As the penetration level of renewables grow, managing, balancing and keeping enough reserves on the grid offers challenges. Part of the solution, currently being implemented in many electricity markets, is the participation of traditionally non-dispatchable renewable energy to balancing services. Active participation of wind power requires accurate minute to hour-scale forecasting. Similar forecasts are crucial to understand the constantly varying need for activation of balancing services to meet departures from production plans. As input to such forecasts, power measurements taken at wind farms are insufficient on their own because their absence in above cut-out (full load) or below cut-in (no production) cases. In both cases, power data fails to indicate forecast trends or biases, which may stem from e.g. high winds or icing. Additionally, in curtailed operation power measurements does not inform on the potential power. Because of this, wind farms should ideally be complemented with independent wind measurements.

With hub heights now exceeding 200m (even 300m in low-wind areas), traditional met masts are inadequate due to costs and required planning permission. Remote sensing tools like SODARs and LIDARs are better suited, offering real-time wind profiles and turbulence data.

However, deploying these instruments in new environments poses challenges:

• Reliability: How is predictability impacted when the data stream is lost?.
• Handling of missing data : When data gaps appear in the observations, what alternatives are there?

To address impacts of weather-related and real-time gaps on forecasting challenges as well as data quality issues, we use the 75-member WEPROG MSEPS ensemble NWP, matching forecasts to recent observations in time and space.

NWP-Measurement Mismatch

NWP models use grid cells, while sensors provide point data—creating a resolution gap. Additionally, NWP models traditionally run hourly, while sensors update in seconds/minutes. We use cross-wavelet analysis to map how phase errors, amplitude errors and resolution differences contribute to the forecast error in time. The analysis is used to study if higher-resolution NWP models at sites with met masts, SODAR, and LIDAR can improve alignment.

Verification Challenges

High-resolution systems are costly, and traditional metrics (MAE, RMSE, etc.) often fail to show clear benefits. With a rule-based skill score matrix we can evaluate different event types and setup phase-error tolerance, measuring both detection predictability and model performance.

The motivation behind the work is the desire to operate wind farms proactively, understand the weather conditions at the wind farm and facilitate participation on intra-hour balancing markets. In line with that aim, we focus on the theoretical benefits and challenges of independent site measurements of wind speed to forecast the available power and investigate skill improvements in forecasts of wind energy and compare the use of remote sensing to traditional met masts.