1. Introduction
In Canada, the cold climate poses a considerable challenge to the wind power industry, and the magnitude and impact of icing on the production of wind plants are nonnegligible. Lacroix (A. Lacroix 2012, personal communication at Quebec’s Sixth Wind Energy Conference) has analyzed electricity generation data from wind plants during May 2010 and April 2011 using real output data and observed wind speeds at nearby weather stations. The result shows that the production loss (defined as the difference between the reference level of power production based on the wind speed from weather stations and the real power production from 24 wind plants across Canada) is highest in New Brunswick and Nova Scotia (26.5%), followed by Québec (12.4%), and Ontario (5.7%) for winter (between November and April). Applying the regional power loss to the regional electrical portfolio mix, Lacroix and Tan (2012) estimated the annual icing-induced production loss of existing wind plants in Canada to be 6.6%, which translates into around $100 million in lost revenue. Atmospheric icing thus results in a huge reduction in power production and also leads to large errors in power forecasts (Choisnard et al. 2012). Therefore, it is important to understand and predict the meteorological conditions favorable for icing events at wind plants, and to provide detailed forecast insight into icing start-up, duration, and shedding time, as well as icing amounts (loads).
In cold climates, in-cloud and precipitation icing are the main processes leading to ice accretion on structures, with the type of ice forming being dependent on the icing process. For example, in-cloud icing usually results in riming through a dry growth process in which supercooled cloud droplets freeze so quickly that air bubbles are trapped and there is no runoff of liquid water. The result is ice with a white and feathery appearance, the density of which ranges from 200 to 900 kg m−3 (International Organization for Standardization 2001) for soft and hard rime, respectively. Precipitation icing can result from freezing rain or wet snow. When freezing rain hits the object surface, a liquid layer is formed on the accretion surface, and freezing then takes place beneath this layer. This wet growth is a slow process, and air bubbles have adequate time to escape. The resultant glaze ice thus appears smooth and transparent, and is evenly distributed with the density ranging between 700 and 900 kg m−3. Wet snow accretion occurs at warmer temperatures (0°–3°C), as snowflakes partly melt in the air before adhering to and freezing on a cold surface. This results in relatively low-density ice ranging between 300 and 600 kg m−3 (Fikke et al. 2007).
Once there are favorable conditions, ice will accrete on cold structures such as wind turbine blades, tower structures, and power lines. The accreted ice can have adverse impacts on, for example, wind turbine operations. Accumulation of ice on wind turbine blades changes the shape and surface roughness of the blades and can lead to an imbalance between the blades. This deteriorates their aerodynamic performance and can result in reduced power production or outright stoppage of the turbines. When the temperature rises because of thermodynamic mechanisms such as solar radiation, warm advection, etc., ice thaws from the turbine blades and drops off (ice shedding). This presents an additional risk of damage to the structures and the environment. Moreover, the random shedding of ice from rotating blades can cause imbalanced ice loads on the wind turbine blades, which increases the wear on the components, accelerates fatigue, and can slow down or even damage the turbines. If the wind turbine is close to a highway or agriculture field, ice throw could be a safety issue that could result in damage and even threaten the loss of life near the turbine site (Morgan and Bossanyi 1996; Morgan et al. 1998; Seifert et al. 2003; Cattin et al. 2007; Sørensen et al. 2012).
Statistical analysis is one way to study icing’s impact on power loss. For example, a statistical analysis of data from two production sites showed a production loss to be correlated with the icing duration time and wind speed (Karlsson et al. 2013). The analysis of field data is invaluable but expensive, and requires long-term observations available only at a limited number of sites. As a result, state-of-the-art numerical models have become useful tools in icing studies, providing high-resolution data temporally and spatially. In recent years, meteorological and icing models have been widely used in icing studies (Hošek 2007; Drage and Hauge 2008; Nygaard et al. 2007a,b, 2011; Dierer et al. 2011; Podolskiy et al. 2012; Yang et al. 2012; Byrkjedal 2012a,b; Soderberg and Baltscheffsky 2012; Bernstein et al. 2012). These studies compared the simulated meteorological fields, cloud properties, and/or ice loads with observations to explore the utility of atmospheric mesoscale and icing models in either in-cloud icing or wet snow events.
Numerical simulation is directly used to study in-cloud icing impacts on wind plant power loss in recent years (Byrkjedal 2012a,b; Soderberg and Baltscheffsky 2012; Bernstein et al. 2012; Karlsson et al. 2013; Turkia et al. 2013; Davis et al. 2014). Turkia et al. (2013) simulate ice accretion on blades and wind turbine power loss. They consider simulations of 20, 200, and 600 min as different icing event durations and interpret these as the onset of an icing event and light and moderate icing, respectively. Their focus is on the simulation of accreted ice amount and its effect on aerodynamic properties of the blades. However, they assume the same weather conditions (e.g., temperature, wind speed, liquid water content, and mean volume diameter) for all three simulated cases, and there is no concurrent ice load and meteorological information for their simulations.
To simulate weather conditions and calculate icing load, a high-resolution version of the Weather Research and Forecasting Model (WRF) and an in-cloud icing model (International Organization for Standardization 2001) are the most commonly used in the literature. Byrkjedal (2012a) used a slightly modified in-cloud icing model that accounts for melting and sublimation processes (calculated from the energy balance on the iced surface). A relation is found between modeled ice load, observed wind speed, and power loss. Based on this relation, the power production with ice is then calculated. Byrkjedal (2012b) used WRF with the Makkonen ice model and compared the forecast power production to three years’ worth of operational data from a wind farm in Sweden. Their results demonstrate the improvement of energy forecasts with icing impacts. With a slightly modified version of the Makkonen model, Soderberg and Baltscheffsky (2012) estimated long-term production losses in icing climates, demonstrating the production loss variability in icing climates. Bernstein et al. (2012) used the Finnish Meteorological Institute’s Local Analysis and Prediction System (LAPS) and “LOWICE” model to study icing on power production for a winter month in 2011. These studies simulate icing on cylinders, whereas Davis et al. (2014) modify the International Organization of Standards (ISO) icing model to account for ice sublimation and ice shedding, and simulate icing on the wind turbine blades. Their coupled model successfully captured the icing periods at a wind farm in Sweden, and the simulation results were demonstrated to be improved compared to the cylinder model simulation.
So far, most of the studies have focused on in-cloud icing impacts from power loss, except for a few studies which also consider the freezing rain or wet snow icing impacts on power loss (Makkonen 1998; Makkonen and Wichura 2010; Dalle and Admirat 2011). However, none of these studies consider the overall impact of the various icing mechanisms on wind power loss generation over a range of icing conditions. This study aims to explore the potential application of an atmospheric model to simulate icing events (freezing rain, wet snow, and in-cloud icing) and to use output from these simulations together with ice accretion models for quantifying the ice load effects of various ice types on power generation in wind plants. This paper proposes a framework for quantitative evaluation of icing impact on wind turbine operations: prediction of the onset and duration of icing events and building a connection between the modeled ice load and power loss of wind plants in an icing event.
The paper is organized as follows. Descriptions of icing events in wind plants in the Gaspé region of Canada are presented in section 2. The model’s description and configuration are addressed in section 3. Section 4 gives the detailed comparison of simulated surface meteorological fields with observations, and comparison of simulated ice load with power loss for two icing events, as well as the summarization of all the studied cases. Section 5 gives the discussions and concluding remarks.
