In this study, the quality of wave data provided by the
new Sentinel-3A satellite is evaluated and the sensitivity of the wave model
to wind forcing is tested. We focus on coastal areas, where altimeter data
are of lower quality and wave modelling is more complex than for the open
ocean. In the first part of the study, the sensitivity of the wave model to
wind forcing is evaluated using data with different temporal and spatial
resolution, such as ERA-Interim and ERA5 reanalyses, the European Centre for
Medium-Range Weather Forecasts (ECMWF) operational analysis and short-range
forecasts, German Weather Service (DWD) forecasts and regional atmospheric
model simulations (coastDat). Numerical simulations show that
the wave model forced using the ERA5 reanalyses and that forced using the
ECMWF operational analysis/forecast demonstrate the best capability over the
whole study period, as well as during extreme events. To further estimate the
variance of the significant wave height of ensemble members for different
wind forcings, especially during extreme events, an empirical orthogonal
function (EOF) analysis is performed. In the second part of the study, the
satellite data of Sentinel-3A, Jason-2 and CryoSat-2 are assessed in
comparison with in situ measurements and spectral wave model (WAM)
simulations. Intercomparisons between remote sensing and in situ observations
demonstrate that the overall quality of the former is good over the North Sea
and Baltic Sea throughout the study period, although the significant wave
heights estimated based on satellite data tend to be greater than the in situ
measurements by 7 to 26
Information on the state of the sea in coastal areas is of great interest, as
waves are a crucial factor for important activities conducted at sea.
Therefore, an accurate wave forecast and hindcast are very important for
marine traffic, recreational activities on the water, urban development near
the coast, ecosystem restoration, renewable energies and offshore management
In many studies, the meteorological input has already been found to be a
crucial factor for conducting good wave forecasts
Another way to increase the accuracy of the modelled significant wave height
is by assimilating the significant wave height measured by satellites into a
first-guess wave field
In the next section, the measured satellite and in situ data as well as the
wind forcing data and the numerical wave model used are described
(Sect.
Here, the ocean wave model WAM is forced using different meteorological input data to evaluate the sensitivity of the model to different wind input spatial and temporal resolutions. Therefore, the numerical model and wind input data used are introduced in this section. Information regarding the in situ measurements used here is also given. Furthermore, the satellite data, especially that of the new Sentinel-3A satellite, are presented.
In this study, wave height data derived from the Jason-2, CryoSat-2 and
Sentinel-3A altimeter missions are used. Jason-2 is a classical pulse-limited
altimeter operating in low-resolution mode (LRM) that was in operation, with
a revisiting time of 10 days, from June 2008 to October 2016
(
The CryoSat-2 satellite, launched in April 2010, is the first space-borne
instrument with synthetic aperture radar (SAR) capabilities. It can operate
in one of three modes, i.e. SAR mode, interferometric SAR (SARIn) mode and
low-rate mode (LRM), following a geographical mask, which is regularly
updated. Compared to conventional pulse-limited (or conventional) altimetry
(CA), SAR altimetry provides a better along-trajectory resolution and a
higher signal-to-noise ratio (SNR). Over the northeastern Atlantic, CryoSat-2
operates in SAR mode. Data collected in SAR mode and processed similarly to
LRM data are called reduced SAR (RDSAR) data. We use CryoSat-2 RDSAR data
(C2-RDSARRADS-1Hz) from the Radar Altimeter Database System (RADS)
(
Sentinel-3A, launched in February 2016, is the first satellite operating
entirely in SAR mode. RDSAR products are also available. Essentially, the
altimeter data are 1-D profiles along the ground track of the satellite, with
a footprint size of 1.5 to 10
Type and availability of the satellite data.
