Val Results at Crossovers
The groundtracks of the Sentinel-3A and Jason-3 intersect at a point (Lat = 34.63893946, Long = 23.98522449) about 20 km south of Gavdos island ( Figure 3). To compare sea-surface heights observed by these two satellites at that crossover location with some confidence, their flyover had to be separated by no more than 2 days.
Between March 2016 and March 2020, this condition was fulfilled 19 times: 9 times the temporal separation of Sentinel-3A and Jason-3 was 1 day and 2 days (48-hours) for the remaining cases. For each pair of altimeters, linear regression was carried out on orbital data to accurately determine the geodetic coordinates of their crossover location at sea. Then, the exact SSH value at the crossover location for each satellite altimeter was computed, taking into consideration the geoid models and Both results in sea surface and transponder analysis corroborate each other during the tandem phase of Sentinel-3A and Sentinel-3B. The analysis continues with crossover Cal/Val results.

Cal/Val Results at Crossovers
The groundtracks of the Sentinel-3A and Jason-3 intersect at a point (Lat = 34.63893946, Long = 23.98522449) about 20 km south of Gavdos island ( Figure 3). To compare sea-surface heights observed by these two satellites at that crossover location with some confidence, their flyover had to be separated by no more than 2 days.
Between March 2016 and March 2020, this condition was fulfilled 19 times: 9 times the temporal separation of Sentinel-3A and Jason-3 was 1 day and 2 days (48-h) for the remaining cases. For each pair of altimeters, linear regression was carried out on orbital data to accurately determine the geodetic coordinates of their crossover location at sea. Then, the exact SSH value at the crossover location for each satellite altimeter was computed, taking into consideration the geoid models and the satellite direction in the region of crossover. Accordingly, 10 sampling points astride the crossover location have been used. These 20 altimeter observation points correspond to a distance of about 5 km along the satellite ground track. Figure 28 presents the outcome of this crossover analysis. It can be seen that the dispersion of the relative bias is of the order of ±4 cm but the average bias has been calculated to be +1.7 cm. This implies that Sentinel-3A measures the SSH +1.7 cm higher compared to Jason-3.
In addition, it seems that the uncertainty of this simplified and localized version of the multi-mission crossover analysis is similar to the FRM uncertainty (Tables 4 and 5) of the sea-surface and transponder calibrations. Note here that in crossover analysis no in-situ measurements were used. the relative bias is of the order of ±4 cm but the average bias has been calculated to be +1.7 cm. This implies that Sentinel-3A measures the SSH +1.7 cm higher compared to Jason-3.
In addition, it seems that the uncertainty of this simplified and localized version of the multi-mission crossover analysis is similar to the FRM uncertainty (Tables 4 and 5) of the sea-surface and transponder calibrations. Note here that in crossover analysis no in-situ measurements were used. This section presented the latest calibration results for Sentinel-3A, Sentinel-3B and Jason-3 satellite altimeters as determined by diverse techniques at the PFAC. Table 7 provides a concise summary of these Cal/Val results. The results of this investigation can be summarized as follows: (1) the bias in sea-surface height determination by the satellite altimeters is within their specification of ±1.5-2.0 cm [17], (2) no This section presented the latest calibration results for Sentinel-3A, Sentinel-3B and Jason-3 satellite altimeters as determined by diverse techniques at the PFAC. Table 7 provides a concise summary of these Cal/Val results.
The results of this investigation can be summarized as follows: (1) the bias in sea-surface height determination by the satellite altimeters is within their specification of ±1.5-2.0 cm [17], (2) no significant drift in altimeters' performance has been monitored, (3) confidence on calibration results has been improved as diverse and independent Cal/Val techniques produce similar results, (4) the difference between sea-surface and transponder Cal/Val methodologies is within their FRM uncertainty, (5) Sentinel-3A measures sea-surface height higher than Jason-3, and (6) Sentinel-3B measures sea-surface height higher compared to Sentinel-3A.

Discussion
This Section reviews calibration results of Section 4 in perspective of other Cal/Val works. It also elaborates upon future research and infrastructure upgrading, scheduled to take place at the PFAC. More specifically, the Sentinel-3 Mission Performance Center (S3MPC) has recently assessed the performance of Sentinel-3A and Sentinel-3B using independent processing of the CDN1 transponder data along with sea-surface calibration results using the Corsica, France permanent calibration facilities [32]. These S3MPC calibration results are compared against the range and sea-surface calibration results obtained at the PFAC. Furthermore, the on-going upgrading of the PFAC as a range and sigma-naught transponder facility is also presented.

