• Brun, R., W. Lewis, P. Perkins, and J. Serafini, 1955: Impingement of cloud droplets on a cylinder and procedure for measuring liquid-water content and droplet sizes in supercooled clouds by rotating multicylinder method. National Advisory Committee for Aeronautics (NACA) Rep. 1215, 43 pp.

    • Search Google Scholar
    • Export Citation
  • Drage, M., and G. Hauge, 2007: Atmospheric icing in a coastal mountainous terrain. Measurements and numerical simulations, a case study. Cold Reg. Sci. Technol., 53 (2), 150161.

    • Search Google Scholar
    • Export Citation
  • Fikke, S. M., J. E. Kristjánsson, and B. E. K. Nygaard, 2008: Modern meteorology and atmospheric icing. Atmospheric Icing of Power Networks, M. Farzaneh, Ed., Springer, 1–30.

    • Search Google Scholar
    • Export Citation
  • Finstad, K., E. Lozowski, and L. Makkonen, 1988: On the median volume diameter approximation for droplet collision efficiency. J. Atmos. Sci., 45, 40084012.

    • Search Google Scholar
    • Export Citation
  • Frohboese, P., and A. Anders, 2007: Effects of icing on wind turbine fatigue loads. J. Phys.: Conf. Ser., 75, 012061, doi:10.1088/1742-6596/75/1/012061.

    • Search Google Scholar
    • Export Citation
  • Gent, R. W., N. P. Dart, and J. T. Cansdale, 2000: Aircraft icing. Philos. Trans. Math. Phys. Eng. Sci., 358, 28732911.

  • Harstveit, K., 2002: Using routine meteorological data from airfields to produce a map of ice risk zones in Norway. Proc. 10th Int. Workshop on Atmospheric Icing of Structures, Brno, Czech Republic, paper 8.1, 8 pp.

    • Search Google Scholar
    • Export Citation
  • Jasinski, W., S. Noe, M. Selig, and M. Bragg, 1998: Wind turbine performance under icing conditions. Trans. Amer. Soc. Mech. Eng. J. Sol. Energy Eng., 120, 6065.

    • Search Google Scholar
    • Export Citation
  • Makkonen, L., 1984: Modeling of ice accretion on wires. J. Climate Appl. Meteor., 23, 929939.

  • Makkonen, L., 1992: Analysis of rotating multicylinder data in measuring cloud-droplet size and liquid water content. J. Atmos. Oceanic Technol., 9, 258263.

    • Search Google Scholar
    • Export Citation
  • Makkonen, L., 2000: Models for the growth of rime, glaze, icicles and wet snow on structures. Philos. Trans. Math. Phys. Eng. Sci., 358, 29132939.

    • Search Google Scholar
    • Export Citation
  • Makkonen, L., and J. Stallabrass, 1987: Experiments on the cloud droplet collision efficiency of cylinders. J. Climate Appl. Meteor., 26, 14061411.

    • Search Google Scholar
    • Export Citation
  • Makkonen, L., T. Laakso, M. Marjaniemi, and K. Finstad, 2001a: Modelling and prevention of ice accretion on wind turbines. Wind Eng., 25, 321.

    • Search Google Scholar
    • Export Citation
  • Makkonen, L., P. Lehtonen, and L. Helle, 2001b: Anemometry in icing conditions. J. Atmos. Oceanic Technol., 18, 14571469.

  • Miles, N., J. Verlinde, and E. Clothiaux, 2000: Cloud droplet size distributions in low-level stratiform clouds. J. Atmos. Sci., 57, 295311.

    • Search Google Scholar
    • Export Citation
  • Morrison, H., and J. Pinto, 2005: Mesoscale modeling of springtime Arctic mixed-phase stratiform clouds using a new two-moment bulk microphysics scheme. J. Atmos. Sci., 62, 36833704.

    • Search Google Scholar
    • Export Citation
  • Morrison, H., G. Thompson, and V. Tatarskii, 2009: Impact of cloud microphysics on the development of trailing stratiform precipitation in a simulated squall line: Comparison of one- and two-moment schemes. Mon. Wea. Rev., 137, 9911007.

    • Search Google Scholar
    • Export Citation
  • Mulherin, N., 1998: Atmospheric icing and communication tower failure in the United States. Cold Reg. Sci. Technol., 27, 91104.

  • Qiang, H., W. Jiahong, and O. Mingyong, 2005: Analysis on accidents caused by icing damage in Hunan power grid in 2005 and its countermeasures. Power Syst. Technol. Beijing, 29, 16.

    • Search Google Scholar
    • Export Citation
  • Rogers, E., T. Black, B. Ferrier, Y. Lin, D. Parrish, and G. DiMego, 2001: Changes to the NCEP Meso Eta Analysis and Forecast System: Increase in resolution, new cloud microphysics, modified precipitation assimilation, modified 3DVAR analysis. NWS Tech. Procedures Bull. 488, 15 pp.

    • Search Google Scholar
    • Export Citation
  • Rogers, E., and Coauthors, 2005: The NCEP North American mesoscale modeling system: Final Eta Model/analysis changes and preliminary experiments using the WRF-NMM. Preprints, 21st Conf. on Weather Analysis and Forecasting/17th Conf. on Numerical Weather Prediction, Washington, DC, Amer. Meteor. Soc., 4B.5. [Available online at http://ams.confex.com/ams/WAFNWP34BC/techprogram/paper_94707.htm.]

    • Search Google Scholar
    • Export Citation
  • Skamarock, W., J. Klemp, J. Dudhia, D. Gill, D. Barker, W. Wang, and J. G. Powers, 2005: A description of the Advanced Research WRF Version 2. NCAR Tech. Note 468, 88 pp.

    • Search Google Scholar
    • Export Citation
  • Sundin, E., and L. Makkonen, 1998: Ice loads on a lattice tower estimated by weather station data. J. Appl. Meteor., 37, 523529.

  • Thompson, G., R. Rasmussen, and K. Manning, 2004: Explicit forecasts of winter precipitation using an improved bulk microphysics scheme. Part I: Description and sensitivity analysis. Mon. Wea. Rev., 132, 519542.

    • Search Google Scholar
    • Export Citation
  • Thompson, G., P. Field, R. Rasmussen, and W. Hall, 2008: Explicit forecasts of winter precipitation using an improved bulk microphysics scheme. Part II: Implementation of a new snow parameterization. Mon. Wea. Rev., 136, 50955115.

    • Search Google Scholar
    • Export Citation
  • Thompson, G., B. Nygaard, L. Makonnen, and S. Dierer, 2009: Using the Weather Research and Forecasting (WRF) Model to predict ground/structural icing. Proc. 13th Int. Workshop on Atmospheric Icing of Structures, Andermatt, Switzerland, Meteotest, 8 pp. [Available online at http://www.iwais2009.ch/index.php?id=44.]

    • Search Google Scholar
    • Export Citation
  • Thorkildson, R., K. Jones, and M. Emery, 2009: In-cloud icing in the Columbia basin. Mon. Wea. Rev., 137, 43694381.

