Research challenges and needs for the deployment of wind energy in atmospherically complex locations

The continuing transition to renewable energy will require more wind turbines to be installed and operated in many new locations on land as well as offshore. The need to have geographic diversity, as well as limited availability of land in historically "good" locations for wind energy, means that wind turbines will also need to be deployed in hilly or mountainous regions, often known as "complex terrain". These areas can also experience challenging weather and climate conditions and may experience instrumentand blade icing that can further impact their operation. This paper – a collaboration between 5 several IEA Wind Tasks and research groups based in mountainous countries – sets out the research and development needed to improve the financial competitiveness and ease of integration of wind energy in hilly or mountainous regions and in regions subject to icing. The focus of the paper is on the interaction between the atmosphere, terrain, land cover, and wind turbines, and covers all stages of a project lifecycle. The key needs include collaborative research and development facilities, improved wind and weather models that can cope with mountainous terrain, frameworks for sharing data and a common, quantitative 10 definition of site complexity.

. Complex terrain can introduce flow inhomogeneity in the measurement volumes of remote sensing devices. In this illustration the terrain (green) causes local speed-up and inhomogenous flow (represented by grey streamlines) through a lidar's measurement volume (represented by the red cone), upwind of a wind turbine.
Research is therefore ongoing into the use of vertical-profiling wind lidar in more complex locations (see e.g., Clifton et al., 2015;Klaas and Emeis, 2021). And, scanning lidar have been seen to be vary useful for measuring wind conditions in complex terrain in research projects (Vasiljević et al., 2017;Menke et al., 2020). However, the experience reported there suggests that such three-dimensional wind scanners must currently be considered research devices that require extensive monitoring and post-processing to obtain usable data. This is unfortunate, given their tremendous measurement capabilities compared to fixed 200 masts. Research is therefore needed into ways to simplify the use of scanning lidar, as well as to process the results. Lidar manufacturers may also need to reduce the cost, weight, and power requirements of their devices to make them easier to deploy.
To truly replace meteorological masts, wind lidar would also have to be able to deliver reliable wind turbulence information.
At this time there is no clear consensus about the ability of wind lidar to do this, or the steps that should be taken to equate 205 wind lidar turbulence to the turbulence derived from a cup anemometer (see e.g. Sathe et al., 2011;Newman and Clifton, 2017;Hofsäß et al., 2018). Although there are many possibilities to post process wind lidar data to retrieve turbulence information, the lack of open data sets prevents these from being tested. We therefore suggest that there is a need for a collection of open data sets consisting of colocated wind lidar and anemometers in well-described locations that could be used to validate data processing methods. 210 The variation in wind energy sites mean that it is unlikely that one type of lidar and one processing approach will work for all sites. But, customised solutions are expensive. Instead we expect to see the development of flexible, digital, modular processes that allow appropriate solutions at each step of the process. This has been seen elsewhere in the wind energy industry and is part of the trend towards greater digitalisation. Research is needed therefore into tools that make it easier to use wind lidar in complex flows, complex terrain, and complex weather. These should leverage available data frameworks such as the e-wind 215 Lidar data format () to build ad-hoc modular processes.

Integrating airborne measurement systems
Although wind lidar partially mitigate the challenges of using meteorological masts in complex terrain, they do not allow high spatial resolution measurements. In contrast, measurement systems on unpiloted aerial vehicles (UAVs) including fixed-wing 9 https://doi.org/10.5194/wes-2022-11 Preprint. Discussion started: 7 February 2022 c Author(s) 2022. CC BY 4.0 License.
aircraft (Rautenberg et al., 2019), helicopters (Hofsäß et al., 2019), and multirotor drones (Molter and Cheng, 2020) can all be used to measure wind vectors and turbulence, as well as other parameters such as air pressure, temperature, and humidity.
However, such systems can usually only measure for short periods of time and only measure at one location.
Research is needed into ways to fly multiple systems simultaneously and autonomously (sometimes known as 'swarms'), and to combine the data from these airborne systems with other data sources. This process, known as 'sensor fusion', 'data fusion', or 'data assimilation' (when used as an input to models) is frequently used in weather forecasting but has not been an 225 active area of research for wind resource assessment.

