1. Introduction
For many years the precise determination of the geoid has been one of the basic problems of physical geodesy. The geoid is the equipotential surface that describes the irregular shape of the Earth and is used as the zero reference for elevation measurements. This would coincide with the Mean Sea Level (MSL), provided that dynamic factors (wind, atmospheric pressure, tide, sea currents, etc.) were to be excluded—that is, if oceans and atmosphere were to be considered in equilibrium. If we extended the MSL below the continents by means of hypothetical canals, we could obtain the whole image of the geoid surface. The precise knowledge of the geoid is very important in several applications, e.g., in investigating petroleum deposits. It is worth noting that the altitudes on the geoid are estimated with respect to the ellipsoid Earth. The methods for determining the geoid can be classified into indirect and direct ones.
Traditionally, the geoid has been determined indirectly either by gravitational or by astro-geodetic measurements in combination with various mathematical transformations (such as the Stokes equations). However, these calculations involve a great amount of uncertainty that can be up to several meters.
The more recent direct methods for geoid determination include marine surface measurements by altimetry satellites [
1] and airplane laser scanners [
2], and “GNSS on floating means” measurements [
3]. The first two methods face double uncertainty, since the calculation of the Global Navigation Satellite System (GNSS) position is required in addition to the distance estimation of the altimetry satellite or plane from the sea surface. Furthermore, they face significant problems in coastal areas and closed seas due to land obstacles. To address these problems, a new method based on direct GNSS measurement on the surface of the sea has been developed. During the decade 2003–2013, a number of research groups have proposed “on-ship” GNSS measurements for the determination of sea surface topography, contributing to the development of this method [
4,
5,
6,
7,
8,
9,
10,
11,
12,
13,
14].
The installation of one or more GPS/GNSS receivers on a navigable medium was based on earlier research efforts involving GPS/GNSS placement on a buoy [
15,
16,
17]. These buoys used GPS/GNSS to measure the dynamic characteristics of the waves and the average sea level of a particular point. Thus, the topography of a marine area could not be assessed. The GPS/GNSS buoy methodology was developed mainly for the construction of early tsunami warning systems [
18,
19].
The goal of this work was the determination of the mean surface topography (MST) of the Ionian and Adriatic Seas. This was attempted by developing an alternative methodology in the context of the “GNSS on floating means” techniques. The methodology followed in the present work is highly promising for various reasons. Firstly, the exact knowledge of the shape of the sea surface may provide quite useful information concerning climate change. It is well known that in most cases the investigation concerning the effect of global warming on the sea level is limited to certain places of the coast, where tide gauges are installed [
20]. However, this approach does not allow investigating the important effects of global warming on the shape of the sea surface. Those effects, to the extent of the author’s knowledge, remain for the moment largely unknown. This issue could be investigated through studies like the present one, repeated at regular intervals. Secondly, an obvious advantage of the present method with respect to the methods based on altimeter satellites is that the hundreds of thousands of passenger and cargo ships traveling worldwide provide a continuous grid of potential measurements. It is actually simpler to place a GNSS on a ship than to launch an altimeter satellite. Moreover, the satellites measure the sea surface at specific points, only along their orbit (
Figure 1), whereas the method “GNSS on ship” allows one to identify possible small changes or anomalies anywhere on the sea surface and to focus at specific points not covered by satellites. Furthermore, the methods based on satellites fail to provide good accuracy in coastal areas [
21]. In these areas, however, the determination of the average marine topography and the geoid is of great scientific interest [
22]. The detection of significant changes in specific parts of the marine topography could reveal interesting geological structures of the marine subsoil related to fissures or oil and gas fields [
23]. This perspective is very important, especially for the study area, because it presents very high seismicity but also strong indications for the existence of hydrocarbon deposits [
24].