In situ observations have great accuracy, but their geographical distribution
is highly inhomogeneous, being mainly along coastal regions of industrialized
countries. Gaps in measurements and other types of inhomogeneities also occur
frequently in in situ observational records
The results of the wave model and the satellite measurements are evaluated
via a comparison with in situ observations at 165 locations. Most of the data
are from the Global Telecommunication System (GTS), which were obtained by
and are archived at the European Centre for Medium-Range Weather Forecasts
(ECMWF)
This data set was augmented with in situ wave buoy data provided by the
Federal Maritime and Hydrographic Agency (Bundesamt für Seeschifffahrt und
Hydrographie, BSH). Figure 1 shows the locations of these in situ data.
Moored wave data buoys are anchored at fixed locations and regularly collect
observations from different atmospheric and oceanographic sensors. Moored
buoys are usually deployed to serve national forecasting needs, to serve
maritime safety needs or to observe regional climate patterns
(
Bathymetry of the model area and locations of the GTS measurements.
The boxes indicate the area of the German Bight (black) and the GTS measurements
in the northern part of the North Sea used for the comparisons in Sect.
The spectral wave model WAM Cycle4.6.2 is used here
To estimate the sensitivity of the wave model to the temporal and spatial
resolutions of the meteorological input, different wind input data are used
(Table
Aside from the wind input provided by the ECMWF, the hindcast coastDat-3
produced by the Helmholtz-Zentrum Geesthacht (HZG) using the Consortium for Small-Scale Modelling Community Land Model (COSMO-CLM)
Horizontal and temporal resolutions of the meteorological input data.
In this section, the sensitivity of the wave model to different wind input data and their different spatial and temporal resolutions is analysed by assessing the general performance of the wave model under different wind forcings over the entire study period (from June to November 2016) and the entire model area. The quality of the simulated significant wave height during an extreme event in September 2016 is analysed in detail.
To study the sensitivity of the wave model simulations to the wind
conditions, WAM is forced using eight different wind data sets, as described
in Sect.
Q–Q scatter plot for measured (GTS wave data) significant wave
height as reference (R) and modelled (WAM) significant wave heights (M)
with
Q–Q scatter plot for measured (GTS wave buoys) wind speeds as
reference (R) and modelled wind speeds (M)
from
When comparing the wind speed with the in situ GTS measurements (Fig.
The general performance of WAM under all different wind forcings is good and fairly similar, especially under normal conditions, where no major differences are found. During extreme events, however, the model simulations tend to be spread out, with the coastDat-3 wind forcing overestimating and the ERA-Interim, ECMWF operational analysis/forecast and ERA5 wind forcings underestimating the large significant wave heights. In the wind data, this cannot be found. The wind is only very slightly underestimated. Particularly, the overestimation of the significant wave height with the coastDat-3 wind forcing cannot be found in the wind data.
The significant wave height (m) of the ensemble for 29 September
2016, 11:00 UTC, as well as the GTS measurements for the model simulations
with
the
As described in the previous section, the modelled significant wave heights tend to spread out during extreme events for different model experiments. Here, a more detailed analysis of data variability during an extreme event is provided. During the study period from June to November 2016, an extreme event occurred on 29 September 2016. The centre of the low-pressure system was located along the coast of Norway. Thus, the highest wind speeds occurred in the northern part of the North Sea, and the corresponding highest significant wave heights could be found in the northern part of the North Sea. At 11:00 UTC, the area with maximum significant wave height coincided with the locations of the GTS measurements. Hence, this event is chosen for further analyses.
In Fig.
When comparing the modelled significant wave height with the GTS measurements
in the northern part of the North Sea
(55
To study the variance of the significant wave height of the eight ensemble
members during the extreme event, an empirical orthogonal function (EOF)
analysis of the extreme event on 29 September 2016, 11:00 UTC, is performed.
The EOF analysis is carried out as described by
Figure
The first EOF of the significant wave height represents 56.16 % of the
total variance of the ensemble. The maximum variance is found in the area of
the maximum significant wave height in the northern part of the North Sea
(Fig.
The maximum of the second EOF of the significant wave height, which
represents 19.31 % of the total variance, is located in the northern
part of the model domain near the coast of Iceland (Fig.
The third EOF pattern shows a dipole in the northern part of the North Sea
(Fig.
