In this paper, surface wind speed and average wind power derived from Sentinel-1 Synthetic Aperture Radar Level 2 Ocean (OCN) product were validated against four weather buoys and three coastal weather stations around Ireland. A total of 1544 match-up points was obtained over a 2-year period running from May 2017 to May 2019. The match-up comparison showed that the satellite data underestimated the wind speed compared to in situ devices, with an average bias of 0.4 m s
With the ever-increasing interest in offshore wind energy, the estimation of the available wind energy over large offshore areas has become necessary. According to the Global Wind Energy Council (Global Wind Council, 2014), offshore wind power costs are expected to reduce by about 45 % by 2050. One factor that can be associated with cost reduction is the increasing availability of accurate remote sensing data over large areas with a high resolution, which can significantly reduce project risk at the site-finding stage (McAuliffe et al., 2018). Moreover, the measurement of offshore wind speed contributes to the understanding of marine phenomena and boundary layer processes. Low-altitude meteorological parameters, such as wind, are therefore key parameters in the modelling of the Earth system.
Several studies have already attempted to assess the offshore wind energy
potential using spaceborne scatterometers, such as ERS-1, ERS-2, NSCAT,
QuickSCAT and ASCAT (Sánchez et al., 2007; Pimenta et al., 2008;
Karagali et al., 2014; Bentamy and Croize-Fillon, 2014; Remmers et al.,
2019). However, the grid spacing of these instruments is at best 12.5 km
In this study, the Sentinel-1 A and B Level 2 Ocean (OCN) product produced by the
European Space Agency (ESA) was validated. This product, derived from SAR
observations, provides measurement of neutral surface wind speed and
direction at 10 m a.s.l. (above sea level) with a grid spacing of 1 km
The aim of this study was to validate and the Sentinel-1 A and B Level 2 OCN product against in situ measurements in Ireland and assess this data ability to describe the wind resources. First the satellite product and the study area are introduced, next the methodology is provided, and finally the results are presented and discussed.
Sentinel-1 A and B are two polar-orbiting satellites equipped with C-band
SAR. This sensor, which records surface roughness, has the advantage of
operating at wavelengths not impeded by cloud cover or a lack of
illumination and can acquire data over a site during day or night in all
weather conditions. The Sentinel-1 Level 2 OCN product includes a component
called Ocean Wind Fields (OWI), which is a ground range gridded estimate of the surface wind speed and direction at 10 m a.s.l., assuming a neutral atmospheric stratification, with a grid spacing of 1 km
Number of Sentinel-1 A and B passes across Ireland over a 2-year
period running from May 2017 to May 2019 with an acceptable quality flag (
Average daily hour of Sentinel-1 A and B passes across Ireland over a 2-year period running from May 2017 to May 2019 with an acceptable quality flag (
Location and characteristics of the weather buoys used in the comparison with Sentinel-1 SAR Level 2 OCN product.
Ireland's Marine Institute operates five offshore weather buoys named M2–M6. Their location is shown in Fig. 3. The data from these were downloaded from the Marine Institute website with a 2-year time series ranging from 1 May 2017 to 1 May 2019. The hourly product corresponds to the wind speed averaged over a period of 10 min every hour at 3 m a.s.l. As a result of extensive maintenance periods, the buoys are not always functional, leading to a lack of measurements in the dataset, up to several months, for some locations. Due to this phenomenon and a poor offshore coverage frequency from Sentinel-1 satellites, the M6 buoy was excluded in the validation analysis.
Location of MetOcean buoys (yellow) and coastal weather stations (green) used in the validation of Sentinel-1 SAR surface winds.
In order to compare Sentinel-1 SAR Level 2 OCN product with this network of
instruments, the in situ buoy measurements were extrapolated from 3 m to 10 m a.s.l. The following log law was used, assuming a neutral atmospheric
stratification (Carvalho et al., 2017):
Three weather stations operated and maintained by Met Éireann, the Irish
weather forecasting service, were used to validate the Sentinel-1 SAR Level 2 OCN wind speeds in coastal areas. These three stations were considered for
the validation analysis because they are located close to the shore (less
than 200 m, see Fig. 3), at a low altitude (
Location and characteristics of the coastal weather stations used in the comparison with Sentinel-1 SAR Level 2 OCN product.
The error
The average wind power density
In this section, a stepwise approach is taken to assess the viability of Sentinel-1 Synthetic Aperture Radar Level 2 OCN product to characterise the long-term offshore wind resource around Ireland. This approach provides an appreciation of the error introduced by some key inherent limitations of the satellite data. The stages in this analysis are summarised in Table 3.
Key stage of the analysis. A stepwise approach was employed to allow an appreciation of the uncertainty introduced by key limitations of the satellite data when deriving long-term wind characteristics.
For each stage the methodology and results are discussed before moving to the next stage of analysis. Overall conclusions are outlined in Sect. 4.
Results of the match-up comparison of satellite-measured wind speeds with in situ measured wind speeds from weather buoys.
