Atmospheric icing and meteorological variables Full scale

Atmospheric icing and meteorological variables–Full scale experiment and testing of modelsThesis for the Dr. Scient. degreeMagne A. DrageFebruary, 2005The University Centrein SvalbardLongyearbyen, NorwayDepartment of GeophysicsUniversity of BergenBergen, Norway

Geophysical InstituteUniversity of BergenAllegt. 705007 BergenISBN 82-8116-006-3ISSN 1502-5519Reports in Meteorlogy and OceanographyNr. 4-2005

PrefaceThis thesis is a part of the Dr. scient degree submitted to the University ofBergen, Norway. The study has been funded by Forsvarsbygg, Norkring,Statnett and Telenor. Thanks to all persons involved for having the faith inthe project in the starting phase. Particularly, I am grateful to ArnfinnJenssen, who initiated the study.For the last three and a half years I have been employed as a research fellowby the University centre in Svalbard (UNIS). The first year I had my dailywork at the Norwegian Meteorological Institute at the Research andDevelopment department, while I have been situated at the University ofBergen the last two and a half year. In this period, I have also been visitingUNIS several times, recently with a one month stay in January 2005. I amthankful to all of them for providing fine working conditions during thisstudy.Support from the operating personal at the field sites has been a key factor forthe success of this work. I would therefore like to bring a special thanks toOdd Rutledal and Øistein Saugerud, for their excellent support during thefield work. I would also like to thank Rune Stenseth for his support in theinitial stage of this study.I will also thank the co-authors of the presented papers for a close cooperation.Thanks to Professor Yngvar Gjessing who has been my main supervisor andhad the faith in my abilities in the starting phase. Thank you for allinspiration to me all the way back from my undergraduate period at UNIS.Thanks to Amanuensis Jan Asle Olseth who has been co-supervisor, forquick response of my question, and encouragement during the study.Last but not least my wife Sigrid deserves thanks for listening to my concernsduring the work process and for putting up with my late nights of work oreven absence for periods of the study.February 2005Magne A. Drage

1. INTRODUCTION 11.1 OBJECTIVES 11.2 SCOPE OF WORK 31.3 STRUCTURE OF THE REPORT 42. ATMOSPHERIC ACCRETION 52.1 IN-CLOUD ICING 62.2 EFFICIENCY COEFFICIENTS 62.3 CLOUD LIQUID WATER CONTENT 102.4 CLOUD DROPLET SPECTRUM 113. THE EXPERIMENTS 193.1 FIELD SITE 193.2 METEOROLOGICAL INSTRUMENTATION 223.3 ICE SCALE 234. PAPERS4.1 SPECIFIC OBJECTIVES OF THE FOUR 25INDIVIDUAL PAPERSPAPER I: INSTRUMENTATION FOR MEASURING 27ATMOSPHERIC ICINGMAGNE A. DRAGE AND TOR DE LANGEReports in Meteorology and Oceanography,University of Bergen, Report No. 2-2005PAPER II: ATMOSPHERIC ICING IN A COASTAL MOUTAINOUS 49TERRAIN. MEASUREMENTS ANDNUMERICAL SIMULATIONS, A CASE STUDY.MAGNE A. DRAGE AND GARD HAUGECold Regions Science and Technology,Accepted subject to revision.

PAPER III. LARGE SCALE MEASUREMENTS AND 73NUMERICAL SIMULATIONS OF IN-CLOUD ICINGAROUND THE RIDGE OF A MOUNTAIN PEAK.MAGNE A. DRAGE AND THOMAS K. THIISPAPER IV: ARCTIC COASTAL CLIMATIC IMPACT ON DESIGN 97CONSTRUCTION AND OPERATION OF THEHAMMERFEST LNG PLANTMAGNE A. DRAGE AND TRULS MØLMANNProceedings of the 17 th InternationalConference on Port and OceanEngineering under Arctic Conditions,Trondheim, Norway5. TESTING OF MODELS 1115.1 BROSVIKSÅTA 1115.1.1 Non-rotating cylinder 1125.1.2 Methods 1125.1.3 Model results 1155.1.4 Rotating cylinder 1165.1.5 Assumptions and sensitivity tests 1175.2 GAUSTATOPPEN 1215.2.1 Modelled ice load between Oct. 17 and Dec. 16, 2003 1215.2.2 Modelled ice load between Jan. 24 and May 05, 2004 1225.3 ICE DETECTOR 1276. SUMMARY AND CONCLUSIONS 1317. REFERENCES 133

1. IntroductionAtmospheric icing on structures is often a serious problem in regionswith cold climate. Large economic costs, as well as human inconvenience,are the result of failures and damage. During the period 1967-1991, a largenumber of structures in Europe, the U.S.A., Canada and Japan have collapsedor been partly damaged due to atmospheric icing (table 1.1). In the U.S.A.alone, there have been about 140 icing-related tower failures over the last 40years. (Mulherin, 1998). Location of structures in elevated areas represents apotential risk for problems related to icing. In order to obtain the bestpossible coverage, Radio - and Telecommunication require high antennas,which often need to be located in mountainous locations. The potential riskof problems related to atmospheric icing is often underestimated during theplanning and building stages in such areas. Thus, ice data are rarely collectedbefore the structure is built. This illustrates the importance of forecastingatmospheric icing, as well as determining design loads. The research historyof the topic ice and snow accretion on structures has been reviewed by Poots(2000).1.1 ObjectivesThe objective of this study has been to obtain a better understandingof the physical processes for atmospheric icing in the boundary layer ofmountainous regions. By use of collected experimental data, the intention hasbeen to test and develop existing methods for prediction of occurrence,duration and intensity of icing. The advantages and limitations of thesemethods have also been studied. A comparison of detailed measurements ofair temperature and humidity versus height, in relation to icingmeasurements, is, in most cases, lacking in literature about icing.1

Table 1.1. Partly damaged or collapsed structures due to icing(Adapted from Mulherin 1988).2

1.2 ScopeThe theory for modelling atmospheric icing by in-cloud icing isbriefly presented in a literature review followed by a discussion on the mainresults of papers I-IV.Measurements of icing on a 1m high rod by an ice scale, as well asmeasurements of air temperature, relative humidity and wind, have beencrucial for the success of this work. These measurements have led to thedevelopment of a method for estimating liquid water content (LWC) andwind speed in remote areas. These are important parameters for ice accretionby in-cloud icing. In-cloud icing has been found to be the type of icing givingthe highest accumulated ice loads. The largest ice-load ever recorded on apower line is 305 kg m -1 . This was recorded on a 22 kV overhead line inVoss, Norway on April 18, 1961 (Figure 1.1).Figure 1.1. Rime on a 22 kV overhead line at Voss, Norway on April 18. 1961. Iceloadrecorded was 305 kg m -1 on each span. (photo: Olav Wist).Measurements of ice loads are sparsely reported both in time and space. Theuse of direct ice load measurement in the design criteria of constructions israre. A statistically meaningful extreme analysis would require data fromseveral years. Therefore, much attention has been paid to estimateatmospheric ice loads using meteorological data from weather stations (e.g.,Haldar et al. 1988; Makkonen and Ahti, 1995; Sundin and Makkonen, 1998;Harstveit, 2002). This approach has the advantage that meteorological dataare available for many years with a relatively good spatial coverage.3

However, an extrapolation of these data to the site of interest is oftennecessary.In this study, measured ice load, icing intensity and duration for a 1m highrod at 10-minute intervals are compared with field weather station data, alsoat 10-minute intervals. Data from synoptic weather stations are also used as acomparison. The synoptic weather station have measurements, every day, at0000 hrs, 0600 hrs and 1800 hrs GMT. A method of estimating in-cloud icingby use of weather station data is presented.1.3 Structure of the reportThe report is divided into seven chapters, where the four papers are presentedin chapter 4. The papers are not presented at the end of the report due to thefact that the equipment, models and methods evaluated in the papers areapplied in chapter 5. The whole report can therefore be read chronologically.4

2. Atmospheric accretionAccretion is defined as the process where ice builds up on the surface ofan object. Different types of icing on structures are recognised, andatmospheric icing is traditionally classified according to three differentformation processes.1) Precipitation icing:a. freezing rain or drizzle;b. accumulation and refreezing of wet snow.2) In-cloud icing, caused by super-cooled water droplets in clouds orfog.3) Hoar frost/sublimation. Direct phase transition from water vapour intoice. Hoar frost is of low density and strength, and normally does notresult in significant load on structures (Makkonen, 1984a).The density of the different types of ice is varying from 200 to 900 kg/m 3(table 2.1).Table 2.1. Typical properties of accreted atmospheric ice (ISO 12494:2001, 2001).TypeiceofDensitykg/m 3AdhesionandcohesionGeneral appearanceColourShapeGlaze 900 strong transparent evenlydistributed/iciclesWetsnow300 to 600 weak(forming)white evenlydistributed/eccentricstrong(frozen)Hard 600 to 900 strong opaque eccentric, pointingrimewindwardSoft rime 200 to 600 low to white eccentric, pointingmediumwindwardIce accretion also depends on the properties of the accreting object itself,described by its shape, size, material and orientation relative to the wind, aswell as the surrounding surface structure. Measurements of ice accretions5

coefficient α 1 becomes less than one when the water droplets follow thestreamlines around the object without colliding (figure 2.1). Small droplets,large object and low wind speeds reduce α 1 .Langmuir and Blodgett (1946) performed a theoretical investigation of waterdroplet trajectories around cylinders. This investigation describes howdroplets collide with the cylinder within a band limited by polar angles –φ toφ. The angle φ is a function of the droplet radius, cylinder radius, air speed,air temperature and pressure. These calculations are computationally timeconsuming and complicated. For practical applications simplifications arenecessary. Assuming that the icing object is cylindrical, the collisionefficiency can be parameterised by the two dimensionless parameters,and2K = ρw⋅ d ⋅ v / 9 ⋅ µ ⋅ D(2.2)2Reφ =(2.3)Kwhere the Reynolds number, Re is given by:ρa⋅ d ⋅ vRe =(2.4)µHere v is the free stream velocity, d is the droplet diameter, D is the cylinderdiameter, ρ w is the water density, µ is the absolute viscosity of air, and ρ a isthe air density.Empirical fit equations to the numerically calculated data for the collisionefficiency as a function of median volume droplet diameter (MVD), windspeed and cross sectional area of a cylinder, are given by Finstad et al.(1988a).whereα = A−0.028−C(B 0.0454)(2.5)1−7

A = 1.066KB = 3.641K−0.00616−0.498C = 0.00637( φ −100)exp( −1.103Kexp( −1.487K0.381−0.688−0.694))Finstad et al. (1998b) have shown that MVD can accurately replace thedroplet diameter, d, without having to calculate α 1 individually for eachdroplet-size category.Variation of the collision efficiency is considerable with varying wind speed,droplet size and cylinder diameter (figure 2.2).DCollision efficiency, α11.00.80.60.40.210 m/s20 m/s30 m/sACollision efficiency, α11.00.80.60.40.2B10 E-6 m20 E-6 m30 E-6 m0.04E-006 8E-006 1.2E-005 1.6E-005 2E-005Median volume droplet diameter, d [m]0.00 1020 30Wind speed, v [m/s]Collision efficiency, α11.00.80.60.40.20.0C10 m/s20 m/s30 m/sIcing intensity (kg/hr m)0.200.160.120.080.040.00Droplet concentration5 E+7 (m3)10 E+7 (m3)15 E+7 (m3)20 E+7 (m3)DWind speed = 10 m/sLWC = 0.0004 (kg/m3)0 0.04 0.08 0.12 0.16Cylinder diameter, D [m]0 0.01 0.02 0.03Cylinder diameter, D [m]0.04 0.05Figure 2.2. A. Collision efficiency, α 1 , as a function of MVD at different wind speeds.Cylinder diameter is 30mm. B. Collision efficiency, α 1, as a function of wind speed, v,at different MVD. Cylinder diameter is 30 mm. C. Collision efficiency, α 1, as afunction of cylinder diameter, D, at different wind speeds. MVD 13 µm. D. Icingintensity as a function of cylinder diameter, at different droplet concentrations. LWCis taken to be 0.4 g/m3, and wind speed to be 10 m/s. All calculations are based on theequations of Finstad et al. 1988a.The collision efficiency increases approximately linearly with increasingMVD and wind speed, while it has an exponential decrease with increasingcylinder diameter. More interesting is the variation in calculated icingintensity with varying cylinder diameter. Given a wind speed of 10 m/s, a8

Ts can be found iteratively by solving the equation of this heat balance, givenby Mazin et al. (2001). The heat budget over the cylinder surface has notbeen estimated in this study. For LWC >0 and air temperature below 0˚C, thegrowth process is not assumed to be affected by bouncing and runoff fromthe accretion surface. Further evaporation and sublimation is not taken intoconsideration. The main focus has been on estimating growth by in-cloudicing, correlated with collision efficiency, wind speed and LWC.2.3 Cloud liquid water contentA method for the calculation of the cloud liquid water content (LWC)is outlined in paper 2. Measurements of relative humidity and air temperatureat a known level(1) in unsaturated conditions are needed. Based upon theassumption that the total mixing ratio (liquid and vapour) of the air isconstant with height, the density, ρ lwc, of LWC (kg/m3) at level z (m a. s. l.)is given byρlwc⎛ e(z)= ε ⋅ ρ d⎜⎝ p11e(z)⎞−⎟p(z)⎠(2.6)where p is the air pressure, e is the water vapour pressure, and ρ d is thedensity of dry air. ε is the constant ratio of the molecular weight for watervapour and dry air, equal to 0.622.The dry adiabatic temperature gradient, Γ d , is, by definition, equal to g/c p ,where g is gravitational force, and c p is specific heat at constant pressure.This gives Γ d equal to 0.0098ºC/m. Measurements on the mountainBrosviksåta gave a measured temperature gradient for unsaturated conditionsequal to 0.0092 ºC/m. The measured temperature gradient for saturatedconditions was 0.0062 ºC/m, while the calculated pseudo-adiabatic lapse ratein the temperature range –10ºC to 0 ºC is 0.0067 ºC/m. The LWC increaseswith height, but with values slightly less than adiabatic. This was expectedwith regards to previous studies (Adapted from Schemenauer et al., 1980.)(figure 2.4).10

Figure 2.4. Maximum, average, and adiabatic liquid water content plotted against heightabove cloud base. (Adapted from Schemenauer et al., 1980.)2.4 Cloud droplet spectrumFinstad et al. (1988b) show that the median volume droplet size(MVD) is the most suitable parameter for estimating the icing intensity givenby equation 2.1. Knowledge of the cloud droplet spectrum is necessary inorder to calculate MVD. Measurements of MVD are not a part of routineweather observations, making reliable approximations necessary. However,given that LWC and droplet concentration are known, the mean volumedroplet size, D mv , is easily calculated by the equation13⎛ 6 V ⎞D mv= ⎜ ⎟(2.7)⎝πN ⎠11

where V is the total volume of cloud liquid water, and N is the dropletconcentration.In most cases the use of mean volume droplet size will lead to anunderestimate of the in-cloud icing intensity, due to the fact that the meanvolume droplet size is smaller than the MVD. Finstad et al. (1988b) comparesMVD and mean volume droplet size with collision efficiency, based ondroplet size data either measured in the field, in icing wind tunnels orestimated from parameterised size distribution. The data were collected from1957 to 1987 (table 2.2). A comparison of MVD with mean volume dropletsizes, based on their results, is presented in figure 2.5. This result gives amethod of estimating MVD based on mean volume droplet size. Applying alinear fit, MVD is a function of mean volume droplet diameter given by theequationMVD = 1.49⋅D mv+ 0.56(2.8)where D mv is mean volume droplet diameter.200Median volume droplet diameter (10 -6 m)160120804000 40 80 120 160 200Mean volume droplet diameter (10 -6 m)Figure 2.5. Mean volume droplet size plotted against median volume dropletsize. Empirical fit (solid line) and 1:1 line (dotted line) are also plotted. Thedroplet size data is either measured in the field, or in icing wind tunnels, orestimated from empirical size distributions, during the period 1957 to 1987.12

Table 2.2. Different measurements/estimates of cloud droplet spectra. (Adapted from Finstadet al. 1988b)13

In this case study, the droplet concentration was assumed to be constantthroughout a cloud layer. The cloud droplet number used is 113 cm -1 , givenby Gjessing and Skartveit (1990). The literature shows convincing evidencethat, at higher altitudes, the droplet spectrum shifts to larger sizes(Schemenauer et al., 1980, Nicholls, 1984, Noonkester, 1984). They showthat in the middle and upper portion of the cloud, the droplet concentrationdecreases with increasing altitude. However, the increase in droplet size dueto decreasing droplet concentration is small compared to the increase indroplet size due to increasing LWC with height. Therefore, we ignore thegrowth in droplet size by coalescence processes, and consider condensationas being the only process leading to the growth of droplets.A droplet that forms on a large condensation nucleus is initially seen to growat a faster rate than droplets with small nuclei, but after reaching a certainradius, the growth rates equal out, regardless of nuclear mass. Furthermore,the droplet radius, r, increases with time according to2r ( t)= r + 2ξ⋅t(2.9)0where r 0 is initial radius and ξ is given byξ = (S −1) /( F + )(2.10)kF dwhere S is ambient saturation ratio, F k is the thermodynamic term associatedwith heat conduction, and F d is the term associated with vapour diffusion.Parcels of dry air are mixed into the cloud layer by entrainment at the cloudtop. The exact nature of this process and the evolution of the droplet sizedistribution are still debated issues. The theory for estimating LWC presentedhere is, therefore, not assumed to be valid in this transition layer. The cloudtop is, therefore, assumed to be above the site of interest where we aremeasuring and estimating icing.14

2.4.1 Multicylinder measurementsThe multi-cylinder method has proven successful in measuring cloud dropletsize and liquid water content (Makkonen, 1992). This system contains of aset of cylinders of different diameters, which rotates at a frequency of 0.2 Hz(figure x). The dimension and weight of the accreted ice is measured after agiven time interval, which depends upon icing intensity. High icing intensityrequires a short measurement interval.The rotating multi-cylinder method was applied during the field experimentsbetween March 28 and April 2, 2003. The cylinders used had an initialdiameter of 1, 5, 10, 20, 50 and 80 mm. The weight and dimension of theaccretions on the separate cylinders was measured after an icing interval of20 minutes. The accuracy of the scale used for these measurements wasestimated to ± 0.1 g. A program given by Finstad (personal communication)developed to calculate the median volume droplet size (MVD) and LWCbased on these measurements was applied. The input parameters in the modelare dimension (length and width) of the ice accretion on each cylinder, thewind speed, the air temperature and the duration of the icing incident. Thereliability of the calculations by this method has been thoroughly tested andverified (Makkonen, 1992).Due to inaccuracies in the wind speed measurements, calculations were madefor varying wind speed at the different cases to illustrate the dependenceupon wind speed (Table 1 and 2). Further, the mean volume droplet size anddroplet concentration are estimated according eq. 2.7 and 2.8. Theseestimates indicate a higher droplet concentration than 113 (cm -3 ). However,the variation is relatively high, from 531 to 1770 (cm -3 ) in case 1, and from291 to 1290 in case 2. A droplet concentration of 113 (cm -3 ) is thereforepossibly to low. However, the concentration is kept constant equal 113 (cm -3 )in this study due to the high inaccuracies and variations in dropletconcentration according the multicylinder experiments (table 1 and 2). Afurther study on the droplet concentration is preferred to investigate thevariation in space and time of droplet concentration.15

Table 1. Case 1: Calculated LWC (kg/m 3 ) and MVD (m) by the multicylindermethod at Gaustatoppen March 29 2003.Wind LWC MVD Mean volume Droplet29.03.2003 speed (kg/m3) (m) droplet size (m) concentration(m/s)(cm-3)1200:1220 hrs1225:1245 hrs1245:1305 hrs1305:1325 hrs1325:1345 hrs1350:1410 hrs1410:14:30 hrs1435:14:55 hrs12 2,30E-04 8,40E-06 5,41E-06 7,41E+0815 1,80E-04 7,70E-06 4,92E-06 7,53E+0818 1,80E-04 5,80E-06 3,61E-06 1,76E+0912 3,20E-04 8,80E-06 5,68E-06 8,97E+0815 2,60E-04 7,80E-06 4,99E-06 1,05E+0918 2,20E-04 7,20E-06 4,58E-06 1,13E+0912 6,40E-04 9,60E-06 6,23E-06 1,38E+0915 5,10E-04 8,70E-06 5,61E-06 1,48E+0918 4,40E-04 7,80E-06 4,99E-06 1,77E+0912 3,60E-04 1,04E-05 6,79E-06 6,11E+0815 2,90E-04 9,30E-06 6,03E-06 6,89E+0818 2,50E-04 8,40E-06 5,41E-06 8,06E+0812 4,50E-04 8,70E-06 5,61E-06 1,31E+0915 3,70E-04 7,80E-06 4,99E-06 1,49E+0918 3,10E-04 7,20E-06 4,58E-06 1,59E+0912 3,50E-04 1,08E-05 7,06E-06 5,31E+0815 2,80E-04 9,70E-06 6,30E-06 5,86E+0818 2,30E-04 9,10E-06 5,89E-06 5,83E+0812 4,00E-04 9,50E-06 6,17E-06 8,91E+0815 3,20E-04 8,60E-06 5,54E-06 9,61E+0818 2,70E-04 7,90E-06 5,06E-06 1,05E+0912 2,40E-04 8,30E-06 5,34E-06 8,02E+0815 1,90E-04 7,50E-06 4,79E-06 8,60E+0818 1,60E-04 6,90E-06 4,37E-06 9,30E+0816