  • Wagner, T., U. Peil, and C. Borri, 2009: Numerical investigation of conductor bundle icing. Proc. Fifth African and European Conference on Wind Engineering, Florence, Italy, International Association for Wind Engineering, 9 pp. [Available online at http://www.iawe.org/Proceedings/5EACWE/058.pdf.]

    • Search Google Scholar
    • Export Citation
  • View in gallery

    Configuration of nested domains used in the WRF simulations. The four nested domains are indicated with white squares, with grid spacing decreasing from 9 km in the outermost domain to 0.333 km in the innermost domain.

  • View in gallery

    Terrain elevation in the innermost domain. The black mesh illustrates the terrain after the process of scaling the topography to match the real mountain height.

  • View in gallery

    Examples of temperature profiles. (a) Measured profile at Sodankylä, 1200 UTC 12 Dec 1994. (b) Corresponding predicted profile at Sodankylä. The lines show the temperature in degrees Celsius (right curve) and the corresponding dewpoint temperature (left curve).

  • View in gallery

    Measured vs predicted cloud SLWC at three different horizontal resolutions for WRF simulations with the EGCP01 cloud microphysics scheme. Case number corresponds to the eight cases listed in Table 1. The markers indicate the point values from the grid box closest to the actual site, extracted in the middle of the RMC exposure time. Thin lines show the range of SLWC from neighboring grid points in time and space, while the thicker parts of the line is drawn between the 25th and the 75th percentile of the neighboring values.

  • View in gallery

    As in Fig. 4, but with the Thompson cloud microphysics scheme.

  • View in gallery

    As in Fig. 4, but with the Morrison cloud microphysics scheme.

  • View in gallery

    MAE of predicted SLWC for simulations with various combinations of cloud microphysics schemes and horizontal resolutions. MAE is calculated on the basis of the deterministic SLWC values.

  • View in gallery

    Measured vs predicted cloud SLWC, all from simulations with grid spacing equal to 0.333 km but with different cloud microphysics schemes. Markers and lines are as explained in Fig. 4.

  • View in gallery

    (top) Time–height cross section from the top of Ylläs. Shading (white/black/red) = cloud LWC, blue contour lines = snow, green contour lines = cloud ice, cyan contour lines = graupel, red contour lines = rain, and gray contour lines = temperature (°C). All hydrometeor species are plotted as mass concentration (g m−3). (bottom) Time series of mass concentration in the lowest model layer. The dark green line with triangle markings indicates accumulated precipitation at the surface. The simulation is initiated at 0000 UTC 8 Feb 1990, with grid spacing = 0.333 km and the Thompson cloud microphysics scheme.

  • View in gallery

    As in Fig. 9, but with the Morrison cloud microphysics scheme.

  • View in gallery

    As in Fig. 9, but with the EGCP01 cloud microphysics scheme.

  • View in gallery

    As in Fig. 9, but for a model simulation initiated at 0000 UTC 12 Dec 1994 using the Thompson cloud microphysics scheme.

  • View in gallery

    As in Fig. 12, but with the Morrison cloud microphysics scheme.

  • View in gallery

    As in Fig. 12, but with the EGCP01 cloud microphysics scheme.

  • View in gallery

    Measured vs predicted values of MVD (μm) using the Thompson cloud microphysics scheme at grid spacing of 0.333 km. (a) MVD calculated with Nc = 100 cm−3 and (b) MVD calculated with Nc = 250 cm−3. The black dots are calculated using the deterministic SLWC values while the lines represent the range of variation from the neighboring grid points.

  • View in gallery

    Measured MVD plotted against measured SLWC. The dashed lines indicate theoretical curves for various droplet concentrations based on the assumed droplet size distribution (2).

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 442 372 46
PDF Downloads 409 367 56

Prediction of In-Cloud Icing Conditions at Ground Level Using the WRF Model

View More View Less
  • 1 The Norwegian Meteorological Institute, Oslo, Norway
  • | 2 Department of Geosciences, University of Oslo, Oslo, Norway
  • | 3 VTT Technical Research Centre of Finland, Espoo, Finland
© Get Permissions
Full access

Abstract

In-cloud icing on aircraft and ground structures can be observed every winter in many countries. In extreme cases ice can cause accidents and damage to infrastructure such as power transmission lines, telecommunication towers, wind turbines, ski lifts, and so on. This study investigates the potential for predicting episodes of in-cloud icing at ground level using a state-of-the-art numerical weather prediction model. The Weather Research and Forecasting (WRF) model is applied, with attention paid to the model’s skill to explicitly predict the amount of supercooled cloud liquid water content (SLWC) at the ground level at different horizontal resolutions and with different cloud microphysics schemes. The paper also discusses how well the median volume droplet diameter (MVD) can be diagnosed from the model output. A unique dataset of direct measurements of SLWC and MVD at ground level on a hilltop in northern Finland is used for validation. A mean absolute error of predicted SLWC as low as 0.08 g m−3 is obtained when the highest model resolution is applied (grid spacing equal to 0.333 km), together with the Thompson microphysics scheme. The quality of the SLWC predictions decreases dramatically with decreasing model resolution, and a systematic difference in predictive skill is found between the cloud microphysics schemes applied. A comparison between measured and predicted MVD shows that when prescribing the droplet concentration equal to 250 cm−3 the model predicts MVDs ranging from 12 to 20 μm, which corresponds well to the measured range. However, the variation from case to case is not captured by the current cloud microphysics schemes.

Corresponding author address: Bjørn Egil Kringlebotn Nygaard, The Norwegian Meteorological Institute, P.O. Box 43, Blindern N-0315 Oslo, Norway. E-mail: bjornen@met.no

Abstract

In-cloud icing on aircraft and ground structures can be observed every winter in many countries. In extreme cases ice can cause accidents and damage to infrastructure such as power transmission lines, telecommunication towers, wind turbines, ski lifts, and so on. This study investigates the potential for predicting episodes of in-cloud icing at ground level using a state-of-the-art numerical weather prediction model. The Weather Research and Forecasting (WRF) model is applied, with attention paid to the model’s skill to explicitly predict the amount of supercooled cloud liquid water content (SLWC) at the ground level at different horizontal resolutions and with different cloud microphysics schemes. The paper also discusses how well the median volume droplet diameter (MVD) can be diagnosed from the model output. A unique dataset of direct measurements of SLWC and MVD at ground level on a hilltop in northern Finland is used for validation. A mean absolute error of predicted SLWC as low as 0.08 g m−3 is obtained when the highest model resolution is applied (grid spacing equal to 0.333 km), together with the Thompson microphysics scheme. The quality of the SLWC predictions decreases dramatically with decreasing model resolution, and a systematic difference in predictive skill is found between the cloud microphysics schemes applied. A comparison between measured and predicted MVD shows that when prescribing the droplet concentration equal to 250 cm−3 the model predicts MVDs ranging from 12 to 20 μm, which corresponds well to the measured range. However, the variation from case to case is not captured by the current cloud microphysics schemes.