2. Icing events occurred in wind plants of the Gaspé region
The current installed wind energy capacity for Canada is 7803 MW. The province of Québec ranks as the second largest producer of wind energy in Canada with 2398 MW, mostly from wind plants on the Gaspé Peninsula, a region of favorable wind conditions. The wind turbines at the Gaspé wind plants are General Electric types, with rated capacities of 1.5 MW. The hub is at a height of 80 m, and the rotor diameter and blade length are 77 and 38 m, respectively. From 30 years (1971–2000) of Canadian Meteorological Center (CMC) climate data and information archive, the average temperature in the Gaspé region is −10.1°C for winter months (December–February). All wind plants in the Gaspé Peninsula thus experience issues related to cold climate, and icing, in particular, is a challenging problem for the operation of these wind plants. Another reason for which wind plants in Gaspé were chosen for this study is that Environment Canada and Hydro-Québec have been collaborating on wind energy forecasting projects since 2007 (Yu et al. 2014). Model performance was analyzed using on-site observation data. It was found that weather extremes (e.g., extremely strong winds from hurricanes or thunderstorms, atmospheric icing during snowstorms) have strong impacts on wind turbine operations that can lead to the failure or damage of the wind turbine. As a continuation of that project, this paper deals specifically with icing events.
Twenty-seven icing events were identified for the period spanning 2008–10. These were selected based on an observed power loss of over 20% of the theoretically generated power and differences in wind speed as measured by heated and nonheated anemometers at a wind plant in Gaspé. These events were confirmed by data analysis and on-site observations (visual inspection when available) for classification according to icing type. (Table 1 lists the average of the meteorological variables and the power loss percentage during these icing events.) Among these events, most were of precipitation icing (freezing rain and/or wet snow), as only three in-cloud icing events were recorded for this wind plant. For all events, the near-surface temperature T3m ranged between 2° and −11°C and the relative humidity RH3m was high, with average values above 90% in most cases; that is, conditions were favorable for icing. For the longest-duration in-cloud icing/freezing-rain case (event 20, a duration of 105 h), the average near-surface air temperature was close to 0°C, and the air was almost saturated with respect to water, with an average relative humidity of 96%. There is significant discrepancy between the wind speed observed from heated (Vc80m) and nonheated (Vnc70m) anemometers, indicating the erroneous measurements from the nonheated anemometers resulting from icing. As indicated in Table 1, more than half of the icing events have resulted in power reduction of more than 50% of the theoretically generated value, as estimated from an empirical wind power curve and average wind power plant wind speed measurements. More details about the empirical power curve are given in Yu et al. (2014) (the power curve is based on the power production and average wind speed from all turbine sites at the plant). This theoretically generated value usually gives a very accurate estimate of power generation without icing. Eight cases (based on 12 events, as event numbers 8–10, 22 and 23, and 24 and 25 occurred sufficiently close in time to be considered single cases) are studied to assess the performance of the atmospheric model on icing simulations, and to propose a framework for quantitative evaluation of icing’s impact on energy production.
The durations of 27 icing events that occurred at one wind plant in the Gaspé region from 2008 to 2010, and average fields during the total icing hours: wind speeds from a heated anemometer (Vc80m) at 80-m height and from a nonheated anemometer (Vnc70m) at 70-m height at a meteorological tower, temperature (T3m), wind speed (V3m), relative humidity (RH3m), and near-surface pressure (P3m) at a height of 3 m from a meteorological tower at the wind plant. The final column is power loss by the wind plant.
3. Model description and configuration
a. Atmospheric model
Meteorological conditions for the icing events are simulated with the GEM-LAM model, a limited-area version of the Canadian Global Environmental Multiscale (GEM) model. Details of GEM are described in Côté et al. (1998a,b). GEM-LAM is a three-dimensional fully elastic nonhydrostatic model, which has been applied to various processes such as cloud studies (Bélair et al. 2005; Milbrandt et al. 2010), radiation processes (Paquin-Ricard et al. 2010), marine fog prediction (Yang et al. 2010), in-cloud icing study (Yang et al. 2012), and wind energy forecasting (Yu et al. 2014).
GEM-LAM uses a terrain-following coordinate system with 56 levels in the vertical and very high resolution in the boundary layer, including 16 levels within the lowest 1-km height (3.5, 10, 21, 35, 57, 87, 128, 180, 240, 311, 392, 482, 581, 691, 813, and 948 m). There are different options available for the physics schemes. Those used in this study are as follow: a solar and infrared radiation scheme (Fouquart and Bonnel 1980; Garand 1983) that is fully interactive with clouds (Yu et al. 1997); a planetary boundary layer scheme based on the time-dependent turbulent kinetic energy with fully implicit vertical diffusion (Mailhot and Benoit 1982; Benoit et al. 1989); a stratified surface layer scheme based on similarity theory (Benoit et al. 1989); aggregated surface variables combining predicted properties of four surface types such as land, glacier, ice, and water (Mailhot and Chouinard 1989); the Kuo convective scheme to model implicit precipitation; and the Milbrandt–Yau mixed-phase scheme for explicit precipitation (Milbrandt and Yau 2005). The Milbrandt–Yau grid-scale precipitation scheme is a two-moment scheme, which includes predictive equations for number concentrations and mixing ratio for water vapor, cloud water, rainwater, ice, snow, and graupel. Cloud condensation nuclei (CCN) concentration is one important factor in calculating the median volume diameter (MVD) of cloud droplets for in-cloud icing; this has been shown in results of Nygaard et al. (2011), Yang et al. (2012), and Davis et al. (2014). There is no difference in CCNs between maritime and continental regions in the current version of GEM-LAM, and cloud droplet concentration is a predetermined value over the entire simulation domain. In the GEM-LAM model, three values of CCN concentration are prescribed for maritime, continent, and polluted continent, respectively, and the value of 200 cm−3 for continent is used for the start-up of the model since the prevailing wind at this wind plant is continental. This was demonstrated to give cloud property fields closer to the observations than was the case using other cloud physics schemes for in-cloud icing cases (Yang et al. 2012).
In this paper, eight sets of numerical experiments were performed for the icing events during 2008–10. All of the experiments are performed over the domain covering 108 × 80 horizontal grid points, with a grid spacing of 15 km, whereas a nested inner domain contains 441 × 321 horizontal grid points with a grid spacing of 2.5 km (Fig. 1). The inner domain covers southern Québec, New Brunswick, and Nova Scotia provinces, including all wind plants installed in the eastern coastal region of Canada. The coarse-resolution (15 km) run of GEM-LAM was initialized with CMC regional forecast data, which were also at 15-km horizontal resolution, providing hourly lateral boundary conditions for the 15-km run. The time steps for the 15- and 2.5-km runs are 450 and 30 s, respectively. The 2.5-km run is initialized 3 h after the coarse-resolution run to allow for model spinup (i.e., to allow for GEM-LAM to adjust to the large-scale forcing). A series of 24-h 2.5-km runs is conducted, with the simulated results for the individual days being combined to build the time series of the modeled fields.
Nested geographical domains (15- and 2.5-km horizontal resolutions for the outer domain 1 and the inner domain 2 within the red box, respectively) for the simulation area over the Gaspé region, with terrain height indicated by the color shadings for domain 1 and the red contours for domain 2. The outer domain provides the lateral boundary conditions for the nested inner domain.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0125.1
b. Ice accretion models
To quantify the icing impact on wind power production, the ice load first needs to be estimated. The meteorological and cloud property fields from the mesoscale model are used to drive icing models to calculate this ice load. Three ice load models were used, respectively, for in-cloud icing, freezing rain, and wet snow processes. To keep the problem simple without losing generality, a static cylindrical line is used as a simple geometry of an object for the ice accretion model. Although the ice load on a static cylinder is not the same as that on the rotating blades of a turbine, it does provide an indication of the severity of the icing events.