Choosing the right measurement instrument
The choice of optimal wind measurement device as well as its location and the measurement time period is a critical part of a wind resource assessment and site operations, and is especially challenging for complex sites due to the increased uncertainties as described above. There is currently no existing guideline, standard or tool available to project planners for doing this.

230
Therefore the development of guidelines, standard or tools for choosing the right measurement instrument would be required to address this challenge.

Higher demands on wind field modelling tools
Wind fields across planned wind farm sites are typically generated using a combination of on-site measurements with some form of flow model. The modelling is used to extrapolate from on-site data from a few locations that might only extend to a 235 limited height above ground, to the tip of the potential wind turbines across the whole site. This is often known as horizontal and vertical extrapolation.
There is broad academic consensus that the linear flow models that are often used in "simple" terrain simply cannot predict the winds and weather at complex sites because of the steep and changing slopes, forestry, and the effects of atmospheric stability. There is hope, however, that models that include additional physics may allow to capture effects such as buoyancy 240 or forest canopy effects (see e.g., Knaus et al., 2017;Letzgus et al., 2018). These models are often described collectively as Computational Fluid Dynamics (CFD) models. In turn, there are many different types and fidelities of CFD models, ranging from Reynolds-Averaged Navier Stokes (RANS) models, Detached Eddy Simulation (DES) models, Large Eddy Simulation (LES) models, and Lattice Boltzmann Method (LBM) models (Schubiger et al., 2020). where LES models are often used in the wind energy community for time-resolved simulations of complex turbulent flows as they offer the ability to resolve 245 turbine-scale flows in realistic terrain (see e.g., Breton et al., 2017).
However, even high-fidelity LES models cannot overcome the greater difficulty of simulating the real flow. As an illustration Figure 5 shows the Root Mean Square Error (RMSE) of 10 m wind speed of the numerical weather prediction model ICON-D2 in March 2021 for sites below 100 m a.s.l., i.e. for rather flat terrain, and for sites above 800m a.s.l., i.e. in hilly and mountainous terrain. The RMSE is only calculated for sites which are accepted by the assimilation system. The wind speed at 250 sites at higher elevations (i.e hilly and mountainous terrain) is less well predicted than sites at lower elevation. These more complex flow models need realistic boundary conditions to deliver accurate results. These may include pressure gradients, surface temperature and moisture conditions, solar radiation or surface heat fluxes, upwind wind profiles, forestry parameters, and other data (Bechmann, 2017). Often these models have many "tuning" parameters, and it is not clear if one set which was successfully used in one case study is equally good for all weather situations at a different place. As a result, 255 complex models are harder to use than simpler models, both in terms of the data required and the knowledge required to assess the results.
Additionally, there is often no clear evaluation data available for such models, and therefore it is difficult for wind resource engineers to decide on the most effective model for a given site. Although software developers often provide site-specific case studies, the lack of a clear definition of complexity and an applicable, relevant comparison metric means that it is difficult to 260 transfer experience from one site to another.
The main research need resulting from the challenge of modelling wind fields at complex sites is to improve atmospheric models. Major improvements for wind energy modelling can be expected from better boundary layer schemes and turbulence models. Schemes for the surface layer are often based on Monin-Obukhov theory, which is strictly valid only for homogeneous sites. Also, in turbulence schemes for atmospheric models horizontal gradients of fluxes and other second order moments are 265 usually neglected compared to vertical gradients. It is not clear at which resolution in complex terrain this simplification is no longer valid. Direct numerical simulation of turbulence is too costly for wind energy assessment.
All models have tuning constants with values obtained in comparing model results to experiments. However, care must be taken not to deteriorate model results when changing them (Sandu et al., 2013). Especially for weather prediction models there can be conflicting interests. One quantity might improve, but, another one might deteriorate. Hence, changing established values must be done carefully though it might be beneficial in complex terrain.
Even at mesh sizes of only 2 or 3 kilometers, the subgrid-scale orographic drag must still be parameterized (Olson et al., 2019). The tuning constants of a sub grid scale scheme depend on the ratio of resolved to unresolved orography which depends on the resolution of the original data used to produce input fields for a subgrid scale scheme. This can be critical for the simulation of winds at typical hub heights.