Corresponding author address: Bjørn Egil Kringlebotn Nygaard, The Norwegian Meteorological Institute, P.O. Box 43, Blindern N-0315 Oslo, Norway. E-mail: bjornen@met.no

1. Introduction

In-cloud icing (also referred to as rime icing) is a weather phenomenon that occurs when an unheated structure is exposed to liquid cloud droplets at a temperature T below the freezing point. It is usually most pronounced in exposed mountainous terrain where the cloud base is frequently located lower than the terrain height such that the mountain peaks are in cloud. In many locations in-cloud icing can persist for many days, or even weeks, and in combination with strong winds the accumulated ice load can become extremely high and cause instabilities, faults, and damages to infrastructure. In the literature several examples are found of such damages to overhead power lines (e.g., Thorkildson et al. 2009; Qiang et al. 2005), reduced efficiency and fatigue damages on wind turbines (Frohboese and Anders 2007; Jasinski et al. 1998), and collapse of telecommunication towers (Mulherin 1998). Also ground structures such as ski lifts, measurement masts, meteorological instruments (Makkonen et al. 2001b), and buildings are often subject to in-cloud icing on the top or near the top of exposed mountains and hills. At a coastal mountainous site in Norway more than 300 kg m−1 of ice were measured in April 1961 on a 66-kV power line (Fikke et al. 2008). Ice loads like that clearly demonstrate that the frequency and magnitude of in-cloud icing need to be carefully taken into account in the design of all ground-based structures in cold regions, both for the mechanical strength of the construction and for functionality and performance under icing conditions. However, long and reliable time series of icing measurements sufficient to assess the icing climatology are rarely available. From a wind energy point of view, the assessment of icing frequency is extremely important as many of the potential onshore wind farm sites in the Nordic countries are located on hilltops and elevated areas. In addition to favorable wind conditions, such sites are very often also exposed to in-cloud icing, and the production loss caused by icing has potentially enormous consequences for the profitability of a wind power plant. Therefore, it is of huge interest for the industry to develop new methods and modeling tools to assess the frequency and intensity of in-cloud icing. In-cloud icing furthermore causes a safety problem to small airplanes and helicopters (Gent et al. 2000).

Several studies on how supercooled water droplets accrete on different structures have been published over the last few decades, both from icing tunnel experiments and from theoretical studies of collision and collection efficiency of water droplets. The theoretical basis for modeling ice accretion on cylinders is described in detail in Makkonen (1984, 2000) and has been verified at a high level of accuracy in controlled laboratory experiments (Makkonen and Stallabrass 1987). Also numerical models for the formation of ice on wind turbine blades are found in the literature (Makkonen et al. 2001a), and Wagner et al. (2009) presented a simulation scheme to model the ice buildup on bundle transmission line conductors. Even though precise models for the ice accumulation on different structures exist, many practical applications of such ice accretion models (IAM) are limited by the lack of reliable meteorological input data. The greatest uncertainty in the meteorological input data is usually related to the supercooled cloud liquid water content (SLWC) and the size distribution of cloud droplets. The SLWC is needed in computing the mass flux of icing particles:
eq1
where is the density of water, D is the droplet diameter, and N(D) is the droplet size distribution. The droplet size distribution is needed in the IAMs to compute the collision efficiency between the droplets and the iced structure. Finstad et al. (1988) showed that instead of calculating the collision efficiency for each size bin in the droplet size distribution, the collision efficiency for the median volume droplet diameter (MVD) alone is a very good representation of the whole spectrum. The MVD is defined so that it splits the droplet size distribution in two halves with respect to mass:
e1
Precise information about SLWC and MVD is very limited in real cases, and therefore they are usually roughly estimated or parameterized based on standard meteorological measurements, as in Sundin and Makkonen (1998) and Harstveit (2002).

Until recent years explicit prediction of in-cloud icing with numerical weather prediction (NWP) models has not been attainable because of coarse model resolution and crude parameterizations of subgrid-scale processes. However, the increase of computing power has made it possible to run NWP models at a grid spacing of 1 km or less, and to incorporate more sophisticated and computationally expensive microphysical processes in the parameterization schemes, as well as more prognostic variables. Drage and Hauge (2007) showed promising results by running the fifth-generation Pennsylvania State University–National Center for Atmospheric Research Mesoscale Model (MM5) at 1-km grid spacing for an icing episode in coastal mountainous terrain in Norway. However, no direct validation of the predicted SLWC and MVD at ground level was made.

In the current study we investigate how well SLWC and MVD measured from the ground can be reproduced in simulations with a state-of-the-art NWP model, namely the Weather Research and Forecasting (WRF) model. We study the effect of different horizontal model resolutions, and investigate the influence of applying different cloud microphysics schemes.

2. Measurements of SLWC and MVD

SLWC and MVD measurements were carried out at the top of Mount Ylläs (67.6°N, 24.3°E) in northern Finland, having an elevation of 719 m. It is a rounded peak, and the highest mountain in a large region with surrounding terrain that ranges in elevation from 150 to 300 m. The cloud physical studies were made in connection with a measurement campaign of ice load on a television tower on Mount Ylläs. Accurate measurements of in-cloud icing were carried out for several years, using a rotating multicylinder instrument (RMC) (Brun et al. 1955). The measurement relies on manually installing and removing the RMC device and weighing the ice samples. Wind tunnel tests and verification demonstrating the preciseness of the technique were presented in Makkonen and Stallabrass (1987). Because of variable weather conditions and the need of manual operation, the measurements were sporadic at irregular intervals. The wind speed required by the RMC method was measured by Hydrotech anemometers heated by 1.5-kW power. This heating power kept them ice free under all conditions, as verified by simultaneous video recordings (Makkonen et al. 2001b).

The data were analyzed by software developed for the purpose (Makkonen 1992). This software includes a data quality control feature that allows selecting the most reliable measurements. For them the accuracy is estimated as 2%, that is, typically ±0.005 g m−3 for the SLWC and ±0.3 μm for the MVD. This high accuracy is possible because of the following:

  • ice accretion on several (up to six) cylinders is used in solving an equation with two unknowns (only two cylinders would be required),

  • low sensitivity of icing rate to MVD for the smallest cylinder,

  • high (and linear) sensitivity of icing rate to SLWC for the smallest cylinder, and

  • high sensitivity of icing rate to MVD for the large cylinders.

For the current study, eight cases were selected, characterized by a moist planetary boundary layer, stratus clouds with base below the mountain top, temperature below 0°C, and absence of precipitation. On the time scale of the exposure time of the RMC, which is typically 60 min for the cases considered, such weather conditions are considered rather stationary and we do not expect the conditions to vary greatly during the RMC measurements. An overview of the observations from all the eight cases is presented in Table 1. The eighth column in the table shows the linear correlation coefficient (r) between the measured ice mass and the ice mass modeled using the SLWC and MVD retrieved by analyzing the RMC data. Column number nine shows the number of cylinders N used in calculating the regression and the correlation coefficient r. The value of N is determined by the criterion that the collision efficiency E (mean for the measurement interval) has been in the valid range of the theory. This range is E ≥ 0.07 (Makkonen and Stallabrass 1987); that is, cylinders with E < 0.07 are always excluded from the analysis, and this determines N. Table 1 shows that for all the selected cases at least three cylinders have been in the valid range and the resulting correlation between measured ice load and modeled ice load using the estimated SLWC and MVD is very good.

Table 1.

Weather data and RMC parameters measured at the Mount Ylläs test site.