1) Ice accretion and melting model for freezing rain
2) Ice accretion model for in-cloud icing
For in-cloud icing, w represents the cloud liquid water content (kg m−3) and V denotes the wind speed (m s−1). In GEM-LAM, part of the cloud liquid water content comes from the cloud mixing ratio in the Milbrandt–Yau explicit precipitation scheme and part comes from the boundary layer clouds in a unified cloudiness–turbulence scheme (Bélair et al. 2005).
The collision efficiency α1 ranges between 0 and 1, and is a function of Langmuir’s parameter and dimensionless Stokes number, which are determined by the cylinder diameter, wind speed, temperature, cloud liquid water content, and MVD of cloud droplets (Finstad et al. 1988). The meteorological fields are output from GEM-LAM, and MVD is diagnosed from the simulated cloud mixing ratio and number concentration. The sticking efficiency α2 is set to 1 since the liquid droplets adhere to the surface when they hit the cylinder instead of bouncing off (Makkonen 2000). The accretion efficiency α3 is also set to unity, which is equivalent to assuming all impacting particles freeze upon impact with the structure. To simplify the problem, the cylinder is assumed to be free to rotate and ice accretes uniformly around the wire. The diameter of the cylinder on which the ice is accreted is assumed to be 1.5 cm.
3) Ice accretion model for wet snow
4. Model results
In our model system, the icing types are categorized by temperature, cloud, and wet snow information at given levels. By definition, freezing rain is the liquid precipitation that falls when surface temperatures are below 0°C. However, on-site observations and data analysis indicate the occurrence of freezing rain despite above-zero temperatures observed near the ground. This is not unusual, as a few studies with observational data have revealed that freezing rain can occur when the surface temperature is slightly above 0°C. For example, a climatological analysis of 15 years’ worth of freezing rain in the Great lakes shows that around 28% of freezing rain occurs when the surface temperature is between 0° and 2°C, and the surface temperature during freezing rain can be as high as 4°C (Cortinas 2000). These are primarily the cases where the atmosphere is warming fast, so that the ground and surface object temperatures remain below 0°C despite above-zero air temperatures. As such, freezing occurs when liquid water comes into contact with the cold ground and object surfaces. Houston and Changnon (2007) analyzed hourly data from 236 first-order stations in the United States for the period between 1928 and 2001, and found around 8% of freezing rain occurred when the surface temperature was above 0°C. Considering the on-site observations and the aforementioned two studies, temperatures below 2°C instead of 0°C at a given model level (87-m height) and liquid precipitation occurrence from GEM-LAM are used in our model system to identify the occurrence of freezing rain. If there is solid precipitation in the model, and liquid water content exists in the falling snow, then the wet snow accretion model is activated to calculate the accretion rates. In-cloud icing is identified in the model when the temperature is below 0°C, and the cloud liquid water content at the given level is nonzero in the air. Once the conditions for any of the icing mechanisms are met, the corresponding icing models are activated to calculate the accretion rate and amount. For the icing models, the increasing diameter of the ice is fed back to calculate the collision efficiency of in-cloud icing; this is not done for freezing rain since, in that case, the ice thickness of the cylinder is independent of the cylinder diameter.
Of the 27 icing events observed between 2008 and 2011, 24 resulted from precipitation and three from in-cloud icing. These cases typically involve some combination of in-cloud icing, freezing rain, and wet snow; most are due to a combination of freezing rain and wet snow. In this study, the accreted ice from all of these processes will be taken into account. For better presenting the model results, detailed discussion and evaluation were done for two selected cases that, respectively, represent typical characteristics of freezing-rain/wet-snow and freezing-rain/in-cloud-icing events. A summarized analysis of the model results for all other cases is then discussed in section 4c.
On-site observation by visual inspection and analysis of observed data indicates a freezing-rain and wet-snow event (event 13 in Table 1) starting on 4 April 2009. This event was selected as case 1 since it had adverse impacts on power generation for a long duration (70 h), and the average power reduction was around 70%. Another long-duration icing event from 14 to 18 February 2010 (event 20 in Table 1) was selected for case 2. For this event, both in-cloud icing and freezing rain occurred, and the impact of ice accreted from both mechanisms was taken into account in the wind power reduction.
a. Case 1: Freezing rain and wet snow
The icing event (event 13) of 4–9 April 2009 (case 1) was reported as a freezing-rain and wet-snow event. It started as freezing rain on 4 April, followed by wet snow that lasted for around 4 days. Theoretically, freezing rain occurs when a layer of subfreezing temperature is below a warm layer. Figure 2 gives the modeled temperature profile at the start time of 100% power loss (0600 UTC 5 April 2009). As indicated, there is a temperature inversion above 600 m, and solid precipitation such as snow and ice crystals can fully melt as they travel through this inversion. The liquid precipitation becomes supercooled when passing through the subfreezing layer below, which is around 350-m thickness. The temperature of these droplets can be assumed to remain below 0°C as they reach the 87-m model height level (i.e., since the passage from the base of the subfreezing level to this level is short compared to the passage through the subfreezing layer itself). Also, the temperature at a height of 87 m is 0.7°C, which is consistent with our definition of freezing rain for temperatures below 2°C in the discussion above. The simulated meteorological fields are compared with observations first, and simulated precipitation and wind speed are then used to calculate the ice accretion amount when the freezing-rain condition is satisfied.
Vertical temperature profile within a boundary layer of 1 km at 0600 UTC 5 Apr 2009 at the wind plant for case 1 (4–9 Apr 2009).
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0125.1
1) Meteorological fields
During 4–10 April 2009, two low pressure systems to the southeast of Gaspé consecutively moved toward and stayed over the eastern coast for a couple of days. Surface pressure at the meteorological tower decreased with the northeast movement of the first cyclone, and then increased until 972 hPa on 7 April with the dissipation of the cyclone (Fig. 3a). The pressure then decreased as a second cyclone system approached from its southwest. Easterly and northeasterly winds brought large amounts of water vapor to the Gaspé region. The 96-h accumulated precipitation from GEM-LAM is just under 60 mm (Fig. 3b), split about equally between liquid and solid precipitation. The simulated temperature and relative humidity are interpolated horizontally to the meteorological tower at the power plant site, and the time series of modeled fields are plotted in Fig. 4 for comparison. The 3-m-height temperature ranged from −5° to 5°C, and near-surface air was almost saturated during the entire precipitation process (Fig. 4). The model captures the temporal evolution trend of temperature and humidity well, with an RMSE of 0.98°C for temperature and a value of 2.2% for relative humidity.
Comparison of (a) near-surface pressure at a meteorological tower site from observations (3-m height) and GEM-LAM (surface) and (b) modeled accumulated (solid, liquid, and total) precipitation for case 1 (4–9 Apr 2009).
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0125.1
Comparison of near-surface (a) relative humidity and (b) temperature at a meteorological tower site from observations (3-m height) and GEM-LAM (7-m height) for case 1 (4–9 Apr 2009).