Table 2. Case 2: Calculated LWC (kg/m 3 ) and MVD (m) by the multicylindermethod at Gaustatoppen April 01 2003.Wind LWC MVD Mean volume Droplet01.04.2003 speed (kg/m3) (m) droplet size (m) concentration(m/s)(cm-3)1030:1050 hrs1055:1115 hrs1120:1140 hrs1145:1205 hrs210:1230 hrs1235:1255 hrs1300:1320 hrs13 5,80E-04 1,43E-05 9,48E-06 3,79E+0816 4,70E-04 1,30E-05 8,58E-06 4,09E+0819 3,90E-04 1,24E-05 8,17E-06 3,91E+0813 8,60E-04 1,14E-05 7,48E-06 1,11E+0916 7,00E-04 1,04E-05 6,79E-06 1,19E+0919 5,90E-04 9,60E-06 6,23E-06 1,27E+0913 5,60E-04 1,39E-05 9,20E-06 3,98E+0816 4,60E-04 1,27E-05 8,37E-06 4,29E+0819 3,80E-04 1,21E-05 7,96E-06 4,10E+0813 6,88E-04 1,20E-05 7,89E-06 7,60E+0816 5,50E-04 1,10E-05 7,20E-06 7,89E+0819 4,60E-04 1,03E-05 6,72E-06 8,04E+0813 4,90E-04 1,35E-05 8,92E-06 3,80E+0816 4,00E-04 1,23E-05 8,10E-06 4,11E+0819 3,40E-04 1,13E-05 7,41E-06 4,50E+0813 5,10E-04 1,37E-05 9,06E-06 3,79E+0816 4,20E-04 1,24E-05 8,17E-06 4,21E+0819 3,50E-04 1,17E-05 7,68E-06 4,17E+0813 4,80E-04 1,32E-05 8,72E-06 3,99E+0816 3,80E-04 1,30E-05 8,58E-06 3,30E+0819 3,20E-04 1,28E-05 8,44E-06 2,91E+0817

3. The experiments3.1 Field siteThe field data presented in this study was collected at two different mountainsites in Norway. The two sites were Brosviksåta (723 m a.s.l., 61˚ 2` N, 5˚9`E) and Gaustatoppen (1883 m a. s. l. 59˚ 51`N, 8˚ 39`E). Brosviksåta issituated on the western coast, while Gaustatoppen is situated inland in thesouthern part of Norway (figure 3.1). Daily weather conditions along theNorwegian coast are primarily dominated by large-scale synoptic systemsmoving in from the west. Such systems result in several periods of icingduring the winter months. Brosviksåta experiences rather short periods oficing (days) both as precipitation icing and in-cloud icing, while the height ofGaustatoppen is favourable for long periods (weeks) of extreme in-cloudicing. In addition to the well-suited climatic conditions, the infrastructure atboth sites makes them ideal for experimental activity. Brosviksåta has a roadall the way to the top, while Gaustatoppen has an elevator inside themountain from 1150 to 1800 m a.s.l. Nevertheless, a key factor for thesuccess of these experimental studies has been the excellent support given bythe operating personnel at both sites.The main purpose of the experiments was to collect full-scale data of themeteorological parameters relevant for atmospheric icing and, at the sametime, measure the icing intensity with suitable equipment. Air temperature,relative humidity, wind speed and wind direction have been measured atdifferent levels along the mountain slope. At the same time, the icingintensity was measured at the mountain peak by an ice scale (Figure 3.2). Allequipment made samplings at 10-minute intervals.19

Figure 3.1. Location of the mountains Gaustatoppen (59º51´N, 08º39´E)and Brosviksåta (723 m a.s.l., 61º 2` N, 5º 9` E),20

Figure 3.2a. Schematic drawing of the measurement set-up at Brosviksåta (723 ma.s.l.). Two weather stations were placed at 718 m a.s.l.,one at 520 and one at 325m a.s.l. The ice scale was mounted on the roof of a building at the top of themountain. The station Takle is situated 12 km east of the base of the mountain.Figure 3.2b. Schematic drawing of the measurement set-up at Gaustatoppen(1882 m a.s.l.). The weather stations were placed at 1811, 1540, 1298 and 1160 ma.s.l. The ice scale was mounted 3 m above the terrain at 1800 m a.s.l.. Thesynoptic weather station Møsstrand is situated 31 km southwest of the mountain.21

3.2 Meteorological instrumentationDipl. Ing Houm and Aanderaa Instruments, Norway, manufacturedthe meteorological instruments used in this study. The sensors wereintegrated in a system consisting of sensors, a sensor-scanning unit, a powersupply and a data storage unit. The system has been extensively tested overseveral years, giving satisfactory results for non-icing conditions. Theperformances of the wind velocity and wind direction sensors were testedFigure 3.3. Iced Aandera weather station at Brosviksåta March 23, 2004 (left), andiced Gill Windobserver II at Gaustatoppen February 26 2002 (right) (photo: Tor deagainst other sensor types by Aasen (1995). Each weather station along theslope of the mountain consisted of a range of these sensors. A heated acousticanemometer (from Gill Instruments) was mounted together with the ice scaleat the mountain peak. All instrumentation for measuring atmospheric icing isdescribed in detail in paper 1 (Drage and de Lange, 2005).The wind speed and wind direction sensors from Aanderaa Instruments werenot suitable for operation under icing conditions. The acoustic anemometerhas a heating element to keep the senor arms free of ice. This heating provedto be insufficient during heavy icing conditions (figure 3.3). A procedure forcalculating the wind speed at the site of interest, based on measurements at alower level and by use of operational model data (HIRLAM10 (HighResolution Medium range Weather Forecast), ECMWF (European Centre ofMedium range Weather Forecasts), was therefore more appropriate.22

3.3 Ice scaleFor measuring atmospheric icing, a system was constructed based onthe requirements outlined in ISO 12494 (2001). The entire measurementsystem is described in detail in paper 1. The ice scale system consists of thefollowing main components: ice scale, data logger, power-supply andconverter units.23

4. Papers4.1 Specific objectives of the four individual studiesThe specific objectives of the four individual studies are presented below,followed by the full version of the papers.Paper I: Instrumentation for Measuring Atmospheric IcingThe paper is concerned with the design, construction and testing ofinstrumentation for measuring atmospheric icing. An ice scale, to measureicing, is constructed according to the requirements and specifications outlinedin ISO 12494 (2001). The ice scale consists of a vertical steel rod with alength of one meter exposed to atmospheric icing. A load cell records thevertical and horizontal load at 10-minute intervals. The equipment is testedand calibrated under controlled conditions in a laboratory.Paper II: Atmospheric Icing in a Coastal Mountainous Terrain.Measurements and Numerical Simulations, a Case StudyDevelopment of a method for estimating vertical gradient of cloud liquidwater content (LWC), from air temperature and humidity at unsaturatedconditions (in respect to water vapour) below the cloud base. A case study oficing is evaluated, where measured icing intensity is compared to estimatesfrom weather station data and numerical simulations. The height of the cloudbase is estimated, and temperature at the site of interest inside the cloud isfound by following an unsaturated gradient below the cloud base, and asaturated gradient inside the cloud. The total mixing ratio (vapour and liquid)is assumed constant throughout the cloud layer.Paper III: Large Scale Measurements and Numerical Simulations of IncloudIcing Around the Ridge of a Mountain PeakA study of local icing variations in the surface layer around the edge of amountain ridge has been carried out. Icing is measured in-situ on 16 stickswith a height of 2 meters and a diameter of 30 mm, for measurements of thelocal variation in icing around the south-eastern part of the ridge atGaustatoppen (1883 m a.s.l.), Norway. The method outlined in paper II isused for calculating variation in LWC with height. Furthermore, a numericalmodel, “Flow 3d”, is used for estimating wind speeds 2 m above the ground.A comparison of icing intensity by in-situ measurements and numericalsimulations is performed.25

Paper IV: Arctic Coastal Climatic Impact on Design Construction andOperation of the Hammerfest LNG PlantAn approach for the evaluation of climatic aspects in relation to design,construction and operation of a plant in an artic coastal area is presented. Thefocus on exploitation of oil and gas resources in the artic sea and coastalregions requires attention to be paid to climatic aspects. Events such as heavysea spray icing and snowdrift must be considered. Fast developing,unpredicted polar lows with high wind speeds and heavy precipitation isanother design criterion. Design, construction and operation of the LNG(liquid natural gas) plant on Melkøya, Hammerfest, Norway, with respect toclimatic conditions are presented in this paper.26

GEOPHYSICAL INSTITUTE - UNIVERSITY OF BERGEN - NORWAYREPORTS IN METEOROLOGY AND OCEANOGRAPHY2 - 2005Magne A. Drage and Tor de LangeINSTRUMENTATION FOR MEASURING ATMOSPHERICICING27

Reports in Meteorology and Oceanography,University of Bergen.Report No. 2-2005ISSN 1502-5519ISBN 82-8116-004-728

1. IntroductionIn regions with severe climatic conditions atmospheric icing on structures is aserious problem. The general effects of icing are increased vertical loads onstructures as well as increased wind drag caused by an increased areaexposed to the wind, leading to a severe increase in wind load. Planning andbuilding requires specifications of expected climatic scenarios. The processof ice build-up on the surface of an object is described as accretion. Accretionresults in the different types of icing on structures. This accretion is afunction of meteorological parameters such as air temperature, wind speed,cloud liquid water content, cloud droplet spectra, etc.Atmospheric icing is traditionally classified according to three differentformation processes:1) precipitation icing:a. freezing rain or drizzleb. accumulation of wet snow2) in-cloud icing which consists of super-cooled water droplets in acloud or fog.3) hoar frost/sublimation. Direct phase transition from water vapour intoice. Hoar frost is of low density and strength, and normally does notresult in significant load on structures.Ice accretion also depends on the properties of the accreting object itself,described by its shape, size, orientation relative to the wind and material, aswell as the surface structure. Measurements of ice accretions must thereforebe specified with respect to devices, procedures, arrangements on site, etc.The rate of accretion by dry growth onto an object by in-cloud icing is givenby the equationdMdt= α 1 ⋅ q LWC ⋅ A⋅v[Kg s -1 ] (1)where q LWC is the liquid water content of the air, flowing with the wind speedvelocity V, towards the cross-sectional area A, of an object. The efficiencycoefficient α 1 represents the collision efficiency. It is the fraction of the totalnumber of droplets in the path of the object that actually collide with thatobject. Small droplets, large cross sections and low wind speeds reduce α 1 .Early results of icing on structures used in continental measurements wereobtained by manual registration of ice thickness and weight of the iceaccretion, or sometimes simply by visual estimates. Few observations and a29

high degree of uncertainty in the existing observations was not satisfactoryfor scientific purposes. Due to this, new techniques of measuring atmosphericicing were developed (Poots, 2000).The early instruments were primarily cylindrical objects such as steelcylinders and horizontal wires. The scientists in the former USSR developeda system of horizontal wires (Nikforov, 1983), while the Europeans mainlyused “the Grunow net” (Grunow and Tollner, 1969). The Grunow net is atube of wire netting which was often installed on the top of precipitationgauges. It recorded water run-off from melting accreted ice. A well knownmethod of estimating the cloud liquid water content and the median volumediameter, is by the rotating multicylinder method (Makkonen and Stallabras,1987, Finstad et al., 1988c).The international standard, ISO 12494, 2001, Atmospheric icing ofstructures, gives recommendations of standardisation of measurements for iceactions on structures. The standard device is a cylinder with the diameter of30 mm which slowly rotates around a vertical axis. The cylinder length is 1m, and should be placed 10 m above the surface. Among the differentvariables, such as density and dimension, which can be measured, the iceload is considered most important. The output from the measurement seriesshould form the basis for extreme value analysis, ranging from a few years toseveral decades. Shorter time series connected to longer series ofmeteorological data, statistically or physically in combination withtheoretical models, will provide important information. Wind measurementsand detailed load recordings of reactions in all directions, vertical andhorizontal, will give the basis for estimating the drag coefficient, C D bycalculation.2. Description of the ice scale systemA system was constructed based on the requirements and specificationsoutlined in the introduction. The entire measurement system is shown in theblock diagram in Figure 2.1. It consists of the following main components;ice scale, temperature and humidity sensor, ultrasonic anemometer, datalogger, and power supply / converter units.Each instrument is described briefly before attention is turned to the ice scaleitself.30

Figure 2.1 Block diagram of the ice scale system.Table 2.1 Measurement ranges and accuracies of the sensors in the icescale system.Parameter Range AccuracyTemperature -40 to 60°C 0.3 °C at 23 °CRelative humidity 0 to 100 % RH 1.5 % at 23 °CWind speed 0 to 70 m/s 2 % at 12 m/sWind direction 0 to 360 degrees 2 % for WS

Temperature and humidity sensorThe Rotronic MP408A sensor is a combined platinum PT100 temperaturesensor and a capacitive relative humidity sensor. It is housed within aradiation screen to minimize the error due to solar heating. The sensors arecalibrated to output a current 4 to 20 mA proportional to the measuringranges. See Table 2.1.Ultrasonic anemometerThe Gill Instruments Wind Observer II is an ultrasonic anemometer (GillInstruments, 2000). The instrument is designed to measure horizontal windspeed along two fixed orthogonal axes by transmitting and receiving sonicsignals. Sampling is done every 25mins, each axis being sampledsequentially. The analogue output isupdated every 1 s, based on an averageof 39 measurements. Wind speed anddirection is calculated based on thetwo wind vector data.Figure 2.2 Gill wind sensor.Data loggerIn order to prevent icing on the windsensor, there are heating elementsinside the probes. The effect of theheating element is 72 Watt. Theseproved to be insufficient in preventingicing under extreme conditions in theareas where the ice scale system wasused. Additional heating elements of90 Watt were attached externally to thesensor housing, with the same result.The data logging system consists of 3, optional 4, Opus 200i loggersproduced by Lufft (Lufft Opus, 2000). The Opus 200i is a universal 2-channel data logger. Each input channel can be configured for eitherresistance, current, voltage, or frequency input. Several loggers may be32

connected in series via a CAN-bus to facilitate additional input channels. Foran ice scale system, either 3 or 4 Opus 200is are required depending onhorizontal forces are measured or not.Each input can operate an actor output channel configured as either a switchor current loop signal. This feature is currently not used for the ice scalesystem, but could in the future be used to control a de-icer when a certainweight is reached.The data logger sampling interval can be set individually for each inputchannel, but the periods are limited to 0.1 s, 1 s, 10 s, 30 s, and 60 s.Storage intervals are also individually selectable, but limited to 0.1 s, 1 s, 10s, 30 s, and 1 minute to 1440 minutes. Stored values can be either average,minimum, or maximum values, or any combination.Data is stored in the internal memory which has a capacity of 30000 valuesfor each logger.Programming and data retrieval is accomplished via a computer with aprogram called SmartControl 1.0 connected to the data logger serialconnector. By connecting a modem to the serial connector remote datadownload is possible via either telephone line or GSM mobile net.The configuration of the Opus data loggers for the ice scale system isoutlined in Table 2.2.Table 2.2system.Configuration of the Opus 200i data loggers in the ice scaleParameter Input Sampling Storage Storageint. int. optionTemperature current 4 to 20 mA 1 s 10 minmeanRelative humidity current 4 to 20 mA 1 s 10 min meanWind speed current 4 to 20 mA 1 s 10 min meanWind direction current 4 to 20 mA 1 s 10 min meanBattery voltage current 4 to 20 mA 1 s 10 min mean & maxIce weight voltage 0 to 20 mV 1 s 10 min mean & maxHorizontal force, x- voltage -40 to 40 mV 1 s 10 min mean & maxdirectionHorizontal force, y- voltage - 40 to 40 1 s 10 min mean & maxdirectionmV33

Power supply and converter unitsThe ice scale system is designed for operation from mains power supply.Several units have a power consumption that makes battery operationdifficult. Heating of the wind sensor is especially power consuming. Thepower converters are included to ensure that the different units are suppliedwith adequate voltages.3. The ice scaleThe construction of the ice scale is based on the recommendations from ISO12494. The ice scale consists of a vertical steel rod with a length of one meterexposed to atmospheric icing (figure 3.1).Figure3.1 Ice scale34

Three load cells, Global Weighing Technologies type MP41/12C3, are placedinside the box to record all reactions on the steel rod. One load cell is locatedon the floor of the box, recording the vertical load. Two load cells are placedperpendicular to each other on the walls of the box, recording horizontal loadand load direction. The solid fixed steel rod which stands on the vertical loadcell has a total length of 1.35 m and a diameter of 15 mm. The rod can freelybe displaced in the vertical direction, but is fastened to the wall to preventdisplacements in the horizontal directions. The horizontal load cells areplaced 20 cm up the shaft of the solid steel rod. This rod passes through thetop of the box through a 30 mm diameter hole. The rod is formed with asharp pointed top. For the recording of ice loads on a stationary construction,a non rotating cylinder should be used for reference measurements accordingto the ISO 12494 standard.For standard reference measurements, a 30 mm diameter hollow steel rod isfed over the outside of the solid rod. Within in the top of the hollow rod is acone enclosure which allows the rod to rotate smoothly, see Figure 3.1.Further a cylindrical enclosure at the bottom of the rotating rod preventshorizontal displacement. During episodes of icing the rod rotates due to themechanical forces of ice and wind action. Free rotation means the rod willturn until minimum drag is achieved. The angle where minimum drag isachieved therefore experiences an increase in wind drag around it. Severalfield observations of icing on the cylinder confirm this theory, showing acylindrical ice accretion on the cylinder (figure 3.2).Testing the vertical load (ice weight)To test the vertical load cell a mechanical collar was constructed. This wasattached to the lower part of the vertical steel rod. One or more high precisioncalibrated loads were placed on this collar to load the vertical load cell, seeFigure 3.3. The testing was performed in an indoor laboratory at atemperature of approximately 21 °C.Output from the load cell is a functionof applied load and the supply voltageto the load cell. The load cell was therefore powered by a laboratory powersupply (Iso-Tech model IPS 2303DD) at several different voltages. Outputfrom the load cell was measured by a digital voltmeter (Keithley 196 systemDMM).35

Figure 3.2. Ice scale with rotating 30 mm cylinder after icing incident at1800 m a.s.l. at Mt. Gaustatoppen, 0925hrs April 2, 2003.Several measurements at each load and supply voltage were done tominimize errors. The average values are presented in Table 3.1. Figures inthe left column are the weights of the free weights used. The steel rod,cylinder, collar, and hinges are partly measured and partly calculated to atotal weight of 7.07 kg.36

Figure 3.3 Testing the Ice scale, vertical load.Table 3.1 Measured output signal [mV] as function of load for differentsupply voltages.Weight [kg] V S = 12.00 V V S = 13.00 V V S = 14.00 V V S = 15.00 V0 0.1831 0.1935 0.2005 0.2103Rod, cyl., 1.9100 2.0521 2.2031 2.3655collar10 kg 4.3597 4.6988 5.0412 5.432615 kg 5.5850 6.0174 6.4793 6.964825 kg 8.0398 8.6642 9.3246 10.032640 kg 11.7160 12.6642 13.5816 14.615350 kg 14.1686 15.3205 16.4278 17.665160 kg 16.6180 17.9597 19.2746 20.723175 kg 20.2898 21.9388 23.5116 25.292990 kg 23.9279 25.8960 27.7359 29.8818100 kg 26.3676 28.5188 30.5346 32.836937

Conducting a linear regression of the measured data results in the coefficientsfor four lines. One line for each supply voltage. These lines were used toestimate the mass of the rod, cylinder, collar, and hinges. The results were7.10, 7.03, 7.15, and 7.16 kg respectively, giving an average of 7.11 kg whilethe measured weight was 7.07 kg. Figure 3.4 shows a plot of themeasurements, while Figure 3.5 shows the measurements deviation from thestraight line.From these results we can calculate the calibration coefficient (named RatedOutput in the data sheet) for each measurement. These are summarized inTable 3.2. Average coefficient is 2.0358 [mV/V] with a standard deviation of0.0052.Table 3.2 Calibration coefficients [mV/V] calculated for different loads andsupply voltagesWeight [kg] V S = 12.00 V V S = 13.00 V V S = 14.00 V V S = 15.00 V10 kg 2.0390 2.0302 2.0256 2.039615 kg 2.0397 2.0299 2.0321 2.040325 kg 2.0415 2.0318 2.0322 2.041840 kg 2.0418 2.0380 2.0306 2.040250 kg 2.0422 2.0389 2.0310 2.039060 kg 2.0420 2.0376 2.0314 2.038975 kg 2.0416 2.0382 2.0288 2.037590 kg 2.0385 2.0368 2.0262 2.0378100 kg 2.0380 2.0350 2.0236 2.0315Average 2.0405 2.0352 2.0291 2.0385Standard dev. 0.0017 0.0036 0.0032 0.0029The operating range of the load cell is 0 to 100 kg. However, the mass of therod, cylinder and collar is 7.07 kg, which means that values of more than 93kg, as shown in figure 3.5, are outside the specified range of the load cell. Inthe laboratory all tests within the specified range had a maximum error of ±0.125 kg.38