275
Accurate numerical schemes are always critical in complex terrain. Especially the calculation of horizontal pressure gradients should not yield spurious circulations in terrain-following coordinates (Zängl, 2012).
A second research need resulting from the challenge of modelling wind fields at complex sites is to develop a decision tool for the optimal choice of wind modelling tool. There is a clear need for software or services that uses consistent rules to set up and run such models, hiding the complexity from the user and thus making it easier for users to adopt them (see e.g., WindNinja 280 [Wagenbrenner et al. (2016)] or WAsP CFD [Bechmann (2017)]). Furthermore, rules-or process-based modelling would give data consumers confidence that the tools have been used appropriately. Recent work involved the development of a decision tool for the optimal choice of WRA tool for a given project at complex sites (Barber et al., 2020a(Barber et al., , b, 2021. However, in order to fully develop an effective decision tool, a much larger set of data related to different site complexities, model set-ups and costs is required. The required skill upgrade, cost of data, and the lack of evaluation data all act as barriers to the adoption of 285 more advanced wind modelling tools. Finally, wind modelling needs to follow repeatable, auditable processes that provide the end user with confidence that the results are trustworthy and based on experience gathered at other sites, rather than each site being an independent study. This will require the wind energy industry develop software and services to consistently set up wind flow models. This can be combined with the data sharing framework discussed in Section 7.2. 290 3.7 Difficulties in predicting future wind climates An important part of a wind resource assessment is predicting the future wind climates. This estimate may need to extend up to 30 years in the future. It is typically carried out by comparing site data to some kind of long-term reference weather data, such as observations from a nearby automated weather station or reanalysis models, and using this to predict the future wind resource. This process is known as measure-correlate-predict (MCP) and takes many forms.

295
Errors depend mainly on the length of the measurements, and the correspondence to the long time series at the reference site. Both sites must have a similar wind climate, e.g. a coastal station with frequent sea breezes can hardly be correlated with an inland mountain site.
Such extrapolations depend upon the regional climate in the future being comparable to the past. However, it is not clear if this will be true as climate change occurs. On complex sites this also concerns changes in long-term and seasonal snow cover in 300 alpine and mountainous regions, which may affect surface temperatures and the associated valley wind systems that contribute to wind energy in some areas. Climate change may also result in changes to the frequency and intensity of storm systems and icing and extreme weather events. While many of the effects of climate change are negative, some may also have positive 12 https://doi.org/10.5194/wes-2022-11 Preprint. Discussion started: 7 February 2022 c Author(s) 2022. CC BY 4.0 License. consequences for wind energy; it is, e.g., possible that turbines at higher elevations might experience less icing in future than previously, raising their energy production.

305
Although some attempts have been made to predict the effects of climate change on wind climates, results suggest that these effects may be strongly localised and site-specific and as such, cannot be captured using today's relatively coarse global climate models (Pryor et al., 2020). Although regional climate models offer higher spatial resolution, they may be out of reach of developers and the wind industry because of the specialist knowledge required to use them.
The first research need related to improving the prediction of future wind climates at complex sites are affordable MCP 310 processes that can account for the complex wind situations and climates found at complex sites. They need to be reasonably easy to use so that they can be applied quickly to different locations as part of the site design process. They also need to be validated using existing sites.
Furthermore, it will be important to develop MCP processes that include the effect of climate change on complex sites.
Because the effects of climate change in mountainous regions are potentially highly localised and site-specific, this might not 315 be an automated process initially and could instead take the form of expert opinions for sites, identifying risks associated with different weather conditions due to climate change. In any case, there are no industry-standard approaches to assessing the impact of climate change on wind energy developments at this time.
The second research need related to improving the prediction of future wind climates at complex sites is to improve methods for estimating the effect of climate change on wind farm performance. (Pryor et al., 2020) 320 4 Project planning Following the resource assessment, an energy yield assessment is carried out by firstly using the results of the resource assessment to choose turbines' locations, and then estimating AEP. This information is then used for financial and risk estimations.
Finally, public acceptance for the project has to be gained for it to proceed. The specific challenges related to these steps include (1) Increased uncertainty of wind turbine performance models; (2) Site-specific wind farm design; (3) Increased financial un-325 certainties; (4) Increased conflict potential between stakeholders. These challenges and the resulting R&D needs are discussed in the next sections.