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0125.1
Figure 5 compares the modeled wind speed and observations from wind turbine sites and a meteorological tower. The observed wind speed in Fig. 5a is the average of all the wind turbines in the wind plant. For comparison, we interpolated the wind speed at a height of 87 m from GEM-LAM grid points to all of the turbine sites, and took the average of the simulated values from all the wind turbines. The standard deviation of wind speed from these turbine sites is shown by the shaded area in Fig. 5. The measured wind speed has a large standard deviation, reflecting high temporal and spatially variability. Unlike the observations, the dispersion of the modeled value at these turbine sites is small. Several factors may explain this low spatial variability in the model simulation: 1) the 2.5-km resolution of the model is not fine enough to resolve the topography of each turbine site, as well as some subgrid-scale processes; 2) the horizontal interpolation from 2.5-km grid points to the turbine sites partly smooths the differences between the turbine sites; 3) the NWP-model-simulated winds do not include the turbulence components (nacelle-observed wind speeds are influenced by turbulence from the turbine’s rotating blades and similarly by the wake from upstream turbines, while the current NWP model does not have the capability of simulating these interactions). The simulated wind speed at 87 m from GEM-LAM is horizontally interpolated to all wind turbine sites and, then, the average wind speed of the wind plant is calculated. The average wind speed from GEM-LAM follows the observations from the heated anemometer well. This can be considered to be a good forecast since the average wind speed of all the wind turbines is used to simulate wind power production successfully.
(a) Comparison of average wind speed at all turbine sites from observations (80-m height; nacelle-mounted anemometers) and GEM-LAM (87-m height) for case 1. Shaded area indicates the standard deviation of wind speed from all turbine sites. (b) Comparison of wind speed at a meteorological tower site from 10-min-average observations (heated anemometer at 70-m height, and nonheated anemometer at 80-m height) and GEM-LAM (87-m height) for case 1 (4–9 Apr 2009). Shaded area of observation indicates the standard deviation of wind speed over 10 min.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0125.1
Figure 5b offers a comparison of the wind speed at a meteorological tower between modeled and observed values. The observed wind speed reports 10-min average values at 70 and 80 m and the model wind speed is at 87 m. Shading in Fig. 5b represents the standard deviation of the observations over 10 min, and the observations at 70- and 80-m height are taken from heated and nonheated anemometers, respectively. Although it is a challenge to compare the instantaneous wind speed at a point observation site, the modeled wind speed at 87 m follows the temporal evolution of the measured value from the heated anemometer well except for the peak values at the start of each 1-day simulation, which are related to downscaling from coarse resolution to high resolution. There is no measurement from the 80-m-height anemometer between 5 and 8 April, indicating that this nonheated anemometer failed because of large amounts of accreted ice.
Figure 6a displays the real power production and theoretically generated power, the latter being calculated from an empirical turbine power curve with average nacelle wind speeds (Yu et al. 2014). The difference between theoretical and real power production is defined as power loss, and the ratio of power loss to the theoretical production (percentage of power reduction) is plotted in Fig. 6b. As seen from the figure, significant power loss is recorded and the maximum power loss is around 80 MW. The power reduction percentage is 100% for 5–6 April, which is consistent with the stoppage of the nonheated anemometer (Fig. 5).
Real and theoretically generated (a) power and (b) power loss (%) at the wind plant for Case 1 (4–9 Apr 2009).
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0125.1
2) Ice accretion and melting
The liquid precipitation is considered to be freezing rain when the modeled temperature at 87-m height is below 2°C. Therefore, the hourly liquid precipitation rate, as well as the temperature and wind speed at 87 m from GEM-LAM, are used as input into the freezing-rain accretion model described in section 3b(1). For wet snow accretion, the simulated solid precipitation, wind speed, and liquid water fraction (the ratio of rainwater mixing ratio to the sum of rainwater, ice crystal, snow, and graupel mixing ratios) in the snow are used to drive the wet snow model. Time series of the accreted ice rate and accumulated ice thickness are presented in Fig. 7. The ice started to accrete at 1000 UTC on 4 April. For the first 24 h, accretion was the result of freezing rain, with wet-snow processes starting to result in ice accretion on the following days. The large precipitation rate resulted in a large ice accretion rate, accounting for most of the 30-mm thickness of the accreted ice. Power reductions occurred sporadically on 4 April, and significant power reduction began on 5 April and lasted for 3 days (Fig. 6).
(a) Simulated icing rate from sources (freezing rain and wet snow) and sinks (melting from radiation) and (b) accumulated ice amount at the wind plant for case 1 (4–9 Apr 2009).
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0125.1
The freezing rain and wet snow determine how much ice will be accreted; however, the heat balance from physical processes such as radiation effect and warm advection results in melting of the ice. Radiation fluxes such as those from longwave and solar radiation are taken into account in the heat balance, to calculate the time when the ice starts to melt and the resultant melting rate due to the net heat balance. Figure 8 depicts the temporal evolution of the downward shortwave flux, the longwave (infrared radiation) energy flux toward ground, the cold surface-upward longwave radiation, and the balance of these three heat fluxes. If the total radiance flux toward the ground is positive, the net flux will melt some of the ice. The melting rate and the resultant thickness of ice are plotted in Fig. 7. Melting due to radiation has nonnegligible impacts for this icing case, with half of the ice mass having been removed for the entire 4 days. The ice accretion amount and power loss percentages are standardized by subtracting their means (17.2 mm and 85%, respectively), and dividing by the standard deviations in Fig. 9. This calculated Z score is a unit-free measure, which is normally used for the comparison of data observations with different units. The start of ice accretion coincides with a significant increase in power loss, indicating positive correlation, and the correlation coefficient is around 0.78. In our simulation, ice accreted on the cylinder remains until the end of the simulation, but the power loss gets small at the end of the simulation. This may be due to the overpredicted ice mass since warm advection is not considered in our current icing model. This warm advection, which is indicated as a significant local temperature increase in Fig. 3, would likely result in more ice melting and less power loss. In our study, heat budget consideration is the simplification of ice thawing, and the sensible heat flux exchange between the warm air and the accreted surface is not considered during the warm advection process. Actually, the situation is in fact far more complex when considering the melting process in rotating turbine blades. For example, turbine blades start generating torque when the accumulated ice is still about 10 mm thick. This could be from ice shedding, which is not considered in our study; for example, small mass melting in the inner part of the accreted ice will indeed generally lead to ice shedding from rotating blades. Davis et al. (2014) used an icing model (iceBlade, which accounts for accretion and ablation mechanisms) to simulate icing on a wind turbine blade, and the results are much improved relative to those for a standard cylinder model. There is no doubt that sophisticated ice load models for rotating blades must be developed to more accurately evaluate the power loss. However, this is beyond the scope of the current study.
(a) Simulated shortwave (left y axis; red) and longwave (right y axis; downward is blue and upward is black) radiation fluxes. (b) Simulated net radiation flux at the wind plant for case 1 (4–9 Apr 2009).
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0125.1
Simulated ice amount and power loss (standardized) at the wind plant for Case 1 (4–9 Apr 2009).
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0125.1
b. Case 2: In-cloud icing and freezing rain
The weather report from the Gaspé weather station, about 30 km away from the wind plant, shows that freezing drizzle, fog, cloudy, and snow occurred intermittently during the period 14–18 February 2010, and the temperature ranged between −5° and 0°C. Observed icing during this time period is considered to be a mixed case of in-cloud icing and freezing rain. Therefore, icing from both processes is taken into account to explore its impact on wind plant power generation.
Near-surface air temperature and relative humidity from GEM-LAM and from observations at a meteorological tower site located inside the wind power plant are compared in Fig. 10. The air is almost saturated, and the temperature ranges between −4° and 3°C, with subfreezing conditions most of the time. This condition is favorable for freezing rain and in-cloud icing. Generally, the simulated temperature underestimates the observations, but it nonetheless follows the temporal evolution of the observed values, and the correlation coefficient between the two is reasonably high (0.78). The simulated and observed relative humidity also agree well with each other, with a correlation coefficient around 0.67. Similar to Fig. 5, Fig. 11 compares the wind speed from GEM-LAM and observations. The model wind speed follows the temporal evolution trend of the observed case at the meteorological tower well. Because of icing impacts, the nonheated anemometer (Fig. 11b) recorded zero wind speed for most of the time. The wind speed for this icing case is medium, ranging mostly from 0 to 6 m s−1, and the theoretically generated power is thus relatively small (Fig. 12a). Accreted ice on wind turbine blades affects the proper operation of turbines, and a large power reduction (as a percentage) is seen for most of the time of this mixed in-cloud-icing and freezing-rain case (Fig. 12b).
