Icing on wind turbines: Causes, risks, and forecasts for short-term power traders

As temperatures drop, wind power production and electricity demand are on the rise, promising attractive revenues for wind power producers and short-term traders. But winter also comes with a critical challenge: turbine icing. This phenomenon can quickly negate the season’s benefits by putting both turbine safety and market performance at risk.

In Northern and Western European markets with high shares of weather-dependent renewables, icing events, though infrequent, are a leading cause of overall imbalance. For instance, Nordic short-term power traders likely remember last year’s increase in icing disruptions. Summing up the overall impact of icing, a forecasting analyst from a renewable energy company we work with said:

“For our onshore wind portfolio, the most expensive ten days caused 50% of the yearly balancing costs. Five of those days are due to icing power losses.“

To help you step into winter prepared, this article explores the complexities of icing, from meteorological factors and turbine mechanics to mitigation strategies and advanced forecasting of icing events.

What causes icing on wind turbine blades?

Icing season typically occurs from November to March but can stretch as early as October and as late as April. For example, Germany experienced an icing event in late April this year.

Icing results from a combination of freezing temperatures (-2°C or lower) and moisture (rain or supercooled droplets). Droplets remain liquid in the atmosphere where they lack a nucleation point, i.e., a surface or particle that would trigger their freezing. As soon as they come into contact with surfaces – such as wind turbine blades – they rapidly freeze.

Meteorological conditions associated with icing in the Netherlands during the 2023-24 winter, with polar air coming from higher latitudes and probability of local precipitation. Source: ECMWF.

Depending on the turbine’s location and altitude, different combinations of temperature and moisture will create varieties or phases of icing:

In-cloud icing forms when supercooled droplets from clouds reach turbine blades and freeze. It includes:

• Rime ice: A rough, frosty layer that forms at below-freezing temperatures (between -4°C and -12°C.), similar to the ice in the back of your freezer.
• Glaze ice: A smoother, heavier ice that forms during freezing rain or at temperatures just below 0°C to -4°C.

Precipitation icing occurs in freezing rain and wet snow, leading to the accumulation of wet snow or black ice.

Offshore wind farms are generally less affected, because salt in the air lowers the freezing point, making it harder for ice to form. Still, a freezing sea remains problematic as drifting ice may reach turbine foundations.

It is worth noting that global temperatures rising due to climate change don’t necessarily mean fewer icing events. Increased storm frequency and moisture levels may lead to more icing in certain regions.

What happens to a wind turbine during icing?

Wind turbine blades are designed in a wing shape to rotate primarily based on lift forces generated by the wind. When they operate in the freezing cold and humid conditions described above, the moisture in the air freezes onto the blades and starts to accumulate, a process known as ice accretion.

Any added mass from the ice alters the blade’s aerodynamics – its shape and weight – thus affecting the interaction with the wind. This adds mechanical strain on the rotor, causing the turbine to produce less electricity than if the blades were ice-free.

As wind speeds increase, the issue worsens. When the rotors spin faster, the blades bend slightly due to the pressure of the wind. However, ice is rigid and doesn’t bend. Instead, it breaks off in large chunks, which can be dangerous for anything or anyone nearby. In severe cases, the ice can cause the turbine to become unbalanced, leading to damage or even a breakdown.

In the long term, repeated ice accretion can damage turbine components, increase maintenance costs, and reduce their lifespan due to the strain on internal mechanics. Beyond the operational impact, the market impact of icing is also considerable, as the next section will show.

Iced wind turbines. Source: Barati-Boldaji, Reza & Komareji, Mohammad. (2017). Techniques of identifying icing and de-icing of wind turbines. Signal Processing and Renewable Energy. 8. 21-28.

What is the impact of icing on power markets?

The extra ice mass on the blades drastically reduces turbine efficiency. Even a light icing event could produce enough surface roughness on turbine blades to considerably decrease their efficiency. When turbines are coated with ice, their power production drops significantly.

Additionally, chunks of ice falling off turbines are a severe safety hazard for the personnel working on wind farms, visitors, or passers-by. To err on the side of caution, wind turbines are often just shut down, manually or automatically, for the entire icing period.

What does the power loss amount to? A field study indicated up to 80% power loss during the icing event and over 20% loss of annual production.

Forecasted versus actual production during an icing event – the output of the turbine (blue) is much lower than forecasted (teal).

Furthermore, the sudden and widespread power production drops caused by icing – often across entire wind parks or even regions – can lead to considerable electricity shortages. To restore balance, grid operators must activate reserves, pushing imbalance prices up. The costs of these interventions are charged to the balancing responsible parties (BRPs). In these instances, the exact financial impact of icing on trading results depends on several factors, such as the availability of ancillary services to compensate for the power loss, the system’s residual load, or the behavior of other market players.

Overall, although widespread icing events affecting hundreds of turbines are rare, the resulting imbalance prices can be extreme. Consequently, short-term power traders can face immense revenue losses if their day-ahead power generation forecast does not account for icing. Based on insights from our customers, we learned that inaccurately forecasted icing events can be responsible for up to 25% of annual balancing costs.

How to detect turbine icing?

To mitigate the impact of icing, most turbine operators offer additional sensors, coatings, or services to detect, prevent, or clear ice on blades, such as by heating rotor blades; one such example is Topwind’s Ice Prevention System. Both fully automatic and human-monitored systems are available.