Table 1.

3. WRF model setup

The mesoscale numerical weather prediction model used in this study is the Advanced Research WRF (ARW) modeling system, version 3.1.1 (Skamarock et al. 2005). Simulations are carried out by applying telescopic nesting of four “levels,” which means that four model domains of increasing resolution are embedded inside each other (Fig. 1). All the four nested domains are initiated at the same time and run simultaneously in one model simulation. To study the effect of different model resolutions, only one-way nesting is applied (from the coarser to the finer grid). The grid spacing increases stepwise by a factor of 3 from 9 km in the outermost domain to 0.333 km in the innermost high-resolution domain. The terrain data used as input to WRF are obtained from the U.S. Geological Survey (USGS) global 30 arc s elevation (GTOPO30) dataset, and the model is configured with 66 hybrid coordinate vertical levels (η levels) for all domains with the model top at 100 hPa. This corresponds to approximately 20 η levels in the lowest kilometer of the atmosphere, and five η levels in the lowest 100 m.

Fig. 1.
Fig. 1.

Configuration of nested domains used in the WRF simulations. The four nested domains are indicated with white squares, with grid spacing decreasing from 9 km in the outermost domain to 0.333 km in the innermost domain.

Citation: Journal of Applied Meteorology and Climatology 50, 12; 10.1175/JAMC-D-11-054.1

Special attention is paid to the model’s ability to resolve local terrain features at the different model resolutions. Since Mount Ylläs is an isolated hill surrounded by a relatively flat region, the modeled height of the hill is strongly dependent on the horizontal resolution applied (Table 2). The intention of using grid spacing as low as 0.333 km is to resolve the local terrain such that terrain-induced vertical motions are captured by the model. Unfortunately no terrain data with higher resolution than the 30 arc s USGS data were available for this region, so even in the 0.333-km domain the height of the hill was suppressed by 82 m. To totally eliminate the height reduction of the hill as a source of error for the icing predictions, an additional scaling of the terrain height was performed in the innermost domain such that the modeled hilltop exactly matches the real height of 719 m. The scaling was done in such a way that all the grid points with land height above a reference height (400 m) were scaled with a factor that was calculated such that the height of Ylläs after the scaling would be exactly 719 m. The terrain before and after the modification is displayed in Fig. 2.

Table 2.

Domain configuration in the WRF simulations. A scaling of the terrain was performed in domain 4 to fit the terrain elevation to the real height of Mount Ylläs. The height before scaling is shown in parentheses.

Table 2.
Fig. 2.
Fig. 2.

Terrain elevation in the innermost domain. The black mesh illustrates the terrain after the process of scaling the topography to match the real mountain height.

Citation: Journal of Applied Meteorology and Climatology 50, 12; 10.1175/JAMC-D-11-054.1

Initial fields and lateral boundary conditions are retrieved from the global European Centre for Medium-Range Weather Forecasts (ECMWF) Interim Re-Analysis data (ERA-Interim), at 38 pressure levels with a temporal resolution of 6 h. All simulations are cold started between 6 and 12 h in advance of the measurement time. To keep the model numerically stable throughout the simulations, a time step of only 0.44 s was required in the innermost domain.

The following parameterization schemes are employed for the subgrid-scale processes: Rapid Radiative Transfer Model (RRTM) longwave and Dudhia shortwave radiation schemes, the Yonsei University (YSU) scheme for the planetary boundary layer, and the Monin–Obukhov surface-layer scheme. Convection is assumed to be explicitly resolved by the model; hence no parameterization scheme is applied for convection. Further description of the parameterization schemes can be found in Skamarock et al. (2005).

Cloud microphysics schemes

For the cloud microphysics three different parameterization schemes have been tested for all cases: the Morrison two-moment scheme (Morrison and Pinto 2005), the Thompson scheme (Thompson et al. 2004, 2008) and the Eta Grid-Scale Cloud and Precipitation scheme (EGCP01) (Rogers et al. 2001), also known as the Ferrier–Eta scheme.

Among the three schemes, the EGCP01 is the most computationally efficient scheme, as it combines all hydrometeors into one variable named total condensate, which is then advected together with water vapor mixing ratio. After advection, the scheme utilizes local storage arrays to separate the total condensate into cloud water, rain, small ice particles, and large ice (snow). The scheme is designed for efficiency but still parameterizes many important mixed-phase processes, and in principle allows supercooled cloud water to exist all the way down to −40°C. Because of its good trade-off between complexity and computational efficiency the scheme is employed in many operational forecasting models, such as the North American Modeling System (Rogers et al. 2005).

The Thompson microphysics scheme is a more sophisticated scheme, with prognostic equations for the mass concentration of five hydrometeors: cloud water, cloud ice, rain, snow, and graupel. It also predicts the number concentration of cloud ice and rain (two-moment ice and rain). The scheme was originally developed to improve explicit prediction of aviation icing, and from there it applies rather sophisticated formulations of mixed-phase processes relative to other bulk microphysics schemes. Special attention is paid to the formulation of the snow category, which has been shown to play a major role for the prediction of SLWC (Thompson et al. 2008).

The Morrison two-moment scheme is the most advanced of the three schemes as it predicts the mass concentration of five hydrometeors (the same as in the Thompson scheme), in addition to the number concentration of four species: cloud ice, snow, rain, and graupel. It is based on the full two-moment scheme described in Morrison et al. (2009); however, the version implemented in the current WRF version does not employ two-moment cloud liquid water as it only predicts the mass concentration. The prediction of both mass mixing ratio and number concentration of four water species allows a more robust description of size distributions, which are important for the parameterization of the individual microphysical interactions and processes.

4. Results and discussion

Eight cases are simulated, with three runs for each case applying the three different microphysics schemes. Output is stored every 30 min at various horizontal resolutions according to the grid spacing in the four nested domains. The simulations are carried out on a 64-bit Linux cluster applying 64 CPUs, and the computing time varied between the simulations because of the different complexities of the microphysics schemes. Compared to the simplest and most economic EGCP01 scheme, the total simulation time was 19% longer with the Thompson scheme and 31% longer with the Morrison two-moment scheme. A typical computing time for a 15-h simulation was 8.5 h when using the EGCP01 scheme.

a. Evaluation of general model performance

Before going into a detailed verification of predicted supercooled cloud water it is useful to investigate how well the model captures the general weather situation. For instance if the predicted temperature and humidity profile differ significantly from reality, it would be meaningless to have any confidence in the predicted SLWC and MVD, simply because the microphysics in WRF are being forced by wrong temperature and humidity profiles.

Unfortunately there are not many data available from the actual site on Ylläs to reveal potential model errors, except instantaneous values of temperature, wind speed, SLWC, and MVD at ground level. However, there is a meteorological observatory at Sodankylä (World Meteorological Organization station code EFSO 02836) approximately 100 km to the east of Ylläs. From this station radiosoundings are available 2 times per day at 0000 and 1200 UTC. A comparison between predicted and measured profiles at Sodankylä shows that the model in general has a good representation of the weather situation in the cases simulated. Both measured and simulated profiles show the air reaching saturation with respect to water at ground level or close to ground level with the moist air masses often trapped inside or below an inversion layer. Some minor discrepancies are found, and the case with largest deviation is shown in Fig. 3. In this case there is a shallow but strong temperature inversion close to the ground that is not captured by the WRF model, causing the modeled temperature at Sodankylä to be approximately 5°C too high at ground level. However, since the entire inversion layer is below the height of Ylläs this error does not affect the predictions on the hilltop where the predicted temperature matches the measurement very well. The overall mean absolute error (MAE) of the temperature predictions for Sodankylä is 1.6°C including the case shown in Fig. 3.