Comparison of (a) near-surface relative humidity and (b) temperature at a meteorological tower site from the observations (3-m height) and GEM-LAM (7-m height) for case 2 (14–19 Feb 2010).
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0125.1
(a) Comparison of average wind speed at all turbine sites from observations (80-m height, nacelle-mounted anemometers) and GEM-LAM (87-m height) for case 1. Shaded area indicates the standard deviation of wind speed from all turbine sites. (b) Comparison of wind speed at a meteorological tower site from 10-min-average observations (heated anemometer at 70-m height, and nonheated anemometer at 80-m height) and GEM-LAM (87-m height) for case 2 (14–19 Feb 2010). Shaded area of observation indicates the standard deviation of wind speed over 10 min.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0125.1
Real and theoretically generated (a) power and (b) power loss (%) for case 2 (14–19 Feb 2010).
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0125.1
For in-cloud icing, the accretion rate α1 is determined by the temperature, wind speed, cloud liquid water content, and MVD of cloud droplets. In our simulation, Milbrandt–Yau’s two-moment microphysics scheme was used to simulate the cloud properties, which predicts the time evolution of the cloud liquid water content and droplet size distribution. Details on the calculation of MVD from the cloud droplet spectrum are given in Yang et al. (2012). The simulated cloud liquid water content and MVD at the first three vertical levels (7, 46, and 72 m) in GEM-LAM are given in Fig. 13. There is almost no cloud liquid water content at the first model level of 7 m, and the value of the liquid water content is smaller than 0.3 g kg−1 at heights of 46 and 72 m. The median volume diameter of the cloud droplets ranges between 0 and 40 μm at these two levels.
(a) Simulated precipitation, (b) liquid water content, and (c) MVD of cloud droplets for case 2 (14–19 Feb 2010).
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0125.1
The hourly output of the liquid water content, the MVD of the cloud droplets, and the wind speed and temperature at a height of 72 m from GEM-LAM are used to drive the ice accretion models described in section 3b. For this icing event, the meteorological conditions were favorable for the freezing-rain and in-cloud-icing mechanisms, and the corresponding ice accretion models were thus activated. The calculated ice rate and amount from in-cloud icing are presented in Fig. 14, as well as those from freezing rain. Because of the relatively small liquid water content, MVD of the droplets, and weak wind speed, the in-cloud icing rate is not large, with the value being up to 1 mm h−1. The accumulated ice due to the riming process is around 1 cm thick, close to the value from the freezing-rain process. Therefore, in the absence of melting, the total accreted ice from both mechanisms over the simulated period is around 2 cm in thickness for a theoretical freely rotating cylindrical power line.
(a) Simulated icing rate from sources (freezing rain and in-cloud icing) and sinks (melting from radiation) and (b) accumulated ice amount at the wind plant for case 2 (14–19 Feb 2010).
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0125.1
The accreted ice increased for the first 12 h as a result of freezing-rain and riming processes and decreased for the next 12 h when the melting process was dominated by a positive downward net radiation flux. The growth rate of accreted ice varied with time, depending on the dominance of the source and sink terms. When ice started to accrete, the power loss increased to 100% during the first 12 h and remained there for the next 2 days while the accreted ice was still on the cold surface. The accumulated ice went to zero as a result of melting between midday 16 February and 17 February; however, there was no power generation because of low wind speeds. Following this, ice amounts increased until the end of the simulation, and further power loss was observed. The power loss then started to decrease despite there being additional accreted ice from riming and freezing. This inconsistency could be related to our model overestimating the ice at the end of the simulation. As can be seen in Fig. 10b, the air temperature rises above 0°C, and the sensible heat from the warm environmental air could possibly melt some of the ice. This process is not considered in our icing model, which accounts for only radiation flux impacts in the heat balance. Nonetheless, Figs. 12 and 14 demonstrate that the icing time corresponds well with the power loss time. For this case, although the net theoretical accreted ice is only around 1 cm, it has direct impacts on power reduction.
c. Results from other cases
Another 6 cases among these 27 observed icing events during 2008–10 are selected for simulations because they have longer duration and/or higher power reduction percentage relative to other cases. Table 2 shows the verification results of modeled surface meteorological fields for these simulated cases. Three model verification scores (bias, root-mean-square error, and correlation coefficient) are calculated to assess the performance of GEM-LAM in simulating the meteorological fields. Note that there is no verification for wind speed in event 6 because of the missing wind speed measurements. For all of these simulated events, the biases for temperature, wind speed, and relative humidity are small, except for events 8–10 in which GEM-LAM overestimated relative humidity, with a bias of 6.44%. This is because the saturated water vapor pressure decreases exponentially with the underestimated temperature. Defined as the ratio between the water vapor pressure and saturated water vapor pressure, the relative humidity of cold air is therefore higher than that of warm air with the same quantity of water vapor. Note that for the cases based on events 20 and 8–10, both of which had long durations of power loss, the temperature is underestimated (with bias close to −1°C). Our modeling system uses one-way coupling, and the latent heat of the freezing process is not accounted for currently. The atmospheric temperature could probably be improved if the freezing impact is fed back to the environmental atmosphere in a two-way coupling to GEM-LAM.
Verification scores of bias, RMSE, and correlation coefficient (r) for simulated temperature (T), relative humidity (RH), and wind speed (V) for all simulated cases. The first column is the event number, and the second column is for icing types, which include freezing rain (FZ), in-cloud icing (RI), and wet snow (WS).
The temporal evolution of temperature, relative humidity, and wind speed is well captured by GEM-LAM for all of these cases (figures not shown), with an overall RMSE average of 1.3°C for temperature, 6.0% for relative humidity, and 2.27 m s−1 for wind speed at the meteorological tower. Also, the correlation coefficients for these variables are high, with an average of 0.84 for temperature, 0.75 for relative humidity, and 0.76 for wind speed. Overall, the performance of GEM-LAM in the simulation of the surface meteorological fields, which are key factors in determining the onset and duration of the meteorological icing, is satisfying and lends support to the study of icing impacts on power generation in wind plants.
Precipitation and cloud properties as well as near-surface meteorological fields at a height of 87 m from GEM-LAM are used to calculate the accretion rates and ice amounts in icing events. By examining all the cases, the icing duration is found to correspond well with that of power loss time for most of the cases (Table 3). The correlation coefficients between ice accretion and power loss values were calculated. Although the values were not large for some events, the positive value for all the cases demonstrates direct icing’s impact on power loss. It should be borne in mind that the calculated icing rate in this study is a meteorological icing rate, which is determined by the current meteorological conditions such as temperature, cloud properties, and precipitation rate. A short time delay (incubation time) exists between the start times of the instrumental and meteorological icing periods, and the instrumental icing is the direct factor impacting wind turbine operation and the resultant power generation (Heimo et al. 2009). The correlation coefficient will be different if the incubation time is considered in our calculation with a lag. Nonetheless, the model system shows good potential for forecasting the icing impacts on wind power. Systematic simulation over long periods is ongoing to assess the full skill of the proposed methodology.