Cross-Examination of Transponder Cal/Val Results
The S3MPC publishes the Sentinel-3 altimeter Cyclic Performance Reports. These reports assess the performance of the scientific instruments on-board Sentinel-3A and Sentinel-3B. They are available online [33]. The S3MPC report for the Cycle No. 56 of the SRAL altimeter of Sentinel-3A contains Sentinel-3A and Sentinel-3B range bias results using the CDN1 Cal/Val transponder. These results, as published by the S3MPC, have been determined independently to the PFAC methodology presented in Section 2. Figure 29 illustrates the range bias results as calculated by the S3MPC and the PFAC for Sentinel-3A Cycles 3-56 and Sentinel-3B Cycles 11-35 as of their launch dates: 16-Feb-2016 and 25-April-2018, respectively.
Remote Sens. 2020, 12, x FOR PEER REVIEW 33 of 39 significant drift in altimeters' performance has been monitored, (3) confidence on calibration results has been improved as diverse and independent Cal/Val techniques produce similar results, (4) the difference between sea-surface and transponder Cal/Val methodologies is within their FRM uncertainty, (5) Sentinel-3A measures sea-surface height higher than Jason-3, and (6) Sentinel-3B measures sea-surface height higher compared to Sentinel-3A.

Discussion
This Section reviews calibration results of Section 4 in perspective of other Cal/Val works. It also elaborates upon future research and infrastructure upgrading, scheduled to take place at the PFAC. More specifically, the Sentinel-3 Mission Performance Center (S3MPC) has recently assessed the performance of Sentinel-3A and Sentinel-3B using independent processing of the CDN1 transponder data along with sea-surface calibration results using the Corsica, France permanent calibration facilities [32]. These S3MPC calibration results are compared against the range and sea-surface calibration results obtained at the PFAC. Furthermore, the on-going upgrading of the PFAC as a range and sigma-naught transponder facility is also presented.

Cross-examination of Transponder Cal/Val Results
The S3MPC publishes the Sentinel-3 altimeter Cyclic Performance Reports. These reports assess the performance of the scientific instruments on-board Sentinel-3A and Sentinel-3B. They are available online [33]. The S3MPC report for the Cycle No. 56 of the SRAL altimeter of Sentinel-3A contains Sentinel-3A and Sentinel-3B range bias results using the CDN1 Cal/Val transponder. These results, as published by the S3MPC, have been determined independently to the PFAC methodology presented in Section 2. Figure 29 illustrates the range bias results as calculated by the S3MPC and the PFAC for Sentinel-3A Cycles 3-56 and Sentinel-3B Cycles 11-35 as of their launch dates: 16-Feb-2016 and 25-April-2018, respectively. The range bias of Sentinel-3A and Sentinel-3B has been determined by the S3MPC to be +6.79 mm and −6.59 mm, respectively [29]. The same biases as presented under FRM methodology in this investigation (Table 7) are, respectively, +4.90 mm and −3.46 mm. Thus, there is an agreement of the two transponder processing techniques at the ±2 mm level. This is within the expected uncertainty of the different and independent methods of data analysis used.

Cross-Examination of Sea-Surface Calibration Results
The Corsica, France calibration facility calculated the sea-surface height bias of Sentinel-3A and Sentinel-3B as +16 mm (S3A) and +19 mm (S3B), respectively [29]. The mean values of all passes calibrated at the PFAC have been (Table 7) −7.93 mm (S3A) and +4.70 mm (S3B), respectively. The deviation (S3MPC minus PFAC) between the calibration results is of the order of +2.5 cm for Sentinel-3A and +1.5 cm for Sentinel-3B. This discrepancy is within the reported FRM uncertainty of the PFAC sea-surface calibration results (±34.5 mm, Table 5).
There are several other reasons that may explain the discrepancy between PFAC and S3MPC such as: (1) the S3MPC Cal/Val results are based on processing baseline No. 2.33 to 2.45 while the PFAC applied the latest PB.2.61, (2) Sentinel-3A Cal/Val results from the S3MPC refer to cycles 1 to 47, but this work is applied till cycle 56, (3) the S3MPC Cal/Val results for Sentinel-3B cover four cycles during its tandem phase with Sentinel-3A and not under its nominal orbit as determined for these PFAC results, and (4) in contrast to transponder Cal/Val where S3MPC used the same data (CDN1 transponder) as the PFAC, sea-surface calibration is carried out at different geographic locations, with different instrumentation and processing methodologies.