Increased uncertainty of wind turbine performance models
The power produced by a wind turbine is a function of wind speed, air density, turbulence intensity, shear, veer and many other factors. Although it is common to apply a site density correction to the power curve according to IEC 61400-12-1 (IEC 330 International Electrotechnical Commission, 2017), the other factors are difficult to account for, and therefore less frequently considered. Studies have shown that such atmospheric conditions can affect the turbine output by 10% or more at the same wind speed, leading to a significant uncertainty in power prediction even if the wind speed and density are known (e.g., Antoniou et al., 2009;Hedevang, 2014;Vanderwende and Lundquist, 2012;Wagner et al., 2011;Wharton and Lundquist, 2012;Clifton et al., 2014;Barber and Nordborg, 2020). Tools that can predict performance at specific sites are therefore required. As well as wind speed, these need to take into account other atmospheric conditions at a turbine's location -such as shear, veer, and turbulence intensity -to estimate the power output at that location. They could be based on experience and leverage data sets from existing power performance tests to generate binned statistics (as explored by the Power Curve Working Group in Lee et al., 2020), or use physicsbased approaches as in the IEC 61400-12-1 standard (IEC International Electrotechnical Commission, 2017). Physics-based 340 approaches have the advantage of being repeatable and easily understood by people, but may not make the best use of the large amount of data available to the wind energy industry.
In contrast, machine learning has the potential to account for the effects of unknown or hard-to-model physics by using power performance data sets to train turbine performance models. These trained models can be used in place of power curves or physics-based models. Studies suggest that power predictions by machine learning tools trained on wind speed, turbulence, 345 and other atmospheric parameters can reduce the error compared to simple power curves (Clifton et al., 2013;Barber and Nordborg, 2020). The application of machine learning methods to real measurement data is on-going (see e.g., Barber et al., in review). It may be possible to leverage data from power performance tests across a fleet to make more accurate machine learning models. However, machine learning approaches suffer from being "black boxes" in that it is often impossible for a human to understand what they contain. This can make it hard to include them in a turbine supply agreement or a warranty, for 350 example.
Collaboration between research and industry is required to develop and test more complex power performance prediction tools that use multiple parameters or machine learning, and mitigate the barriers to their adoption.

Site-specific wind farm design
Wind farms are usually designed with the goal of minimising the long-term cost of energy from the site, usually termed the 355 Levelised Cost Of Electricity (LCOE). This is done by optimising the number, size, and layout of turbines on site to maximise energy production and minimise operating costs (Clifton et al., 2016). Accurate wind field models ( §3.6) and long-term wind climate data ( §3.7) are essential to this; they are foundational for the process of estimating turbine energy production or longterm viability.
Wind turbine energy production at a site is a function of the energy that can be harvested by a turbine, and the losses from 360 that turbine. Wind resource data can be used to predict the energy available from a turbine using power curves (with and without adjustment for turbulence, shear, and veer) or aeroelastic models (e.g. NREL's FAST and others), while other models are required to predict the wakes from those turbines and their impact on downwind machines. It is also important to account for losses due to environmental effects such as blade soiling, and the formation of ice on the turbine blades or instrumentation.
Knock-on effects such as turbine shutdowns to minimise ice throw, or slower maintenance in challenging weather should also 365 be included in the plant energy yield assessment process.
Current wind turbine performance models are designed around inflow angles, shear, and turbulence that lie within standard ranges (defined in e.g., the IEC 61400-12-1 standard). Although there has been some effort to develop power curves that cover a wider range of conditions, these have not been widely adopted or tested openly for complex sites ( §5.3). Wakes from turbines from more complex terrain, where it is possible that increased turbulence and inclined flows may lead to faster dissipation (Menke et al., 2018).
The challenge is therefore to provide wind farm designers with the information that they need to optimise a wind plant at a complex site. This includes appropriate wind fields and an icing climatology for the location, turbine performance models that can account for non-standard operating conditions, and wake models that capture the effect of complex flow and terrain on 375 wakes.
Many different wake models exist and many have been validated for use in simple terrain. However, it is not clear how well these models perform in complex terrain or in complex weather situations. Validated wake models would allow increased confidence in energy yield analysis carried out in complex terrain locations. Wake models could be validated through field measurements, for example combining data from from met masts, wind lidars and wind turbines (Menke et al., 2018). This 380 data would also allow the creation of new wake models. These improved wake models could be used to give better predictions of the wind resources available to downwind turbines.