Start-up times and duration of icing (simulation) and power loss (observation), and correlation coefficient (r) between icing amount and power loss for the simulated events.
5. Discussion and concluding remarks
This paper explored the possibility of using atmospheric and icing models to predict ice duration and amounts for various icing mechanisms, and to build the connection between simulated ice amounts and power loss in wind plants. Two long-duration ice events during the period 2008–10 at a wind plant in the Gaspé region of Canada were studied in detail. The synoptic background was first analyzed, and an atmospheric mesoscale model (GEM-LAM) was used to study two icing cases in which accreted ice is from various icing mechanisms. The simulated surface fields (temperature, relative humidity, and wind speed) follow the observations closely. These fields, together with cloud properties and precipitation from an advanced microphysics scheme, were used to drive icing models (respectively for freezing rain, wet snow, and in-cloud icing), leading to estimates for the total ice amount on a rotating cylinder. A sink term was added to the existing icing models to account for the ice-melting process due to radiation, and the developed icing models are then coupled to GEM-LAM. Our coupled model successfully captured the start-up of ice accretion, and the icing duration is consistent with the power loss period, with positive correlation between ice accretion and real power loss.
Simulation results from 6 other of the total 27 icing events were summarized, and GEM-LAM performs well overall in simulating the evolution of the near-surface fields for all of the events. On average, the RMSEs for temperature, relative humidity, and wind speed are 1.3°C, 6.0%, and 2.27 m s−1, respectively, for all of these cases. Positive correlations were found between icing episodes and the reduction of wind power production in the wind plant, although the correlation coefficients are not high for some cases. Note that the simulated icing rate is the meteorological icing rate, where the power loss is directly related to the instrumental icing and the wind turbine control parameters. Instrumental icing is determined not only by atmospheric conditions, but also the surface shape, roughness, and characteristics of the accreting objects. With more sophisticated ice accretion models, the ice amount on rotating blades will be much better evaluated than that on the cylinder, and the correlation coefficient between the ice amount and power loss is anticipated to be higher.
It was noticed that significant power reduction occurs even when there are small amounts of accreted ice in most of the icing cases, as it is known that very small amounts of accreted ice load can result in large power loss percentages, and even the slightest amount of surface roughness has the potential of reducing energy output from wind turbines by 20% (Rong and Bose 1990). Turkia et al. (2013) used different meteorological conditions to simulate the ice accretion and determined that the average ratio of ice mass on the standard wind turbine blade (5 MW) could be 20 times that on the static cylinder in the same weather conditions. The ice accretion calculation in this study is based on a rotating static cylinder, and the ice amount on the turbine blades could be much larger, or at least certainly different, than that on the cylinder. In this first attempt to combine atmospheric and ice accretion models to simulate power loss, a cylindrical geometry for the ice accretion calculations was assumed. It seems plausible that the relatively low values obtained for ice accretion may be related to this simplification. Consideration of rotating blades with more realistic shapes, as well as more sophisticated ice accretion models, may yield higher accretion amounts.
Note that in our study, the supercooled precipitation freezes onto the cold surface of the wind turbine blades at a height of 80 m, resulting in icing and power loss after. As there is no direct observation for 80-m height air temperature nor of the turbine blade temperature, we used a threshold of 2°C air temperature for freezing-rain activation. This works for this freezing case, but it could potentially lead to false freezing precipitation in periods without icing. Accurate freezing-rain forecasts are not only related to near-surface air temperature but are also dependent on the vertical profile of temperature in the boundary layer, the thickness of the temperature inversion, and the surface temperature of the object upon which freezing occurs. As pointed out by Cortinas (2000), if the warming is fast and the ground and surface object are still below freezing, freezing rain can occur even if the air temperature is above 0°C. Therefore, taking into account vertical temperature profiles and object surface temperature could lead to improvements in forecasts of freezing rain.
Our icing models consider radiant heat flux, and the net radiant flux is used to melt the accreted ice. When there are various icing mechanisms, this radiation effect is accounted for only one time to avoid replication. But the situation is far more complex when considering the melting process in rotating turbine blades. For example, radiative melting of a small amount of ice in the inner part of the accretion will probably lead to large ice throw from rotating blades. Additionally, mechanical processes such as vibration of the wind turbine, erosion by wind, and centrifugal forces when the turbine is turning can lead to the removal of ice mass. These processes play an important role in ice thawing and throwing, the study of which is beyond the scope of this study.
It is our belief that clearly quantifying icing impacts on power reduction will involve accounting for all the processes discussed above. Despite the limitations and uncertainty in our simulation results, this study is an initial step to exploring icing impacts on power generation over a range of icing conditions, and provides an exemplary framework for the quantitative evaluation of icing’s impact on wind turbine operations and a possible improvement in wind power forecasting under icing conditions.
Acknowledgments
The authors are grateful to the two anonymous reviewers for their constructive and valuable suggestions. This research is supported by the Canadian Federal Government ecoEnergy Innovation Inititative (ecoEll).
REFERENCES
Bélair, S., J. Mailhot, C. Girard, and P. Vaillancourt, 2005: Boundary layer and shallow cumulus clouds in a medium-range forecast of a large-scale weather system. Mon. Wea. Rev., 133, 1938–1960, doi:10.1175/MWR2958.1.
Benoit, R., J. Côté, and J. Mailhot, 1989: Inclusion of a TKE boundary-layer parameterization in the Canadian regional finite-element model. Mon. Wea. Rev., 117, 1726–1750, doi:10.1175/1520-0493(1989)117<1726:IOATBL>2.0.CO;2.
Bernstein, B. C., J. Hirvonen, E. Gregow, and I. Wittmeyer, 2012: Experiences from real-time LAPS-LOWICE runs over Sweden: 2011–2012 icing season. Winterwind Int. Wind Energy Conf., Skellefteå, Sweden, Swedish Wind Power Association, 21 pp. [Available online at http://www.slideshare.net/WinterwindConference/3a-bernstein-lapslowice.]
Byrkjedal, O., 2012a: Mapping of icing in Sweden—On the influence from icing on wind energy production. Winterwind Int. Wind Energy Conf., Skellefteå, Sweden, Swedish Wind Power Association, 20 pp. [Available online at http://www.slideshare.net/WinterwindConference/3a-byrkjedal-icingkvt.]
Byrkjedal, O., 2012b: The benefits of forecasting icing on wind energy production. Winterwind Int. Wind Energy Conf., Skellefteå, Sweden, Swedish Wind Power Association, 24 pp. [Available online at http://www.slideshare.net/WinterwindConference/3-forecast-kvt.]
Cattin, R., A. Heimo, S. Kunz, G. Russi, M. Russi, M. Tiefgraber, S. Schaffner, and B. E. Nygaard, 2007: Alpine Test Site Guetsch: Meteorological measurements and wind turbine performance analysis. Proc. 12th Int. Workshop on Atmospheric Icing of Structures (IWAIS), Yokohama, Japan, IWAIS, 6 pp. [Available online at http://www.meteotest.ch/cost727/media/IWAIS2007_paper.pdf.]
Chaine, P. M., and G. Castonguay, 1974: New Approach to Radial Ice Thickness Concept Applied to Bundle Like Conductors. Vol. 4, Industrial Meteorology, Environment Canada, 22 pp.
Choisnard, J., S. Antic, J. Bourrent, and C. Brabant, 2012: Centralized wind power forecasting system in Québec: Operational products in cold climate environment. Proc. EWEA Annual Event, Copenhagen, Denmark, European Wind Energy Association. [Available online at http://proceedings.ewea.org/annual2012/proceedings/ewec.php?id=571.]