The region under consideration extends from the Gulf of Patras to the northern Ionian Sea and the southern Adriatic Sea up to the port of Brindisi. It covers 250 km north–south and 150 to 200 km east–west. A schematic representation of the region studied is illustrated in
Figure 1 (red region) together with the trajectories of the TOPEX/POSEIDON, Jason-1, Jason-2, Envisat and ERS-2 satellites in the eastern Mediterranean region.
The mean sea level in the region studied, approached satisfactorily by the estimates of the geoid, has changed considerably, creating differences in height of several meters, in some cases larger than 10 m. For example, the EGM96 model estimates a sea level in the area near to the port of Brindisi as 14.2 m higher than that estimated in the sea area near the port of Patras, with respect to the ellipsoid reference [
25]. The study of surface sea surveying in the above area following the new approach is of interest for several reasons. Specifically, the few tide recording stations do not allow developing a clear picture concerning the eventual changes in sea level due to medium-term meteorological factors or to even more permanent fluctuations due to climate change. Thus, in many cases we cannot know whether the sinking or rising shores observed in the area are due to geological causes, meteorological fluctuations or both. Additionally, the trajectories of the TOPEX/POSEIDON, Jason-1, Jason-2, Envisat and ERS-2 altimetry satellite missions intersect over the area studied (
Figure 1, red). However, the estimates of altitude satellites are subject to significant uncertainties ranging from five to ten centimeters in the assessment of marine topography near coastal areas or areas with extensive coastline [
26].
Although the idea of placing a GNSS receiver on a passenger ship for recording a single track has been suggested [
8], the technique presents significant weaknesses as far as the uncertainties are concerned [
3]. Developing an approach of repeated measurements on the same route offers the potential of significantly reducing uncertainties and improving the methodology. Thus, the main approach of the present contribution is to carry out multiple measurements (passes) for the same maritime route over several months in order to include all possible conditions. Adopting this approach is expected to lead to higher accuracy estimations for marine topography, especially when compared with older methodologies. In the present work we have used the “Patra – Igoumenitsa – Brindisi” line for developing our methodology, using measurements from the “Ionian Queen” passenger ship over a period of 6 months. Using our method, we studied the change of the sea level altitude along the ship’s route, to determine the sea surface topography and compare our estimates to those of the geoid EGM96 model.
Two alternative methods were used to resolve our data, differential GNSS (D-GNSS) and precise point positioning (PPP). The main difference between these methods is that the first one requires a fixed GNSS reference station at a short distance from the moving receiver, in order to remove the atmospheric noise from the data [
27]. This fixed station must be on land. The greater the distance from land, the poorer the accuracy. Thus, despite the fact that the D-GNSS method potentially provides data with much higher accuracy than any other method [
28], its use away from the shore is quite problematic. On the contrary, the PPP method does not require a fixed station and is ideal for marine topography. However, its accuracy still remains lower than D-GNSS, though it is constantly improving [
29]. In the present work, a combination of both methods was adopted, to constitute a model approach for studying marine topography.
3. Methodology
3.1. General Procedure for Calculating the Mean Sea Level (MSL)
The calculation of the sea altitudes is inevitably quite complex because it is based on multiple crossings for marine topography. The ultimate goal of the analysis of the time series which were received from the GNSS solutions was the determination of the mean sea level (MSL). In this section the methodology for calculating the MSL using GNSS receiver’s altitude measured at each moment is presented and analyzed.
For each separate route, the
measured geometric altitude H (i,t) of the receiver on the ship at time
t and position
i is defined by the equation:
where
: The height of the receiver relative to the sea surface.
: The effect of astronomical and meteorological tide.
: The effect of waves.
: The total error of the measurement (random and systematic) of the geometric altitude of the receiver.
: The requested (unknown) geometric altitude of the MSL at position i.
The above parameters refer to a position i at time t.