Fig. 3.
Fig. 3.

Examples of temperature profiles. (a) Measured profile at Sodankylä, 1200 UTC 12 Dec 1994. (b) Corresponding predicted profile at Sodankylä. The lines show the temperature in degrees Celsius (right curve) and the corresponding dewpoint temperature (left curve).

Citation: Journal of Applied Meteorology and Climatology 50, 12; 10.1175/JAMC-D-11-054.1

b. Validation of predicted SLWC

Measured values of cloud SLWC at certain time slots are compared to the corresponding predicted values at 719 m MSL. In the innermost domain this height corresponds to the lowest model level, while in the coarser domains it corresponds to levels located higher above the ground because the height of the hill is reduced because of smoothing effects. In these coarser domains the predicted SLWC is linearly interpolated between the vertical levels, to match the height of the real mountain.

For each simulation one deterministic SLWC value is extracted from the grid point on the very top of the hill for a point in time centered in the middle of the exposure time of the RMC measurement. These deterministic values are shown by symbols (dots, triangles, and diamonds) in Figs. 46. Model uncertainty/variation is added to the graphs using lines representing the range of SLWC from neighboring grid points. In the horizontal direction SLWC is extracted at ±one grid box from the original point, giving nine SLWC values. All values are interpolated vertically to the height of 719 m. In addition, values are extracted at ±30 min in time. This corresponds approximately to extraction at the beginning, in the middle of and at the end of the RMC exposure time. In total, this procedure provides a distribution of 27 SLWCs with their full range visualized by thin lines (Figs. 46), and the range between the 25th and the 75th percentile (first and third quartile) visualized by thicker lines.

Fig. 4.
Fig. 4.

Measured vs predicted cloud SLWC at three different horizontal resolutions for WRF simulations with the EGCP01 cloud microphysics scheme. Case number corresponds to the eight cases listed in Table 1. The markers indicate the point values from the grid box closest to the actual site, extracted in the middle of the RMC exposure time. Thin lines show the range of SLWC from neighboring grid points in time and space, while the thicker parts of the line is drawn between the 25th and the 75th percentile of the neighboring values.

Citation: Journal of Applied Meteorology and Climatology 50, 12; 10.1175/JAMC-D-11-054.1

Fig. 5.
Fig. 5.

As in Fig. 4, but with the Thompson cloud microphysics scheme.

Citation: Journal of Applied Meteorology and Climatology 50, 12; 10.1175/JAMC-D-11-054.1

Fig. 6.
Fig. 6.

As in Fig. 4, but with the Morrison cloud microphysics scheme.

Citation: Journal of Applied Meteorology and Climatology 50, 12; 10.1175/JAMC-D-11-054.1

Figures 46 show that in general the predicted SLWC increases with increasing model resolution. At spacing of 0.333 km the model successfully predicts icing conditions in all cases when the Thompson or the Morrison microphysics scheme is used, and in seven out of eight cases when switching to the EGCP01 scheme (deterministic values). When grid spacing is increased to 3 km the model predicts no supercooled cloud water at all in many of the simulations, most evident when the EGCP01 microphysics is applied (Fig. 4) where only one out of eight simulations predicts cloud water at the site. In most cases the model underestimates the SLWC in the 3-km domain, while the predictive skill increases as the horizontal resolution increases. When the highest resolution is applied, the average results suggest a slight overestimation, most evident in the simulations with the Morrison microphysics scheme (Fig. 6). The best match with measurements is achieved when using the Thompson scheme in combination with the highest resolution. These characteristics are confirmed both by the deterministic values and by the bars representing the first and third quartile. Figure 7 shows that the MAE of predicted SLWC computed based on the deterministic values clearly decreases with increasing horizontal resolution. A similar calculation (not shown) based on the median values of the neighboring grid points gave slightly higher MAE for the 1-km results but no significant change to the 3-km and the 0.333-km results.

Fig. 7.
Fig. 7.

MAE of predicted SLWC for simulations with various combinations of cloud microphysics schemes and horizontal resolutions. MAE is calculated on the basis of the deterministic SLWC values.

Citation: Journal of Applied Meteorology and Climatology 50, 12; 10.1175/JAMC-D-11-054.1

1) Sensitivity to horizontal resolution

The reason for the large dependency on horizontal resolution is to a large extent related to orographic lifting of air over the hilltop. As explained in the previous chapter the height of the hill is strongly dependent on the horizontal resolution (Table 2). The main source of the cloud water measured on the top of Ylläs is by condensation when moist air is forced to ascend on the upstream side of the hill. At grid spacing of 3 km the modeled hill is 367 m MSL, which is approximately 160 m higher than the surrounding terrain, while in the highest-resolution domain (and also in reality) this difference is roughly 510 m. Not surprisingly the results suggest that a realistic representation of the terrain is crucial for a successful prediction of icing conditions on isolated hills like Ylläs. This might not be as apparent for locations where other forcing mechanisms are more important, for example, predicting icing for tall structures in flat terrain or offshore.

2) Sensitivity to cloud microphysics scheme

Figure 7 shows that in addition to a general improvement when increasing the resolution, the results are quite sensitive to the choice of microphysics scheme. The Thompson and the Morrison simulations have a slight positive bias, 0.04 and 0.08 g m−3, respectively, while the EGCP01 runs have a negative bias of −0.13 g m−3 when the highest resolution is used. In general, we obtain the best match with observations when the Thompson microphysics is applied, with an MAE of 0.08 g m−3, as compared with 0.13 g m−3 for the Morrison runs and 0.15 g m−3 for simulations with the EGCP01 scheme, shown in Fig. 7.

With regard to further development and improvement of the microphysics schemes it would be useful to understand why the results differ between the microphysics schemes, and to identify which assumptions and parameterizations that are important for the successful prediction of in-cloud icing at ground level.

By performing a qualitative examination of the three-dimensional evolution of the hydrometeors for all the different model simulations we are able to point out some features of the schemes that seem to be responsible for some of the differences in the SLWC predictions, and hence are important for the successful prediction of icing conditions. There is especially one case in which the results from the three schemes differ greatly. At 0900 UTC 8 February 1990 the measured SLWC was 0.43 g m−3 while predicted values were 0.0, 0.42, and 0.50 g m−3 for simulations with the EGCP01, Thompson, and Morrison schemes, respectively (Fig. 8). Figure 9 shows that the run with the Thompson scheme initially creates a supercooled liquid cloud in the lowest levels and a cloud consisting of ice particles above 5000 m, producing some light snow. After about 1 h the snow falling from above gets into contact with the low-level liquid cloud, and riming on the snowflakes starts as the snowflakes collide with the cloud droplets and deplete the cloud water. In a short time period the riming results in the production of a small amount of graupel (heavily rimed snow crystals). As the concentration of snow increases it starts to reach the ground producing almost 5-mm water equivalent precipitation before 0600 UTC. At this point the amount of snow falling from above slowly decreases, and after 0700 UTC no more snow crystals are seeding the top of the low-level cloud. Terrain-induced vertical motions continue to produce cloud water on the upstream slope of the hill, hence the low-level liquid cloud is reestablished and the amount of supercooled cloud water increases and is quite close to the measured value at 0900 UTC.