Cortinas, J., Jr., 2000: A climatology of freezing rain in the Great Lakes region of North America. Mon. Wea. Rev., 128, 3574–3588, doi:10.1175/1520-0493(2001)129<3574:ACOFRI>2.0.CO;2.
Côté, J., S. Gravel, A. Méthot, A. Patoine, M. Roch, and A. Staniforth, 1998a: The operational CMC–MRB Global Environmental Multiscale (GEM) model. Part I: Design considerations and formulation. Mon. Wea. Rev., 126, 1373–1395, doi:10.1175/1520-0493(1998)126<1373:TOCMGE>2.0.CO;2.
Côté, J., J.-G. Desmarais, S. Gravel, A. Méthot, A. Patoine, M. Roch, and A. Staniforth, 1998b: The operational CMC–MRB Global Environmental Multiscale (GEM) model. Part II: Results. Mon. Wea. Rev., 126, 1397–1418, doi:10.1175/1520-0493(1998)126<1397:TOCMGE>2.0.CO;2.
Dalle, B., and P. Admirat, 2011: Wet snow accretion on overhead lines with French report of experience. Cold Reg. Sci. Technol., 65, 43–51, doi:10.1016/j.coldregions.2010.04.015.
Davis, N., A. N. Hahmann, N.-E. Clausen, and M. Žagar, 2014: Forecast of icing events at a wind farm in Sweden. J. Appl. Meteor. Climatol., 53, 262–281, doi:10.1175/JAMC-D-13-09.1.
Dierer, S., R. Cattin, S. C. Müller, B. E. K. Nygaard, P. Steiner, and B. Calpini, 2009: Modeling the risk of icing in Switzerland. Proc. 13th Int. Workshop on Atmospheric Icing of Structures (IWAIS), Andermatt, Switzerland, IWAIS, 7 pp. [Available online at http://iwais.compusult.net/html/IWAIS_Proceedings/IWAIS_2009/Session_3_cost_727_wg1/session_3_dierer.pdf.]
Dierer, S., R. Oechslin, and R. Cattin, 2011: Wind turbines in icing conditions: Performance and prediction. Adv. Sci. Res., 6, 245–250, doi:10.5194/asr-6-245-2011.
Drage, M. A., and G. Hauge, 2008: Atmospheric icing in a coastal mountainous terrain. Measurements and numerical simulations, a case study. Cold Reg. Sci. Technol., 53, 150–161, doi:10.1016/j.coldregions.2007.12.003.
Elíasson, A. J., E. Thorsteins, H. Ágústsson, and Ó. Rögnvaldsson, 2011: Comparison between simulations and measurements of in-cloud icing in test spans. Proc. 14th Int. Workshop on Atmospheric Icing of Structures (IWAIS), Chongqing, China, IWAIS, 7 pp. [Available online at http://iwais.compusult.net/html/IWAIS_Proceedings/IWAIS_2011/Papers/A4_5_231.pdf.]
Elíasson, A. J., H. Ágústsson, G. M. Hannesson, and E. Thorsteins, 2013: Modeling wet-snow accretion: Comparison of cylindrical model to field measurements. Proc. 15th Int. Workshop on Atmospheric Icing of Structures (IWAIS), St. John’s, NL, Canada, IWAIS, 9 pp. [Available online at http://www.landsnet.is/library/Skrar/Landsnet/Upplysingatorg/Frettir/2013/16.09.2013/IWAIS%20-%20Modeling%20wet-snow%20accretion_2013-09-02.pdf.]
Fikke, S., and Coauthors, 2007: COST 727: Atmospheric icing on structures, measurements and data collection on icing: State of the art. MeteoSwiss, 110 pp.
Finstad, K. J., E. P. Lozowski, and E. M. Gates, 1988: A computational investigation of water droplet trajectories. J. Atmos. Oceanic Technol., 5, 160–170, doi:10.1175/1520-0426(1988)005<0160:ACIOWD>2.0.CO;2.
Fouquart, Y., and B. Bonnel, 1980: Computation of solar heating of the Earth’s atmosphere: A new parameterization. Beitr. Phys. Atmos., 53, 35–62.
Garand, L., 1983: Some improvements and complements to the infrared emissivity algorithm including a parameterization of the absorption in the continuum region. J. Atmos. Sci., 40, 230–244, doi:10.1175/1520-0469(1983)040<0230:SIACTT>2.0.CO;2.
Heimo, A., R. Cattin, and B. Calpini, 2009: Recommendations for meteorological measurements under icing conditions. Proc. 13th Int. Workshop on Atmospheric Icing of Structures (IWAIS), Andermatt, Switzerland, IWAIS, 5 pp. [Available online at http://iwais.compusult.net/html/IWAIS_Proceedings/IWAIS_2009/Session_4_cost_727_wg2/session_4_heimo.pdf.]
Hošek, J., 2007: Icing measurements at Milešovka and their comparison with reanalysis and mesoscale model outputs. Proc. 12th Int. Workshop on Atmospheric Icing of Structures (IWAIS), Yokohama, Japan, IWAIS, 5 pp. [Available online at http://iwais.compusult.net/html/IWAIS_Proceedings/IWAIS_2007/Category3%20Field%20Observations,%20Data%20Gathering%20and%20Instrumentation/Lecture/Hosek.pdf.]
Houston, T. G., and S. A. Changnon, 2007: Freezing rain events: A major weather hazard in the conterminous U.S. Nat. Hazards, 40, 485–494, doi:10.1007/s11069-006-9006-0.
International Organization for Standardization, 2001: Atmospheric icing of structures. ISO 12494, Geneva, Switzerland, 56 pp.
Jones, K. F., 1996: Ice accretion in freezing rain. CRREL Rep. 96-2, 23 pp. [Available from Cold Regions Research and Engineering Laboratory, 72 Lyme Rd., Hanover, NH 03755-1290.]
Jones, K. F., 1998: A simple model for freezing rain ice loads. Atmos. Res., 46, 87–97, doi:10.1016/S0169-8095(97)00053-7.
Karlsson, T., V. Turkia, and T. Wallenius, 2013: Icing production loss module for wind power forecasting system. VTT, 20 pp. [Available online at http://www.vtt.fi/inf/pdf/technology/2013/T139.pdf.]
Lacroix, A., and M. Tan, 2012: Assessment of wind energy production penalties due to cold climate in Canada. Natural Resources Canada Tech. Rep., 35 pp. [Catalog No. M154-68/2012E-PDF; ISBN 978-1-100-21299-9.]
Mailhot, J., and R. Benoit, 1982: A finite-element model of the atmospheric boundary-layer suitable for use with numerical weather prediction models. J. Atmos. Sci., 39, 2249–2266, doi:10.1175/1520-0469(1982)039<2249:AFEMOT>2.0.CO;2.
Mailhot, J., and C. Chouinard, 1989: Numerical forecasts of explosive winter storms: Sensitivity experiments with a meso-α scale model. Mon. Wea. Rev., 117, 1311–1343, doi:10.1175/1520-0493(1989)117<1311:NFOEWS>2.0.CO;2.
Makkonen, L., 1984: Modeling of ice accretion on wires. J. Climate Appl. Meteor., 23, 929–939, doi:10.1175/1520-0450(1984)023<0929:MOIAOW>2.0.CO;2.
Makkonen, L., 1996: Modeling power line icing in freezing precipitation. Proc. Seventh Int. Workshop on Atmospheric Icing of Structures (IWAIS), Chicoutimi, QC, Canada, IWAIS, 195–201.
Makkonen, L., 1998: Modeling power line icing in freezing precipitation. Atmos. Res., 46, 131–142, doi:10.1016/S0169-8095(97)00056-2.
