Mapping and Forecasting Icing on Structures
More and more wind energy projects are situated in areas affected by icing, such as Canada, Scandinavia and the Alps. Icing causes a decrease in power output and an increase in risk (e.g. by ice throw). For that reason collecting information about icing conditions is a crucial part of site assessment in many regions of the world. Additionally, icing forecasts are important when operating a wind farm under icing conditions in order to reach optimal performance. Icing itself is difficult to measure and a dense observation network is mostly missing. Numerical weather forecast models have the potential to fill the gap. In this article, a coupled model system consisting of a weather forecast model and an icing algorithm that simulates ice accretion on a cylindrical structure is presented. The system’s ability to simulate icing events is investigated. Additionally, the use of model simulations for mapping icing frequencies is shown. Finally, other information gained from model simulations and future challenges are discussed.
By Silke Dierer and Rene Cattin, Meteotest, Thomas Grünewald and Michael Lehnin, WSL Institute for Snow and Avalanche Research and Philippe Steiner, Federal Office of Meteorology and Climatology, Switzerland
Weather forecast models offer a relatively cheap and efficient way of receiving information about atmospheric conditions – even in remote areas. Additionally, they allow the possibility of predicting future developments. The aim of the studies presented is to evaluate the potential of using the new generation of high-resolution weather forecasting models (1–3 km grid size), coupled with an icing algorithm, for mapping icing risk as well as for icing forecasting purposes. Ice load is simulated using the icing algorithm proposed by Makkonen (2000), which is driven by the wind, temperature and cloud water information of weather forecast models. This method of icing modelling became feasible in recent years due to increased computing resources and improved accuracy of cloud microphysical schemes. A brief overview of the models is given in the following sections.
The Icing Algorithm
The icing algorithm describes ice accretion on a cylindrical structure caused by in-cloud icing (Figure 2). Based on information about temperature, liquid water content (LWC), wind speed and volume number concentration of droplets, the liquid water mass flux hitting the cylindrical structure is calculated. The part of liquid water that contributes to icing is described by several coefficients. The collision efficiency describes the ratio of droplets that actually hit the object while small particles are transported around the obstacle (Figure 2). The sticking efficiency describes the ratio of droplets hitting the object that adhere minus the particles that bounce from the surface. The accretion efficiency describes the part of the adhered droplets that contributes to the rate of icing when the heat flux is strong enough to cause sufficient freezing. The result is the change of ice mass per time step.
Weather Forecast Models
Numerical weather forecast models have undergone strong development in the last few years. Today’s operational high-resolution forecast models have grid sizes in the range of 1–3 km. This resolution allows a good description of the terrain as well as atmospheric gradients. Additionally, physical and numerical parameterisations in the models have been improved. This allows the use of numerical weather forecast results in applications that were impossible until recently, such as icing simulations. Two state-of the-art weather forecast models are used in the presented studies: WRF is a high-resolution weather forecast model mainly developed by NCAR (USA) used with grid sizes down to 800 m. COSMO is a high-resolution weather forecast model that is in operational use in several countries (e.g. Germany, Switzerland and Italy). For the present studies the Swiss operational COSMO-2 model with a grid size of 2.2 km is used.
Forecasting Icing on Structures – Case Studies
Eleven icing periods during winter 2008/09 were simulated for three sites in Switzerland. All sites are situated in complex terrain (Jura, Pre-Alps, Inner Alps), Guetsch in the Inner Alps being the most complex one. The comparison of simulated and measured ice mass shows that most of the icing events are captured by the model (Figure 3). The model has a good potential to describe the frequency of icing (Figure 4). Ice load is predicted less precisely (Figure 4); this might be connected to limited knowledge about the volume number concentration of droplets, which strongly affects the simulated ice mass. Additionally, the ice mass measurements used for comparison are uncertain within a range of about 0.5 kg/m. Sensitivity studies show that for the Jura and the Pre-Alps grid sizes of around 2–3 km are sufficient, while grid sizes of around a few hundred metres are needed in the very complex terrain found in the Inner Alps.
Mapping Icing Frequencies
A map of icing frequencies for Switzerland (2007–2009) at 2.2 km grid size has been created in order to identify regions where icing might affect wind energy developments (Figure 5). COSMO-2 analysis data has been used to drive the icing algorithm. Analysis data has the advantage that model results can be pulled closer to reality by the assimilation of measurement data. Simulated icing frequencies are compared to those derived from measurements at IMIS stations. The model reproduces the general icing climatology quite well; the icing frequency predictions averaged over all stations and the two years of data agree well with the measurements. Still, local icing conditions which vary within a few kilometres, such as in the Inner Alps, cannot be reproduced by a model at this resolution, which is better at providing more regional-scale information. For moderately complex terrain, like the Jura, evaluations show a good representation of the measured icing frequencies between 20 and 50 days/year at 50 m height.