Fig. 8.
Fig. 8.

Measured vs predicted cloud SLWC, all from simulations with grid spacing equal to 0.333 km but with different cloud microphysics schemes. Markers and lines are as explained in Fig. 4.

Citation: Journal of Applied Meteorology and Climatology 50, 12; 10.1175/JAMC-D-11-054.1

Fig. 9.
Fig. 9.

(top) Time–height cross section from the top of Ylläs. Shading (white/black/red) = cloud LWC, blue contour lines = snow, green contour lines = cloud ice, cyan contour lines = graupel, red contour lines = rain, and gray contour lines = temperature (°C). All hydrometeor species are plotted as mass concentration (g m−3). (bottom) Time series of mass concentration in the lowest model layer. The dark green line with triangle markings indicates accumulated precipitation at the surface. The simulation is initiated at 0000 UTC 8 Feb 1990, with grid spacing = 0.333 km and the Thompson cloud microphysics scheme.

Citation: Journal of Applied Meteorology and Climatology 50, 12; 10.1175/JAMC-D-11-054.1

From Fig. 10 we see that the corresponding simulation with the Morrison microphysics scheme behaves in a very similar manner as the one with the Thompson scheme. Some differences can be noticed though: The Morrison scheme produces a much higher concentration of cloud ice, and the riming of snow crystals does not result in the production of graupel. However, the differences do not have any significant influence on the predicted SLWC.

Fig. 10.
Fig. 10.

As in Fig. 9, but with the Morrison cloud microphysics scheme.

Citation: Journal of Applied Meteorology and Climatology 50, 12; 10.1175/JAMC-D-11-054.1

The simulation with the EGCP01 scheme (Fig. 11) develops quite differently from the other two schemes. As the initially produced low-level supercooled cloud is seeded with snow/ice crystals, all cloud water is depleted and the low-level cloud becomes completely glaciated at 0600 UTC. In contrast to the other two schemes the cloud remains glaciated also after the seeding of ice/snow crystals from higher altitudes stops. The scheme continues to produce light snow and fails to predict icing conditions at the measurement time. The reason for this contrasting behavior is the structure of the EGCP01 scheme with advection of only one water category, namely the total condensate, and local storage arrays to distribute the total concentrate into different hydrometeor classes. When the condensate at the top of the mountain is categorized as 100% in the ice/snow category, all condensate entering this grid box by advection will be considered ice. The only way to gain liquid cloud water in the grid box is by condensation (melting of cloud ice is not an issue in this particular case), but since condensation mainly occurs on the upstream slope of the hill, the condensate on the very top remains in the ice category. This example illustrates one shortcoming of the simplest and least computationally expensive scheme compared to the other two schemes.

Fig. 11.
Fig. 11.

As in Fig. 9, but with the EGCP01 cloud microphysics scheme.

Citation: Journal of Applied Meteorology and Climatology 50, 12; 10.1175/JAMC-D-11-054.1

Another important result is that all the simulations with the EGCP01 scheme produce slightly less cloud water by condensation than those using the other two schemes. Also lower contents of ice and snow particles are found when using the EGCP01 scheme. These features are exemplified in the time–height cross section plots shown in Figs. 1214. In this particular case all the three microphysics schemes produce a thin supercooled cloud in the lowest model levels that persists throughout the whole simulation. The Thompson and Morrison schemes produce slightly more snow and cloud ice, which results in mixed-phase conditions at ground level during the last few hours of the simulation, while the EGCP01 on average has slightly lower SLWC and does not produce any snow/ice reaching the ground as found in the two other schemes. At 1100 UTC the measured SLWC was 0.10 g m−3 while the predicted values were 0.02, 0.12, and 0.14 g m−3 for the EGCP01, Thompson, and Morrison schemes, respectively. The differences found in this example are quite characteristic for many of the other simulations, and summarized over all the cases we find that the SLWC at the top of Ylläs is generally lower when using the EGCP01 scheme compared to the Thompson and Morrison schemes. A second important difference is that the EGCP01 scheme produces less cloud ice and less frequent seeding of the orographic cloud at Ylläs, causing less frequent events of precipitation reaching the ground relative to the Thompson and the Morrison scheme. This is to some extent compensated by more rapid glaciation and production of snow when mixed-phase events occur. Hence domain-averaged values of accumulated precipitation do not differ significantly between the schemes.

Fig. 12.
Fig. 12.

As in Fig. 9, but for a model simulation initiated at 0000 UTC 12 Dec 1994 using the Thompson cloud microphysics scheme.

Citation: Journal of Applied Meteorology and Climatology 50, 12; 10.1175/JAMC-D-11-054.1

Fig. 13.
Fig. 13.

As in Fig. 12, but with the Morrison cloud microphysics scheme.

Citation: Journal of Applied Meteorology and Climatology 50, 12; 10.1175/JAMC-D-11-054.1

Fig. 14.
Fig. 14.

As in Fig. 12, but with the EGCP01 cloud microphysics scheme.

Citation: Journal of Applied Meteorology and Climatology 50, 12; 10.1175/JAMC-D-11-054.1

Even though the Thompson scheme and the Morrison scheme behave very similarly in the simulations, there are small variations that result in rather significant differences in the SLWC validation at the top of Ylläs (Fig. 7). The main reason for the higher MAE with the Morrison scheme is that SLWC is overestimated in many of the cases. Whether or not this is a weakness of the Morrison scheme is not straightforward to judge from the following reason: there is a sink term for cloud water that is not taken into account in the WRF model simulations, which is observed in reality, namely droplet deposition on the ground (in-cloud icing/riming on trees etc.). In some cases this process might cause a significant loss of cloud water close to the ground, especially in exposed forested areas. However, the magnitude is highly uncertain and no parameterization of such a process has so far been documented. Keeping this in mind, we could expect an overprediction of SLWC close to the ground by the current model version, and thus it is difficult to draw firm conclusions when comparing the two schemes. We also note that there is a slight tendency to overestimate the high SLWC measurements and to underestimate the low SLWC measurements, particularly when using the Thompson or the Morrison scheme (Figs. 5 and 6). Possible reasons for this behavior have not been attainable to identify based on the limited number of cases in this study. Further research with more cases would be necessary to address this issue.