Figure 5 shows a schematic representation of the main parameters of Equation (1). It should be noted again that the parameter
H(t,i) is deduced from the GNSS recordings, whereas the parameter
hGNSS is calculated by means of geometric characteristics of the ships taking into account the draught of the ship. Moreover, the estimations of the effects of astronomical and meteorological factors of the tide were based on GNSS measurements in ports and static solutions in order to determine the geometric altitude
H (
i,t) in ports.
The determination of all parameters of Equation (1) and the estimation of their uncertainties allows approaching the desired altitude
Ψ(
i) together with its precision. This can be done for each route separately. The contribution of the present approach, however, is that it uses multiple paths and thus it can determine altitudes
Ψ(
i) more accurately than methods using single paths. To calculate a reliable altitude at each point
Ψ(
i) for each of the aforementioned parameters of Equation (1), we should into account the uncertainties which can affect the final results. For the calculation of these parameters, a step-by-step analysis was followed, which is described in the next sections. Starting with the parameter
hGNSS, it was calculated by means of the geometric characteristics of the ships and corrected by taking into account the draught of the ship (
Section 3.2). Continuing, the impact of astronomical tide, the meteorological factor of the tide and the estimation of the effect of waves (
Section 3.3) are presented. Finally, we are presenting our methodology relevant to the solution of the GNSS raw data for determining the
H(
i,t) values, following the differential GNSS (D-GNSS;
Section 3.4) and precise point positioning (PPP;
Section 3.5) methods.
Before the next subsections, it is important to note that the multiple-paths approach developed in this work had to address the following methodological problem: The ship, on each separate route, although driven by an automatic satellite navigation system (using GNSS), does not always follow an identical route. It deviates in the horizontal plane.
Figure 6 illustrates an indicative area, near to the port of Brindisi, with nine different crossings. It is seen that the different routes, although somewhat parallel, may vary by tens of meters. The deviation may be even greater. An example is illustrated in
Figure S1. It concerns eight different routes (11/2–19/2, route IDs 13–21,
Table S1). In view of the above for determining the altitude
H (
i,
t), we cannot use a point on an individual route (e.g., route 1,
Figure 6), but instead must use a “mean point” on the straight section that intersects almost vertically all the different paths at corresponding points. These “mean points” constitute the “principal path” illustrated in
Figure 4. It is important to note that the differences in the values of
H (
i,
t) determined at these corresponding points are actually negligible, and this allows us to draw conclusions for a broader zone around the “principal route.”
3.2. Correction of the Altitude of the Receiver Relative to the Sea Level (hGNSS) to Take into Account the Cargo Changes after Each Loading and Unloading of the Ship
As already mentioned, the GNSS receiver was stationary on the ship (
Figure 3). However, its altitude in relation to the sea level (light blue in
Figure 5) was changed at each port due to cargo changes after each loading and unloading. This required a suitable correction for each route.
The recordings of the draughts at positions A and B (
Table S2) were adjusted by means of a simple linear interpolation at the projection point of the GNSS receiver (see
Figure 3) according to the equation:
where
is the requested height of the receiver relative to the sea surface;
is the known height of the receiver relative to the ship’s bottom level horizontal keel;
is the measured ship draughts at control points A and B (see
Figure 3);
and
are horizontal distances from the A and B points of the depth-meters in relation to the projection point of the GNSS receiver on the AB section of the horizontal keel;
is the draught altitude at the intersection point corresponding to the projection point of the GNSS receiver on the AB section of the horizontal keel.
Figure 7 illustrates the geometric calculation of the antenna height relative to the plane of the sea surface.
Based on the accuracy of the A and B draught measurements, as provided by the ship’s instrument specifications, the error of the determined parameter
hGNSS has been estimated. The calculations are presented in
Appendix A and resulted to an error equal to 0.76589 cm.