Figure 3.4 Measured output signal as a function of applied load(exclusive rod, cylinder, and collar) for different supply voltages.Figure 3.5 Test values deviation from linear fit.39

To verify this, data from Brosviksåta between March 9 and June 6, 2004 wasexamined. In this period, the ice scale was fitted with a fixed cylinder ofdiameter 14 cm. The total mass of the rod, cylinder etc was measured to be19.5 kg.Within this period, data was extracted when the air temperature was above2.0 °C. This was done to ensure that there was no ice on the ice scale. Themean output from the ice scale was calculated to be 19.509 kg with astandard deviation of 0.027 kg. Maximum output from these data was 19.584kg and minimum output was 19.391 kg. The extracted data set consisted of5572 10 minutes average values.Testing the horizontal forcesTesting of the horizontal forces was done according to Figure 3.6, and asdescribed below.A rope was fastened to the massive steel rod at four different heights. Therope was led via a low friction wheel at a right angle to a free hanging loadwith a mass of 10 kg. The load was connected at different heights along therod and the rope was, at all times, ensured to be perpendicular to the rod. Itwas thereby possible to simulate different horizontal forces to the load cell.Rotating the ice weight allowed all four axes to be tested (x-direction, y-direction, press and strain).Applied force to the horizontal load cells was calculated by mechanicalmomentum equations. Output signal from the load cell and supply voltagewas recorded. It was noted that for the horizontal forces there was asubstantial hysteresis in the output signal. This was most likely caused by themechanical construction of the ice scale and bending of the rod. Based onthese observations, it was decided that two recordings would be made foreach applied force. One from increasing loads and one from decreasing loads.This was supposed to be an indication of the ‘worst case’ values, giving avalue of the uncertainty of the measurement of the horizontal forces. Refer toTable 3.3 for measured data. Note that the distance indicated in the leftcolumn is from the top of the rod.40

Figure 3.6 Testing the Ice scale, horizontal forces.Table 3.3 Measured output signal [mV] as a function of applied force.Supply voltage is 12V.Distance X-dir. Y-dir.fromtop[cm]Press,incr.Press,decreas.Strain,incr.Strain,decreas.Press,incr.Press,dcreas.Strain,Incr.Strain,dcreas.100 3.57 3.77 3.50 3.63 3.45 3.75 3.52 3.7675 6.29 6.76 6.22 6.51 6.33 6.80 6.26 6.6850 9.11 9.67 9.02 9.67 9.20 9.91 9.15 9.7025 11.96 12.58 11.79 12.69 12.13 12.52 11.97 12.5641

For the x-direction, the data is plotted in Figure 3.7. Four lines in the plotindicate the data listed in Table 3.3. The fifth line is a linear fit of all thesedata. The deviation of the measured data with respect to the linear fit isplotted in Figure 3.8. The deviation is recalculated to force/gravity.Corresponding data for the y-direction is plotted in figure 3.9 and 3.10. Injust the same way as for the vertical load cells, these data can then be used tocalculate the calibration coefficients (RO) for the horizontal load cells. Thisis indicated in Table 3.4. Average RO for the x-direction is 1.951 [mV/V],standard deviation is 0.063. For the y-direction corresponding values are1.961 [mV/V] and 0.068.Table 3.4 Calibration coefficients [mV/V] calculated for different loads.Distance X-dir. Y-dir.fromtop[cm]Press,incr.Press,decreas.Strain,incr.Strain,decreas.Press,incr.Press,dcreas.42Strain,Incr.Strain,dcreas.100 1.894 1.987 1.863 2.005 1.916 1.978 1.891 1.98475 1.942 2.002 1.867 2.002 1.905 2.052 1.894 2.00850 1.889 2.030 1.868 1.955 1.901 2.042 1.880 2.00625 1.951 2.060 1.913 1.984 1.885 2.049 1.923 2.055All these tests were conducted in an indoor laboratory at room temperaturearound 21 °C. The supply voltage was 12.0 V.Tests with different supply voltages (12.0 V, 13.0 V, 14.0 V, and 15.0 V) forapplied force at one height and different angles (x- and y-direction) showedthat the error from variable voltage was negligible.From Figures 3.8 and 3.10, it can be seen that the deviation in the data for thehorizontal force is one order of magnitude greater than that of the verticalload. As the three load cells are of the same type, the deviation is believed tobe due to the mechanical construction of the ice scale and the bending of therod.According to the datasheet, the load cells are temperature compensated in therange -10 to +40 °C. No facilities were available to test this.To be able to compensate for varying supply voltage, a voltmeter wasincluded inside the box of the ice scale. Output from this voltmeter was acurrent 4 to 20 mA, representing a voltage 0 to 15 V. This current was loggedby the logging system, thus allowing voltage variations to be compensatedfor.

Figure 3.7 Measured output signal from x-direction horizontalload cell for different forces.Figure 3.8 Deviation from linear fit.43

Figure 3.9 Measured output signal from y-direction horizontal load cellfor different forces.Figure 3.10 Deviation from linear fit.44

4. Application of the Ice scale system.An ice scale system was installed on top of a building on mount Brosviksåta(723 m) on the western coast of Norway. This was run during the 2002/2003winter, using a 14 cm cylinder on the ice scale. The period from March 20until March 25, 2003 was selected to illustrate the features of the ice scalesystem. Data from this period is plotted in Figure 4.1.On the morning of March 20, the wind direction turned from north-east tosouth. This was accompanied by decreasing temperature and increasingrelative humidity. From noon on March 21, ice accretion started to build upon the cylinder, reaching a maximum of 4.5 kg in the morning of March 23.The weight indicated in Figure 4.1 is the net weight of the accreted ice only.These data are further described in Drage and Hauge (2004).45

Figure 4.1 Data from Ice scale system from Brosviksåta March 20 to 26,2003.46

5. ConclusionA total of three prototype ice scales and two complete systems have beenbuilt. They have been run for, all together, 34 months at different locations,both in mountainous regions in southern Norway and a coastal area in thenorth of the country. The ice scale systems have proved to provide useful andreliable data for testing and verifying icing models.The mass of the ice scale from the field measurements, with no load, wascalculated to be 19.509 kg. This showed a deviation of 0.009 kg from whatwas measured in the laboratory. The standard deviation, as calculated in thelaboratory, was 0.027 kg.For horizontal force tests of the ice scale indicated that ice weight can bemeasured with an absolute accuracy of approximately 0.125 kg.The mean output from tests in the laboratory suggests an absolute accuracybetween 20 N and 500 N for a horizontal force.Based on these experiences, a few improvements can be suggested. The icescale itself is quite heavy, approximately 60 kg. It should be possible toconstruct a box with sufficient strength and a substantially lower weight. Itcould be useful to add a de-icer to the ice scale. This could be controlled bythe data logger and all ice should be removed when a certain weight isreached. In severe icing conditions, the de-icer attached to the wind sensorshave proved inadequate. A different wind sensor or better de-icer should beconsidered. Choosing a data logger with more inputs would eliminate theneed for three or four smaller loggers, thus simplifying the system.47

6. ReferencesGill Instruments – WindObserver II Ultrasonic Anemometer User Manual,1390-PS-0004 Issue 05, November 2000.Grunow, J. and Tollner H. (1969). Fog deposition in high mountains. Archivfur meteorologie, geophysik und bioklimatologie, Ser. B, Vol. 17, pp. 201-218 (in German).GWT Global Weighing Technologies GmbH – Load cell Type MP41/12C3calibration certificate.Finstad, K. J., Lozowzki, E. P. and Makkonen, L., 1988c. On the medianvolume diameter approximation for droplet collision efficiency. J. Atmos.Sci. Vol 45. 4008 – 4012.ISO – ISO 12494 – Atmospheric icing of structures, First edition 2001-08-15.Lufft – Opus 200/300 Version 12/2000 Hardware Standard.Lufft – SmartControl 1.0 User Guide, March 2000.Makkonen, L. and Stallabras, J. R., 1987. Experiments on the cloud dropletcollision efficiency of cylinders, J. of Clim. Appl. Meteor., 26: 1406 –1411.Nikiforov, E. P. (1983). Icing related problems, effect of line design and iceload mapping. Proceeding of first international workshop on atmophericicing of structures. U. S. A. Cold region reserch and engineering laboratory,Special report 83-17, pp. 239-245.Observator Instruments BV – OMC-4xx Temperature sensors / radiationscreenPoots, G., 2000. Ice and snow accretion on structures. Phil. Trans. R. Soc.Lond. Series A, Vol. 358, 2799 – 3033.www.rotronic.co.uk – MP100/400.48

ATMOSPHERIC ICING IN A COASTALMOUNTAINOUS TERRAIN.MEASUREMENTS AND NUMERICALSIMULATIONS, A CASE STUDYCold Regions Science and Technology, Accepted subject to revisionMagne A. DrageGeophysical Institute, University of Bergen/The University Center onSvalbard, NorwayGard HaugeStorm Weather Center ASBergen, NorwayABSTRACT: Icing on structures occurs as rime ice, clear ice or wet snowdeposit. Reliable forecasts of duration and intensity of this icing requiresprognoses of standard meteorological parameters, in addition to more specificparameters such as the density (ρ LWC ) of cloud Liquid Water Content (LWC).Icing conditions on the mountain Brosviksåta (723 m a.s.l., 61º 2` N, 5º 9` E),on the western coast of Norway were investigated from March 21-24, 2003.A nonrotating vertical steel rod mounted on a scale was used to measure theaccumulated ice load. Air temperature, relative humidity and wind weremeasured at three levels along the mountain slope. The maximum build-up ofice, in this case study, was measured to 4.5 kg on a 1 m high 0.14 m diameterrod. Comparison of measured ice-growth rate and calculated ρ LWC gave acorrelation coefficient of 0.85.A mesoscale atmospheric model (MM5) has also been tested at a highhorizontal resolution (1km) in order to evaluate its ability to reproduceweather conditions where freezing occurs. Comparison from the directmeasurements and calculations, with results from MM5, gave 58% of themeasured accumulated ice growth. Further studies of real-time cases on areal-time system at a coarser model resolution will reveal its capability forforcasting freezing events.49

Keywords: Atmospheric icing; Liquid water content; Ice load; Simulations1. IntroductionAtmospheric icing has severe economical and technological consequencesfor human activities. It occurs frequently in sub arctic and artic climates aswell as exposed locations at a certain height above sea level. Convincingevidence of its effects comes from the long list of human activities that haveoccasionally been disrupted, such as aircraft operations, telecommunicationnetworks, power transmission lines, roads and railways (Poots, 2000). Icingdirectly onto structures occurs as rime ice, clear ice or wet snow deposit.Reliable icing forecasts require meteorological data of standard parameterssuch as air temperature, relative humidity, wind speed, wind direction andturbulence, in addition to more specific parameters such as median volumedroplet size and liquid water content of the air masses concerned.Daily weather conditions along the Norwegian coast are primarily dominatedby large scale synoptical systems moving in from the west. Large variationsin weather conditions occur over just a few kilometers due to the fjords andcoastal mountains creating a complex topography that directly influencesatmospheric circulation patterns. To give a realistic reproduction of weatherconditions during icing events, a weather prediction model of high horizontaland vertical resolution is necessary. The mesoscale model MM5 version3.6.3 applied in this study (hereafter MM5) is thoroughly tested under suchconditions and is assumed to give a realistic representation of theatmospheric conditions during freezing events. MM5 is the most widely usedatmospheric research model in the world. The development of this modelingsystem is continuous and improvements are made on a monthly basis. Thequality of the MM5 forecasts has, through many studies of real-timeforecasting systems, shown to be as reliable as other atmospheric models inmost situations (Grell et al., 1994).Several methods and physical models exist for estimating the flow patternaround constructions as well as the shape and formation of ice upon them.These methods and models have been mostly concerned with cylinders(Chaine and Skeates, 1974, Makkonen, 1984, 1996, Haldar et al., 1996). Inthe past, a number of experiments have been conducted to establish arelationship between atmospheric icing and measured standardmeteorological parameters (e.g. Ahti and Makkonen, 1982, Makkonen andAhti, 1995, Lott and Jones, 1998, Sundin and Makkonen, 1998). Thelimitations of these methods are that standard (routinely observed)50

meteorological parameters are necessary but have shown not to be sufficientto describe the icing process. Here, a new attempt of establishing arelationship between atmospheric icing and routinely measuredmeteorological data has been performed.The first attempt of applying a numerical boundary-layer model to predictLWC for icing calculations was made by Vassbø et al. (1998). They used theHIRLAM (High Resolution Limited Area Model) over Finnish topography.This was however a study which used very coarse horizontal and, not least,vertical resolution. The horizontal resolution was 22 km by 5.5 km. Themodel had only 3 layers below 627 m, which reproduced the planetaryboundary layer poorly.The improvement of mesoscale weather prediction models of high horizontaland vertical resolution opens for the possibility of being able to forcast icing.It was important to establish a relationship between actual in-cloud icing andthe output from a real-time mesoscale weather prediction model. Thefollowing task was therefore undertaken: the results from a real-time icemonitoringsystem and the measurements of standard meteoroligicalparameters at Brosviksåta (723 m a.s.l.) were compared with the MM5numerical simulations of the weather conditions. In this study, MM5 applied38 layers in the vertical, where 17 of them were placed in the lowestkilometer above the ground in order to reproduce realistic results in theboundary layer. Icing was measured using an ice scale mounted at themountain peak, in addition to 4 weather stations at 3 different levels leadingdown to 325 m a.s.l.. All stations measured air temperature, relativehumidity, wind speed and wind direction. Data from a synoptic weatherstation located in the nearby area were also included in the measurementdataset.2. Study site and data2.1 Measurement setupThe mountain Brosviksåta (723 m a.s.l., 61º 2` N, 5º 9` E, see also figure 1)is situated to the south of the Sognefjorden outlet on the western coast ofNorway. The mountain faces open sea within the sector SSW to NNW. Thisregion is highly affected by the passage of frontal systems moving up theNorth Atlantic Ocean. Air masses related to these frontal systems are oftenhumid and have air temperatures favorable for atmospheric icing (-15 < TºC< -1) during the winter season (November-April).51

Figure 1: Left figure shows the 3 nested MM5 domains within the European region withthe Atlantic sea on the western boundary of the 9 km domain. The figure on the righthand side shows the 1 km MM5 topography along with the coastline. Brosviksåta isindicated on the figure with an X.The measurement setup consisted of four weather stations (two at 718 andone each at 520 and 325 m a.s.l.) and one ice scale (733 m a.s.l.). Theweather stations were of the conventional type (Aanderaa Instruments, seehttp://www.aanderaa.com), consisting of a rotating cup anemometer for windspeed and a typical wind vane sensor for wind direction measurements. Thisequipment is not designed to operate during icing conditions. Nevertheless,an icing period could be detected from the time when the anemometerstopped rotating and the wind vane remained in a fixed position.Measurements from the site showed that the ice accumulation started at 1100hrs March 21, while the two anemometers at the mountain top stoppedrotating at 1750 hrs and 1820 hrs, respectively. In addition, air temperatureand relative humidity were measured at all levels. The accuracy of the airtemperature and relative humidity measurements are ±0.1C° and ±2%,respectively. The observation interval for all the weather stations was 10minutes. In addition, precipitation data from the synoptic station Takle (38 ma.s.l.), situated 12 km east of the mountain base, are included. This station isoperated by the Norwegian Meteorological Institute.Icing was measured by an ice scale, consisting of a 1 m high verticalnonrotating steel rod of 14 cm diameter, measuring ice loads of up to 150 kg.A laboratory calibration test of the ice scale system gave a precision andstandard deviation of 0.1 kg and 0.027 kg, respectively (Drage and de Lange,52

2004). This scale, with a logging interval of 10 minutes, was mounted 10meters above the ground on the roof of a building (figure 2). The effect of thebuilding on the ice scale was estimated to be negligible. Using a two monthrecord at Mt. Brosviksåta with no icing, the precision of the instrument wasfound to be within the range of the laboratory test. During that period thewind speed reached more than 20 m/s.Figure 2. Ice scale mounted at Mt. Brosviksåta. The length of the cylindrical rod is 1 m andthe diameter is 14 cm.53

2.2 Weather prediction modelThe atmospheric mesoscale model MM5 was developed by PSU(Pennsylvania State University) and NCAR (National Centre forAtmospheric Research). It is a mesoscale modeling system that includesadvanced atmospheric physics. It is widely used for real-time weatherforecasts, air quality investigations and hydrological studies (Warner et al.,1991, 1998, Grell et al., 1994, Mass and Kuo, 1998, Chatfield et al., 1999,Chang et al., 2000, Mass et al., 2002). MM5 is based upon a set of equationsfor a fully compressible and non-hydrostatic atmosphere. Consequently it ispossible to run the model at fine horizontal and vertical scales correspondingto the meso γ-scale (O(1)km).Initial and lateral boundary conditions were obtained from the EuropeanCentre for Medium-Range Weather Forecasting (ECMWF) model with 0.5degree distance between the grid points, and nested down to 9 km, 3 km andfinally 1 km (figure 1). 38 vertically unevenly spaced full-sigma levels wereplaced in the vertical, with the highest density in the lowest 1500 meters.This allows the planetary boundary layer to be well represented in the model.MM5 was initiated as a "cold start" with no pre-forecast spin up period orassimilation of additional observations. This could however easily have beendone, but observations are already assimilated into the analysis provided bythe ECMWF, and we do not wish to use the same observational backgrounddata twice. By experience, the assimilation of observations would not haveany large effect in this situation because conventional observation sites aresparse both in time and space during the time studied here.MM5 has a wide variety of different options for the parameterization of subgridprocesses (Grell et al., 1994). Here, we have applied the turbulencescheme based on Hong and Pan (1996), coupled with a simple soil diffusionmodel. For moisture, an explicit scheme was applied, including super cooledwater, ice, rain, snow and vapour (Reisner et al., 1998). Cumulusparameterization based on Kain and Fritsch (1993) was used for the 9 kmdomain, but not for the 3 km and 1 km domains. Topography and land-usewere derived from the 1 km USGS (United States Geological Survey) dataset(Eidenshink and Faundeen, 1998). Finer topography grid was available fromNorwegian sources. However, in this study the horizontal grid distance in theMM5 model was 1 km. A higher resolution would only have been smoothedto a 1 km model topography. Regarding the limitations of the physical subgridparameterizations, the 1 km resolution is seen as the present day limit ofthe MM5 system. Further information on the model system can be found inGrell et al. (1994).54

3. Methods for calculating in-cloud icingThe rate of icing (dM/dt) onto an object is given by the equationdMdt= α1 α 2α3 ⋅ ρ LWC ⋅ A⋅V[Kg·s -1 ] (1)where ρ LWC is the density of liquid water content of the air, flowing with thewind velocity, V, towards the cross-sectional area, A, of the object.Efficiency coefficients α 1 , α 2 and α 3 represent processes that reduce the rateof icing (Brun et al., 1955, Lozowski, 1983, Makkonen and Stallabras., 1987,Finstad et al., 1988). These factors vary between 0 and 1.3.1 Efficiency coefficientsα 1 represents the collision efficiency of the particles, i.e. α 1 is the ratio of theflux density of the particles that hit the object to the flux density of particlesin the cross sectional area upstream of the object. The collision coefficient α 1becomes less than one when the water droplets moving towards an objectfollow the streamlines around it without colliding. Small droplets, largeobjects and low wind speeds reduce α 1 . Langmuir and Blodgett (1946), andFinstad et al. (1988) conducted a theoretical investigation of water droplettrajectories around cylinders. This investigation describes how droplets hitthe cylinder within a band limited between polar angles –φ and φ (Figure 3).The angle φ 0 is a function of the droplet radius, cylinder radius, air speed, airtemperature, and pressure. The equations are empirically fitted to the resultsof the collision efficiency, as given by Finstad et al. (1988) and Makkonenand Stallabras (1987). Here, collision efficiency is taken as being a functionof median volume droplet diameter, wind speed and cylinder diameter.Variation of the collision efficiency is considerable. This theory is only validfor cylinders, which often makes it inapplicable in nature. Therefore, thistheory of estimating the collision efficiency is limited to be valid only in thebeginning of the ice incident discussed here. During time, the ice will form avane against the wind, in which case the theory for a cylinder becomesinapplicable.55