c. Prediction of MVD

The current versions of the three cloud microphysics schemes predict only the mass mixing ratio of the cloud water, and apply a fixed number concentration for cloud droplets throughout the simulations. A droplet size distribution is employed to describe how droplet mass is distributed with respect to size. Based on the droplet concentration and the assumed size distribution, characteristic droplet sizes can be diagnosed, such as the MVD, which is required for further postprocessing in IAMs. The Thompson scheme applies a generalized gamma distribution for the cloud water (Thompson et al. 2008):
e2
where N(D) is the number of droplets per unit volume, N0 is the intercept parameter, and μ is the size distribution shape parameter given by
eq2
with Nc representing the droplet concentration given as droplets per cubic centimeter. The parameter λ in (2) is the slope parameter
e3
where ρw is the density of water. By solving (1) with respect to MVD using the size distribution in (2) we obtain a diagnostic relation for the MVD (Thompson et al. 2009):
e4

From the simulations with the Thompson microphysics scheme we diagnose the MVD from (4) and compare with the measurements. Figure 15a shows that on average the MVD is overestimated by the model (7 out of 8 cases based on the deterministic SLWC values) when the Thompson microphysics scheme is used with a droplet concentration of Nc = 100 cm−3. Since there was basically no bias (0.04) in the SLWC prediction, this result suggests that 100 droplets per cubic centimeter is a too- low value. The result corresponds well to the recommendations found in Thompson et al. (2008) where the user is encouraged to set the droplet concentration according to the environment, with 100 cm−3 as a typical value for very clean or maritime air, and 250 cm−3 or higher for continental or polluted air. This clear difference between marine and continental air is also well supported by empirical data (Miles et al. 2000). Figure 15b shows the same comparison, but now with the predicted MVD calculated with Nc changed to 250 cm−3, which is a more reasonable value for Ylläs, which must be considered a continental site. By increasing the droplet concentration the average result is improved and the systematic overprediction of MVD is clearly reduced. By choosing Nc = 400 cm−3 the bias is completely eliminated from the results.

Fig. 15.
Fig. 15.

Measured vs predicted values of MVD (μm) using the Thompson cloud microphysics scheme at grid spacing of 0.333 km. (a) MVD calculated with Nc = 100 cm−3 and (b) MVD calculated with Nc = 250 cm−3. The black dots are calculated using the deterministic SLWC values while the lines represent the range of variation from the neighboring grid points.

Citation: Journal of Applied Meteorology and Climatology 50, 12; 10.1175/JAMC-D-11-054.1

From Fig. 15 we can also see that even though we are able to predict the average MVD by setting the proper droplet concentration, the model does not seem to capture any of the case to case variation in MVD. By plotting measured MVD against measured SLWC (Fig. 16) we find the measurements randomly distributed on the scatterplot, and we do not see any sign of correlation between the two variables. This contradicts with the assumption used in the predictions where the relation: MVD ~ (LWC)1/3 is applied [by combining (3) and (4)], as illustrated in Fig. 16 with dashed lines drawn for different values of Nc. If we assume that the cloud water follows the size distribution given in (2), we find large variation in the true droplet concentration, with two measurements corresponding to Nc smaller than 100 cm−3 and four measurements corresponding to Nc larger than 500 cm−3. The result implies that a one-moment prediction of cloud water is not adequate for predicting the variation in droplet size for the cases considered here. This is most likely related to different wind directions and advection of air masses with different concentrations of cloud condensation nuclei. However, no such relation to local wind direction could be identified based on the eight cases considered here. It should be mentioned that the variation in Nc might be totally different for icing predictions on a coastal site, where in-cloud icing only occurs with winds from a certain sector, where maritime air is advected and lifted until saturation is reached. At such sites the droplet concentration would be expected to vary less between icing events, hence a more pronounced correlation between droplet size and SLWC than found here would be expected (Miles et al. 2000).

Fig. 16.
Fig. 16.

Measured MVD plotted against measured SLWC. The dashed lines indicate theoretical curves for various droplet concentrations based on the assumed droplet size distribution (2).

Citation: Journal of Applied Meteorology and Climatology 50, 12; 10.1175/JAMC-D-11-054.1

5. Summary and conclusions

We have examined how well episodes of in-cloud icing at ground level can be simulated by a state-of-the-art NWP model. Eight cases were simulated with the WRF model and predicted values of SLWC and median volume droplet size were validated against RMC measurements at Mount Ylläs, located in northern Finland.

The overall results suggest that horizontal resolution is a key element for successful prediction of icing at ground level, because terrain-induced vertical motions are the main forcing for the production of cloud water at this site. When the highest horizontal resolution is applied with grid spacing of 0.333 km the model is able to capture all the icing events when the Thompson or Morrison cloud microphysics scheme is used. We obtain the best match between measured and predicted SLWC when the highest resolution is applied in combination with the Thompson microphysics scheme, resulting in a mean absolute error of only 0.08 g m−3. There are at least two reasons why such good results can be obtained for this particular site. First, the main forcing for the production of SLWC is the orographic lifting of moist air on the upstream slope of the hill. This forcing is rather strong and a successful simulation is dependent on a good representation of the local terrain (horizontal resolution). Second, many of the icing events are related to pure liquid low-level stratus and/or orographic clouds without interaction with ice particles. More mixed-phase conditions would have increased the number of interaction terms in the prognostic calculation of cloud water, and hence made the SLWC predictions more challenging.

Among the three different microphysics schemes tested the study addresses one limitation of the simplest and most economic scheme (EGCP01) that is partly responsible for a lower score than the other two schemes in the prediction of SLWC. While the Thompson and the Morrison schemes employ explicit advection of each hydrometeor category, the EGCP01 scheme only advects total condensate and uses local storage arrays to diagnose the relative contribution to each hydrometeor class. In one of the cases this difference plays a particularly large role as the EGCP01 fails to predict the measured amount of supercooled cloud water, and instead diagnoses all the condensate in terms of cloud ice and snow. This is opposite to the other two schemes that predict a pure supercooled liquid cloud more in accordance with the measurements.

By using the droplet size distribution and droplet concentration that is assumed in the microphysics schemes together with the predicted SLWC we are able to predict the droplet size in terms of MVD. The results suggest that a droplet concentration that is typical of a continental site provides an MVD that in the mean corresponds well with the measurements. The study also suggests that explicit prediction of the variation of MVD from case to case requires a prediction of the variation in droplet concentration for cloud water. This is a limitation of all the three microphysics schemes tested here, which only predict one moment of cloud liquid water (mass mixing ratio). For practical application of the model, this will introduce an uncertainty in the predicted icing intensity for single icing events. However, since there is no bias in MVD when using the proper droplet concentration, a fixed number concentration may be adequate for climatological studies of in-cloud icing. It is a subject for future experiments to test whether full two-moment schemes are able to explicitly predict the variation in droplet concentration, for improvement of explicit prediction of icing intensity.

Acknowledgments

The authors thank the Top-Level Research Initiative of the Nordic Council for funding through the IceWind project. This work was also supported by The Norwegian Water Resources and Energy Directorate (NVE) and Statnett, Norway’s national main power transmission grid owner and operator. The authors also thank Gregory Thompson for valuable input. The third author was supported by Tekes via the IEA Ice Action project. The RMC data were collected by the staff of the Finnish Broadcasting Co., Distribution Department (now Digita Oy).

REFERENCES

  • Brun, R., W. Lewis, P. Perkins, and J. Serafini, 1955: Impingement of cloud droplets on a cylinder and procedure for measuring liquid-water content and droplet sizes in supercooled clouds by rotating multicylinder method. National Advisory Committee for Aeronautics (NACA) Rep. 1215, 43 pp.