3.3. Calculation of Impact of Astronomical Tide and Meteorological Factor of the Tide, htide: Static Calculation of Altitude in Ports
The determination of the impacts of the astronomical and meteorological factors of the tide,
, on the requested parameter,
Ψ(
i), allows removing their influence. To determine such an impact, we used to a static solution in the ports. The ship remained each day in the ports of Patras or Brindisi for a period of 4 h and this allowed solving statically the altitudes in all cases. The time period chosen for resolution (2.5 h) was less than 4 h in order to correspond to the time that the ship was completely empty and not in the process of loading or unloading. Taking into account the magnitude of the tide and the rate of its change in the area [
32] we considered that at this period the sea level remained constant, and thus a static solution could provide a reliable estimate of the altitude of the sea level in ports. The static solution of altitude in ports was performed following the PPP methodology.
Figure 8 illustrates the results of the static resolutions in both ports. The mean values and the standard deviation values are also illustrated. Representative results are also illustrated in
Table S3.
Since our measurements cover almost one full semester, and they take into account all possible weather conditions, which justifies the variations observed, we considered that the average sea level altitude (mean values in
Figure 8) is an estimate of the MSL in the ports (see
Figure 5). Thus, we have obtained two points of “control and correction” in the two ports that are the “extremes” of each route. The obtained MSL values and the individual values resulted from the static resolutions at both ports, and were used in order to correct the altitude estimation curves for each route (linear correction). Essentially, the effect of the tide,
h (i,t), was eliminated without using data from the tide gauges, given that they were not in operation. After this correction, our differential resolutions did not show any significant differences at a given point (
i) for different routes.
Recently, in 2017, we had the opportunity to confirm the aforementioned mean sea level at the port of Patra. In effect, navy services provided us the opportunity to calculate the coordinates of the benchmark GNSS. The elevation of the benchmark was linked to the sea-level recorder—which during our experiments was, unfortunately, out of order. However, the recent measurements have confirmed our calculation for the average value of sea topography at the port of Patras. This demonstrates the accuracy of the above described methodology for obtaining the mean sea level using the GNSS of the ship.
To confirm twice the static GNSS results above, our estimates of the sea level altitude in the ports were compared with the estimates of the DTU10 global ocean tide model. Tide model estimates are usually preferred but have the disadvantage of being indirect estimations of the sea level altitude. The DTU10 model’s differences from the estimations of our GNSS measurements were very small (<16 cm), except for three cases where the difference was greater, and those routes were removed. These differences can be considered negligible, if we consider that the estimated accuracy of the model is about 20–25 cm [
33].
Since the calculations of the impacts of the astronomical and meteorological components of tide were not done using modern time-series via tidal collectors, but using the limit values measured at ports, the uncertainty of this factor cannot be accurately estimated. However, the basic component of the astronomical tide (M2) expected in the area does not exceed a maximum range of 5–7 cm [
34]. Therefore, it seems plausible to assume that the astronomical tide is eliminated to such an extent so that its effect is not actually detectable. If, for example, despite the aforementioned linear correction, in some segments of the path, e.g., in the middle section, the astronomical tide adds or removes 1–2 cm from our estimation, it is very difficult to be detected. This is because the uncertainty estimated of each separate route covers this range. In other words, the astronomical component cannot contribute beyond this range.
In our previous works we examined in detail the relationship between meteorological and astronomical tides in some parts of the Mediterranean [
35]. The ratio of meteorological to astronomical tides in these areas is very high (up to 40). This makes the effect of the astronomical factor practically insignificant [
36]. At the same time, the most important meteorological factor in terms of its effect on the average sea level is completely random [
37].
In contrast, the meteorological component of tide is related too much wider ranges than those of the astronomical component; in extreme conditions up to 50 cm [
32]. However, the change of its amplitude per time unit is much smoother. Thus, the aforementioned linear correction is sufficient to confront the meteorological component of the tide. Obviously, the presence of noise cannot be excluded. However, the use of data of a 6-month period, during which almost all weather conditions occur, randomizes any uncertainty.