φFigure 3. Air streamlines of droplet trajectories around a cylindrical objectα 2 represents the efficiency of collection of those particles that hit the object,i.e. α 2 is the ratio of the flux density of the particles that stick to the object tothe flux density of the particles that hit the object. The collection efficiency,α 2, is reduced from one when the particles begin to bounce from the surface.Particles are considered to have stuck either when they are permanentlycollected or when their residence time is sufficient to have affected the icingrate. As an example, the icing rate can be affected when there is heatexchange between surface and particle. The collection efficiency is assumedequal to 1 for in-cloud icing (Ahti and Makkonen, 1982).α 3 represents the efficiency of accretion, i.e. α 3 is the ratio of icing to the fluxdensity of the particles that stick to the surface. The heat released by freezingdepends on the wind speed, ρ LWC , as well as droplet size and diameter of theobject. The efficiency of accretion reduces from 1 when the heat flux fromthe accretions is too small to cause sufficient freezing to incorporate allsticking particles into the accretion. In such a case, part of the mass flux ofthe particles is lost from the surface by run-off (Makkonen, 1996). At somespecific ρ LWC or wind speed the released heat of freezing will increase thesurface temperature (T s ) to 0ºC. The minimum value of ρ LWC at which T sreaches 0ºC is called the Ludlam limit (Ludlam, 1951). T s can be founditeratively by solving the equation of the heat balance over an ice surface,given by Mazin et al. (2001). However, this shedding of water from thesurface is often ignored when the air temperature is below 0ºC (Sundin andMakkonen, 1998). Thus, α 3 is set equal to 1 in this case study.56

As the wind direction was nearly constant for this case study, the crosssectional area of the cylinder was assumed to be constant. The super cooledcloud droplets only hit the cylinder on the windward side, creating an icevane facing the wind. This has been confirmed by visual observation. Nomeasurements of droplet size were performed during the case study.Therefore, by assuming that the number of droplets per volume was constant,the droplet size is only correlated to the estimated ρ LWC . From equation 1, thecollision coefficient can be found using the observed ice-growth rate andestimates for wind speed and ρ LWC:( dM )α = dtA ⋅V⋅ ρ(2)LWCObservations show that the collision coefficient during icing varies fromapproximately 0.04 to 0.13 in this case study (figure 4), with the exception ofthe high values in the beginning of the period, which were associated withprecipitation.Figure 4. Measured ice load and calculated collision coefficient based on equation 2, duringthe icing incident March 21 – 24 2003.57

3.2 Determination of cloud base and air temperatureReliable values for air temperature and humidity are needed for estimation ofwater vapor pressure and density of dry air at certain heights above sea level.Measurements do not normally exist at any actual icing site, making reliableestimates of air temperature crucially important. A lifted volume of air closeto the slope of the mountain is mixed with ambient air. However, this processis assumed to be near adiabatic. Air temperature at different heights for thevolume of air that is lifted is assumed to follow the function;T z = T − γd⋅ z −;z ≤ z c (3.a)( )1( z1)T ( z) = T − ⋅ z − z − γ ⋅ ( z − z )1 d(c 1) wcγ ;z > z c (3.b)where T1is the temperature at the lower station, z 1 and z c is the height of thelower weather station and the cloud base, respectively. γ is defined as –dT/dz,where γ d is the temperature gradient for unsaturated conditions (below cloudbase) and γ w is the temperature gradient for saturated conditions (inside thecloud). Earlier work suggests γ w = 0.54 ºC/100m and γ d = 0.85ºC/100m(Harstveit, 2002). Estimates from measurements at Brosviksåta suggest γ w =0.62ºC/100m and γ d = 0.92 ºC/100m, with standard deviations of 0.07 and0.08, respectively (figure 5).Figure 5. Measured temperature gradients for saturated (left) and unsaturated (right)conditions for wind speed above 5 m/s, wind direction in the sector 180 – 45 Deg. Besidesaverage gradients (dT/dz), standard deviations and number of samples (Nr. Samp.) aregiven.58

To exclude situations of high stability and weak mixing in the boundarylayer, a lower wind speed limit of 5 m/s at the weather station at 325 m a.s.l.was chosen. Furthermore, only wind directions between South (180º) viaNorth to East (45º) were chosen due to large scale turbulence created by thetopography in the sector 45º – 180º degrees. Based on these criteria the valueof 0.92 ºC/100m for unsaturated conditions indicates a slightly stableboundary layer, with a vertical temperature gradient less than adiabatic. Thisis consistent with previous observations of a smaller increase in ρ LWC withheight than the adiabatic (Nicholls, 1984, Noonkester, 1984).Before the temperature gradients defined above can be used, the height of thecloud base must be determined. The air temperature and pressure at the lowerlevel 1 are T 1 and p 1 , respectively, with mixing ratio w 1 . This air is cooled bylifting at the temperature gradient for dry conditions, γ d (defined above), untilits adiabat intersects the vapor line defined by w s = w 1 . The air temperature atthis level is T c . An analytical approximation for T c , which must be solved byiteration, is given by:⎡1 kAε ⎛ T ⎞ ⎤1T = ⎜ ⎟ ⎥c B / ln⎢wp(4)⎢ 1 ⎝ Tc⎠ ⎥⎣⎦where A, B, k are constants, respectively 2.53·10 8 kPa, 5.42·10 3 K and 0.286(Rogers and Yau, 1989). w, T 1 , and p 1 are known at the unsaturated lowerlevel 1. T c is therefore the air temperature at cloud base. Cloud base height isestimated by using the temperature gradients in equation 3 above, either byfollowing the gradient for unsaturated conditions from below (3a), or byfollowing the gradient for saturated conditions from above (3b).T1 −Tz+ γ d ⋅ z1z = ; z c > z 1 (5a)cγ dz + γ w ⋅ z1T1 Tzc=−; z c < z 1 (5b)γ wEquation 5a can be used during unsaturated conditions, and equation 5bduring saturated conditions.3.3 Cloud liquid water contentMixing ratio w is defined as the mass of water vapour (M v ) per unit mass ofdry air (M d );M v ρ vw = = ≈ ε ⋅(6)ρMddep59

where p is the air pressure, e is the water vapour pressure, ρ v is the watervapour density, and ρ d is the density of dry air. ε is the constant ratio of themolecular weight for water vapour and dry air, equal to 0.622. The saturationvapour pressure e s can be fitted to within 0.1% over the temperature range -30ºC ≤ T ≤ 35ºC by the empirical formula⎛ 17.67 ⋅ T ⎞e s ( T ) = 6.112 exp⎜⎟ (7)⎠⎝ T + 243.5where e s is in mb and T in degrees Celsius (Bolton, 1980).The mixing ratio defined above is constant with height for a volume of airthat is lifted dry adiabatically without any entrainment. In our case,measurements show that the temperature gradient for dry conditions atBrosviksåta is 0.92 ºC/100m, while the dry adiabatic temperature gradient, Γ= g/c p = 0.98/100m. This gives a difference in the temperature gradients ofonly 0.06 ºC/100m, and the mixing ratio along the mountain slope istherefore assumed to be constant with height. This gives the relationship;wee121 = ε = w2= ε(8)p1p2where the indices 1 and 2 refers to the lower (1) and the upper (2) levels,respectively.A reduction in the mixing ratio at level 2 indicates saturation with respect towater vapour. Based upon the assumption that the total water content of theair is constant with height, the total mixing ratio, w tot , is the sum of thesaturation mixing ratio (w s ) and the liquid water content mixing ratio (w LWC )at level 2, i.e.:ws2 = wtot− wLWC 2(9)The total mixing ratio equals the mixing ratio at unsaturated conditions, inthis case at the lower level 1:w tot = w 1(10)That means no fall out by rain and no exchange of water vapor between theair masses and the slope of the mountain.Combining equation (8), (9) and (10) gives the mixing ratio of LWC at level2:ρ LWC2= w − w(11)ρ d 21s2C ombining equation (6) and (11) gives ρLWC in a cloud given by the equation⎛ eρ LWC2= ε ⋅ ρ 2⎜ 1 e2⎞d −⎟(12)⎝ p1p2⎠with the cloud base lying between the lower (1) and upper (2) level. Belowthe cloud base, at level 1, water in the air consists only of water vapour.60

Inside the cloud, at level 2, the water in the air consists of a mixture of waterv apour and cloud liquid water.Liquid water content of stratus clouds is usually in the range of 0.05 to 0.25g/m 3 , while in stratocumulus and deep nimbostratus the values are usuallylower than 1 g/m 3 (Rogers and Yau, 1989). Detailed observations of themicrophysical structures of marine stratus clouds (Nicholls, 1984,Noonkester, 1984), which are of importance in this case study, show averageliquid water content over horizontal layers increasing with height, but withvalues less than adiabatic. This increase of water content with height isaccounted for by an increase in droplet size rather than concentration ofcloud drops. This is due to the fact that the growth process is dominated bycondensation rather than coalescence. Because of this, the droplet spectrumin stratus clouds is relatively narrow, as expected with growth bycondensation.3.4 Wind speed estimatesWind speed measurements were carried out along the slope of Brosvisåta at325, 520 and 718 m a.s.l. Observations of wind speed were taken 2 metersabove the ground, and given as 10 minute mean values. The wind sensorsused were rotating cup anemometers, which stopped rotating when icing tookplace. The wind speed ratio between level 2 and level 1, (V 2 /V 1 ) as afunction of a lower limit of wind speed at level 1, was analyzed.A data selection was performed with respect to air temperature, winddirection and wind speed. The air temperature must be higher than +1.5 ºC at718 m a.s.l. to exclude all the icing incidents. The wind direction was in thesector 160 – 180 degrees (SSE–S) at the lower weather station at 325 m a.s.l.for the actual icing incident investigated in this case study. Based on one yearof data collection at the site, with the wind direction and air temperaturelimits specified above, the results are shown in figure 6. This shows that theratio decreases with increasing wind speed at 325 m a.s.l. from 2.25 at windspeed 4 m/s to 2.0 at wind speed 11-12 m/s. In this work, a lower limit of 5m/s was chosen for level 1, which gives a ratio equal 2.2 and a standarddeviation of 0.41.61

Figure 6. Wind speed ratio, (left y-axis) and the standard deviation (right y-axis) as afunction of wind speed at the lower level, with wind direction in the sector 160-180˚.4 Results and discussion4.1 Measurements and calculation of atmospheric ice accretionThe icing incident in this study took place March 21-24 2003, with amaximum build-up of ice measured to approximately 4.5 kilograms (figure7). In the beginning of the period, the air temperature was -3ºC with a nearlinear increase until it reached 0ºC on March 23. A sudden drop in ice load atthe end of the period can be explained by the increase of temperature above0ºC. The rate of precipitation was estimated from the synoptic weatherstation Takle, situated 12 km to the east of the mountain base, withmeasurements taken at 0600hrs and 1800hrs daily. Precipitation accumulatedduring the icing incident was 6.3 mm, with only three observed periods ofprecipitation (figure 7).The first period of precipitation measured at Takle, from 0600hrs to 1800hrsMarch 21, was associated with a fast increase in ice load. This indicates thatthe ice load might have been a combination of in-cloud icing andprecipitation icing. The ice load increases further by approximately 3kilograms during the period from 0600hrs March 22 to 0600hrs March 23,when no precipitation was recorded. This period was therefore assumed onlyto be an in-cloud icing incident. An increase in air temperature above 0ºC atthe mountain top as well as precipitation during the period 0600hrs to62

1800hrs March 23 was associated with a sudden decrease in ice load due tomelting and ice falling off from the scale.The ρ LWC in figure 7 was determined by eq. 12 using the data recordedautomatically by the weather stations along the mountain slope. Negativevalues in the beginning of the period were probably caused by low windspeeds and therefore a boundary layer that was not well mixed. Wind speedincreases as the frontal system approaches. The method for determining theρ LWC given by eq. (12) can therefore be applied.A plot of the ice growth and the ρ LWC during the icing incident from 0800 hrsMarch 21 until maximum ice load was reached on March 23 shows a goodcorrelation of the two independently measured values (figure 7). The onlyexception is found in the beginning of the icing period. This can be explainedby the fact that precipitation occurred in the same period, and the ice loadwas therefore a combination of in-cloud ice and precipitation ice. Decreasingice growth rate towards zero around 2400 hrs March 22 was associated withthe decreasing ρ LWC (until it reached zero). Likewise, the decrease in ρ LWC ataround 0400 hrs March 23 was associated with a decrease in ice growth rateat the same time. A statistical correlation between the ice growth rate and theρ LWC , from the beginning of the ice growth until it reached maximum weight,gives a correlation coefficient of 0.73. By excluding the period in thebeginning, associated with precipitation, a higher correlation coefficient of0.85 is found.63

Wind speed (m/s)Ice growth rate (g/10 min)Ice load (kg/m)201000.040.02043210ABCDE3.51.42.46420-2-40.60.40.20Air temp. [DegC] Liquid Water Content (g/m3)20.0321.03 22.03 23.03 24.03 25.03Date (dd.mm)Figure 7. A. Wind speed calculated at 723 m a.s.l. B. Air temperature estimated (solid line) andmeasured (dotted line) at 723 m a.s.l. C. Ice growth rate (g/m 10 min) 3 hrs running mean atthe ice scale. D. Estimated liquid water content at 723 m a.s.l. E. Ice load measured (solid line)and estimated (dotted line) (kg/m) at the ice scale. Lower step plot shows 12 hrs accumulatedprecipitation, where the numbers above indicate the amount of precipitation in mm.64

Ahti and Makkonen (1982) and Sundin and Makkonen (1998) estimate thein-cloud ice load, M i (N m -1 ), for each icing events as,M 10 −3i= 5.5 ⋅ v T g(13)iiwhere 5.5⋅10 -3 is an empirical constant, v i is the wind speed (m s -1 ) at thelevel of interest, T i is the event duration (hrs), and g is the acceleration ofgravity. Furthermore, the criteria for in cloud icing to occur is simply that theheight of cloud base is lower than the height of interest and an airtemperature in the range 0 to –15 °C. According to this theory, in-cloud icingintensity (kg m -1 s -1 ) is independent of ρ LWC . As mentioned above, however,measurements and calculations of the icing incident at Mt. Brosviksåtaindicate a strong correlation between the ice growth rate and ρ LWC (figure 7).This may also indicate a relationship between the ρ LWC and the collisionefficiency. Assuming collision efficiency equal to zero when ρ LWC is zero anda linear increase of the collision efficiency creating an ice load equal themeasured ice load at the end of the icing period, the following formula isobtained:αwhere ρ LWC is in (g m -3 ). The ice growth is assumed to stop when the airtemperature reaches 0ºC. This result indicates a strong connection betweenthe measured and calculated ice growth based on a collision efficiencydetermined only by the ρ LWC (figure 7). This simple model for estimating thecollision efficiency ignores the effects of changes in wind speed, shape andarea of cylinder as well as the number of droplets per volume air. Theindependence of wind speed is explained by the fact that the wind speed isapproximately constant during the icing incident studied here (figure 7).Further testing against similar cases of in-cloud icing is needed in order toverify the reliability of this model.1 = 0.225⋅ ρLWC(14)4.2 MM5 resultsA qualitative evaluation of the data from the MM5 model run during theicing incident was performed. The modeled synoptic situation given by theMM5 model corresponded well with the observed weather data during theperiod in question. Wind speeds and temperatures in MM5 corresponded wellwith the observations near Brosviksåta as well as the synoptical observationsat Takle. Main surface winds came from S to SW, with winds from W to NWin the higher model levels (500 – 700 hPa). At the same time, the modelgives little or no precipitation during this period, though the cloud base wasbelow the mountain top at Brosviksåta.The model performs a smoothening of the topography. This is why theground level in the model was at a different level to the actual terrain in the65

area. Maximum height of the mountain was found to be 554 m a.s.l. in themodel topography at 1 km resolution (see also figure 1) whereas the actualheight of the mountain is 723 m a.s.l. A correction of model data from 554 ma.s.l. to 723 m a.s.l. was necessary for determining icing. The modeled airtemperature at the top of the mountain was found by following thetemperature gradient for saturated conditions, found equal to γ w =0.62, from554 m up to 723 m (figure 8). It can be observed that the simulated airLiquid Water Content (g/m3)Ice load (kg/m)0.60.40.2432100ACBDMeasured/estimatedSimulated21.03 22.03 23.03 24.03Date (mm.dd)Figure 8. A. Wind speed and B LWC estimated from measurements (solid line) at 723 ma.s.l and simulated by MM5 (dotted line) at 554 m a.s.l., respectively. C. Air temperatureat 723 m a.s.l. measured (solid line) and simulated (dotted line). D. Ice load measured bythe ice scale (solid line) and simulated by MM5 (dotted line).24201612840210-1-2-3-4Wind speed (m/s)Air temperature (Deg.C)66

temperatures show a good correlation with the measured air temperatures,with an increase from -3 ºC in the beginning of the period until it reached +1ºC at the end of the period. The model slightly underestimates airtemperature, with an average error of 0.45 ºC. The correlation coefficientbetween measured and simulated air temperatures is 0.94 in this case study.ρ LWC estimated from MM5 results and from the measurements during theicing incident are also given in figure 8. The model gave data at one-hourintervals. The surface effects on the model results at 723 m a.s.l. weresignificantly reduced by the smoothening of the topography in MM5. Amodel height of 554 m a.s.l. was therefore used for the ρ LWC . Due to this, ithas been assumed that the ρ LWC was underestimated when the height waskept to 554 m a.s.l. Data from the MM5 model gave a ρ LWC in the same sizeof order as the ρ LWC determined from measurements (figure 8). As expected,MM5 slightly underestimates the ρ LWC , with an average difference of 0.05g/m 3 . The correlation coefficient of these ρ LWC values equalled 0.72.Simulated ice growth was calculated by use of eq. 1. α 2 and α 3 were set equalto one, and α 1 was estimated by eq. 14, as in chapter 4.1 above. Values ofρ LWC , wind speed and temperature were taken from the results of the MM5-simulation. According to figure 8, the simulated ice growth wasunderestimated.5 Summary and conclusionsReliable forecasts of duration and intensity of icing on structures as in-cloudice requires data of standard meteorological parameters, in addition to morespecific parameters such as the density of the liquid water content of the air(ρ LWC ). Icing conditions on the mountain Brosviksåta (723 m a.s.l., 61º 2` N,5º 9` E), at the western coast of Norway was investigated in the period March21-24, 2003 in this study. Measurements of ice growth on a one meter highnon-rotating steel rod, of 0.14 m diameter, at the mountain peak showed amaximum build-up of 4.5 kg during that period. The type of icing wasidentified as mainly in-cloud icing by dry growth (rime ice), due to the factthat little precipitation was measured during the case study period.A method for calculation of ρ LWC has been described. The method onlyrequires measurements of air temperature, air humidity and wind speed at aknown level under unsaturated conditions. These parameters were measuredat three different levels along the slope of the mountain. The beginning and67

end of the icing period was determined to a high degree of accuracy with thismethod. Reliable prognoses of ρ LWC will greatly improve the procedures offorecasting duration and intensity of in-cloud icing. Furthermore, testing ofthis method on similar cases will be necessary in order to verify its accuracyand reliability.A comparison of measured ice growth rate (kg/s) by the ice scale andestimated ρ LWC (kg/m 3 ) by use of the lower weather station gave a correlationcoefficient of 0.85 for the period identified as in-cloud icing. Based upon thisstrong correlation a simple model of estimating the coefficient of collisionefficiency has been developed. However, further testing of this model is alsoneeded.A mesoscale model (MM5) was tested at a high horizontal resolution (1km),to evaluate its ability to reproduce weather conditions where freezing occurs.The MM5 forcast identified the start and end times of the icing event with ahigh degree of accuracy. On the other hand, the accuracy of the simulatedicing intensity was not that good. The MM5 calculations for ice growth gavea value 58% of the actual meausred accumulated ice growth. More events oficing should be studied to determine the overall accuracy at a 1 kmresolution. However, simulations with the high horizontal resolution used inthis case study, are very time consuming. It is not taken for granted that ahigh resolution forcast would be more accurate than a forcast using a coarserresolution. A study by Mass et al. (2002) shows no increase in accuracy ofprecipitation when going from 12 to 4 km resolution. Further studies of realtimecases on real-time systems at coarser model resolutions will thereforereveal MM5s capability for making reliable daily forecasts of freezingevents.ACKNOWLEDGEMENTSForsvarsbygg, Norkring, Statnett and Telenor funded this study. The authors’wish to thank Odd Rutledal for inestimable field support.68

REFERENCESAhti, K. and Makkonen, L., 1982. Observation on rime formation in relationto routinely measured meteorological parameters, Geophysica, 19(1): 75– 85.Bolton, D., 1980. The computation of equivalent potential temperature. Mon.Wea. Rev. 108; 1046 – 1053.Brun, R. J., Lewis, W., Perkins P. J. and Serafini, J. S., 1955. Impingement ofcloud droplets on a cylinder and procedure for measuring liquid-watercontent and droplet sizes in super cooled clouds by rotating multicylindermethod, NACA, Lewis Flight Propulsion Laboratory, Cleveland, Ohio,Rep. No. 1215, 43 pp.Chaine, P. M. and Skeates, P., 1974. Ice accretion handbook (FreezingPrecipitation), Industrial Meteorology Study VI, Environment Canada,Toronto, Canada.Chang, D., Jiang, L. and Islam, S., 2000. Issues of soil moisture coupling inMM5: Simulation of the diurnal cycle over the FIFE area, J. ofHydromet., 1(6): 477-490.Chatfield, R., Vastano, J., Li, L., Sachse, G. and Connors, V., 1999. TheGreat African Plume from biomass burning: generalizations from a threedimensionalstudy of TRACE A carbon monoxide, J. Geophys. Res.,103(D21): 28059-28077.Drage, M. A. and de Lange, T., 2004. Description of an ice scale formeasuring atmospheric icing. Meteorological report series university ofBergen, 2004. In press.Eidenshink, J. and Faundeen, J., 1998. The 1-km AVHRR global landdataset: first stages in implementation, Int. J. of Rem. Sens, 15: 3443-3462.Finstad, K. J. and Lozowski, E. P. and Gates E. M.,1988. A computationalinvestigation of water droplet trajectories, J. Atmos. and Oce, Tech., 5:160 – 170.Grell, G., Dudhia, J. and Stauffer, D. 1994. A Description of the Fifth-Generation Penn State/NCAR Mesoscale Model (MM5), NCARTechnical Report Note TN-398, National Center for AtmosphericResearch, Boulder Colorado, US.69