A Second Transponder in Gavdos
The European Space Agency FRM for the Sentinel-6 (FRM4S6) project is developing a prototype transponder to be installed and operated as part of the PFAC at Gavdos, called GVD1 Cal/Val, to ensure redundancy (FRM standard requirement) in transponder calibrations for Sentinel-6. This transponder is specifically responding to the need of Copernicus Sentinel-6. The second transponder is to calibrate not only the range but also the backscatter coefficient (sigma-naught, σ 0 ) of radar altimetry.
The backscatter coefficient (sigma-naught, σ 0 ) is a parameter that is useful for all Earth observing radars. It controls the efficiency of imaging radars when land deformations are observed at the mm level. It is used in radar scatterometers to determine wind speed and direction on the ocean surface and moisture determination in soil. It is also applied to radar altimeters to provide accurate measurements of the sea level, wave heights and wind speed over the ocean. Some ocean and meteorological models make use of satellite altimeter-derived wind speed and significant wave height observations. Indeed, the longest and most complete record of global ocean sea state is derived from satellite altimetry. This transponder will be established at the GVD1 Cal/Val site under a triple crossover of Sentinel-6 (descending Pass No. 18, and ascending No. 109) and Sentinel-3A (descending Pass No.335) ( Figure 30). The future research direction of the PFAC is to provide two alternative, independent techniques to perform range and sigma-naught calibration primarily for the upcoming Sentinel-6 [34]. Transponder and sea-surface Cal/Val is already carried at the PFAC, whereas the GVD1 transponder Cal/Val is to operate in late 2020 ( Figure 31). The future research direction of the PFAC is to provide two alternative, independent techniques to perform range and sigma-naught calibration primarily for the upcoming Sentinel-6 [34]. Transponder and sea-surface Cal/Val is already carried at the PFAC, whereas the GVD1 transponder Cal/Val is to operate in late 2020 ( Figure 31).

Conclusions
This work has provided a thorough assessment of the performance of Sentinel-3A, Sentinel-3B and Jason-3 satellite altimeters using the Permanent Facility for Altimetry Calibration of the ESA. Performance is monitored thorough land transponder and sea surface techniques at different locations and settings, as well as crossover analysis. Cal/Val results are accompanied by their uncertainty as calculated in the context of the new ESA strategy of Fiducial Reference Measurements.
The performance of all satellites appears to be within their design requirements, exhibiting altimetry biases of ±1.5-2.0 cm. Figure 32 shows all results in a boxplot where all satellite altimeter biases, as presented in Table 7, are compared. It appears that transponder Cal/Val results in Sentinel-3A and Sentinel-3B are less dispersed (inter quantile range is smaller) than those in Jason-3. This may be attributed to the SAR mode of Sentinel-3 on one hand, but also to the implementation of Level-0 data in SAR calibration instead of Level-2 data carried out in Jason-3. Evaluation of the transponder and sea-surface calibration methodologies indicates that they both present an FRM uncertainty of the order of ±3.0-3.5 cm.

Conclusions
This work has provided a thorough assessment of the performance of Sentinel-3A, Sentinel-3B and Jason-3 satellite altimeters using the Permanent Facility for Altimetry Calibration of the ESA. Performance is monitored thorough land transponder and sea surface techniques at different locations and settings, as well as crossover analysis. Cal/Val results are accompanied by their uncertainty as calculated in the context of the new ESA strategy of Fiducial Reference Measurements.
The performance of all satellites appears to be within their design requirements, exhibiting altimetry biases of ±1.5-2.0 cm. Figure 32 shows all results in a boxplot where all satellite altimeter biases, as presented in Table 7, are compared. It appears that transponder Cal/Val results in Sentinel-3A and Sentinel-3B are less dispersed (inter quantile range is smaller) than those in Jason-3. This may be attributed to the SAR mode of Sentinel-3 on one hand, but also to the implementation of Level-0 data in SAR calibration instead of Level-2 data carried out in Jason-3. Evaluation of the transponder and sea-surface calibration methodologies indicates that they both present an FRM uncertainty of the order of ±3.0-3.5 cm. Remote Sens. 2020, 12, x FOR PEER REVIEW 37 of 39 Calibration results derived by these two calibration techniques as well as relative calibration given by the crossover analysis all agree to within ±1 cm. Sentinel-3 results are in agreement with results given by other Cal/Val groups, as previously described.
A major upgrade of the PFAC is currently under way. The target missions for this altimetry Cal/Val service would be primarily the Copernicus Sentinel-6 and Sentinel-3, while other international missions, such as Jason-3, HY-2, SWOT, CryoSat-2, etc., may benefit as well.