Finally, a few sentences concerning the effect of waves (h
wave): Reinking et al. [
8], faced with a problem similar to that of the present work, analyzed the dynamics of a ship with a similar size to the “Ionian Queen.” Although their work concerned a marine area with worse wave conditions (the Atlantic Ocean), the authors concluded that the dynamic movement of the ship is not actually affected by the vertical component of the gravitational ripple. Thus, the negligible displacement of a ship of the magnitude of “Ionian Queen” almost eliminates the effect of small-scale waves. Moreover, taking into account that our data cover a period of almost six months and more than 140 passes, as well as the high frequency of sampling on each route (1 Hz), it seems plausible to assume that the effect of the gravitational wave is zero, as it is randomized.
3.4. Solution of the GNSS Raw Data for Determining the H(i, t) Values Following the D-GNSS Method
The data obtained from the GNSS recordings were classified into separate routes illustrated in
Table S1. Each route corresponds to a passage of the ship in the line Patra–Igoumenitsa–Brindisi or Brindisi–Igoumenitsa–Patra. These 142 routes were solved by all six reference stations (
Figure 4). However, fifteen routes were not used because they could not be resolved by the base stations, as they showed unresolved uncertainties. Moreover, seven additional routes cannot be solved by the LECC station. Thus, 120 of the 142 routes were finally utilized. For improving the accuracy of the results, a static solution of the six reference stations and a correction of their coordinates were completed prior to carrying out the kinematic solution.
Figure 9 depicts the horizontal coordinates of the ship’s principal route as solved by the PONT-based D-GNSS method (see also
Figure 4).
The solutions were calculated using Leica Geo Office software (LGO). In order to confront the signal distortion due to the atmospheric delay and take into account the large length of the maritime route, a suitable ionospheric model was automatically adopted by the software. This was based on the updates of the orbital satellite corrections. A full ambiguity-fixed solution was chosen. Since the recordings of our data exceeded 45 min, the calculation of the ionospheric model was chosen from the LGO software. This option has the advantage that the calculations of the model are consistent with the particular conditions prevailing in the region. The software’s ability to calculate the troposphere model was also utilized. This option calculates variations of troposphere delay between the moving and the reference receivers. Different solutions were tested with respect to the degree of resolution of their uncertainties, and those segments in which the uncertainties had not been resolved were rejected.
The distances between the six reference stations range from a few to several hundred kilometers from the ship’s crossing points. According to the literature, the distances of the recording points from the reference stations in the D-GNSS method should not be greater than 50 km, though calculations have shown that a distance of 60 km does not lead to additional uncertainty greater than 2 cm [
38]. At higher distances, the noise may become quite large. This is exemplified in
Figure S2. It concerns an area in Patra Bay where the data have been resolved by different stations. It is obvious that the noise level increases with the distance of the station from the region mentioned. However, it does not exceed 5 cm. The aforementioned effect of distance is more clearly illustrated in the examples shown in
Figure 10.
It is evident that the noise level is generally increased with the distance of the base station from the part of the ship’s journey in focus. In this study, we have generally adopted the critical distance of 50 km in order to ensure high reliability. However, this distance is rather specific, as resolving can be done successfully over somewhat larger distances. An overview of the parts of the route situated inside the distance of the 50 km from a given base station is illustrated in
Figure 11. It is seen that the area south of Igoumenitsa and a large area between Corfu and Brindisi are located at distances greater than 50 km from base stations. In view of the above, we are accepting the individual solutions of a given base station, for a given part of the route on the basis of their proximity. For example, the individual solutions from the LECC station were used for the part of the route inside the upper–left circle of
Figure 11. From the parts of the route being outside the radius of 50 km, we have used the closest neighboring base station from a given point of the route.
An example of the solution following this approach is illustrated in
Figure 12.