Haldar, A. P., McComber, P., Marshall, M. A., Ichac, M., Goel, A. andKatselein, M., 1996. Validation of ice accretion models for freezingprecipitation using filed data, Proceeding 7 th International Workshop onAtmospheric Icing of Structures, pp. 189 – 194.Harstveit, K., 2002. Using routine meteorological data from airfields toproduce a map of ice risk zones in Norway. Proceeding 10 th InternationalWorkshop on Atmospheric Icing of Structures, CD: Session 8-1.Hong, S. and Pan, H., 1996. Nonlocal Boundary Layer Vertical Diffusion in aMedium Range Forecast Model, Mon. Weather Rev., 124: 2322-2339.ISO 12494, 2001. ISO (the International Organization for Standardization)12494 – Atmospheric icing of structures.Kain, J. S. and Fritsch, J. M., 1993. Convective parameterization formesoscale models: The Kain-Fritsch scheme. The representation ofcumulus convection in numerical models, K. A. Emanuel and D. J.Raymond, Eds., Amer. Meteor. Soc., 246 pp.Langmuir, I. and Blodgett, K. B., 1946. A mathematical investigation ofwater droplet trajectories. Collected Works of Irwing Langmuir,Pergamon Press, 10: 348 – 393.Lott, J. N. and Jones, K. F., 1998. Using U. S. weather data for modeling iceloads from freezing rain, Proceeding 8 th International Workshop onAtmospheric Icing of Structures, pp. 157 – 162.Lozowski, E. P., Stallabras, J. R. and Hearty, P. F., 1983. The icing of anunheated, nonrotating cylinder, Part I: A simulation model, Part II: Icingwind tunnel experiments, J. Climate Appl. Meteor., 22: 2053 -2074.Ludlam, F. N., 1951. The heat economy of rimed cylinder. Quart. J. Roy.Meteor. Soc., 77: 663 – 666.Makkonen, L., 1984. Modeling of ice accretion on wires, J. Climate Appl.Meteor., 23: 929 – 939.Makkonen, L. and Stallabras, J. R., 1987. Experiments on the cloud dropletcollision efficiency of cylinders, J. of Clim. Appl. Meteor., 26: 1406 –1411.Makkonen, L. and Ahti, K. 1995. Climatic mapping of ice loads based onairport weather observations, Atmos. Res., 36: 185 – 193.Makkonen, L., 1996. Modeling power line icing in freezing precipitation,Proceeding 7 th International Workshop on Atmospheric Icing ofStructures, pp. 195 – 200.Mass, C. and Kuo, Y., 1998. Regional real-time numerical weatherprediction: Current status and future potential, Bull. Amer. Meteor. Soc,79: 253-263.Mass, C., Ovens, D., Westrick, K. and Colle, B., 2002. Does Increasinghorizontal resolution produce more skilful forecasts? Bull. Amer. Meteor.Soc, 83: 407-430.70

Mazin, I. P., Korolev, A. V., Heymsfield, A., Isaac, G. A. and Cober, S. G.,2001. Thermodynamics of icing cylinder for measurements of liquidwater content in supercooled clouds, J. Atmos. and Oce, Tech., 18: 543 –558.Nicholls, S., 1984. The dynamics of stratocumulus: aircraft observations andcomparisons with a mixed layer model, Q. J. R. Meteor. Soc., 110: 783 –820.Noonkester, V. R., 1984. Droplet spectra observed in marine stratus cloudlayers, J. Atmos. Sci., 41: 829 – 845.Poots, G., 2000. Ice and snow accretion on structures. Phil. Trans. R. Soc.Lond. A (200), 358: 2799 – 3033.Reisner, J., Rasmussen, R. J. and Bruintjes, R. T., 1998. Explicit forecastingof supercooled liquid water in winter storms using the MM5 mesoscalemodel, Quart. J. Roy. Meteor. Soc., 124B, 1071-1107.Rogers, R. R. and Yau, M. K., 1989. A short course in cloud physics,Pergamon Press.Sundin E. Makkonen L., 1998. Ice loads on a lattice tower estimated byweather station data. J. Appl. Meteoror. Vol. 37, pp. 523-529.Vassbø, T., Kristjansson, J. E., Fikke, S. M. and Makkonen, L., 1998. AnInvestigation of the Feasibility of Predicting Icing Episodes usingNumerical Weather Prediction Model Output. Proceeding 8 thInternational Workshop on Atmospheric Icing of Structures.Warner, T., Kibler, D. and Steinhart, R., 1991. Separate and coupled testingof meteorological and hydrological forecast models for the SusquehannaRiver Basin in Pennsylvania, J. Appl. Meteor., 30: 1521-1533.Warner, T., Peterson, R. A. and Treadon, R. E., 1998. A tutorial on lateralboundary conditions as a basic and potential serious limitation to regionalnumerical weather prediction, Bull. Amer. Meteor. Soc, 78: 2599-2617.71

LARGE-SCALE MEASUREMENTS ANDNUMERICAL SIMULATIONS OF IN-CLOUD ICINGAROUND THE RIDGE OF A MOUNTAIN PEAKMagne A. DrageUNIS/UIBThomas K. ThiisNorwegian Building Research InstituteABSTRACT: Atmospheric icing by in-cloud icing has been measured aroundthe ridge of Mt. Gaustatoppen (59°51’, 08°N39’E, 1882 m a.s.l.) in Norway,during a 5-day period. Sixteen sticks of 2 m height and 3 cm diameter wereplaced around the edge of the mountain ridge. A finite volume CFD(Computational Fluid Dynamics) solver was used to simulate the wind flowand the rate of icing around the top 270 meters of the mountain.Measurements and simulations show that even small variations in thelocation of the sticks around the ridge of a mountain peak, can cause largevariations in accreted ice on the sticks. A basic understanding of the air flowaround isolated mountain peaks is vital to understand how complextopography and altering wind direction can influence icing intensity. In thiscase, use of a micro scale numerical model to describe the wind field aroundthe mountain peak, or measurements at the location for a short time period,proved to give valuable information. A study of the icing around a buildinglocated at the peak indicate that local sheltering may reduce the in cloud icingby up to 100 %.1. INTRODUCTIONIn regions with severe climatic conditions, the planning and building of newconstructions calls for special attention to be paid with respect to atmosphericicing. In many cases the most severe form of icing is in-cloud icing,especially in elevated areas. The expected maximum amount of in-cloudicing is an important design criterion within the building industry,communications and energy distribution. The largest recorded iceload on apower line is 305 kg m -1 . This was measured on a 22kV overhead line closeto the Norwegian village Voss in 1961 (Makkonen, 2000). In general, it hasnot been the practice to collect ice data before a construction is erected in anenvironment favourable of in-cloud icing. This can often result in operational73

problems and failures, as shown by the collapse of 13 TV towers in theUnited States alone due to ice loads during a 20-year period (Sundin andMakkonen, 1998). A power line in western Norway, which experiencedserious icing and failure, was re-erected in a parallel section 48 m lower thanthe original line. The new line was almost completey sheltered against icingby the hill above (Raastad, 1958). In this context, the location of theconstruction on a mountain peak or slope is of great importance since icing isstrongly dependent upon wind speed, wind direction, cloud liquid content andcloud droplet spectra. Full scale measurements of icing are sparsely reported.The main features can be recognised in wind tunnel experiments, but suchexperiments suffer from scaling problems and a lack of information inconnection with local variations of climatologically parameters.In-cloud icing, which is the main process studied in this field measurementprogram, occurs when super cooled cloud droplets collide with a surface andfreeze spontaneously. This is the type of icing which gives the highestaccumulated ice load on structures (Makkonen and Ahti, 1995). The rate oficing onto an object by in-cloud icing by dry growth is given by the equationdMdt=1 2α3α α ⋅ q ⋅ A ⋅V[Kg·s -1 ] (1)LWCwhere q LWC is the liquid water content of the air, flowing with the wind speedvelocity V, towards the cross-sectional area A, of the object. The efficiencycoefficient α 1 represents the collision efficiency, which is the fraction of thetotal droplets in the path of the object that collide with the object, given byFinstad et al. (1988a) and Makkonen and Stallabras (1987). Small droplets,large cross sections and low wind speeds reduce α 1 . α 2 and α 3 represent thecollection and accretion efficiency, respectively. They are assumed to beequal to 1 for in cloud icing, according to earlier field measurements by, forexample, Ahti and Makkonen (1982) and Sundin and Makkonen (1998).The meteorological input parameters in eq.1 are liquid water content (LWC),air temperature and wind speed. Measurement of these parameters does notnormally exist at any actual icing site, making reliable estimates cruciallyimportant. The method for calculating these estimates are given by Harstveit(2002) and Drage and Hauge (2004).Estimates in this study were made using measurements from Gaustatoppenover a one-year period. These estimates suggested a temperature gradient fordry conditions γ d = 0.86 ºC/100m, and a temperature gradient for wetconditions γ w = 0.52 ºC/100m, with standard deviations of 0.16 and 0.11,74

espectively. The air temperature at cloud base T c was found solving byiteration the analytical approximation given by:⎡1 / kAε⎛ T ⎞ ⎤⎜1T ⎢ ⎟ ⎥c = B / ln⎢(2)wp1⎝ T c ⎠ ⎥⎣⎦where A, B, k are constants, respectively 2.53·10 8 kPa, 5.42·10 3 K and 0.286.w, T 1 , and p 1 are mixing ratio, air temperature and air pressure known at anunsaturated lower level 1(subscript 1) (Rogers and Yau, 1989). Furthermore,the cloud base height was estimated by using the temperature gradientsdefined above, either by following the gradient for unsaturated conditionsfrom below, or by following the gradient for saturated conditions from above.By measuring air temperature and relative humidity at level 1 below cloudbase, it is possible to estimate the liquid water content (LWC) of the cloud acertain height above the cloud base. This method is described by Drage andHauge (2004):⎛ e⎜1 e−⎝ p1p⎞⎟⎠ρ LWC = ε ⋅ ρ d 22(3)2with the cloud base lying between the lower (subscript 1) and upper(subscript 2) levels. The constant ε is the ratio of the molecular weight forwater vapour and dry air, equal 0.622. The density of dry air at level 2 isgiven by ρ d2 . The water vapour pressure and the air pressure at the two levelsare given by e 1 , e 2 and p 1 , p 2 respectively. This equation is based upon theassumption that the total water content of the air is constant with height.According to Sundin and Makkonen (1998), icing is forecasted when the airtemperature is below 0 ºC and the relative humidity is higher than a criticalvalue, or that the cloud base is below the level of the site in interest.Furthermore, the icing intensity is simply assumed to be a function of thewind speed.However, literature shows convincing evidence that LWC increases withheight in a cloud layer. This increase in LWC is accounted for by an increasein droplet size rather than droplet concentration. In fact, droplet concentrationis observed to be approximately constant throughout a stratus cloud layer,whereas droplet size increases monotonically with height (Rogers and Yau,1989). Assuming a constant droplet concentration, droplet size is only afunction of LWC, which increases with height above the cloud base. Given aLWC dependent upon height above cloud base in eq. 3, it is now possible toestimate the icing intensity onto a structure given in eq. 1 as a function ofheight above sea level. These calculations also require estimates of windspeed and collision efficiency.75

A Computational Fluid Dynamics (CFD) three-dimensional solver, describedin the next chapter, calculates the local wind speed and the icing rate. Thecollision efficiency in eq. 1 is dependent upon wind speed, diameter of thestructure (cylindrical), and median volume droplet diameter (Finstad et al.1988b). Assuming a constant droplet concentration, the mean volume dropletdiameter is calculated as a function of height above the cloud base.During icing with approximately constant wind direction, the ice will form anice vane towards the wind, in which case the theory for a cylinder discussedabove is generally inapplicable. One approach to this problem is to plot theicing intensity for varying cylinder diameters. In this theoretical case anLWC of 0.4 g/m3 and a wind speed of 10 m/s were chosen, and four cases ofdifferent droplet concentration have been plotted (figure 1). Measurementsand observations during the icing incident studied here showed that the icevane created a peak towards the wind. The width of the peak was observed todecrease as the ice vane grew. This can be interpreted as a decreasingeffective cylinder diameter as the icing continues. Applying the equations byFinstad et al. (1988a) and assuming a droplet concentration of 10 E+7 (m3)and an initial cylinder diameter of 3 cm, figure 1 shows that the icingintensity decreases rapidly when the effective cylinder diameter decreasesbelow 0.02 m.Icing intensity (kg/s hr m)0.200.160.120.080.040.00Droplet concentration5 E+7 (m3)10 E+7 (m3)15 E+7 (m3)20 E+7 (m3)Wind speed = 10 m/sLWC = 0.0004 (kg/m3)0 0.01 0.02 0.03 0.04 0.05Cylinder diameter (m)Figure 1. Theoretical icing intensity on a cylinder with varying cylinderdiameter. Wind speed and LWC are 10 m/s and 0.4 g/m3, respectively. Thecomputations are based upon the equations given by Finstad et al. (1988a).76

2. STUDY SITE AND METHODSThe field experiment was performed at the mountain Gaustatoppen,Telemark, Norway. The mountain has a peak of 1883 m a.s.l. and a SE-NWleading ridge (figure 2). The mean level of the terrain surrounding themountain within a distance of 10 km is estimated to be approximately 950 ma.s.l. The peak is exposed to rather harsh climatic conditions, with a meanannual air temperature of -4.3 ºC. It experiences 177 days per year with windspeeds stronger than 10 m s -1 and 14 days per year stronger than 20 m s -1 .The peak is exposed to heavy icing in all wind directions except from anortherly direction. The maximum ice load on a horizontal rod exposed at thepeak has been measured to 284 kg/m (Finstad et al. 1988c).2.1 MeasurementsThe experiment lasted for 5 days from March 28 to April 2, 2003, which wasa sufficient length of time to form significant ice accretion on the sticks. Theaccumulated precipitation at the nearby synoptic station of Møsstrand wasmeasured to 6.7 mm during the experiment. This station is operated by theNorwegian Meteorological Institute, and is located 977 m a.s.l. 31 km west ofthe mountain peak.To study variation of icing in the terrain, sixteen sticks of 2 m height and 3cm diameter were placed around the edge of the mountain ridge, from analtitude of 1745 to 1845 m.a.s.l (figure 3). The sticks were placed such thatone profile was along the ridge of the mountain (hereafter profile 1), and twoprofiles were perpendicular to the ridge of the mountain (hereafter profile 2(upper) and profile 3 (lower)). The distance in the terrain between the sticksfollowing a profile was 40 meters. The accreted ice on the sticks wasmeasured two times during the experiment. The length of the ice vane at0.10, 0.65, 1.30 and 2.00 meters above the terrain was measured along withthe total weight of accreted ice (figure 4).An attempt of measuring wind speed and wind direction was performed byapplying a heated sonic anemometer at 1800 m a.s.l. Icing onto thetransducers caused irregularities in the wind speed data set, and a filtering ofthese data was therefore necessary. A supplement of data from handheldanemometers and from HIRLAM10 (High Resolution Limited Area Model10km) was also included in the data set.77

Figure 2. Location of the mountain Gaustatoppen (59º51´N, 08º39´E), 1883 meter above sealevel in southern Norway.100 mUTM coordinateFigure 3. Topography plot of stick location around the southern ridge of the mountainGaustatoppen. The equidistance is 10 m.78

The data from the weather station at Møsstrand was included for determiningthe large-scale wind direction during the field experiment. Measurements atthis station are taken daily at 0700 hrs, 1300 hrs and 1900 hrs GMT.The accumulated precipitation for 12-hour periods is also measured everydayat 0700 hrs and 1900 hrs GMT. The analysis output data of wind speed andwind direction from the operational model HIRLAM10 at 850 hPa was alsoincluded as a supplement and verification to the meteorological data from thestation at Møsstrand and as a basis for estimating the wind speed atGaustatoppen 1800 m a.s.l.In addition, a series of measurements by a Rotating Multi-Cylinder (RMC )were performed in order to estimate cloud liquid water content and medianvolume droplet size given by the method of Finstad et al (1988b). The icingonto the multicylinder was measured after six incidents of icing, each of 20minute durations between 1000 hrs and 1400 hrs on April 1 st . Based uponthese measurements the average LWC at 1800 m a.s.l. was estimated to 0.48(g/m3), with a standard deviation of 0.11 (g/m3). Using the weather stationmeasurements from 1160 m a.s.l and the method described by Drage andHauge (2004), the LWC at 1800 m a.s.l was estimated as 0.45 (g/m3), with acorresponding standard deviation of only 0.002 (g/m3).Figure 4. Measurements of the length of the ice wane to the left, and a demonstration of atypical ice accretion on a stick to the right.79

At the site, the temperature, humidity and wind speed were measured by fourautomatic weather stations along the mountain slope, located at 1811, 1540,1298 and 1150 m a.s.l. The wind speed sensors were rotating cupanemometers. The sampling interval for all the equipment was ten minutes.In addition, the icing was measured on a one meter high rotating cylindricalrod with diameter 3 cm placed at 1800 m a.s.l. The rod was placed on a loadcell which measured the ice load every 10 minutes (Drage and de Lange,2005). The ice was distributed on the rod in a circular shape due to rotation,and the variation in the ice load with height was negligible (figure 5).Figure 5. Ice accretion on the ice scale April 1. 2003. The length of the rod is1 m, and the diameter without ice is 3 cm.80

2.2 Numerical simulationsA finite volume CFD solver (Cham inc., 2004) was used to solve the windflow around the top 270 meters of the mountain. The Reynolds AveragedNavier—Stokes equations, closed with the RNG (ReNormalized Grouptheory) k-ε turbulence model were solved for a three dimensional cartesiangrid (Yakhot et.al, 1992). The RNG model combines the eddy viscosityconcept with the statistics of small-scale turbulence. The effects of smallscaleturbulence are represented by means of a random forcing function in theNavier Stokes equation. The RNG procedure systematically removes thesmall scales of motion from the governing equations by expressing theireffects in terms of larger scale motions and a modified viscosity (Versteegand Malalasekera, 1995). The method is frequently used in the assessment ofwind power turbine sites.In this study, the grid size was 150x140x49 cells with grid refinement closeto the area where the measurements were performed. Within this area, thegrid size is 7x7x5 meters. To improve the accuracy of the flow simulation atthe fluid/solid boundary, the Partial Solution Algorithm (PARSOL) wasapplied (Cham, 2004). This algorithm calculates the intersections of themountain surface with the cell edges and ensures that the terms in thealgebraic representations of the conservation equations are properlymodified. The rate of icing was calculated for each cell with equation 1. Theterrain model was constructed on the basis of map data with a 5 mequidistance. The roughness of the terrain is assumed to be 0.01 m (Stull,1988). The flow solver has been used at two different wind episodes. Table 1gives an overview of the simulation cases.The wind direction applied in the model is considered to be constant duringeach experiment. Measurements and observations show that the winddirection is within ± 15 degrees during each experiment at the same time asthe ice load is increasing according to the ice scale (figure 6). The wind speedapplied in each simulation case is roughly estimated by measurements atGaustatoppen at 1160 m a.s.l.Table 1. Overview of the numerical simulation casesCase no. Wind direction [deg] Wind speed [m/s] Duration (hrs)1 260 17,5 482 190 11,5 5281

3. RESULTSThe temperature during this field experiment from March 28 to April 2ranged from -2.4 ºC to –12.2 ºC. Based on this, all icing was assumed to be ofthe type in-cloud icing, according to existing models for estimating icing onstructures (Sundin and Makkonen, 1998).The accumulated ice on the sticks located around the southern ridge(figure 3) was measured after two incidents of icing (table 1). The firstmeasurement period lasted 48 hours, from 1500 hrs March 28 to 1500 hrsMarch 30, while the second measurement period lasted 52 hours, from 1600hrs March 30 to 2000 hrs April 1. The third icing incident that gave accretionof ice on the sticks around the building lasted 13.5 hours, from 2100 hrs April1 to 1030 hrs April 2. Measurements from the Møsstrand synoptic weatherstation, situated 31 km to the west of Gaustatoppen, and results from theHirlam10 model at 850 hPa showed a westerly wind during the first incident,a southerly wind during the second incident and a north-westerly wind duringthe third incident. The precipitation during the whole experiment wasassociated with the passage of two weather systems. Their passage isindicated in figure 6 by a period of no precipitation midway through theexperiment.To study the effect of a construction with respect to in-cloud ice, eight stickswere placed around a building at the peak, as described in figure 7a. Theheight of the building was 2.3 m, and the lengths of the walls were 2.5 and3.5 meters. The ice accumulation on the sticks placed around the buildingwas measured after the third icing incident (figure 7b). The sticks on thewestern and southern side of the building gave approximately the sameamount of ice accumulation. A smaller amount was recorded on the northernside, and there was no accumulation at all on the eastern lee side. The distinctdifference in the accumulation of ice on the sticks around the buildingillustrates how the rate of icing may vary considerably around a singlelocation. Sticks 5 and 6 were on the leeward side to the south-east of thebuilding and experienced no icing during the experiment, while sticks 3 and 4on north-east side were partly in the lee zone, and therefore had reducedicing.82