Increased financial uncertainties
All of the previous factors lead to uncertainty in the potential income from a planned wind energy project.
Electricity from wind energy is usually sold through long-term energy supply contracts with a customer. If the contracts are 385 too expensive, the wind farm owner risks being underbid by another supplier. Therefore, the developer is under pressure to drive the cost of energy as low as possible. However, if these contracts are too cheap (i.e. energy is sold at less than the cost to produce it), the owner risks losing money. To protect against such risks, the project financiers can increase the interest rates on any loans, which in turn increases the project cost and the LCOE.
Project developers typically mitigate these risks by carrying out extensive and detailed pre-construction studies. While these 390 may be more expensive at complex sites than are required in simple terrain, they can reduce the uncertainty enough to reduce the overall project costs and thus justify the extra expense, especially if the site has a high capacity factor. However, there are no guidelines or standards for doing this.
In order to approach the challenge of planning and financing with uncertainties, a guideline for dealing with additional risk related to complex sites is recommended. This would allow project developers to mitigate the risks by carrying out extensive 395 and detailed pre-construction studies in a standardised and agreed-upon way.

Increased conflict potential between stakeholders
Developing and operating wind energy projects involves a large number of stakeholders. As well as those directly involved with the development, they affect local residents, visitors, and people further away through visual impact, shadow flicker, sound, traffic, and other mechanisms.

400
The acceptance of wind farms by stakeholders is one of the major barriers to the adoption of wind energy. Acceptance must be considered for all wind farm developments, both on land and offshore. Experience suggests that wind farm acceptance can be increased through appropriate and sympathetic wind farm visual and acoustic design (Hübner et al., 2019), coupled with positive stakeholder engagement (Pohl et al., 2018). These challenges may become harder at complex sites because hilly or mountainous regions may be important for tourism or recreation, wildlife, or other uses, leading to potential conflict between 405 stakeholders (see e.g., Straka et al., 2020).
Also, it is possible that the physical processes linked to social acceptance may be harder to predict in complex terrain or at complex sites. Sound propagation from wind turbines is fairly well understood over flat and uniform terrain in uniform wind conditions and can be modelled with some accuracy. In contrast the physical effects of complex terrain or patchy landcover on sound propagation are less well understood and sound reflection by terrain or damping by forestry have only recently started 410 to be explored (see review in Hansen and Hansen, 2020).
Securing public acceptance is thus one of main challenges the development of wind energy has to face in the next decades. This is part of the growing need to obtain public acceptance -and even more important support -for the far-reaching technological changes connected to the transformation to a carbon-neutral energy generation and the associated social and economic impact. Developing wind energy in complex terrain is just one focus point where, e.g. the prominent and highly visible siting 415 of wind turbines on peaks and ridges in mountainous regions, may evoke concerns about landscape conservation and touristic and recreational uses. Technical measures such as reducing and managing of the wind turbine's sound and light emissions or changes in turbine design and wind park layout may contribute to a certain degree to the alleviation of these concerns.
However, social acceptance of wind energy in complex terrain might also grow from ongoing social transformations through policy making, fostering of the public understanding of the need for renewables, and the personal participation and benefit from 420 renewable energy projects. One of the initiatives on this interface between technology and social research is the IEA UsersTCP, which also has a big focus on the social acceptance of clean energy technologies.

Wind turbine design
Complex sites pose challenges for wind turbine design due to the complex flow conditions. This includes (1) Increased importance of quantifying the operating conditions at complex sites; (2) Higher complexity of input conditions for wind turbine 425 modelling; (3) Identifying the freestream wind speed for power performance measurements; (4) Identifying the freestream wind speed for mechanical loads testing; (5) Taking into account icing in the design. These challenges and the resulting R&D needs are discussed in the next sections.