With only a minor risk of icing, a turbine operator might decide to keep the turbine running, while, in the case of a severe risk, automatic systems may shut down the turbine automatically. Often, there is a human-in-the-loop when deciding when to resume operation.

While these systems are helpful in mitigating the performance and safety challenges of icing, they’re not ideal for short-term power traders, as they don’t forecast imbalance volume. Furthermore, signals from on-blade sensors will arrive after icing has occurred, and bids on the day-ahead market have already been submitted. In some cases, asset managers will only alert traders about a shutdown due to an icing event unless contractually obliged to do so.

Therefore, a robust day-ahead trading strategy should account for icing to minimize volumetric risk resulting from the power loss, ideally providing forecasts five days in advance. However, while such solutions are available in Nordic countries, they remain limited in Western Europe, where traders often rely on weather forecasts and personal experience to predict icing events.

Why are icing forecasts difficult to get right?

Forecasting icing is not commonly available because it is a challenging problem to solve. Three main factors contribute to this complexity.

Limitations of weather models

Weather conditions leading to icing are highly specific, localized, and rapidly changing. These factors are typically not accurately incorporated into most weather forecasts as their focus lies more on surface levels rather than specifically and fully observing the wind turbine levels. Moreover, while some weather models offer variables to automatically diagnose freezing rain, these aren’t always reliable; some weather models are also known to raise many false alarms.

A perfect solution would require sensors and detailed modeling of turbine properties – aerodynamics, height, materials, energy density carried by the materials – as well as local conditions, which is prohibitively expensive.

Data scarcity

Because icing events are infrequent – only expected a few times each year during winter months – there’s not a lot of data covering their occurrence and resulting power losses at turbines or wind park levels.

The variable duration of icing events

Finally, even if predicting icing was straightforward, it would not be sufficient to solve the problem for power producers or traders; they’d still need to know how long the icing event will last to determine when turbines can be on or at full capacity again. The length of an icing event covers the period of ice growth, retention, and melting or removal and can range anywhere from an hour to a day.

All of the above accounted for, how do we know if an icing forecast is actually valuable in a short-term trading operation?

How to evaluate an icing forecast?

An icing prediction is a binary signal; the event is either expected to occur or not. The matrix below provides a useful mathematical tool for evaluating this prediction. The formulas use the number of true/false positives or negatives to return three metrics, each providing insight into different aspects of a prediction model’s performance.

• Precision: How many of the things you predicted as “positive” (e.g., icing events) were correct?
• Recall: Out of all the actual positive cases, how many did you successfully predict?
• Accuracy: The overall correctness of your model, showing how often the predictions (positive or negative) were right.

Calculation of Precision, Recall, and Accuracy in the confusion matrix. Source: MA, Jun, et al. (2019). Analyzing the Leading Causes of Traffic Fatalities Using XGBoost and Grid-Based Analysis: A City Management Perspective.

Nevertheless, a well-performing icing model does not guarantee optimal short-term trading results. For trading decisions, a “good enough” icing forecast isn’t about being right on average but about predicting the icing events with high stakes on power markets. Large imbalance volumes during large price spreads are very costly, so having an accurate forecast precisely at these times is critical.

In other words, even if a forecast occasionally returns false positives, using it in a trading strategy is worth it as long as the cost of false positives is outweighed by the benefits of being right when it matters most. Thus, the evaluation of an icing forecast ultimately comes down to its value to the bottom line.

To explore the topic of forecast assessment further, also download our white paper on evaluating metering-point power generation forecasts.

How does Dexter forecast icing?

In October 2023, we launched an icing feature for our Wind Power Generation Forecast, developed to support short-term trading decisions. Our data-driven approach uses over 500 TB of weather forecasts, turbine sensor data, park-level load measurements, and turbine-level weather parameters to deliver:

• (Probabilistic) icing forecasts;
• Impact forecasts with automated corrections for performance loss;
• Alerting to promote vigilance and risk management.

As explained earlier in this article, forecasting the impact of icing on short-term power markets remains too challenging for automation to fully address; thus, human intervention is still necessary. For more context into the human-machine balance, have a look at this article on our human-in-the-loop approach in wind power forecasting.

Integrating our machine learning-based and human-backed icing feature into their winter trading strategies, our customers have seen tangible savings in balancing costs last year, and we expect even better results this season.

Are you ready for icing season?

In this blog, we’ve outlined the main facts to consider regarding icing this winter. To summarize:

• Icing on wind turbines occurs in colder climates between October and April as a result of freezing temperatures and moisture in the air;
• As ice accumulates on turbine blades, their aerodynamics change, affecting the interaction with the wind and causing power production loss;
• The power loss is reflected in short-term power markets through spikes in imbalance prices, leading to high balancing costs;
• Icing events are difficult to predict due to limitations in weather models, lack of data, and variable duration;
• Short-term wind power traders should incorporate icing forecasts in their strategy to minimize their imbalance volumes during large price spreads.

At Dexter Energy, we understand that a successful trading cycle isn’t just about accurately predicting the weather but about making the right decision at the right market moment. Get in touch to ensure your strategy is equipped with a reliable power forecast for managing icing events this winter.