Liquid water content (g/m 3 )Ice load (Kg/m)0.500.250.0012840ABCDPeriod 10.4 0.22.5Period 20.33.2Period 30.1201612840360270180Wind speed (m/s)Wind direction (Deg)28.03 29.03 30.03 31.03 01.04 02.04Date (dd.mm)Figure 6. A. Simulated wind speed at 850 hPa in Hirlam10. B. Calculated LWC at 1800m a.s.l. C. Wind direction at 850 hPa in Hirlam10 (dotted line with symbol) plottedagainst measured wind speed at Møsstrand. D. Measured ice load on the ice scale at1800 m a.s.l. The step plot is measured 12hrs accumulated precipitation (mm) atMøsstrand, indicated by the numbers above the histograms.83

3North2m4121mWind8657Figure 7a. Measurement-setup around the building at the mountain peak. Thelength, width and height of the building are, 2.5 m, 3.5 m and 2.3 m respectively.0.20.16Ice weight (kg)0.120.080.0401 2 3 4 5 6 7 8Stick numberFigure 7b. Accumulated ice accretion on the sticks around the building.84

Numerical simulations of the wind conditions during the actual icingincidents by the CFD solver, show strong local variations (figure 8). Anormalization of the simulated surface wind at each stick to the inlet windspeed in the model, gives an indication of the relative change in wind speedrelated to location (figure 9). These wind speed ratios at each stick isconsidered constant through the experiments.6 6 3 5 8 0 06 6 3 5 6 0 06 6 3 5 4 0 06 6 3 5 2 0 06 6 3 5 0 0 06 6 3 4 8 0 06 6 3 4 6 0 06 6 3 4 4 0 06 6 3 4 2 0 0WINDDIRECTION6 6 3 4 0 0 0479400 479600 479800 480000 480200 480400 480600 480800 481000 481200Figure 8. Simulated wind speed (m/s) in simulation case 1. Inlet wind speed is17,5 m/s. The scales on the axis are in meter.85

Wind speed ratio (WSstick / WS inlet)1.61.20.80.40Incident 1Incident 2Profile 11 3 7 8 10 14 15 161378101415160 80 160 2401860182017801740Height a. s. l. (m)1.6Profile 218601.20.80.42 3 45618201780Height a. s. l. (m)01740Wind speed ratio (WSstick / WS inlet)1.61.20.80.402 3 4 5 6Profile 39 10 11 12 13Stick number0 80 160 2409 10 11 12130 80 160 240Distance betweensticks (m)1860182017801740Height a. s. l. (m)Figure 9. Simulated wind velocity ratio in the two different simulation cases. Locationof the sticks related to height and horizontal distance is plotted to the left.86

0.03Profile 1MeasuredSimulated11860Ice growth (Kg/m hr)0.020.01037810141516182017801740Height a. s. l. (m)1 3 7 8 10 14 15 160 80 160 240Ice growth (kg/m hr)0.030.020.01Profile 2MeasuredSimulated2 3 456186018201780Height a. s. l. (m)017402 3 4 5 60 80 160 240Ice growth (kg/m hr)0.030.020.01Profile 3MeasuredSimulated9 10 11 1213186018201780Height a. s. l. (m)017409 10 11 12 130 80 160 240Stick numberDistance betweensticks (m)Figure 10. Plot of measured and estimated ice growth rate on each stick in theterrain during icing incident 1. Locations of the sticks related to height andhorizontal distance is plotted to the left.According to the model, during incident 1, with the winds from the west,there is an increase in wind speed at the ridge and a sharp decrease in windsspeed on the leeward side. The wind speed increases with increasing heightdue to the forcing of the air masses around the ridge. Moving towards theleeward side gives a decrease in modelled wind speed of 90 % and a decreasein measured ice growth towards zero on sticks 5 and 6, indicated clearly byprofile 2 (figure 10). With southerly winds, as during incident 2, there is no87

such simulated speed up at the ridge, or any sharp decrease in wind speed atthe leeward side of the ridge as in incident 1 (figure 9). A slight speed up atthe ridge is most apparent in profile 2, while profile 3 shows approximatelyno variation. The last part of this study is concerned with the testing of themodels presented in papers 2 and 3. Ice load data collected at Mt. Brosviksåtaand Mt. Gaustatoppen during the winter 2003/2004 is evaluated.During the first incident of icing, the measurements show a decrease in icingintensity with increasing height in profile 1, from 0.022 Kg m -1 hr -1 at stick16, to 0.002 Kg m -1 hr -1 at stick 1 (figure 10). Likewise there is a decreasingtrend in intensity from the west to the east side of the ridge. Themeasurement of the sticks in profile 2 and 3 gave maximum intensity forlocations on the upstream-western side of the ridge, with 0.008 Kg m -1 hr -1 atstick 2, and 0.027 Kg m -1 hr -1 at stick 9. At the same time stick number 5 and6 at the downstream-eastern side of the ridge of profile 2 had no accretion ofice, and stick 13 in profile 3 had 0.005Kg m -1 hr -1 .Simulated icing intensity during incident 1 corresponds well with measuredintensity (figure 10). The input parameters for this simulation were estimatedLWC at 1800 m a.s.l given by the method by Drage and Hauge (2004). Thewind speed was estimated at each stick by applying the wind speed ratiogiven by the CFD-solver, to the wind speed at 850 hPa in Hirlam10. LWCwas estimated each ten minutes, while HIRLAM10 gave data only each sixhours. A linear interpolation of wind speed to ten minute values wastherefore performed. The droplet concentration was assumed constant withheight above cloud base, equal to 113 cm -3 , given by Gjessing and Skartveit(1990). Furthermore, a calculation of the collision efficiency at each stickwas performed using the method given by Finstad et al. (1988a) andassuming a constant cylinder diameter equal to 3 cm. The decreasing trend inicing intensity towards the east of profiles 2 and 3 can be identified in boththe measurements and estimates (figure 10). An overestimate of icingintensity at stick 2, 3 and 4 of profile 2 is evident. At the same time, stick 5and 6 have practically no estimated ice, identical to what was measured. Evenmore interesting is the increasing trend in intensity in profile 1 down slope,both in the measured and estimated values. The method described hereoverestimates the icing intensity by an average of 35 %.During the second incident of icing, there was no pronounced difference inicing intensity between the east and the west side of the ridge for profiles 2and 3, as it was observed in incident 1 (figure 11). Stick number 2 and 5 inprofile 2 show an ice intensity of 0.036 Kg m -1 hr -1 and 0.029 Kg m -1 hr -1 ,respectively, while the intensity at stick number 9 and 13 in profile 3 is88

shown to be 0.030 Kg m -1 hr -1 and 0.033 Kg m -1 hr -1 , respectively. It shouldbe mentioned here that the measured data from stick number 6 in profile 2 ismissing in the second incident. Both profile 2 and 3 show maximum icegrowth on the edge of the mountain ridge represented by stick 3 and 10, withice growth of 0.054 Kg m -1 hr -1 and 0.065 Kg m -1 hr -1 , respectively. Inaddition, profile 1 shows no clear trend in ice accumulation with increasingheight, varying from 0.029 Kg m -1 hr -1 at stick 8 to 0.054 Kg m -1 hr - 1 atstick3.0.1Profile 1MeasuredSimulated11860Ice growth (Kg/m hr)0.080.060.040.02037810141516182017801740Height a. s. l. (m)1 3 7 8 10 14 15 160 80 160 240Ice growth (kg/m hr)0.10.080.060.040.02Profile 22 3 456186018201780Height a. s. l. (m)017402 3 4 50 80 160 2400.1Profile 31860Ice growth (kg/m hr)0.080.060.040.029 10 11 121318201780Height a. s. l. (m)017409 10 11 12 13Stick number0 80 160 240Distance betweensticks (m)Figure 11. Plot of measured and estimated ice growth rate on each stick in theterrain during icing incident 2. Locations of the sticks related to height andhorizontal distance is plotted to the left.89

Observations indicate that the “diameter” of the ice vane decreases with time(Figure 4). Thus, the icing intensity also decreases with time, according tofigure 1. The estimates of icing intensity performed in this study are based onthe assumption that the cylinder diameter is a constant 3 cm throughout bothicing incidents. An overestimate of ice intensity should therefore be expectedin both icing incidents. The difference between measured and estimated icingintensity should also be expected to increase with increasing accumulated iceload on the sticks. This is confirmed in this study where an increase inaccumulated ice loads from incident 1 (figure 10) to incident 2 (figure 11)gave a larger overestimate of icing intensity.The simulated instantaneous icing intensity on the whole mountain peak isshown in figure 12a and 12b, for selected situations during the twosimulation cases. This simulated icing intensity is indicating the variation inintensity relative to wind speed and wind direction.For both cases calculated cloud base temperature of –6.7°C and cloud baseheight of 1505 m a.s.l, was chosen, resulting in a LWC of 0.22 (g/m3) at1800 m a.s.l.The fact that the simulated wind speed decreases slower than the ice growthrate might be explained by the variation in cloud liquid water content aroundthe ridge. Icing onto the ground on the windward side of the ridge is a sink inthe cloud LWC. In addition, the downward motion on the leeward side leadsto evaporation.The Median Volume Droplet diameter (MVD) has proven successful as theparameter to be applied when calculating the collision efficiency given theequations by Finstad et al. (1988b). No continuous measurement of thedroplet spectrum or droplet concentration was performed in this study. Usingmean volume droplet diameter would in most cases lead to an underestimateof the collision efficiency (Finstad et al, 1988b). In nature, two factors can beconsidered to reduce the difference between mean and median volumedroplet size. First, the droplet size spectrum gets narrower, as the dropletsgrow by condensation of steady supersaturation (Rogers and Yau, 1989).Second, the spectrum is measured to have a majority of large droplets, givinga MVD more equal the mean volume droplet size (Schemenauer et al. 1980).The mean volume droplet size is therefore applied in this study, even if aslight overestimation should be expected in most cases.90

6 6 3 5 8 0 06 6 3 5 6 0 06 6 3 5 4 0 066 3 5 2 0 06 6 3 5 0 0 06 6 3 4 8 0 06 6 3 4 6 0 06 6 3 4 4 0 06 6 3 4 2 0 0WIND DIRECTION6 6 3 4 0 0 0479400 479600 479800 480000 480200 480400 480600 480800 481000 481200Figure 12a. Simulated icing intensity dM/dt (kg/s) in simulation case 1.LWC is 0.22 (g/m3) at 1800 m a.s.l. The scales on the axis are in meter.91

Figure 12b. Simulated icing intensity, dM/dt (kg/s) in simulation case 2. LWC is 0.22(g/m3) at 1800 m a.s.l. The scales on the axis are in meters.92

4 DISCUSSIONThe ice accumulations on the sticks at Gaustatoppen showed a strongdependence upon the actual location of the sticks. The main reason for thiswas the local variation in wind speed related to the topography. At winddirections perpendicular to the mountain ridge, the increase in wind speedover the ridge thereby causes an increase in ice load on the sticks located onthe windward side of the ridge. This was verified by the first icing incident,both by measurement and simulation.At wind directions parallel to the mountain ridge, there were no apparentzones showing increases or decreases in wind speed. Both measurements andsimulations showed an increase in ice load on the sticks on the edge of theridge, but no obvious relationship with height was identified.The reason why the icing rate decreased rapidly with decreasing wind speedcan be explained by equation 1. The wind speed and the LWC affect thisequation in two ways. These parameters are both used directly in equation 1,but are also used to recalculate the collision coefficient α 1 . It is easily shownthat a reduction in wind speed or LWC will give a reduction in the collisionefficiency.5. CONCLUSIONThe present measurements have shown that even small variations in thelocation of sticks around the ridge of a mountain peak can cause largevariations in accreted ice onto these sticks. Existing methods for estimatingdesign load for construction in such harsh environments are based upon theactual height of the site above sea level, together with climatologically datafrom synoptic weather stations or airports. To improve the estimates of in-icing in the vicinity of the mountain peak several efforts may becloudundertaken. In this case, the application of micro scale numerical models, likecomputational fluid dynamics, to describe the wind field around the mountainpeak, or measurements at the location for a short period of time, gavevaluable information. The study of the icing around the building has shownthat local sheltering might reduce the in cloud icing by up to 100 %. A basicunderstandingof the air flow around isolated mountain peaks is vital to93

understanding how complex topography and changing wind directionstrongly influence icing intensity.AKNOWLEDGEMENTSForsvarsbygg, Norkring, Statnett and Telenor funded this study. The authorswish to thank Øistein Saugerud and Øyvind Leikvin for inestimable fieldsupport.REFERENCESCham inc., 2004, Phoenics 3.5 user guideDrage, M. A. and Hauge G., 2004. Atmospheric icing in a coastalmountainous terrain – measurements and numerical simulations – a casestudy. Cold Regions Sc. and Tech., In Press.Drage, M. A. and de Lange, T., 2005. Instrumentation for measuringatmospheric icing. Reports in meteorology and oceanography, Report No.2-2005.Finstad, K. J., Lozowzki, E. P. and Gates, E. M., 1988a. A computationalinvestigation of water droplet trajectories. J. Atmos. Oceanic Technol. Vol5. 160 – 170.Finstad, K. J., Lozowzki, E. P. and Makkonen, L., 1988b. On the medianvolume diameter approximation for droplet collision efficiency. J. Atmos.Sci. Vol 45. 4008 – 4012.Finstad, K., Fikke, S. M., and Ervik M., 1988c. Meteorological and cloudphysical observations of atmospheric icing events on Gaustatoppen.Proceeding 4 th International Workshop on Atmospheric Icing ofStructures, 61 - 64.Gjessing, Y. T., Skartveit, A., Utaaker, K., 1990. Vurdering av sikt- ogvindforhold på Hurumåsen. Meteorological Report Series University ofBergen. Nr. 1, 1 – 49.Harstveit, K., 2002. Using routine meteorological data from airfields toproduce a map of ice risk zones in Norway. Proceeding 10 th InternationalWorkshop on Atmospheric Icing of Structures, CD: Session 8-1.Makkonen, L. and Stallabras, J. R., 1987. Experiments on the cloud dropletcollision efficiency of cylinders, J. of Clim. Appl. Meteor., 26: 1406 –1411.Makkonen, L. and Ahti, K. 1995. Climatic mapping of ice loads based onairport weather observations, Atmos. Res., 36: 185 – 193.94

Makkonen, L., 2000. Models for the growth of rime, glaze, icicles and wetsno won structures. Phil. Trans. Roy. Soc., Nr. 1776, Vol. 358 2913 –2939.Sundin, E. and Makkonen, L. 1998. Ice loads on a lattice tower estimated byweather station data. J. Applied Meteorology, Vol. 37(5), 523 – 529.Raastad, H., 1958. Probes icing on overhead lines. Electric light and power,42 – 51.Rogers, R. R. and Yau, M. K., 1989. A short course in cloud physics,Pergamon Press.Schemenauer, R. E., Macpherson, J. I., Isaac, G. A., and Strapp, J. W., 1980,Canadian participation in HIPLEX 1979. Report APRB 110 P 34,Atmospheric Environment Service, Environment Canada, 206 pp.Stull, R. B., 1988. An introduction to boundary layer meteorology. KluwerAcademic Publishers.Versteeg, H.K., Malalasekera, W., 1995 An introduction to computationalfluid dynamics – The finite volume method, Longman Scientific&TechnicalYakhot, V., Orszag, S.A., Thangam, S., Gatski, T.B., Speziale, V.G., 1992,Development of turbulence models for shear flows by a double expansiontechnique, Phys. Fluids A, Vol. 4, no. 7, pp. 1510-152095

ARCTIC COASTAL CLIMATIC IMPACTON DESIGN CONSTRUCTION ANDOPERATION OF THE HAMMERFESTLNG PLANTM. A. Drage 1 and Truls Mølmann 21 University of Bergen, Bergen, Norway2 Barlindhaug Consult AS, Tromsø, NorwayABSTRACTLinde AG, Germany, is constructing a Liquid Natural Gas (LNG) plant forStatoil on the island Melkøya outside Hammerfest, Norway. A study of theclimatic factors has been carried out to give input to design, construction andoperational aspects of the Hammerfest LNG Plant. Theoretical modelling, onsite measurements and observations of icing and snow conditions during the1997 – 1998 and 2001 - 2002 winter seasons have been performed. Sea sprayicing must be considered for a 25 m zone from the shoreline, includingjetties. Maximum sea spray ice thickness has been estimated to 75 cm.Eccentric loading situations have to be taken into account. Atmospheric icingcaused by rain or wet snow causes eccentric loads, and the maximum builduphas been estimated to 12 cm of ice. A sudden increase in the mean wind tomore than 25 m/s due to polar lows passing Melkøya, with a frequency ofapproximately one polar low about each 2 years has been expected.INTRODUCTIONThe Hammerfest LNG plant is situated on the island of Melkøya (70° 38’North, 23° 33’ East, area 3 km 2 , highest point 70 m a.s.l). This paper resultsfrom studies and evaluations of climatic data obtained from the NorwegianMeteorological Institutes synoptic weather stations at Fruholmen Lighthouse,and Hammerfest Radio (1957-1998), and to some extent meteorologicalmeasurements and observations on the site during the 1997 – 1998 and 2001– 2002 winter seasons. This icing and snow drift study defines the expectedfrequency and severity of specific winter climate events, such as extreme97

wind, icing and snow drift, which should be considered in design,construction and operation of the Hammerfest LNG Plant.CLIMATIC FACTORSIcing types and propertiesIcing is classified according to the different formation processes:1. sea spray icing2. atmospheric icing (precipitation icing, in-cloud icing, negativesublimation).According to NORSOK N-003, the density of sea spray ice is defined in theorder of 850 – 900 kg/m 3 . Typical properties of accreted atmospheric ice areglaze (900 kg/m 3 ), wet snow (300-600 kg/m 3 ), hard rime (600-900 kg/m 3 )and soft rime (200-600 kg/m 3 ) (ISO 12949).Ice loadsIce loads on structures at Melkøya are a function of parameters such as:• the icing process (sea spray, wet snow, freezing rain etc)• surface temperature of the construction/ structure• the slope and direction of the surface• distance from shoreline, etc.Uneven distributions of ice must be considered. Ice caused by sea spray, wetsnow and rain will cover all surfaces facing the wind. It may be assumed thatthe ice will cover half the circumference of tubular structures.Design ice loads are based on operational requirements for the removal of iceaccumulated during an icing period. Icing values are then to be considered asmaximum design values for one single icing incident and can in this contextbe considered as a 100- year return period load.Weather dataAll analyses are primarily based on weather data from the synoptic weatherstations Fruholmen Lighthouse (1957 – 1998) and Hammerfest Radio (1957– 1987). The Hammerfest Radio meteorological station is locatedapproximately 3 km South East of Melkøya, 70 m above sea level. TheFruholmen Lighthouse is located 45 km in northerly direction. Weatherstatistics from neither of these locations fully represent the conditions atMelkøya. Evaluations based on local knowledge, Finite Element Model(FEM) simulations and to some extent measurements have therefore beenmade to describe the meteorological conditions of significance to the98

construction and operation of an LNG plant. Table 1 gives an overview of theannual and seasonal weather conditions at Fruholmen Lighthouse.Snow statisticsPrecipitation has never been recorded at Melkøya, and the 2001-2002 wintermeasurements gave only limited snow data. The nearest locations with longtime records of snow precipitation are Hammerfest Radio and FruholmenLighthouse. The Fruholmen Lighthouse data are assumed to be morerepresentative for the Melkøya conditions than Hammerfest Radio because ofless local topography effects. Figure 1a shows that the main wind directionfor snow events is in the 340-10 sector. The events with freezing rain or wetsnow come from the 250-10 sector. For these wind directions the airtemperatures at Fruholmen can be regarded as representative for Melkøya.Table 1: Overview of the annual and seasonal weather conditions at FruholmenLighthouse. The analysis is based on data from 1957 – 1987 for air temperature andwind conditions, and on data from 1957 – 2001 for precipitation.annual winter spring summer autumnMean wind speed [m/s] 9,0 10,5 8,9 7,3 9,2Standard deviation [m/s] 4,8 5,3 4,6 3,9 4,9Median [m/s] 8,2 10,3 8,2 7,2 8,2Over 10 m/s [%] 37,9 50,1 37,5 23,9 40,2Maximum wind speed [m/s] 35,0 35,0 31,9 25,7 34,0Mean air temperature [ºC] 2,8 -2,1 0,8 8,7 3,8Min air temperature [ºC] -17,4 -17,4 -14,3 -1,9 -9,1Max air temperature [ºC] 26,1 11,0 21,7 26,1 16,6Mean precipitation [mm] 1098,5 315,3 196,5 217,3 356,4Min precipitation [mm] 673,4 136,9 82,4 87,7 117,6Max precipitation [mm] 1630,4 571,5 350,8 426,9 618,1Snow driftThe shortest distance between the island Melkøya and the main islandKvaløya is about 100 m. This fetch is too long for any transport of driftingsnow onto Melkøya. Our measurements during the winter season 2001-02indicate that about 50 % of the snow precipitation will be transported awayfrom the island. Hence, Melkøya is probably the best location in the regionfor the LNG plant as regards the potential problems of snow and driftingsnow.99

all precipitation330030rain330030330030330030300603006030060300602700 900 1800902700 700 1400902700 200 400 600 800902700 100 200 30090240120240120240120240120210180150210180150210150180Zero celcius degrees210150180-5 celcius degreesfreezing rain330030snow330030330030330030300603006030060300602700 200 400902700 300 600902700 20 40902700 1 2 390240120240120240120240120210180150210180150210150180-10 celcius degrees210150180-15 celcius degreesFigure 1a. Wind direction duringprecipitation events at FruholmenLighthouse (1957 – 1987). Rain T >2°C, freezing rain and wet snow 2°C