Makkonen, L., 2000: Models for the growth of rime, glaze, icicles and wet snow on structures. Philos. Trans. Roy. Soc. London, 358A, 2913–2939, doi:10.1098/rsta.2000.0690.
Makkonen, L., and B. Wichura, 2010: Simulating wet snow loads on power line cables by a simple model. Cold Reg. Sci. Technol., 61, 73–81, doi:10.1016/j.coldregions.2010.01.008.
Milbrandt, J., and M. K. Yau, 2005: A multimoment bulk microphysics parameterization. Part II: A proposed three-moment closure and scheme description. J. Atmos. Sci., 62, 3065–3081, doi:10.1175/JAS3535.1.
Milbrandt, J., M. K. Yau, J. Mailhot, S. Belair, and R. McTaggart-Cowan, 2010: Simulation of an orographic precipitation event during IMPROVE-2. Part II: Sensitivity to the number of moments in the bulk microphysics scheme. Mon. Wea. Rev., 138, 625–642, doi:10.1175/2009MWR3121.1.
Morgan, C., and E. Bossanyi, 1996: Wind turbine icing and public safety—A quantifiable risk? Proc. BOREAS III: Wind Energy Production in Cold Climates, Saariselkä, Finland, Finnish Meteorological Institute.
Morgan, C., E. Bossanyi, and H. Seifert, 1998: Assessment of safety risks arising from wind turbine icing. Proc. BOREAS IV: Wind Energy Production in Cold Climates, Hetta, Finland, Finnish Meteorological Institute, 113–121. [Available online at http://virtual.vtt.fi/virtual/arcticwind/boreasiv/assessment_of_safety.pdf.]
Nygaard, B. E. K., 2009: Evaluation of icing simulations for the “COST 727 icing test sites” in Europe. Proc. 13th Int. Workshop on Atmospheric Icing of Structures (IWAIS), Andermatt, Switzerland, IWAIS, 5 pp. [Available online at http://iwais.compusult.net/html/IWAIS_Proceedings/IWAIS_2009/Session_3_cost_727_wg1/session_3_nygaard.pdf.]
Nygaard, B. E. K., S. M. Fikke, L. Elvertrø, and K. Harstveit, 2007a: Modeling icing in exposed mountain terrain. Proc. 12th Int. Workshop on Atmospheric Icing of Structures (IWAIS), Yokohama, Japan, IWAIS, 4 pp. [Available online at http://iwais.compusult.net/html/IWAIS_Proceedings/IWAIS_2007/Category2_Modeling_Simulation_of_Icing/Lecture/Nygaard_1.pdf.]
Nygaard, B. E. K., J. E. Kristjánsson, E. Berge, and L. Makkonen, 2007b: Using NWP models to simulate in-cloud atmospheric icing episodes. Proc. 12th Int. Workshop on Atmospheric Icing of Structures (IWAIS), Yokohama, Japan, IWAIS, 4 pp. [Available online at http://iwais.compusult.net/html/IWAIS_Proceedings/IWAIS_2007/Category2_Modeling_Simulation_of_Icing/Lecture/Nygaard_2.pdf.]
Nygaard, B. E. K., J. E. Kristjánsson, and L. Makkonen, 2011: Prediction of in-cloud icing conditions at ground level using the WRF Model. J. Appl. Meteor. Climatol., 50, 2445–2459, doi:10.1175/JAMC-D-11-054.1.
Nygaard, B. E. K., B. Egil, H. Ágústsson, and K. Somfalvi-Tóth, 2013: Modeling wet snow accretion on power lines: Improvements to previous methods using 50 years of observations. J. Appl. Meteor. Climatol., 52, 2189–2203, doi:10.1175/JAMC-D-12-0332.1.
Paquin-Ricard, D., C. Jones, and P. A. Vaillancourt, 2010: Using ARM observations to evaluate cloud and clear-sky radiation processes as simulated by the Canadian regional climate model GEM. Mon. Wea. Rev., 138, 818–838, doi:10.1175/2009MWR2745.1.
Podolskiy, E. A., B. E. K. Nygaard, K. Nishimura, L. Makkonen, and E. P. Lozowski, 2012: Study of unusual atmospheric icing at Mount Zao, Japan, using the Weather Research and Forecasting model. J. Geophys. Res., 117, D12106, doi:10.1029/2011JD017042.
Rong, J. Q., and N. Bose, 1990: Power reduction from ice accretion on a horizontal axis wind turbine. Proc. 12th British Wind Energy Association Conf., Norwich, UK, British Wind Energy Association, 77–86.
Seifert, H., A. Westerhellweg, and J. Kröning, 2003: Risk analysis of ice throw from wind turbines. Proc. BOREAS VI: Wind Energy Production in Cold Climates., Pyhä, Finland, Finnish Meteorological Institute, 18 pp. [Available online at http://www.windaction.org/posts/13298-risk-analysis-of-ice-throw-from-wind-turbines#.VV1FNyg_tnE.]
Soderberg, S., and M. Baltscheffsky, 2012: Long-term estimates and variability of production losses in icing climates. Winterwind Int. Wind Energy Conf., Skellefteå, Sweden, Swedish Wind Power Association, 17 pp. [Available online at http://www.slideshare.net/WinterwindConference/longterm-estimates-and-variability-of-production-losses-in-icing-climates-stefan-sderberg-magnus-baltscheffsky-weathertech-scandinavia.]
Sørensen, J. D., J. N. Sørensen, and J. Lemming, 2012: Risk assessment of wind turbines close to highways. Proc. EWEA Annual Event, Copenhagen, Denmark, European Wind Energy Association, 9 pp. [Available online at http://proceedings.ewea.org/annual2012/allfiles2/969_EWEA2012presentation.pdf.]
Turkia, V., S. Huttunen, and T. Wallenius, 2013: Method for estimating wind turbulence production losses due to icing. VTT Technology 114, 44 pp. [Available online at http://www.vtt.fi/inf/pdf/technology/2013/T114.pdf.]
Wakahama, G., 1979: Experimental studies of snow accretion on electric lines developed in a strong wind. J. Nat. Disaster Sci., 1, 21–33.
Wareing, J. B., and B. E. K. Nygaard, 2009: WRF simulation of wet snow and rime icing incidents in the UK. Proc. 13th Int. Workshop on Atmospheric Icing of Structures (IWAIS), Andermatt, Switzerland, IWAIS, 3 pp. [Available online at http://iwais.compusult.net/html/IWAIS_Proceedings/IWAIS_2009/Session_3_cost_727_wg1/session_3_wareing.pdf.]
Yang, D., H. Ritchie, S. Desjardins, G. Pearson, A. MacAfee, and I. Gultepe, 2010: High-resolution GEM-LAM application in marine fog prediction: Evaluation and diagnosis. Wea. Forecasting, 25, 727–748, doi:10.1175/2009WAF2222337.1.
Yang, J., K. F. Jones, W. Yu, and R. Morris, 2012: Simulation of in-cloud icing events on Mount Washington with the GEM-LAM. J. Geophys. Res., 117, D17204, doi:10.1029/2012JD017520.
Yu, W., L. Garand, and A. P. Dastoor, 1997: Evaluation of model clouds and radiation at 100 km scale using GOES data. Tellus, 49A, 246–262, doi:10.1034/j.1600-0870.1997.t01-1-00006.x.
Yu, W., and Coauthors, 2014: An operational application of NWP models in a wind power forecasting demonstration experiment. Wind Eng., 38, 1–21, doi:10.1260/0309-524X.38.1.1.
