Dual scanning lidar (DSL), combining two long-range scanning lidar devices with their beams intersecting at a measurement point of interest, is a promising measurement approach in particular for offshore wind resource and site assessment. Although this approach can in principle be applied to any location that is within the range of the scanning lidars, i.e. currently up to 10-15 km, the measurement accuracy of such a setup has not been verified under representative offshore conditions at long range yet. A suitable verification typically requires an offshore met mast at the target location that can provide traceable reference measurements not just for mean wind speed and direction but also turbulence intensity (TI), gusts and extreme winds, to cover the entire range of site conditions that should be covered in the test.
With the Centre for the Testing of Environmental Sciences Technology (C-Test) utilizing the National Offshore Anemometry Hub (NOAH) close to Newcastle, UK, we have found a suitable site. This site was used for a DSL performance verification test from April to August 2025 involving three scanning lidars of type Vaisala 400s, the results of which are presented and discussed in this contribution. The devices were set up in a triangle with about 6-7 km between each scanning lidar and the NOAH meteorological mast to the East and about the same distance between the North and South test pad, respectively, with both locations being onshore situated directly at the coast. At the North pad, we had installed two scanning lidar devices of the same type but different generations, which when combined with a 400S located on the south pad forms a dual scanning lidar pair.
The results of the test correspond to the first of its kind for this distance, being publicly presented and published. They confirm that DSL are indeed capable of providing accurate and precise wind resource data at this distance. The same applies to TI data which requires a more careful consideration of the system setup and configuration. DSL performance is further discussed in comparison to that of floating lidar systems, which are considered as de-facto standard measurement tool for offshore wind resource assessment in the wind industry but with limited capability/trust with regard to TI.
As a last aspect, we analyse the measurement performance under storm conditions, focusing on storm Floris that was captured early August 2025. For the analysis, we not only have wind speed measurements at our disposal, but also detailed additional atmospheric and oceanographic measurement from a ceilometer, a microwave radiometer, and a wave buoy, that are part of C-Test or NOAH, respectively.

Norway’s fjords present major opportunities for the offshore wind industry, particularly to ensure safer and more efficient assembly of floating wind turbines. This sector relies on deep-water harbours that possess several advantages: The rapid increasing depth of a fjord allows to host the wind turbine with its floater close to the shore during the assembly phase; the sheltering characteristics of the harbour site reduces the number of days where work must stop for bad weather which leads to an average installation time of 10 days per turbine; the assembly crane can be on a fixed part, improving the stability; and the harbour can be connected to logistical hub for the transport of essential materials. The development of more deep-water harbours, similar to the one built for Hywind Tampen at Wergeland Base (Sløvågen, Norway), is crucial to reducing the average assembly time and meeting the goal of 30 GW of offshore wind power by 2040 set by Norway. This highlights the need for efficient methods to identify protected locations. Numerical weather models struggle in mountainous regions because of their coarse resolution. High-fidelity wind simulations using CFD are realistic. However, they are computationally intensive and require specialised expertise.

This study investigates the use of WindNinja (Wagenbrenner et al.,2023), a lightweight and accessible wind modelling tool, to downscale mesoscale wind data and identify sheltered areas in fjords. WindNinja has 2 solvers: one based on the conservation of mass only, while the second one uses a CFD Reynolds Averaged Navier Stokes (RANS) approach at a low computational cost. This study focuses on the CFD solver since it can predict wakes and recirculation areas.

WindNinja was validated using three met-masts equipped with 2 to 3 ultrasonic anemometers from the Bjørnafjord and compared, using the 10-minute mean wind speed, to the NORA3 mesoscale dataset (Haakenstad et al., 2021). NORA3 was used to initialise WindNinja at the border of the computational domain for events corresponding to the strongest storms recorded in the area between 2016 and 2020.

The results showed that the CFD solver postprocessed by a 2D Gaussian filter significantly reduced the mean absolute bias, from 2.9 ± 2.5 m.s−1 to 1.7 ± 1.4 m.s−1, corresponding to a 40% improvement compared to NORA3. The study highlights some of the limits of mast-mounted anemometers as in-situ wind sensors. The validation of WindNinja is challenged by the uncertainties coming from the impact of the local terrain on the measurement data.

While several limitations remain, such as the use of a constant and homogeneous roughness length or the initialisation scheme, which cannot be gridded for the CFD solver, the study demonstrates that WindNinja offers a practical, low-cost solution for early-stage study of wind conditions in fjords to identify areas subject to the development of deep-water harbours.