Fig. 2. Overall plan of snow drift areas on the Hammerfest LNG-plant.Elevated structures with ground clearance (constructions on poles) arerecommended because they create higher wind speeds close to the groundwhich cause the erosion of deposited snow. The following comments withrespect to estimated snow drift formations are related to the defined zones onthe Hammerfest LNG plant site plan.• A: Slug catcher area. The pipeline system will in principle collect driftingsnow, but the rate of deposition will be strongly reduced due to thehorizontal orientation of the pipes and ground clearance. The groundsurface underneath the pipeline “roof” should be as smooth as possible.• B: Breakwater. The breakwater will collect snow on the lee (South) side.Side slopes less than 1:4 will reduce or remove the problems along thetoe of the fill.• C: Piperacks/ Air separation unit/ Sea water outlet. The area consists of avariety of installations, structures and buildings. Snow drifts will formover the whole area. Bigger installations like buildings will generallyhave snow deposition on the lee side (South). Small depositions may alsobe expected 2-3 m from walls parallel to the N-S direction.101

• D: Smaller tanks. Snow drifts will form between the tanks in the E - Wdirection while erosion will take place in the N - S direction.• E: Process barge. Heat loss from the process equipment will reduce theproblems of snow drift because of melting and run off. Heavy snowstorms, however, may fill voids and openings in exposed areas(predominantly facing North).• F: LNG/ LPG/ condensate storage tanks. Snow drifts will form betweenthe tanks in the E – W direction while erosion will take place between thetanks in the N – S direction. Constructions on the top of the tanks mayinitiate formation of snow drifts as described in section 4.4.• G: Jetties. Installations on the jetties may collect snow on lee (South)side.• H: Future expansion area. The high South facing rock faces will havelarge depositions of snow at plant level. At the base of the rock face thethickness of the snow deposition may be of several meters. Overhangingsnow drifts will also form on the top.• I: Admin building/ Camp area. The parking area will give limited snowdrift problems if cars are aligned as indicated and no permanent obstaclesare placed on the open square. Buildings will experience snow driftformations on the lee side and doors and openings should generally beplaced on East or West facing walls. The administration building willcollect snow on the SE side with a snow drift starting on the NW corner.• J: Roads. Road constructions should be elevated from adjacent terrain.This will reduce deposition of snow on the road itself. The road along thebog deposition and breakwater will experience frequent snow driftformations. The problem is reduced if side slopes of adjacent terrain areless than 1:6. There should also be an open space between roadembankments and adjacent fill or cut sections to provide for snowdeposition when snow clearing is necessary.Polar lowsLow-pressure systems of different scales and of different intensities appearfrequently in the Northern oceans. The definition of a polar low is that itforms in cold maritime air, has diameter of between 200 and 500 km andwind speeds in excess of 15m/s. Polar lows only form and develop over seain arctic regions in the winter season. As soon as they make landfall they dieout rapidly. A polar low is rather short-lived, typically from less than a dayup to 2 days, and it is difficult to predict its path, intensity and duration.However, one way of discovering a polar low even at its early stage is byinspection of satellite images.102

The passage of a polar low at a specific location can take several hours. Asthe low approaches a location, there is a sudden increase in the mean wind tomore than 25 m/s, with gusts of hurricane force. During the passage of thelow, the wind direction changes by about 180º. The very sudden windincrease and change in wind direction within just a few hours arecharacteristic for polar lows. Also characteristic is that the wind is strongestnear the surface, and decreases gradually upwards. When a polar lowapproaches there is a rather strong temperature increase. Within a few hours,the temperature will typically increase from around -10 ºC up to a fewdegrees above freezing. After the passage, the temperature will gradually fallagain. The frequency of polar lows passing Melkøya during the last 20 yearsis approximately one polar low every second years.MODEL STUDY OF WIND CONDITIONS AROUND MELKØYAMeteorological data has been collected at Melkøya for 2 years, which shouldprovide an indication of the typical conditions. The data have been collectedusing standard weather stations, in addition to specialized equipment toaccess icing. The main purpose of the model study was to complement theobservations by providing more detailed information about the threedimensionalstructure of meteorological fields, for example: wind profiles,horizontal distribution of the wind field around Melkøya, vertical velocities,turbulent kinetic energy and turbulent length scale, and stability of theatmosphere close to the surface.A secondary objective was to evaluate the reproducibility of the SYNOPstation at Fruholmen Lighthouse as compared to those at Melkøya. Theanalysis consists of 3 nested models, HIRLAM 10, MC2 and SAFRA. Anexample from SAFRA is given in Fig. 3.103

Fig. 3. The east case 3.3.2000, SAFRA: East-west profile over Melkøya for Turbulent KineticEnergy (TKE) and turbulent lenght scale (TLS). The horizontal and vertical scales are in km.Melkøya is situated between 3 and 4 in the figure.SEA SPRAY ICING ON MELKØYASea spray icing on Melkøya normally occurs when a high pressure system istransporting cold air from inland out to the coastal area. Weather data fromFruholmen Lighthouse shows that during the last 40 years there has beenonly one event lasting as long as 36 hours with air temperatures less than –10°C and strong southerly winds. It should be noted that the air temperatureat Melkøya is expected to be in the order of 2°C lower than at Fruholmenduring weather situations like this. Nevertheless, it is a fact that thecombination of low air temperatures and strong winds is rather rare in theregion. Normally all cold air which is stored inland will be transported away104

during a period of 12 – 24 hours giving weather situations with surface windsstronger than 10 m/s.Observation in 1998During the winter of 1998 a major sea spray icing build up was recorded atMelkøya. Sea spray icing as severe as documented in Figure 4a has neverbeen observed before. Since this may be considered as a worst case sea sprayicing scenario, it has been important to be able to model the situation usingthe available long term meteorological data (primarily from FruholmenLighthouse). Then, all other sea spray icing situations can be predicted withinacceptable tolerances both with respect to severity and frequency. Thephotograph was developed in March 1998, but no written documentationexists as to exactly when it was taken.Observation in 2002Due to the 1998 documented near shore sea spray ice build up, it was statedas important in Multiconsult/Barlindhaug’s work scope to try to documentsimilar incidents and three video cameras where mounted to observe seaspray icing during the 2001 – 2002 winter. Figure 4 b) shows sea spray icingon January 26, 2002 on the same spot as shown in the 1998 photograph.During the period from January 23 to 27, 2002, there were two days withtemperatures in the range of – 9 º C to – 11 º C and wind from south with windspeeds up to 18 m/s. This weather situation lead to an ice growth rate in theorder of 30 cm in a zone 25 m from the shoreline during 48 hours, andmaximum ice thickness was measured to approximately 30 cm on January27.a) b)Fig. 4. Icing caused by sea spray on the existing breakwater on Melkøya in:a) 1998 (left) and b) 2002 (right). The same gate structure is observed on both photos.(Photo: a) Ragnhild Normann – Spring 1998 and b) Ola Brandt – January 26, 2002).105

Modelling sea spray using Mertin’s diagramMertin’s diagram for icing has been used for a rough estimate of ice growthrates on Melkøya (Mertins, 1968). The estimates are based on the followinginput from Fruholmen Lighthouse: wind (1957-1987), air temperatures(1957–2000), and sea surface temperatures (1968-2000). The Mertin’sdiagram categorizes icing as:(C)Air temperature (Deg. C)-14-10-6-2LightIcingHeavy IcingModerateIcingNo IcingExtreme Icing• Light icing1 – 3 cm/24 hours• Moderate icing:4 – 6cm/24 hours• Heavy icing:7 – 14 cm/24 hours• Extreme icing:≥ 15 cm/ 24 hour10 14 18 22 26 3034(m/s)Wind Speed (m/s)Fig. 5. Diagram used for icing calculations due to sea spray (based on Mertin’s diagram,1968).Frequency analysisA frequency analysis of the Fruholmen data showed a main trend of sea sprayicing from southerly directions (Fig. 1b). It should be noted that simultaneouswind measurements at Fruholmen Lighthouse and Melkøya showed that windfrom the South East sector may be stronger at Melkøya than at FruholmenLighthouse. This means a slight underestimate of the frequency of incidentsat Melkøya. The frequency of wind speed stronger than 15 m/s and airtemperatures lower than -8 ºC at Fruholmen Lighthouse during the period,has been estimated to 2.45 per year (1957-1987).106

The frequency analysis gave incidents of heavy icing (7-14 cm/24 h) in 13winters of the 43 years of data record from 1957 to 2000. Only in 1966 and1998 were there as many as three days during the year with heavy icing,while January 1999 showed the longest continuous period with heavy icing;24 - 30 hours. For 1998 the analysis shows heavy icing on February 9 (7PM), February 14 (1 PM and 7 PM) and February 15 (1 AM). The 2002 icingincident was estimated using Mertin’s diagram. The results showed an icingrate of “extreme icing” (> 15 cm/ 24h). The ice accumulation estimated withMertin’s diagram fit well with observed icing as documented in the photosfrom the 2002 incident. This leads to the conclusion that estimates of icingbased on Mertin’s diagrams conform well with the icing conditions on thebreakwater of Melkøya. Mertin’s diagram was therefore applied in this study.ATMOSPHERIC ICINGBoth theoretical calculations and on-site measurements of atmospheric icingat Melkøya have been carried out. In the initiating phase of the icing study itwas assumed that surface temperatures of the structures will be higher than -15°C. Three cooled cylinders with different surface temperatures wheremounted (Fig. 6). The general conclusion is that atmospheric icing will createless significant ice accumulation with a frequency of incidents less than thatof sea spray icing. There will be negligible atmospheric icing defined asdeposition (negative sublimation) or accretion (freezing of super cooledcloud drops) on surfaces with temperatures equal to ambient air temperatures.Icing on structures caused by freezing rain or wet snow accretion will occuron surfaces predominantly facing South West to North. Even though no iceof this type was measured on the cylinder with surface temperature equal toambient air temperature, statistical evaluation of climatic data supports theconclusion of maximum ice build up to 12 cm.107

Fig. 6. Riming measurement cylinders at Melkøya, with surfacetemperatures -15°C (left cylinder), 5°C lower than ambient airtemp. (middle cylinder), and -5°C (right cylinder).CONCLUSIONSA study of icing and snow drift conditions at Melkøya has been carried out inorder to review and supplement the Statoil Metocean Design Criteria. Thestudy consists of theoretical modelling and to some extent on-sitemeasurements and observations of icing and snow conditions during the 2001– 2002 winter.Observations during the 1998 winter show an excessive sea spray ice buildup on the old Melkøya breakwater, which indicates that icing problems canbe far more severe than normally expected on near shore installations inNorthern Norway. The study focused on describing the weather situationleading to the documented ice build up, and based on this, preparingpredictions as to the frequency and severity of the icing.Melkøya is very exposed to wind which reduces potential snow and snowdrift problems at the LNG plant. Wind conditions has also been modelled andmeasured. Results were used as input for the estimates of icing and snowconditions. In addtion, they were used as a basis for a general review of thedefined wind profile for wind load predictions.108

The major conclusions of the study are as follows:• potential icing problems must be addressed in the plant operationalprocedures• all calculation are based on the assumption that all weather exposedplant equipment, components, structures or buildings will havesurface temperatures equal to ambient air temperature• sea spray ice has to be considered for a 25 m zone from the shoreline(including jetties). No sea spray ice build-up will be considered morethan 25 m from shoreline. Maximum sea spray ice thickness isestimated to 75 cm. Eccentric loading situations have to be taken intoaccount• atmospheric icing caused by rain or wet snow will cause eccentricloads, and the maximum build-up will be 12 cm of ice• due to the compact and complicated geometry of the process area,falling ice will form only limited design input• maximum design icing values are estimated• all structures are in general to be designed according to theNorwegian Standard series NS 3491 – Basis of design and actions onstructures; part 3: Snow Loads• overhanging snow drifts may form on top of tall rising structures likecolumns or storage tanks. Traffic zones or installations should beavoided or sheltered beneath exposed areas• blowing snow will tend to fill voids in open structures andinstallations• polar lows passing approximately every 2 year will give a suddenincrease in the mean wind speed to more than 25 m/s.109

REFERENCESInternational Standardization Organization, ISO 12494:2001(E),Atmospheric Icing of Structures.Mertins, H.O., (1968). Icing on fishing vessels due to spray. London,Mar.Cosr., 38, pp 128-130.Norsk Standard NS 3491- 1. Basis of design and actions on structures; Part 1:Densities, self weight and imposed loads. Norsk Standardiseringsforbund(Norwegian Standardization Organization) Oslo, Norway, 1. Edition,December 1998, 22 p.Norsk Standard NS 3491- 3. Basis of design and actions on structures; Part 3:Snow loads. Norsk Standardiseringsforbund (Norwegian StandardizationOrganization) Oslo, Norway, 1. Edition, March 2001, 37p.NORSOK STANDARD N-003. Actions and Action Effects. NorwegianTechnology Standards Institution. Oslo, Norway, 1 st of February 1999, p.81.Statoil Metocean Design Criteria. Hammerfest Terminal at Melkøya –StatoilReport # 98S97*5191 – Rev. No.: 5. Stavanger, Norway, 16 th . February2001, 53 p.110

5. Testing of modelsThe last part of this study is concerned with the testing of the modelspresented in papers 2 and 3. Ice load data collected at Mt. Brosviksåta andMt. Gaustatoppen during the winter 2003/2004 is evaluated.5.1 BrosviksåtaAt Brosviksåta data were continuously recorded between October 10, 2003and May 1, 2004, and several incidents of icing occurred (figure 5.1). The icescale was mounted with a 14 cm diameter one-meter high non-rotating rodduring the period October 7, 2003 to February 23, 2004. On February 23,2004, a rotating cylinder of 3 cm diameter replaced the 14 cm cylinder. Forall cases, the ice was assumed to have fallen off when the air temperature washigher or equal to 0 ºC.Measured ice-load 07.10.2003 - 01.05.200412Ice load (kg)84015.10 12.11 10.12 07.01 04.02 03.03 31.03 28.04Date (dd.mm)Figure 5.1: Recorded ice load on a one-meter high rod at Brosviksåta 723 ma a.s.l.during the period October 10, 2003 to May 1, 2004.111

5.1.1 Non-rotating cylinderA non-rotating rod was chosen due to the fact that most constructions likemasts, buildings, antennas etc. are rigid. Equations for calculation of collisionefficiency with a cylinder are given by Finstad et al. (1988a) and aretherefore generally not applicable in these cases. In theory, the ice issupposed to create a vane on the windward side of a non-rotating cylinder,which was confirmed by inspection on February 23, 2004. The shape anddirection of the vane is strongly dependent upon variations of the winddirection. Brosviksåta, which is situated on the coast, experiences a highvariation in air temperature, wind speed, wind direction, precipitation rateand humidity. The incidents of icing are however of relatively short duration(days), due to the fact that the air temperature often rises above 0 ºC at theend of an icing incident. As a result, the ice then falls off.Examination of the different icing incidents with the non-rotating cylindershows that the wind direction is stabile during any given icing incident withinan interval of ±15 degrees.The width of the ice vane decreases as the ice forms, creating a peakpointing towards the wind. This can be interpreted physically as a decreasingeffective cylinder diameter as the ice continues to form. The theoreticalcalculation of change in collision efficiency with varying cylinder diameter isgiven in figure 2.2 in chapter 2. Given a wind speed of 10 m/s, a LWC of 0.4g/m 3 and a droplet concentration of 100 pr/m 3 , this figure shows that the icingintensity (kg/m hr) has a maximum when the cylinder diameter isapproximately 3 cm, decreasing rapidly with decreasing cylinder diameterand decreasing slowly with increasing cylinder diameter. Assuming aconstant diameter of 14 cm, the collision coefficient variations are a functionof wind speed, LWC and droplet concentration. During an in-cloud icingincident with only minor changes in wind direction, the width of the ice vanewill decrease. Thus, an underestimate of icing is expected until the width ofthe ice vane decreases towards the critical diameter given by figure 2.2. Anoverestimate should be expected when the wind direction changes during anicing incident.5.1.2 MethodsLWC is estimated by the method described in the introduction and also indetail in Drage and Hauge (2004) (paper 2). For a fixed diameter of thecylinder, and assuming a constant droplet number of 113 drp/m 3 , the icingintensity will vary as a function of LWC and wind speed. For a constantdroplet concentration, an increase in wind speed or LWC will both lead to anincrease in collision efficiency, according to the theory by Finstad et al.112

(1988a) (figure 2.2). In the following three methods for ice-load estimates areoutlined.Method 1:Wind speed, droplet concentration, droplet size (Median Volume Droplet)and the dimensions of the icing object, all control the collision efficiency.Assuming constant droplet concentration and object dimensions, wind speedand droplet size are the controlling parameters. For a constant dropletconcentration, the MVD is only a function of LWC.A simple approach for calculating the collision efficiency might be expressedby the formulaα = k ⋅ ρ 1v(5.1)LWC ⋅where α is the collision efficiency, k 1 is a constant (g/m 2 s), ρ LWC is the liquidwater content (g/m 3 ) and v is the wind speed (m/s). In this model the collisioncoefficient increases linearly with increasing wind speed and/or LWC. Anincrease of LWC or wind speed by a factor of two will similarly increase thecollision coefficient by a factor of two. This is a slight overestimatecompared with the theoretical method of Finstad et al. (1988a). The constantk 1 is adjusted to give the best fit against the measured icing at Brosviksåtaduring the winter 2003/2004, and was found equal to 0.0225 (g/m 2 s).Method 2:Evaluation of an icing incident at Brosviksåta March 20-25 2003 by Drageand Hauge (paper 2) gave a collision efficiency dependent only on LWC:α = k2⋅ ρ LWC(5.2)where α is the collision efficiency, k 2 is a constant (m 3 /g), and ρ LWC is theliquid water content (g/m 3 ). On this occasion, the best-fit constant, k 2 , wasfound equal to 0.225. Evaluation of data for the winter 2003-2004 gave a kequal to 0.255 (m 3 /g), an increase of 13.3 % from the value found in paper 2.Method 3:An approach to the problem of estimating in-cloud icing on a non-rotatingcylinder, is to assume that it is still a cylinder but that the cylinder diameter isdecreasing as a function of accumulated ice. The width of the ice vanedecreases with increasing ice load, when wind direction is assumed to beapproximately constant during the icing incident. This method is notapplicable if the wind direction changes during the icing incident.113

Measurements of the width and weight of the ice vane on the 14 cm diametercylinder on Brosviksåta February 23, 2004, and on the 3 cm diametercylinders at Gaustatoppen April 1, 2003 are presented in figure 5.2. AtBrosviksåta the accumulated ice load on the 14 cm diameter cylinder was 4kg, while the width of the vane was 10 cm at cylinder surface and 3 cm at thetip of the vane. At Gaustatoppen the accumulated ice load on the 3 cmdiameter stick was approximately 4 kg, while the width of the vane was 3 cmat the cylinder surface and 2 cm at the tip of the vane. Based on this the icevane at Brosviksåta is assumed a further ice growth from 4 to 8 kg and at thesame time a decrease in the width of the vane from 3 to 2 cm (figure 5.2).0.1MeasuredLogarithmic fit0.08Width of ice vane (Cyl. dia, D) (m)0.060.040.0200 2 4 6 8Ice load (kg)Figure 5.2. Measured width of ice vane (labels) and a best logarithmicfit (dotted line), based on measurements at Brosviksåta andGaustatoppen.A best-fit logarithmic function is given asD = −1.44⋅ ln( x)+ 5(5.3)where D is the width of the ice vane, and x is ice load in kilograms. Applyingthe equations given by Finstad et al. (1988a) will result in a negative collisioncoefficient for large cylinder diameters. This is obviously wrong and aminimum value of 0.01 is therefore chosen, according to Harstveit (2002).An iteration procedure recalculates the cylinder diameter and accumulated iceat each sampling interval.114