The fourth EOF explains 7.71 % of the total variance. This EOF reveals the
larger-scale differences in the synoptic situation and therefore in the
wind fields, which are also reflected in the wave field. In the wind forcing
data, the exact location of the centre of the low-pressure system and
therefore the area of light wind differs, which also leads to different wave
heights off the coast of the northern part of Norway. In addition, due to the
different strengths of the wind fields in the wind forcings, the significant
wave height west of Ireland in the Atlantic as well as off the coast of
Norway is larger relative to that east of Great Britain due to the fetch
conditions (Fig.
In order to estimate the difference between the model simulations with hourly
and 6-hourly wind forcing during the whole time period, a temporal EOF over
the difference between the model simulations with hourly and 6-hourly ERA5
wind forcings is conducted. Here, no dominant EOF can be found, since the
first EOF has an explained variance of 3.13
Further investigation of the magnitude in significant wave height of the
respective peak is required, since this is the largest difference between the
ensemble members. Time series extracted from the ensemble members are
compared to the time series of the GTS measurements (Fig.
Figure
The model experiment with the 6 h ERA5 wind forcing yields the lowest
significant wave heights for 29 September 2016 (Fig.
The peak in the observed significant wave height is best illustrated by the
model simulation with the hourly ERA5 wind forcing (Fig.
A few days earlier, two smaller wave height peaks occur. The first one on 27 September 2016 is overestimated by all of the model experiments, although the corresponding peak in the wind speed is captured well by the model simulations with the hourly ERA5 and ECMWF operational analysis/forecast wind forcings. The 6 h wind forcings capture this peak very well, but due to the wind speed being high 3 h prior to and after the peak, the simulated significant wave height is too high. The model simulation with the hourly DWD forecast wind forcing is the most successful at reproducing the significant wave height peak, although the estimated wind speed is lower than the observed wind speed. The second peak, which occurred on 28 September 2016, is best matched by both model simulations with the ECMWF operational analysis/forecast wind forcing. Both simulations with ERA5 wind forcings slightly underestimate the significant wave height peak. All other simulations overestimate the significant wave height.
During normal conditions both before and after the peaks, the results of all model simulations are very similar.
Time series of significant wave height (m) as modelled by WAM with
different wind forcings and GTS measurements within the northern part of the
North Sea (55
From the analyses above, it can be concluded that during extreme events, the wave model results are quite sensitive to the wind forcing. Hence, high-quality wind data are needed to improve the ability to predict the sea state.
For our area of interest, a higher temporal resolution of the wind forcing is
more important than a higher spatial resolution. Although the spatial
resolution of the DWD forecast and coastDat-3 is higher than that for ERA5
and the ECMWF operational analysis, the wave model simulations using the
latter two increase the model capabilities. However, clearly better results
can be found via model simulations with hourly wind forcing than via those
with 6 h wind forcing. This conclusion differs from that of the study on the
Black Sea by
In this section, the quality of the newly available Sentinel-3A satellite
data is assessed and compared to that of older satellite data. The focus in
this study is on coastal areas, where the quality of both the satellite and
the model data tends to deteriorate. Also, the quality of the
Sentinel-3A data is analysed based on the relative orientation of the
coastline and satellite heading, varying metocean conditions and the wind
direction relative to the satellite flight direction. In this section, when
comparing satellite data with the simulated significant wave height, the
model simulation with the ERA5 wind forcing is used, as this simulation,
along with that with the ECMWF operational analysis/forecast wind forcing,
produced the best results during both extreme events and normal conditions
(Sect.
To estimate the overall performance of the different satellite products
during the entire study period and over the study area, scatter plots of the
in situ measurements versus remote sensing measurements are analysed
(Fig.
Q–Q scatter plots of measured significant wave height – in situ GTS
(R) versus remote sensing data (M) of
Scatter index between satellite and modelled significant wave
heights along the satellite tracks for
To analyse the spatial distribution of the quality of the satellite data, the
SI between the modelled and measured significant wave heights along the
satellite tracks within each grid box is calculated for Jason-2 and
Sentinel-3A SAR (Fig.