Quantification of operating conditions at complex sites
Operating conditions at sites in complex terrain often fall outside "typical" values. Historically wind energy standards focuson 430 the operating conditions found in development areas such as those found in northern Europe or the American mid west, giving rise to a few standard turbine classes. Small deviations from the operating conditions in such sites are captured using special classes. The design conditions for each special class have to be determined on a case-by-case basis, requiring extra measurements or modeling of the site and extra effort by the turbine OEM, and thus raising costs. And, the lack of understanding of the operating conditions at complex sites results in a combination of mechanically conservative designs (i.e. with larger safety factors), but may also result in unexpected component failures.
Since then there have been efforts to develop guidelines or standards for wind energy developments in cold climates 1 , but in general the trend has been to consider complex terrain to be unique sites and require local measurements of operating conditions as well as extrapolation to the plant life cycle. This is problematic as it add costs for the developer and turbine supplier, and slows down the development process.

440
Research is needed to develop tools that can cheaply, accurately, and quickly define operating conditions over complex terrain. These tools also need to account for the effect of forestry and be capable of predicting complex weather associated with the site. This could include realistic time series or spectra of wind resources and complex weather, akin to the standard operating conditions defined in the IEC 61400 family of standards.

Wind turbine modelling 445
Wind turbine design is carried out in the wind energy industry according to the IEC 61400-1 standard (IEC International Electrotechnical Commission, 2019a). In this standard, so-called "load cases" are defined. These refer to particular combinations of external conditions and wind turbine operational status, which have to be considered when simulating wind turbine performance in the design phase.
These simulations are particularly challenging to set up and carry out for turbines at complex sites due to the higher turbu-450 lence intensity, shear, veer, temporally varying temperature gradients and extreme changes in wind speed and wind direction.
Validating wind turbine aeroelastic modelling requires accurate, high-resolution information about the inflow to a test turbine, coupled with data from loads and electrical sensors. Meteorological towers that are tall enough are hard to build and operate, while assumptions need to be made about the structure of atmospheric turbulence. Ground-based wind lidar can be used in some cases, but are not ideal in complex terrain situations.

Power performance testing
Power performance testing according to IEC 61400-12 (IEC International Electrotechnical Commission, 2017) is done as part of the certification process of a new wind turbine type. Power performance testing relates the power produced by a wind turbine to the free-stream wind conditions. Power performance testing in simple, flat terrain using upwind masts or vertically-profiling remote sensing devices is covered 460 by the IEC 61400-12-1 standard (IEC International Electrotechnical Commission, 2017). This standard specifically excludes winds from directions where there are steep slopes or obstacles from the power performance database. This is because in these conditions it is extremely challenging to identify an appropriate free-stream wind speed, as there may be terrain-induced speed-up or slow-down effects on the flow. As a result, there is no widely-recognised way to perform a power performance test in complex terrain.
Investigations suggest that it may be possible to fit wind speed measurements made by a nacelle-or spinner-mounted wind lidar looking forward into the turbine's induction zone to a model of the induction, and use this model to estimate the freestream wind speed (Borraccino et al., 2017). This approach would allow a power curve of power versus free-stream wind speed but has not been widely tested, or standardised.
The recently published IEC 61400-50-3 standard (IEC International Electrotechnical Commission, 2019b) for the use of 470 nacelle-mounted lidar for power performance testing describes the use of wind lidar to measure the turbine inflow wind speed.
The wind is required to be measured at more than 2D upwind of the turbine. Modern wind turbines can have rotors with diameter D of more than 150 m and so this could require wind measurements at well over 300 m upwind. However, complex flow conditions could introduce significant flow variation between the measurement point and the turbine (Figure 6), and so it is not clear that the method can be reliably used in complex terrain.

Mechanical loads testing
The certification process of a new wind turbine type requires mechanical load measurements. These are carried out according to to IEC 61400-13 (IEC International Electrotechnical Commission, 2015). The wind measurements required for this testing are covered by the IEC 61400-12 standard discussed above, and therefore the same challenges apply. Additional challenges to mechanical loads testing at complex sites relate to the complex behaviour of the loads on the rotor blades due to effects such 480 as shear and veer.
In order to help solve the challenges related to power performance testing and mechanical loads testing at complex sites, field measurements on large wind turbines situated at complex sites are required. This would allow an improved understanding of the actual behaviour of operating wind turbines in the field, enabling OEMs and researchers to improve their design tools and thus optimise design.