5.1.3 Model resultsFigure 5.3 presents observed ice-loads and ice loads estimated by the threemethods described above.6Des 2003Ice-scaleMethod-1Method-2Method-3Ice load (kg)42Ice load (kg)001.12 06.12 11.12 17.12 22.12 27.12 31.12Date (dd.mm)1284Feb 2004Ice-scaleMethod-1Method-2Method-3001.02 06.02 11.02 17.02Ice load (kg)201510January 2004Ice load - measuredMethod-1Method-2Method-35001.01 06.01 11.01 17.01 22.01 27.01 01.02Date (dd.mm)Figure 5.3. Measured and simulated ice load on a one meter high 14 cm diameternon-rotating cylinder at Brosviksåta at 723 m a.s.l. from December to February2003/2004.115

Comparison the results of three methods presented in figure 5.3, indicatesthat method 2 gives the best result. Method 1 is sometime overestimating andsometime underestimating the ice growth. This can be explained by therelatively high standard deviation in the wind speed ratio. Icing onto the windspeed sensor at the lowest level result in an underestimate of the ice growth.Calculated collision efficiency in method 2 is independent of wind speed.This method is estimating the ice growth better than method 1. A simpleexplanation is that the wind speed is relatively constant during an icingincident, and is therefore not affecting the collision efficiency.Method 3 is generally overestimating the ice growth. An adjustment of themethod of decreasing ice vane width is probably needed for a better fit.5.1.4 Rotating cylinderThe 3 cm cylinder rotates freely. Free rotation means the rod will turn untilminimum drag is achieved, creating a cylindrical shape of ice accretion.Several field observations confirm this theory, showing a cylindrical iceaccretion on the cylinder (Drage and Lange, 2004) (paper 1). The diameter ofthe cylinder, D, increases with increasing ice load, and is by geometry givenby the formulaD = 2 ⋅m2+ r cρ ⋅ h ⋅πi(5.4)where m is the mass of ice in kilograms, ρ i is the density of ice, h is thelength of the cylinder, and r c is the radius of the initial cylinder without ice.The type of ice by in-cloud icing is assumed to be a combination of soft andhard rime, dependent upon wind speed, air temperature and LWC during theicing incident. Calculation of density based upon measurements of size,shape and weight on March 23, 2004, gave a density equal to 500 kg/m 3 . Thecollision efficiency was calculated according to the equations given byFinstad et al. (1988a). Here, measurements of wind speed, relative humidityand air temperature are sampled every 10 minutes at a known level belowcloud base. The amount of ice accreted on the cylinder during a 10-minuteperiod is calculated. This amount of ice is added to the total amount of ice,and a new cylinder diameter, D, is calculated. The new cylinder diameter, D,is then used to recalculate the collision efficiency used in the next 10 minuteperiod. This procedure was repeated during the whole measurement periodwith the 3.0 cm rotating rod, from February 23 to May 1, 2004 (figure 5.4).116

0.63020100Wind speed (m/s)LWC (g/m3)0.40.20108840-4-8Air temp. (DegC)Ice load (kg)642MeasuredEstimated024.02 02.03 09.03 16.03 23.03 30.03 06.04 13.04 20.04 27.04Date (dd.mm)Figure 5.4. Estimated wind speed (m/s), LWC (g/m 3 ) and air temperature (°C) plottedtogether with measured and simulated ice load on a one meter high 3 cm rotating rod atBrosvisåta 723 m a.s.l. from February 24, to May 05, 2004.5.1.5 Assumptions and sensitivity testsLWC is calculated by equation 2.6 (chapet 2), assuming that the actual site ofinterest is above the cloud base. When the height of the cloud base is abovethe height of interest, this results in a negative LWC, which is simplyreplaced by a LWC equal to zero. A plot of the two independently estimated117

and measured parameters, cloud base height and relative humidity, at themountain top Brosviksåta in January 2004, indicates a negative correlation(figure 5.5). The plot shows that for cloud base above the mountain top, therelative humidity increases as the cloud base lowers, and vice versa. Astatistical correlation of these two parameters, from October 07, 2003 to May01, 2004, gives a correlation coefficient equal to –0.90.2500January 200410080604020Rel. hum. 723 m a.s.l. (%)Cloud base (m a.s.l.)20001500100050001.01 06.01 11.01 17.01 22.01 27.01 01.02Date (dd.mm)Figure 5.5. Measured relative humidity (%) at 723 m a.s.l. at Brosviksåta plotted againstestimated cloud base height, during January 2004.Variations in wind speed, droplet concentration and ice density are tested toevaluate the effect on estimated ice load. Among these three parameters, themost uncertain one is droplet concentration. A droplet concentration of 113pr/cm3 was successfully used during a field experiment in eastern Norway(Gjessing and Skartveit, 1990).118

A. Wind speedOn Brosviksåta, the wind speed at 723 m a.s.l., is, on average, higher than at325 m a.s.l. by a factor of 2.2 (paper 2). Three cases of ice load have beenestimated using a wind speed ratio of 1.4, 2.2 and 3.0 (figure 5.6). The ratioof 1.4 gives 40%, while the ratio 3.0 gives 179%, of the ice load given byusing wind ratio 2.2. Compared with field measurements, the best ice loadestimates appeared using wind speed ratio of 2.2.B. Ice densityThe ice density was measured equal to 500 kg/m 3 at February 23, 2004.Variations in ice density result in variations in cylinder diameter, whichthereby affects the collision efficiency. Three cases of ice load have beenestimated using a ice density of 200, 500 and 800 kg/m3 (figure 5.6). An icedensity of 200 kg/m 3 gives 77%, while an ice density of 800 kg/m 3 gives104%, of the ice load using ice density 500 kg/m3. The figure indicates thatat Brosviksåta, a density of 200 kg/m 3 is too low, while a change in densityfrom 500 to 800 kg/m 3 gave only a minor change in ice load.C. Droplet concentrationA droplet concentration of 113 pr/cm 3 is assumed a good estimate in lack ofmore field observations. It is regularly used to calculate droplet size, andthereby collision coefficient (Harstveit, 2002). Three cases of ice load areestimated using a droplet concentration of 73, 113 and 153 droplets m -3(figure5.6). A droplet concentration of 73 m -3 gives 143%, while a dropletconcentration of 153 m -3 gives 78%, of the ice load using a dropletconcentration of 113 m -3 . More interesting here is the fact that a dropletconcentration of 73 m -3 is a better match to the measured results. A dropletconcentration of 113 m -3 might be an overestimate for the average conditionson Brosviksåta.119

Ice load (kg)8642Ice load (kg)MeasuredWind speed ratio 3.0Wind speed ratio 2.2Wind speed ratio 1.40Ice load (kg)8642Ice load (kg)MeasuredIce density 800 kg/m3Ice density 500 kg/m3Ice density 200 kg/m30Ice load (kg)8642Ice load (kg)MeasuredDrp. - cons. 73 cm3Drp. - cons. 113 cm3Drp. - cons. 153 cm3024.02 02.03 09.03 16.03 23.03 30.03 06.04 13.04 20.04 27.04Date (dd.mm)Figure 5.6. Measured and estimated icing with the 3.0 cm rotating rod, from February 23to May 1, 2004, for varying wind speed ratio (upper), ice density (middle) and dropletconcentration (lower).120

5.2 GaustatoppenAt Gaustatoppen data were recorded continously between October 17 andDecember 16, 2003, and between January 24 and May 5, 2004. The ice scale,with a 3 cm rotating rod, was mounted at the peak and several incidents oficing occured (figure 5.7 and 5.8)..5.2.1 Modelled ice load between October 17 and December 16, 2003.Modelled ice load, from the model described in chapter 5.1.4 above andin paper 2, are plotted in figure 5.7. Data, from the station at 1160 m a.s.l.,were applied in order to estimate LWC, air temperature and wind speed at1800 m a.s.l. on Gaustatoppen. It should be noted that the maximumcalibrated load of 100 kg (paper1) was reached on November 28, 2003. Thismakes the accuracy of the measurements from 100 to 150 kg questionable,even if the maximum recordable load is 150 kg.200160Ice load - measuredIce load - model 1Ice load (kg/m)1208040018.10 28.10 07.11 17.11 27.11 07.12Date (dd.mm)Figure 5.7: Measured and simulated ice load on the 3 cmdiameter rotating rod at 1800 m a.s.l at Gaustatoppen during theperiod October 17 to December 16, 2003.121

A. Wind speedThe wind speed on Gaustatoppen, at 1800 m a.s.l., is, on average, higher thanat 1160 m a.s.l. by a factor of 1.9, with a standard deviation of 0.6. Thesevalues were obtained from one year of wind speed data. Incidents with airtemperature ≤0ºC, and relative humidity was ≥96%, were excluded due to thepossible effect of icing onto the sensors. During the period October 17 toDecember 16, 2003, the wind speed sensor stopped several times. This wasidentified as icing onto the sensors, and a linear interpolation betweenmissing data was therefore carried out.B. LWCLWC was estimated by the method described in paper 2. Occasionally, therelative humidity at the lower, (1160 m a.s.l.) station was 100%. Anunderestimate of LWC was then expected in such conditions.Figure 5.7 shows that the model detect all the icing incidents as well as thetrend in the ice growth with good accuracy. The model drop in ice loadNovember 04 correspond well with a sudden decrease in measured ice load.However, the ice load measurements indicates that not all ice is falling off.5.2.2 Modelled ice load between January 24 and May 05, 2004.Results of ice load by the model described in chapter 2 above and inpaper 2, is plotted in figure 5.8. Data from the weather station at Møsstrand,977 m a.s.l., was applied in order to estimate LWC and air temperature.Wind speeds on Gaustatoppen, at 1800 m a.s.l were estimated usingECMWF-model data (European Centre of Medium Range Weather Forecast),with analysis data was given every 6 hours.A. Wind speedWind speed was measured by the sonic wind speed sensor (Gill-instrument,paper 1), from May 30 up until June 26, 2002, at 1800 m a.s.l. atGaustatoppen. A comparison of the modelled (HIRLAM10) wind speed atthe 850-hPa level and the measured wind speed at 1800 m a.s.l. showed thatthe wind speed at 1800 m a.s.l. was on average, higher than that at 850 hPaby a factor of 1.24, with a standard deviation of 0.55. Measured andrecalculated simulated wind speed of this period is plotted in figure 5.9.122

B. LWCThe method for estimating LWC was based upon the assumption of anadiabatic increase in LWC with height. The difference between estimated andactual LWC increases with increasing stability. Applying the method forestimating LWC for situations of high stability result in the LWC beingoverestimated, given the theory outlined in paper 2.Modeled ice load during this period shows a high overestimation (figure 5.8).This is partly due to the fact that no sublimation is considered. The trend inthe ice growth is fairly good reproduced. The ice is modeled to fall of whenthe air temperature at 1800 is eqaul or higher than 0˚C. The high difference inelevation from 977 m a.s.l. at Møsstrand to 1800 m a.s.l. at Gaustatoppenmakes the model sensitiv when the actual air temperature at the top is close to0˚C. This results sometimes in modeled ice falling off like April 06, while themeasured shows no such drop in ice load.123

16012010.50LWC (g/m3)Ice load - measuredIce load - model 1LWC (g/m3)Ice load (kg/m)8040025.01 14.02 06.03 26.03 05.04 25.04Date (dd.mm)Figure 5.8: Estimated LWC (g/m 3 ) and measured and simulated ice load on the 3cm diameter rotating rod at 1800 m a.s.l at Gaustatoppen during the period January24 to May 5, 2004.124

25Wind speed 1800 m a.s.l. - measuredWind speed simulated - EC-model 850 hPaWind speed at Gaustatoppen 1800 m a.s.l. (m/s)2015105031.05 04.06 08.06 12.06 16.06 20.06 24.06Date (dd.mm)Figure 5.9. Wind speed (m/s) measured by Gill-sonic anemometer and simulated byHIRLAM10-850hPa during the period May 31, to June 27 2002.LWC on the mountain top (1800 m a.s.l), estimated from measurements fromGaustatoppen at 1160 m a.s.l. and from Møsstrand at 977 m a.s.l. betweenApril 02 and May 05 is plotted in figure 5.10. Estimated wind speed is alsoincluded in the figure. LWC estimated from weather data at Møsstrand was,in most cases, higher than LWC estimated from the station at 1160 m atGaustatoppen. This can be explained by the effects mentioned above, such ashigh stability or that the station at 1160 m was above the cloud base.Noteworthy is the estimated LWC from the Møsstrand-data from April 30 toMay 2, 2004. No LWC was estimated from the station on Gaustatoppen at1160m. At the same time, a low wind speed was simulated from EC-850 hPa-125

data. Nevertheless, the comparison shows a good correlation betweenestimated LWC by the two different methods.LWC Møstrand (g/m3)LWC GT(g/m3)020Wind speed(m/s)1LWC (g/m3)0.5003.04 13.0423.04 03.05Date (dd.mm)Figure 5.10. Estimated LWC (g/m 3 ) by the weather station at Møsstrand 977 ma.s.l. (blue curve) and Gaustatoppen 1160 m a.s.l. (red curve), plotted wind speed(m/s) simulated by EC-model.126

5.3 Ice detection criteriaA simple test to identify icing conditions on Gaustatoppen between October18 and December 16, 2003, and on Brosviksåta between February 01 andFebruary 23, 2004, is presented in figure 5.11, 5.12 and 5.13. Therequirements for icing conditions is only estimated positive LWC and airtemperature

1601120Ice detectorIce load (kg/m)80Ice load (kg/m)Ice detector400008.11 14.11 20.11 26.11 01.12 07.12 13.12Date (dd.mm)Figure 5.12 Measured ice load and detected icing conditions measured by the icescale and identified by the ice detector between November 08 and December 16,2003 at 1800 m a.s.l. at Gaustatoppen.128

121.2Ice load (kg/m)840.80.4Ice detector001.02 06.02 11.02 16.02 21.02Date (mm.dd)0Figure 5.13 Measured ice load and detected icing conditions measured by the icescale and identified by the ice detectoion between February 01 and February 23, 2004at 723 m a.s.l. at Brosviksåta.129

.130

6. Summary and conclusionThe previous chapters reveal that the research in the area of atmosphericicing by in-cloud icing has aspects that have to be further investigated in thefuture. The theoretical models of ice growth on a cylindrical object are welldocumented as isolated phenomena, such as in a controlled laboratory.However, the limitation of these methods is the need of detailed information,such as the median volume droplet size, and thereby the cloud droplet sizespectrum. The cloud droplet concentration is another parameter that is highlyuncertain. Further investigation is therefore needed.A total of three prototype ice scales for measuring atmospheric icing havebeen built, tested and calibrated. They have been run for, all together, 34months at different locations, both in mountainous regions in southernNorway and a coastal area in the north of the country. The ice scale systemshave proved to provide useful and reliable data for testing and verification oficing models. For vertical force, tests of the ice scale indicated that ice weightcan be measured with an absolute accuracy of approximately 0.125 kg. Thestandard deviation, as calculated in the laboratory, was 0.027 kg.The density (ρ LWC ) of cloud liquid water content (LWC) is an importantparameter for estimating icing on structures as in-cloud icing. A method forcalculation of ρ LWC has been described. The method only requiresmeasurements of air temperature, air humidity and wind speed at a knownlevel at unsaturated conditions. These parameters were measured at differentlevels along the slope of the mountains Brosviksåta (723 m a.s.l.) andGaustatoppen (1882 m a.s.l.). The beginning and end of the icing period wasdetermined to a high degree of accuracy with this method. Reliable prognosesof ρ LWC will greatly improve the procedures of forecasting duration andintensity of in-cloud icing.A mesoscale model (MM5 identified the start and end times of the icingevent with a high degree of accuracy. On the other hand, the accuracy of thesimulated icing intensity was not that good. It is not taken for granted that ahigh resolution forecast would be more accurate than a forecast using acoarser resolution. Further studies of real-time cases on real-time systems atcoarser model resolutions will therefore reveal the capability of MM5 formaking reliable daily forecasts of freezing events.131

Measurements of accumulated ice on a set of sticks around the ridge of amountain peak show that even small variations in the location can cause largevariations in accreted ice. Existing methods for estimating design load forconstruction in such harsh environments are based upon the actual height ofthe site above sea level, together with climatologically data from synopticweather stations or airports. To improve the estimates of in-cloud icing in thevicinity of the mountain peak several efforts may be undertaken. In this case,the application of micro scale numerical models, like computational fluiddynamics, to describe the wind field around the mountain peak, ormeasurements at the location for a short period of time, gave valuableinformation. A basic understanding of the air flow around isolated mountainpeaks is vital to understand how complex topography and changing winddirection strongly influence icing intensity.Observations during the 1998 winter show an excessive sea spray ice buildup on the old Melkøya breakwater, which indicates that icing problems canbe far more severe than normally expected on near shore installations inNorthern Norway. The study focused on describing the weather situationleading to the documented ice build up. Based on this, preparing predictionsas to the frequency and severity of the icing can be done.Melkøya is very exposed to wind which reduces potential snow and snowdrift problems at the Liquid Natural Gas (LNG) plant. Wind conditions hasbeen modelled and measured. Results were used as input for the estimates oficing and snow conditions. In addtion, they were used as a basis for a generalreview of the defined wind profile for wind load predictions.132

7. REFERENCESAasen, S. E., 1995. The Skipheia Wind Measurement Station –Instrumentation, Wind speed profiles and Turbulence spectra, Dr. Scientthesis, University of Trondheim, ISBN-82-90896-89-1.Ahti, K. and Makkonen, L., 1982. Observation on rime formation in relationto routinely measured meteorological parameters, Geophysica, 19(1): 75– 85.Finstad, K. J. and Lozowski, E. P. and Gates E. M.,1988a. A computationalinvestigation of water droplet trajectories, J. Atmos. and Oce, Tech., 5:160 – 170.Finstad, K. J., Lozowski, E. P. and Makkonen L., 1988b. On the medianvolume diameter approximation for droplet collision efficiency. J. Atmos.Sci. 45: 4008 - 4012.Gjessing, Y. T., Skartveit, A. and Utaaker, K., 1990. Vurdering av sikt- ogvindforhold på Hurumåsen. Meteorological Report Series University ofBergen. 1: 1 – 49.Haldar, A. P., McComber, P., Marshall, M. A., Ichac, M., Goel, A. andKatselein, M., 1996. Validation of ice accretion models for freezingprecipitation using filed data, Proceeding 7 th International Workshop onAtmospheric Icing of Structures, pp. 189 – 194.Harstveit, K., 2002. Using routine meteorological data from airfields toproduce a map of ice risk zones in Norway. Proceeding 10 th InternationalWorkshop on Atmospheric Icing of Structures, CD: Session 8-1.ISO 12494, 2001. ISO (the International Organization for Standardization)12494 – Atmospheric icing of structures.Langmuir, I. and Blodgett, K. B., 1946. A mathematical investigation ofwater droplet trajectories. Collected Works of Irwing Langmuir,Pergamon Press, 10: 348 – 393.Ludlam, F. N., 1951. The heat economy of rimed cylinder. Quart. J. Roy.Meteor. Soc., 77: 663 – 666.Makkonen, L., 1984a. Modeling of ice accretion on wires, J. Climate Appl.Meteor., 23: 929 – 939.Makkonen, L. and Stallabras, J. R., 1987. Experiments on the cloud dropletcollision efficiency of cylinders, J. of Clim. Appl. Meteor., 26: 1406 –1411.Makkonen, L. and Ahti, K. 1995. Climatic mapping of ice loads based onairport weather observations, Atmos. Res., 36: 185 – 193.133

Makkonen, L., 1996. Modeling power line icing in freezing precipitation,Proceeding 7 th International Workshop on Atmospheric Icing ofStructures, pp. 195 – 200.Mazin, I. P., Korolev, A. V., Heymsfield, A., Isaac, G. A. and Cober, S. G.,2001. Thermodynamics of icing cylinder for measurements of liquidwater content in supercooled clouds, J. Atmos. and Oce, Tech., 18: 543 –558.Mulherin, N. D., 1988. Atmospheric icing and communication tower failurein the United States. Cols regions Science and Technology, 27: 543 –558.Nicholls, S., 1984. The dynamics of stratocumulus: aircraft observations andcomparisons with a mixed layer model, Q. J. R. Meteor. Soc., 110: 783 –820.Noonkester, V. R., 1984. Droplet spectra observed in marine stratus cloudlayers, J. Atmos. Sci., 41: 829 – 845.Poots, G., 2000. Ice and snow accretion on structures. Phil. Trans. R. Soc.Lond. A (200), 358: 2799 – 3033.Schemenauer, R. S., Macpherson, J. I., Isaac, G. A. and Strapp, J. W., 1980.Canadian participation in HIPLEX 1979. Report APRB 110 P 34,Atmospheric Environment Service, Environment Canada, 206 pp.Sundin E. Makkonen L., 1998. Ice loads on a lattice tower estimated byweather station data. J. Appl. Meteoror. 37: 523-529.134