We present an empirical evaluation of boundary layer wind speeds estimated from the C-band synthetic aperture radar (C-SAR) sensor of the Sentinel-1 satellites for the purpose of offshore wind resource assessment. We compare SAR-based estimates and ERA5 reanalysis model predictions of wind speeds with in situ lidar measurements from 20 m to 260 m above sea level at an offshore wind site in the North Sea. Validation results show that SAR-based wind speeds have lower bias than wind speeds from the ERA5 reanalysis and comparable root mean square error (RMSE) and mean absolute error (MAE). We further show that despite the low temporal resolution of the Sentinel-1 wind speed estimates, a wind speed distribution estimated over an archive of Sentinel-1 data is closer to the in situ wind distribution than a corresponding ERA5-based distribution, as measured by different distance measures used to contrast probability density functions.

The method we use to derive SAR-based wind speed is the CMOD5.N wind retrieval algorithm, which is routinely used to produce 10 m wind products from Sentinel-1 C-SAR data (Hersbach, 2008; Mouche & Vincent, 2019) for the Copernicus Marine Service. We extrapolate its 10 m wind speed output to different heights by use of a logarithmic or a power law wind profiles and Monin-Obukhov similarity theory, whose atmospheric stability related parameters are computed with fields from the ERA5 reanalysis (Hersbach et al., 2020) as input. Our analysis shows that wind speed estimates from SAR data show similar performance under different atmospheric stability regimes, as there is no significant variation in error measures and bias with Obukhov length, which we compute for each SAR data acquisition time based on concurrent ERA5 data. No clear seasonality or dependency of the wind speed levels can be observed for either of the performance measures.

The observed accuracy and precision install trust in the SAR-based wind speed estimates. Our assessment of wind speed distribution estimates further demonstrates the value of SAR data as an additional source of wind resource information to the reanalysis models traditionally used. We also present new machine learning algorithms that allow us to perform bias corrections and compensate scale and sensor dependencies on the distribution estimates. The method we use is to establish optimal transport matrices that define transformations between 1) in-situ wind speed data and ERA5 reanalysis wind speed data during the lidar campaign; and 2) ERA5 reanalysis wind speed data and SAR wind speed data over the entire window of Sentinel-1 acquisitions. These transformations capture the biases between the different data sources, deviations that can be attributed both to their different temporal and spatial scale (aggregation levels) and the inherent differences between model predictions, and remote sensing-based estimates, and in-situ measurements.

The design of offshore wind turbines requires the evaluation of normal and extreme current conditions to properly assess the loads on the structure. For fixed offshore wind turbines, the contribution of current to fatigue loads is generally considered small and negligible (IEC 61400-3-1, 2019). However, for floating offshore wind turbines with mooring systems and dynamic power cables, current-induced forces and vibrations can significantly contribute to fatigue and extreme loads, making accurate current predictions essential.

Assessment of metocean conditions at proposed wind farm sites is typically performed by referencing existing observation data in the vicinity or by conducting new in-situ measurements. However, sea current observations are scarce and often only available for short periods, making it difficult to assess long-term variability and extreme current velocities. Numerical ocean models are promising for this purpose, but few studies have validated their prediction accuracy using in-situ observations. Studies of wind speeds and wave heights have focused not only on model prediction accuracy but also on the distribution of extreme values (e.g. Ishihara and Yamaguchi, 2015). However, similar studies on sea currents are very limited.

In this study, current velocities predicted by the numerical ocean model JCOPE-T (Japan Coastal Ocean Predictability Experiment) are validated using in-situ observation data from an acoustic Doppler current profiler (ADCP) at the Fukushima Floating Offshore Wind Farm Demonstration Project (Fukushima FORWARD). The annual mean current speed, current rose distribution, and directional mean current speed predicted by JCOPE-T show favorable agreement with the observation data. However, the extreme current speed with a 1-year return period based on monthly maximum predictions by JCOPE-T underestimates the observed value by approximately 20%.

To improve this underestimation, the 21-year JCOPE-T dataset is analyzed to compare the extreme value distributions using monthly and annual maximum current speeds. The two follow a Gumbel distribution and show good agreement, which implies that the corrected monthly maximum current speeds can be extended to predict the annual maximum current speed. A comparison of observed and predicted extreme value distributions of monthly maximum current speeds over a three-year period shows that the predictions capture the observed trend but exhibit systematic biases. A double bias correction method for extreme current speeds is proposed using the ratio of observed to predicted values of the mean and standard deviation of monthly maximum current speeds over a three-year period. The corrected extreme current speed with a 1-year return period is in perfect agreement with the value calculated from the observed extreme current speeds.

The extreme current speed with a 50-year return period based on the proposed correction method is validated using the maximum surface current speed over a 42-year period reported by the Japan Oceanographic Data Center (JODC). The extreme surface current speed with a 50-year return period estimated directly from JCOPE-T is underestimated by 11.0% compared to the observed value, but the proposed correction method reduces the underestimation to 0.3%.