    • Search Google Scholar
    • Export Citation
  • Drage, M., and G. Hauge, 2007: Atmospheric icing in a coastal mountainous terrain. Measurements and numerical simulations, a case study. Cold Reg. Sci. Technol., 53 (2), 150161.

    • Search Google Scholar
    • Export Citation
  • Fikke, S. M., J. E. Kristjánsson, and B. E. K. Nygaard, 2008: Modern meteorology and atmospheric icing. Atmospheric Icing of Power Networks, M. Farzaneh, Ed., Springer, 1–30.

    • Search Google Scholar
    • Export Citation
  • Finstad, K., E. Lozowski, and L. Makkonen, 1988: On the median volume diameter approximation for droplet collision efficiency. J. Atmos. Sci., 45, 40084012.

    • Search Google Scholar
    • Export Citation
  • Frohboese, P., and A. Anders, 2007: Effects of icing on wind turbine fatigue loads. J. Phys.: Conf. Ser., 75, 012061, doi:10.1088/1742-6596/75/1/012061.

    • Search Google Scholar
    • Export Citation
  • Gent, R. W., N. P. Dart, and J. T. Cansdale, 2000: Aircraft icing. Philos. Trans. Math. Phys. Eng. Sci., 358, 28732911.

  • Harstveit, K., 2002: Using routine meteorological data from airfields to produce a map of ice risk zones in Norway. Proc. 10th Int. Workshop on Atmospheric Icing of Structures, Brno, Czech Republic, paper 8.1, 8 pp.

    • Search Google Scholar
    • Export Citation
  • Jasinski, W., S. Noe, M. Selig, and M. Bragg, 1998: Wind turbine performance under icing conditions. Trans. Amer. Soc. Mech. Eng. J. Sol. Energy Eng., 120, 6065.

    • Search Google Scholar
    • Export Citation
  • Makkonen, L., 1984: Modeling of ice accretion on wires. J. Climate Appl. Meteor., 23, 929939.

  • Makkonen, L., 1992: Analysis of rotating multicylinder data in measuring cloud-droplet size and liquid water content. J. Atmos. Oceanic Technol., 9, 258263.

    • Search Google Scholar
    • Export Citation
  • Makkonen, L., 2000: Models for the growth of rime, glaze, icicles and wet snow on structures. Philos. Trans. Math. Phys. Eng. Sci., 358, 29132939.

    • Search Google Scholar
    • Export Citation
  • Makkonen, L., and J. Stallabrass, 1987: Experiments on the cloud droplet collision efficiency of cylinders. J. Climate Appl. Meteor., 26, 14061411.

    • Search Google Scholar
    • Export Citation
  • Makkonen, L., T. Laakso, M. Marjaniemi, and K. Finstad, 2001a: Modelling and prevention of ice accretion on wind turbines. Wind Eng., 25, 321.

    • Search Google Scholar
    • Export Citation
  • Makkonen, L., P. Lehtonen, and L. Helle, 2001b: Anemometry in icing conditions. J. Atmos. Oceanic Technol., 18, 14571469.

  • Miles, N., J. Verlinde, and E. Clothiaux, 2000: Cloud droplet size distributions in low-level stratiform clouds. J. Atmos. Sci., 57, 295311.

    • Search Google Scholar
    • Export Citation
  • Morrison, H., and J. Pinto, 2005: Mesoscale modeling of springtime Arctic mixed-phase stratiform clouds using a new two-moment bulk microphysics scheme. J. Atmos. Sci., 62, 36833704.

    • Search Google Scholar
    • Export Citation
  • Morrison, H., G. Thompson, and V. Tatarskii, 2009: Impact of cloud microphysics on the development of trailing stratiform precipitation in a simulated squall line: Comparison of one- and two-moment schemes. Mon. Wea. Rev., 137, 9911007.

    • Search Google Scholar
    • Export Citation
  • Mulherin, N., 1998: Atmospheric icing and communication tower failure in the United States. Cold Reg. Sci. Technol., 27, 91104.

  • Qiang, H., W. Jiahong, and O. Mingyong, 2005: Analysis on accidents caused by icing damage in Hunan power grid in 2005 and its countermeasures. Power Syst. Technol. Beijing, 29, 16.

    • Search Google Scholar
    • Export Citation
  • Rogers, E., T. Black, B. Ferrier, Y. Lin, D. Parrish, and G. DiMego, 2001: Changes to the NCEP Meso Eta Analysis and Forecast System: Increase in resolution, new cloud microphysics, modified precipitation assimilation, modified 3DVAR analysis. NWS Tech. Procedures Bull. 488, 15 pp.

    • Search Google Scholar
    • Export Citation
  • Rogers, E., and Coauthors, 2005: The NCEP North American mesoscale modeling system: Final Eta Model/analysis changes and preliminary experiments using the WRF-NMM. Preprints, 21st Conf. on Weather Analysis and Forecasting/17th Conf. on Numerical Weather Prediction, Washington, DC, Amer. Meteor. Soc., 4B.5. [Available online at http://ams.confex.com/ams/WAFNWP34BC/techprogram/paper_94707.htm.]

    • Search Google Scholar
    • Export Citation
  • Skamarock, W., J. Klemp, J. Dudhia, D. Gill, D. Barker, W. Wang, and J. G. Powers, 2005: A description of the Advanced Research WRF Version 2. NCAR Tech. Note 468, 88 pp.

    • Search Google Scholar
    • Export Citation
  • Sundin, E., and L. Makkonen, 1998: Ice loads on a lattice tower estimated by weather station data. J. Appl. Meteor., 37, 523529.

  • Thompson, G., R. Rasmussen, and K. Manning, 2004: Explicit forecasts of winter precipitation using an improved bulk microphysics scheme. Part I: Description and sensitivity analysis. Mon. Wea. Rev., 132, 519542.

    • Search Google Scholar
    • Export Citation
  • Thompson, G., P. Field, R. Rasmussen, and W. Hall, 2008: Explicit forecasts of winter precipitation using an improved bulk microphysics scheme. Part II: Implementation of a new snow parameterization. Mon. Wea. Rev., 136, 50955115.

    • Search Google Scholar
    • Export Citation
  • Thompson, G., B. Nygaard, L. Makonnen, and S. Dierer, 2009: Using the Weather Research and Forecasting (WRF) Model to predict ground/structural icing. Proc. 13th Int. Workshop on Atmospheric Icing of Structures, Andermatt, Switzerland, Meteotest, 8 pp. [Available online at http://www.iwais2009.ch/index.php?id=44.]

    • Search Google Scholar
    • Export Citation
  • Thorkildson, R., K. Jones, and M. Emery, 2009: In-cloud icing in the Columbia basin. Mon. Wea. Rev., 137, 43694381.

  • Wagner, T., U. Peil, and C. Borri, 2009: Numerical investigation of conductor bundle icing. Proc. Fifth African and European Conference on Wind Engineering, Florence, Italy, International Association for Wind Engineering, 9 pp. [Available online at http://www.iawe.org/Proceedings/5EACWE/058.pdf.]

    • Search Google Scholar
    • Export Citation
Save