Comparison of the data quality within the first 10
To quantify this, the statistical values within the first 10
Due to the way satellite altimeter data are processed, the data quality can
deteriorate in the vicinity of coastlines, particularly for passes from land
to ocean. To test how much the satellite measurements over the study area are
affected by this problem, the flights are separated into onshore and offshore
flights, with onshore flights passing from the ocean to the shore and
offshore flights passing from the shore to the ocean. For the analysis here,
again, only measurements within the first 10
Another assessment of the quality of the data measured by the satellites can
be carried out by analysing their quality in terms of the fetch conditions.
To test this, Sentinel-3A SAR data within the German Bight
(53.23
Comparison of the data quality, organized by onshore and offshore
flights, for Sentinel-3A SAR. Only measurements taken within the first
10
In previous studies, e.g.
Comparison of the data quality, organized by long- and short-fetch situations within the German Bight, for Sentinel-3A SAR.
Comparison of the data quality, organized by the wind direction relative to the satellite flight direction, for Sentinel-3A SAR.
The newly available Sentinel-3A data yield better results for coastal areas
compared to the data quality of older satellites such as Jason-2 and
CryoSat-2. Especially within the first 10
To enhance the quality of the significant wave height data of the ensemble
mean, the satellite measurements and the ensemble of the modelled significant
wave height are combined to produce a best-guess wave field using the EOFs. A
more detailed explanation of this method, which is based on a maximum a
posteriori approach, can be found in
Best guess of the significant wave height of the ensemble (coloured), together with the Sentinel-3A track (line) and the GTS measurements (dots), on 29 September 2016 at 11:00 UTC.
In this study, the sensitivity of the wave model to wind forcing data with
different spatial and temporal resolutions is tested. The analysis shows that
the general performance of WAM for all different wind forcings is good and
fairly similar. Especially during normal conditions, no major differences can
be found. During extreme events, however, the model simulations tend to be
spread out, with the model simulation with the coastDat-3 and DWD wind
forcings tending to overestimate the significant wave height and the model
simulations with the ECMWF operational analysis/forecast, ERA-Interim and
ERA5 wind forcings tending to underestimate the high significant wave
heights. The EOF analysis shows that the largest difference between the
model simulations is the magnitude of the peak significant wave height, with
a difference of 2.92
Furthermore, the quality of the newly available Sentinel-3A data is assessed
in comparison with data from older satellites, i.e. Jason-2 and CryoSat-2.
The general performance is good and fairly similar between all satellite
products, although all products tend to overestimate the in situ significant
wave height measured within the GTS. The analysis of the spatial
distributions of the satellite data quality reveals better results for
Sentinel-3A over coastal areas than for Jason-2. Especially within the first
10
In the last section, where the ensemble and satellite data are merged, the carrying out of bias correction before assimilating satellite data into a wave model is shown to be necessary. Also, for an extreme event, satellite data can be used to guide an ensemble towards a better best-guess wave field, though it cannot be used to strictly force the ensemble towards the satellite data, as they are not accurate enough compared to the in situ measurements.
The WAM model code can be found at
Mean value:
Errors:
Standard deviation of the errors:
Root mean square error:
Scatter index:
Bias:
Correlation:
The authors declare that they have no conflict of interest.
This article is part of the special issue “Coastal modelling and uncertainties based on CMEMS products”. It is not associated with a conference.
This publication has received funding from the European Union's H2020 Programme for Research, Technological Development and Demonstration under grant agreement no. H2020-EO-2016-730030-CEASELESS. Luciana Fenoglio acknowledge the support of the European Space Agency (ESA) within the project SAR Altimetry Coastal & Open Ocean Performance (SCOOP). The authors would like to thank Beate Geyer for providing coastDat-3 wind data. The article processing charges for this open-access publication were covered by a Research Centre of the Helmholtz Association.Edited by: Agustín Sánchez-Arcilla Reviewed by: two anonymous referees