Sentinel-1 SAR Level 2 OCN surface wind data and in situ wind data were
co-located in space and time. Since the grid spacing of this product is very
high (1 km
For all buoys, the wind speed correlation with the remotely sensed data at a
1 h time interval was around 0.99, which showed that the time
difference between the satellite and in situ data does not introduce a
significant source of error. Therefore, in the time domain, each in situ
measurement with a corresponding satellite measurement performed within a 30 min time interval was selected for the analysis. Another factor in this
respect is that Sentinel-1 SAR Level 2 OCN spatial averaging at the
resolution of 1 km
The bias for all available data was found to be
Results of the match-up comparison of satellite-measured wind speeds with in situ measured wind speeds from coastal weather stations.
Statistical representation of the Sentinel-1 Level 2 OCN error against weather buoy data as a function of SAR wind speeds
The bias was found to be wind speed dependent. Figure 4 (left) shows that the bias was stronger at small wind speed values and reduced as the wind speed increased. This is consistent with the fact that Sentinel-1 SAR uses the sea state in order to estimate surface winds. Indeed, low wind speeds do not necessarily cause a significant effect on the sea state and, consequently, the instrument does not always accurately estimate the surface winds. This problem is already well known and often leads to an unrealistically high number of very low wind speed values. This can be seen on the scatterplot in Fig. 4 (right), which also confirmed the results related to the bias.
As expected, the satellites also underestimated the wind power. The average
error in the wind power was 6 % for the weather buoys and 13 % for the
coastal weather stations, respectively (Figs. 5 and 6). Since the wind
power is proportional to the cube of the wind speed, a higher error (
Wind speed histograms of Sentinel-1 SAR Level 2 OCN with Weibull fit (red curves)
Wind speed histograms of Sentinel-1 SAR Level 2 OCN with Weibull fit (red curves)
The main limitation of satellite remote sensing to accurately assess the offshore wind resource derives from their reduced temporal coverage and revisit time at a given location. Since wind speeds can have strong daily variations, the impact due to the lack of intra-diurnal measurements needs to be investigated. To do so, for each match-up between the satellites and the in situ instruments, all the in situ measurements from that 24 h period were added to the in situ data before computing the statistics (Fig. 7). The bias and the error on the wind power assessment were increased on average by 9.14 % across the 7 sites as shown in Table 8. It can be concluded that the lack of intra-diurnal satellite data has a relatively small impact on the results. Since the satellites pass different locations at different times of day, some in situ locations were more affected than others. However, the increase of error in the wind power due to intra-diurnal variability was always below 7 % of the total wind power.
Comparison of wind speed long-term statistics obtained from the four weather buoys with the ones obtained from the SAR data. These values are the results of the match-up comparison exercise and are used to evaluate the accuracy of the satellite data.
Comparison of wind speed long-term statistics obtained from the three coastal weather stations with the ones obtained from the SAR data. These values are the results of the match-up comparison exercise and are used to evaluate the accuracy of the satellite data.
Average wind speed off Ireland over a 2-year period running from May 2017 to May 2019 retrieved using the Sentinel-1 SAR Level 2 OCN product. Satellite tracks are visible, particularly in the northeast. These are an artefact of the analysis.
Increase in the bias and the error on the wind power when intra-diurnal data of in situ measurements are taken into account, compared with the same results obtained for the match-up comparison.
In this section all the available in situ data over the 2-year period of
study were taken into account, including days for which there was no
satellite pass. In order to compare statistics derived from the same time
periods, the histograms of in situ data were computed using all of the
available periods and the histogram of satellite data with satellite
measurements available during these periods (see Figs. 5 and 6 for the Weibull distribution fits and Tables 6 and 7 for the corresponding parameters and wind powers). These figures showed that, although the histograms produced from the satellite data exhibited important discrepancies compared to the one produced from the in situ data, the SAR measurements were nonetheless sufficient to correctly estimate the Weibull laws describing wind speed statistics (in red for Sentinel-1 Level 2 OCN and in green for in situ devices in the figures). The analysis revealed a strong overall agreement between the in situ and SAR wind speed distributions, as can be seen in Tables 9 and 10. The Weibull parameters and the corresponding wind powers had very similar results, with wind power errors below
Comparison of the long-term wind speed statistics produced from the weather buoy data with those produced from the SAR data at the same locations. These values evaluate the accuracy of the satellite-derived data to provide the correct long-term average wind statistics.
Comparison of the long-term wind speed statistics produced from the coastal weather station data with those produced from the SAR data. These values evaluate the accuracy of the satellite-derived data to provide the correct long-term average wind statistics.