The Model’s Added Value
The case studies show that the model system is a suitable tool for mapping, as well as for forecasting, icing frequency. With grid sizes of 2–3 km good results are achieved in moderately complex terrain. In very complex terrain, such as the Inner Alps, higher resolutions down to a few hundred metres are necessary to get good accuracy at local level. However, if the main interest is maximum ice load the model results are still very uncertain; one reason for this is the uncertainty of the cloud droplet volume number, which is mostly unknown. Better estimates are expected to come from weather forecast models in the next few years. Case studies in other geographical regions have also shown promising results (Nygaard, 2009). Therefore, icing maps derived from weather forecasting models are a good way to determine the icing frequency at planned wind parks and can contribute to a better wind resource and risk assessment, and a more accurate energy yield prediction.
Future Challenges
The above findings are based on a limited number of test cases and so need verification by running more simulations in different geographical regions. A shortcoming of the existing model that needs to be addressed is the simulation of icing when it is affected by very local conditions. Increasing the resolution of weather forecast models to a grid size of a few hundred metres could be a solution, but this development is unlikely in the next few years due to the limitations imposed by computing resources. Possibly post-processing methods using subgrid-scale topography might help to produce results incorporating local effects better. The current studies simulate accretion on a cylindrical structure in order to make comparisons with field measurements. Ice accretion on wind turbine blades is a more complex process. Models describing icing on wind turbine blades exist and could be coupled with numerical weather forecasts. Finally, the use of icing and wind forecasts with respect to the behaviour of a wind park under icing conditions will be one of the next challenges.
Biography of the Main Author
Silke Dierer studied meteorology at the University of Hamburg and for her PhD investigated polar mesocyclones. Between 1998 and 2004 she was a scientist working on mesoscale weather modelling at the Meteorological Institute, University of Hamburg, and this was followed by studies and work at, first, ISAC-CNR, Rome, then the Federal Office for Meteorology and Climatology, Zurich. Since 2007, Dr Dierer has become more involved in the field of wind energy and weather modelling, and is now responsible for weather models at Meteotest, Bern.{/access}
More and more wind energy projects are situated in areas affected by icing, such as Canada, Scandinavia and the Alps. Icing causes a decrease in power output and an increase in risk (e.g. by ice throw). For that reason collecting information about icing conditions is a crucial part of site assessment in many regions of the world. Additionally, icing forecasts are important when operating a wind farm under icing conditions in order to reach optimal performance. Icing itself is difficult to measure and a dense observation network is mostly missing. Numerical weather forecast models have the potential to fill the gap. In this article, a coupled model system consisting of a weather forecast model and an icing algorithm that simulates ice accretion on a cylindrical structure is presented. The system’s ability to simulate icing events is investigated. Additionally, the use of model simulations for mapping icing frequencies is shown. Finally, other information gained from model simulations and future challenges are discussed.By Silke Dierer and Rene Cattin, Meteotest, Thomas Grünewald and Michael Lehnin, WSL Institute for Snow and Avalanche Research and Philippe Steiner, Federal Office of Meteorology and Climatology, Switzerland
{access view=!registered}Only logged in users can view the full text of the article.{/access}{access view=registered}Icing has a strong effect on the planning and the operation of wind turbines (Laakso et al. 2009): it influences the aerodynamics of the blades and causes production losses (Figure 1). Moreover, additional ice loads lead to extreme and fatigue loads. Iced wind measurement sensors at the wind turbine’s nacelle lead to erroneous behaviour and security stops. Finally, ice throw represents a significant safety risk for pedestrians and service personnel. For that reason detailed knowledge on frequency and duration of icing events as well as maximum ice loads are crucial components of site assessment in regions prone to these conditions. Additionally, the inclusion of icing within wind forecasts is important when operating a wind farm because it allows operators to optimise performance and reduce risk.
A Model Simulating Icing on StructuresWeather forecast models offer a relatively cheap and efficient way of receiving information about atmospheric conditions – even in remote areas. Additionally, they allow the possibility of predicting future developments. The aim of the studies presented is to evaluate the potential of using the new generation of high-resolution weather forecasting models (1–3 km grid size), coupled with an icing algorithm, for mapping icing risk as well as for icing forecasting purposes. Ice load is simulated using the icing algorithm proposed by Makkonen (2000), which is driven by the wind, temperature and cloud water information of weather forecast models. This method of icing modelling became feasible in recent years due to increased computing resources and improved accuracy of cloud microphysical schemes. A brief overview of the models is given in the following sections.
The Icing Algorithm
The icing algorithm describes ice accretion on a cylindrical structure caused by in-cloud icing (Figure 2). Based on information about temperature, liquid water content (LWC), wind speed and volume number concentration of droplets, the liquid water mass flux hitting the cylindrical structure is calculated. The part of liquid water that contributes to icing is described by several coefficients. The collision efficiency describes the ratio of droplets that actually hit the object while small particles are transported around the obstacle (Figure 2). The sticking efficiency describes the ratio of droplets hitting the object that adhere minus the particles that bounce from the surface. The accretion efficiency describes the part of the adhered droplets that contributes to the rate of icing when the heat flux is strong enough to cause sufficient freezing. The result is the change of ice mass per time step.