It concerns the final configuration of a “combinatorial” solution using the segments that we consider as most appropriate for the path ID 71, taken as an example. It should be stressed that the synthesis of the solutions was not automated. It was done by examining the suitability of the individual solutions from each station with respect to the noise level. In this example (path ID 71), the solutions of the VLSM and SPAN stations were not used because in the parts c–e their signals had more noise than PONT. A comparison of the results illustrated in
Figure 12 with those illustrated in
Figure 10 and
Figure S3 clearly shows that the above described approach leads to significant decrease of the noise level.
To decrease the noise level further, we proceeded to a noise correction with the Savitzky–Golay filter. The filter is based on the repeated adjustment of a polynomial utilizing the least squares method [
39].
Figure 13 (top) illustrates the part of the Patra–Igoumenitsa route resolved by the RLSO station and the estimate of the moving average (red line). By subtracting the original signal from the estimates of the moving average,
Figure 13 (bottom) was obtained. The left-hand side of the figure (white) represents part of the route near the settlement station, justifying the quite small standard deviation. In contrast, the differences on the right-hand side of the figure (gray), representing part of the route at a significant distance from the settlement station, are characterized by larger standard deviation. This was to be expected, taking into account the considerations of the previous paragraph.
The technique of the Savitzky–Golay moving average filter does not only correct the noise of GNSS measurements, but also the wave dynamic effect. As already mentioned, the wave action on a big ship is practically negligible because it is eliminated by the dynamic behavior of the ship’s construction; only waves higher than 4 m and longer than 10 m can exert a significant effect [
8]. However, even in this case the action of the filter permanently eliminates this dynamic phenomenon.
3.5. Solution of the GNSS Raw Data for Determining the H(i,t) Values Following the PPP Method
As an alternative to the D-GNSS method, all route data were also resolved by the PPP method. The PPP method does not need reference stations, and thus it is at an advantage regarding the sections of the route that are at long distances (<50 km) from these stations. As the track logs were large in volume and could not be resolved with the current features of the editor, they were split into shorter records. In this way the routes were divided in half and doubled in their number. In
Figure 14 shows a solution with the PPP methodology. It concerns the two sections in which a given route has been divided.
The steps followed in the previously described D-GNSS analysis were also followed here except for base stations, as this is not necessary in the PPP method. These are illustrated in
Figure 15. The figure concerns the “Patra–Igoumenitsa“ route with code ID1, taken as an example. Three other examples concerning the routes with codes ID3, ID11 and ID141 are illustrated in the
Figures S3–S5.
In all cases, the curve in part A of each of the aforementioned figures describes the raw data. Three of the four examples concern the middle of winter (end of January, ID1, ID3 and ID11) with adverse weather conditions, while the fourth (ID141) concerns mid-summer with very smooth weather conditions (see
Table S1). The examples were chosen to render visible the large difference in the range of noise we encounter when the weather changes. It is evident that the range of noise is much greater in the passes with code numbers ID1, ID3 and ID11 illustrated in
Figure 15,
Figures S3 and S4 than that in the route ID141 illustrated in
Figure S5.
Part B of each of the aforementioned figures presents the curve with the original data; the curve of mean values (red curve) obtained by applying the Savitzky–Golay method [
38]; and the two light-blue curves, which indicate a band filter selected to have a total width of 1 m beyond which each value was deducted. It can be seen that the band filter does not remove any value, as the range of all values is smaller than the bandwidth, in the last example (
Figure S5B) referring to good weather conditions.
In part C, of each of the aforementioned figures, the
hGNSS receiver elevation correction at port check points described in
Section 3.2 is illustrated. The height of the receiver on the vessel measured at the ports was removed from the data in order to generate the height of the sea topography.
Finally, in part D of the aforementioned figures, the linear correction concerning the influence of astronomical and meteorological components of tide,
, which is based on calculations of altitude in ports described in
Section 3.3, is illustrated. With this correction, as explained earlier, the effect of the tide was removed.