Design for icing
Icing impacts the turbine in several ways and these effects should also be taken into account in turbine design. IEC-61400-1 (IEC International Electrotechnical Commission, 2019a) outlines a number of issues caused by icing that need to be taken into account in turbine design. These include reduced turbine performance due to blade icing, unequal ice distribution on wind turbine blades leading to unequal loads and increased vibrations, ice shedding from blades, icing effects on wind measurements increased sound levels and prolonged standstills.
These conditions cause issues for turbine control due to ice accretion altering the blade aerodynamics. Turbine control during icing events can have different and competing goals, depending on operator objectives and local regulations. The priority can be chosen to maximise production, to be minimise risks or to minimise additional loads on the turbine components. Additionally, icing conditions might require extra instrumentation or changes in materials. Active icing mitigation systems such as blade 495 heating will often require changes in turbine design.
An important requirement when designing a turbine to operate in icing conditions is to understand and quantify how ice builds up on the turbine blades, and how this will affect the turbine aerodynamics. This would need to be taken into account when doing simulations during turbine design.
There are existing solutions for these issues. For example, icing on the blades can be mitigated by a blade heating system, 500 anemometers are available on the market that function better in icing conditions, and the risks caused by ice shedding and the issues with increased noise levels can be taken into account when planning the site. The IEA Wind TCP Task 19 report "Available technologies for wind power in cold climates" lists the state-of-the-art solutions that exist in the market (Lehtomäki, 2016). In 2019 Task 19 did a survey on experiences with blade heating and other cold climate solutions and found that many people working in the field still feel that there is room for improvement in the maturity and reliability of these solutions 505 (Godreau and Tete, 2020). This need for continued testing is part of the reasoning behind establishing the Nergica test centre in Canada, and the RISE cold climate test centre in Sweden.
Many of the solutions for icing need to be designed in to wind turbines and wind plants. For example, safe operation in icing conditions, and optimal blade heating control, will require reliable ice detection. Any ice detection method should be able to react quickly to icing conditions and also be able to tell when icing conditions and active ice accretion end in order to optimise 510 turbine and blade heating control. In addition, if ice detection is done for safety reasons it's important to be able to tell when blades are ice free.
More detailed icing models are being constantly developed. These models are mainly being validated against wind tunnel measurements (Son and Kim, 2020). Measurements of water droplets during icing events would be very useful. Actual measurements of ice shapes from an operating turbine are required to validate these models.

515
There is a large uncertainty related to icing conditions and the icing of turbine air foils. The year-over-year variation of icing conditions can be large and will introduce a large uncertainty in operations. The impact of icing on turbine production also has a large variation that further introduces uncertainty in any estimates on production in an icing climate site. More research is needed to reduce the margin of error in forecasting and modelling production losses (Strauss et al., 2020).
In order to determine the need for icing mitigation, the existence of icing conditions at the site needs to be determined early 520 during site prospecting. The methods for converting these pre-construction measurements into estimates on production losses still have room for improvement. Also, icing conditions need to be known before construction starts in order to determine the need for a blade heating system and the specific operating envelope of a blade heating system (Roberge et al., 2019). complexity impacts different aspects of the development and operation of wind energy facilities in different ways. Therefore the criteria used to assess complexity for flow modelling using WAsP or power performance testing are not interchangeable, and should not be used for other applications.
Furthermore, many studies have shown that other modelling tools do not have the same limitations as WAsP (e.g., Tabas et al., 2019). As a result, it only makes sense to use an absolute value of RiX as a division between "complex" and "not 660 complex" terrain when using WAsP, but not as a general metric for all modelling tools.
A binary definition of "complex" or "not complex" is difficult to translate into project uncertainties or risks, which are usually assessed on an continuous scale from 0-100% (uncertainties) or at least in 3-4 different categories (risks). This binary definition is in turn difficult to translate into "go" or "no-go" decisions, or to use for deciding which tool or workflow to use for a particular project.

665
As an illustration of these challenges, imagine three wind farm development sites in complex terrain. One is a sparsely wooded and moderately uniform slope, another is a mountain ridge, while another is in a mountain pass (Figure 7). Experience suggests that each site will have different pre-and post-construction challenges related to the interaction of the terrain and weather: -A uniform slope is subjectively relatively simple terrain, but is likely to experience a diurnal cycle of up-and down-670 slope winds, driven by surface heating and cooling. As a result, wind flow models are required that can generate such buoyancy-driven flows.