Weather Forecast Models
Numerical weather forecast models have undergone strong development in the last few years. Today’s operational high-resolution forecast models have grid sizes in the range of 1–3 km. This resolution allows a good description of the terrain as well as atmospheric gradients. Additionally, physical and numerical parameterisations in the models have been improved. This allows the use of numerical weather forecast results in applications that were impossible until recently, such as icing simulations. Two state-of the-art weather forecast models are used in the presented studies: WRF is a high-resolution weather forecast model mainly developed by NCAR (USA) used with grid sizes down to 800 m. COSMO is a high-resolution weather forecast model that is in operational use in several countries (e.g. Germany, Switzerland and Italy). For the present studies the Swiss operational COSMO-2 model with a grid size of 2.2 km is used.
Forecasting Icing on Structures – Case Studies
Eleven icing periods during winter 2008/09 were simulated for three sites in Switzerland. All sites are situated in complex terrain (Jura, Pre-Alps, Inner Alps), Guetsch in the Inner Alps being the most complex one. The comparison of simulated and measured ice mass shows that most of the icing events are captured by the model (Figure 3). The model has a good potential to describe the frequency of icing (Figure 4). Ice load is predicted less precisely (Figure 4); this might be connected to limited knowledge about the volume number concentration of droplets, which strongly affects the simulated ice mass. Additionally, the ice mass measurements used for comparison are uncertain within a range of about 0.5 kg/m. Sensitivity studies show that for the Jura and the Pre-Alps grid sizes of around 2–3 km are sufficient, while grid sizes of around a few hundred metres are needed in the very complex terrain found in the Inner Alps.
Mapping Icing Frequencies
A map of icing frequencies for Switzerland (2007–2009) at 2.2 km grid size has been created in order to identify regions where icing might affect wind energy developments (Figure 5). COSMO-2 analysis data has been used to drive the icing algorithm. Analysis data has the advantage that model results can be pulled closer to reality by the assimilation of measurement data. Simulated icing frequencies are compared to those derived from measurements at IMIS stations. The model reproduces the general icing climatology quite well; the icing frequency predictions averaged over all stations and the two years of data agree well with the measurements. Still, local icing conditions which vary within a few kilometres, such as in the Inner Alps, cannot be reproduced by a model at this resolution, which is better at providing more regional-scale information. For moderately complex terrain, like the Jura, evaluations show a good representation of the measured icing frequencies between 20 and 50 days/year at 50 m height.
The Model’s Added Value
The case studies show that the model system is a suitable tool for mapping, as well as for forecasting, icing frequency. With grid sizes of 2–3 km good results are achieved in moderately complex terrain. In very complex terrain, such as the Inner Alps, higher resolutions down to a few hundred metres are necessary to get good accuracy at local level. However, if the main interest is maximum ice load the model results are still very uncertain; one reason for this is the uncertainty of the cloud droplet volume number, which is mostly unknown. Better estimates are expected to come from weather forecast models in the next few years. Case studies in other geographical regions have also shown promising results (Nygaard, 2009). Therefore, icing maps derived from weather forecasting models are a good way to determine the icing frequency at planned wind parks and can contribute to a better wind resource and risk assessment, and a more accurate energy yield prediction.
Future Challenges
The above findings are based on a limited number of test cases and so need verification by running more simulations in different geographical regions. A shortcoming of the existing model that needs to be addressed is the simulation of icing when it is affected by very local conditions. Increasing the resolution of weather forecast models to a grid size of a few hundred metres could be a solution, but this development is unlikely in the next few years due to the limitations imposed by computing resources. Possibly post-processing methods using subgrid-scale topography might help to produce results incorporating local effects better. The current studies simulate accretion on a cylindrical structure in order to make comparisons with field measurements. Ice accretion on wind turbine blades is a more complex process. Models describing icing on wind turbine blades exist and could be coupled with numerical weather forecasts. Finally, the use of icing and wind forecasts with respect to the behaviour of a wind park under icing conditions will be one of the next challenges.
Biography of the Main Author
Silke Dierer studied meteorology at the University of Hamburg and for her PhD investigated polar mesocyclones. Between 1998 and 2004 she was a scientist working on mesoscale weather modelling at the Meteorological Institute, University of Hamburg, and this was followed by studies and work at, first, ISAC-CNR, Rome, then the Federal Office for Meteorology and Climatology, Zurich. Since 2007, Dr Dierer has become more involved in the field of wind energy and weather modelling, and is now responsible for weather models at Meteotest, Bern.{/access}