3.6. Outline of the Analysis Procedure
In this section, the analysis procedure that has been followed to render it more comprehensible is briefly visualized. The basic steps of the methodology are illustrated in
Figure 16 in the form of a flowchart. The raw GNSS data are those obtained from ship’s routes (kinematic raw data) and those obtained from the ground stations. The first step involved static solutions in the base stations, whereas the second step was a kinematic solution of all routes from all base stations following the D-GNSS method. Then two different calculations were done in parallel. One concerned the application of the D-GNSS method, and the second, the application of the PPP method. Concerning the D-GNSS application, it was first examined whether the D-GNSS (LGO) software resolved the uncertainties relevant to a given route. If that had not happened, the results were discarded. For the routes where the uncertainties were resolved, a combinatorial solution was chosen (
Section 3.4,
Figure 10). After this selection, the procedure involved numerical filtering (
Section 3.4,
Figure 13), followed by the removal of static heights of the receiver (
Section 3.2, Equation (3),
Figure 7) and then the linear tidal correction described in
Section 3.3.
A parallel process was followed with the PPP method. The only exception was that the procedure did not involve the aforementioned combinatorial solution, but a simple solution from the beginning. Once all the paths were resolved, the ones that had been solved were selected. Then the procedure involved noise removal by numerical filtering, the removal of static heights of the receiver (
Section 3.2, Equation (3),
Figure 7) and then the linear tidal correction described in
Section 3.3 (see also
Figure 15 and
Figures S4–S6).
Having achieved the estimates of the Ψ(i) values from both methods, we proceeded to a final assessment of the marine topography by calculating the dynamic means both between the individual routes–resolutions and between the different methods. This allowed constructing the map of the mean marine topography for the area studied.
4. Estimation of the Mean Sea Level in the Area Studied
On the basis of what were described above, the final assessment of the mean marine topography in the studied area was achieved (
Figure 17).
This assessment was obtained while taking into account jointly the final calculations performed using the D-GNSS and PPP methods. It is worth noting that for each
Ψ(i) (line segment vertically intersecting all the different paths; see
Figure 6), the average of all estimates from all the crossings for both methods was calculated. When creating the map illustrated in
Figure 17, as an application area on the horizontal plane, we considered a zone that encompasses all the ship’s routes and covers an area of tens of meters up to a few kilometers on either side of the principal route. It is evident that we can distinguish very significant altitude differences of up to 16 m. Moreover, the marine topography in the region shows a nearly constant slope of 4 cm/km in the N–S direction.
A final estimation of the accuracy of the contour map (parameter u(i,t) in Equation (1)) was obtained while taking into account the following: (a) the accuracy of calculating the height of the
hGNSS receiver based on the draught measurements (
Section 3.2), (b) the mean range of the standard deviation of the differences of the kinematic data from the mean value, as it was derived by the moving average filter (e.g.,
Figure 13,
Figure 15B,
Figures S4B, S5B and S6B) and (c) the standard deviation of static estimations at ports (
Figure 8). Factors α and β were thought to contribute linearly to the final estimate of uncertainty. Concerning factor c, the linear correction described in the
Section 3.3 eliminates completely the factors of the meteorological component. However, it would be rather exaggerated to consider complete elimination of this parameter, as we cannot assess the tidal effect upon travel, except in the ports. Thus, adopting a more rigorous approach, we assumed an uncertainty of 2 cm as a representative value of the tide in this area. By taking into account the influence of factors a, b and c and using the well-known error transmission law, an accuracy of 3.31 cm was obtained for the MSL in the area studied (σ
hMSL = 3.31 cm).
Finally, the results of the proposed methodology are compared to those of the geoid model EGM96 (
Figure 18). Taking into account the accuracy of the EGM96′s own estimates and other estimates of the mean sea level, the differences are relatively small (up to 50 cm). The most significant difference is between points I and O, north of Corfu, where our estimate appears to be significantly lower than the estimates of EGM96, up to a maximum of 48 cm.