The results show that the percentage error of the average wind power was
lowest for the coastal weather stations. This may indicate that they could
be more reliable than weather buoys, perhaps due to the presence of waves
and the relatively low altitude of the buoys. This finding must be treated
with caution given the relatively low number of weather stations included in
this study. It is possible that the error in offshore locations could be
overestimated due to inaccuracies with the weather buoy data, although there
is no possibility of proving this with certitude. The validation of the
Level 2 OCN product should be further investigated in coastal areas since
land contamination and coastal topography can introduce bias. Another
interesting feature is that the bias observed in the match-up comparison
seemed to disappear in this climatological analysis. The main difference
between the match-up comparison and the analysis performed here arises from
including in situ data even when satellite data were not available. In this
study, satellite data can be unavailable for two reasons: no data were
recorded as a consequence of the relatively low revisit time of the
satellite, or the data recorded were discarded if it was flagged as “bad
quality”. The former should not have any effect on the long-term statistics
since an increase in sample size will result in a better Weibull
distribution. However, the latter might actually introduce an artificial
bias in the match-up comparison by limiting it to a specific type of
situation in which satellite measurements are easier to perform. For
example, if good quality flags are more likely to correspond to turbulent
situations, then the different scales at which the measurements are
performed (10 min for in situ devices and 1 km
Wind power off Ireland over a 2-year period running from May 2017 to May 2019 retrieved using the Sentinel-1 SAR Level 2 OCN product. Satellite tracks are visible, particularly in the northeast. These are an artefact of the analysis.
In this section, the use of the Sentinel-1 Level 2 OCN product to assess
wind resources around Ireland at 10 m a.s.l. with a 1 km
Seasonal average wind speed off Ireland over a 2-year period running from May 2017 to May 2019 retrieved using the Sentinel-1 SAR Level 2
OCN product: winter
Seasonal wind power off Ireland over a 2-year period running from May 2017 to May 2019 retrieved using the Sentinel-1 SAR Level 2 OCN product: winter
In terms of wind power, the results logically revealed a similar pattern
with an increased heterogeneity, due to the fact that the wind power is
connected to the cube of the wind speed (Fig. 8). The northwest area had
an average wind power of 700 W m
The seasonal averages of wind speed and wind power showed expected trends of low and strong winds typical of the summer and winter seasons, respectively (Figs. 9 and 10). Autumn was also associated with strong winds, which correlated to the cyclonic activity in the North Atlantic Ocean ending their trajectory in this area of western Europe. The wind climate during spring was much more moderate than that of autumn.
As shown in Figs. 7 to 10, the tracks of the satellites were still visible. This discrepancy can be related to several factors, such as instrument bias associated with the incidence angle, difference in the number of samples (Fig. 1) affecting the quality of the Weibull fits, or simply a difference in the average time of the day at which the satellites pass (Fig. 2) resulting in a different impact of the intra-diurnal variability. Unfortunately, no clear correlation was found between these factors and the anomalies on the maps. It was only found that the edges of the swaths have more unrealistic values, which could be due to the incidence angle or the instrument thermal noise. As a consequence, a margin of 5 pixels (roughly equivalent to 5 km) was removed from the swaths before creating the maps. The areas with less observations also had a less reliable assessment of the mean wind speed and power; however, this limitation should disappear in the future as more samples become available. It can be concluded that the accuracy was dependent upon location, which is a factor that should be considered when using Sentinel-1 SAR data; this is shown to be particularly the case at the edge of swaths, and users should be aware of this limitation and filter the data accordingly.
Measurements from the Sentinel-1 Level 2 OCN product were compared with
measurements from four weather buoys and three coastal weather stations
located around Ireland. The match-up comparison indicated that the
satellites underestimated the in situ data by 0.4 m s
The fact that the satellites always pass at the same hour of the day,
limiting their ability to record the intra-diurnal variability, was
investigated and its effects on the long-term statistics were found to be
minor. Finally, the error in the average wind power was found to be on the
order of 10 % and 5 % for weather buoys and coastal weather stations,
respectively. This result was quite remarkable given the fact that the wind
power is proportional to the cube of the wind speed, which strongly enhances
the original error from the wind speed. Maps of the average wind speed and
wind power around Ireland were presented with a resolution of 1 km
Future studies could focus on the combined use of SAR and scatterometer-measured wind speed in order to create climatologies constructed using a longer period than the 2-year period of this study. This could be particularly interesting to more accurately estimate the offshore wind energy resource. Another important application in the future would be to modify the acquisition mode in coastal areas for the satellites carrying SAR, in order to obtain the required information to estimate the wave heights. This information, only available in open seas with Sentinel-1, would be useful to correlate the wind and wave energy and thus provide a more detailed description of the marine environment for optimising offshore wind farm siting.
Data sets are available upon request by contacting the corresponding author.
LdM did the main part of the research and wrote the paper. TR provided some insights about wind speeds around Ireland and spaceborne scatterometer measurements. ROC helped with the preparation of the figures. CD supervised the work as an experienced researcher and contributed to answering referees comments.
The authors declare that they have no conflict of interest.
The authors would like to thank the Marine Institute for providing the offshore weather buoy data, Met Éireann for the coastal weather station data and ESA for the Sentinel-1 SAR Level 2 products. The authors also would like to thank the European Regional Development Fund (ERDF) INTERREG Atlantic Area Project ARCWIND for assuming the publication costs of this study.
This research has been funded by Science Foundation Ireland (SFI), Brockmann Consult GmbH and by the European Regional Development Fund (ERDF) INTERREG Atlantic Area Project ARCWIND.
This paper was edited by Andrea Hahmann and reviewed by two anonymous referees.