MM5 Contrail Forecasting in AlaskaMartin Stuefer, Xiande Meng and Gerd Wendler

Geophysical Institute, University of Alaska, Fairbanks

1. AbstractFifth-generation mesoscale model (MM5) is being used for forecasting the atmosphericlayers of aircraft condensation trail (contrail) formation. Contrail forecasts are based on aconventional algorithm describing the adiabatic mixing of aircraft exhaust withenvironmental air. Algorithm input data are MM5 forecasted temperature and humidityvalues at defined pressure or sigma levels, and an aircraft relevant contrail factor that isderived statistically from a contrail observation database.For comparison purposes we introduce a mean overlap (MO), which is a parameterquantifying the overlap between forecasted contrail layers and contrail layers derivedfrom radiosonde measurements. Mean overlap values are used as test for the altitude andthickness of forecasted contrail layers. Contrail layers from Arctic MM5 and AirforceWeather Agency (AFWA) MM5 models agree well with contrail layers derived fromcorresponding radiosonde measurements for certain forecast periods; a steady decrease ofthe MO shows a decrease of contrail forecast accuracy with the increasing forecastperiod. Mean overlaps around 82% indicate reasonable results for the 48 hours forecasts.Verification of MM5 with actual contrail observations shows slightly better performanceof Arctic MM5. A possible dry bias might occur in humidity measurements at lowtemperature levels due to temperature dependence errors of the humidity sensor polymer,which might also affect forecasts of humidity of the upper troposphere or lowerstratosphere. Despite this fact, our contrail verification study shows hit rates higher than82% within forecast periods up to 36 hours using Arctic MM5.

2. IntroductionAircraft contrails are mostly formed inthe upper troposphere and in some casesmay remain for several hours. Persistentcontrails that have a long narrow cloudlike appearance are easily identifiedshortly after formation. A distinctionbetween contrails and naturally formedcirrus clouds may be difficult after a

------------------Corresponding author address: Dr MartinStuefer, Geophysical Institute, University ofAlaska Fairbanks, 903 Koyukuk Dr., P.O.Box 757320, Fairbanks, AK 99775

while because of wind drift and relatedspreading. Sausen et al. (1998) reporteda comparatively significant contributionof contrails to the total cirrus-cloudcoverage. Persistent contrails influencethe radiation balance of the atmosphereand hence have a possible effect onsurface climate (Meerkötter et al. 1999,Minnis et al. 1999). Aircraft contrails arealso of importance for the military,which in times of expensive stealthdevelopment is interested in non-formation of contrails for preservation ofan aircraft’s ‘low observability’.

STUEFER ET AL.

2

The intent of this paper is to verifycontrail forecasts that can be usefulespecially to military pilots for flightplanning purpose. The accuracy of themodel forecasts is discussed withoperational contrail observations andwith contrail predictions derived fromradiosonde measurements directly.Dichotomous contrail forecastingapplications were developed byspecifying critical temperatures whichserve as threshold to separate if acontrail will be formed or not. Schmidt(1941) and Appleman (1953) describedoriginally the thermodynamics of an airparcel that is influenced by theentrainment of moist and warm exhaustgases. Schumann (1996), Schrader(1997) and Jensen (1998) publishedmore recent reviews and explanations ofthe physics involved in contrailformation processes. We calculated thecontrail layers by comparing criticaltemperatures for contrail formation withforecasted temperatures. Following wegive a summary of the physics and thederivation of critical temperatures forcontrail formation.

3. Critical Temperature TheoryAppleman (1953), Schrader (1996),Schumann (1996) and others publishedsimilar equations showing the contrailformation theory. Efforts are ongoing forimplementation of the derived algorithmin operational contrail forecast-models;the U.S Airforce Weather Agency(AFWA) uses the algorithm in theirJETRAX contrail forecast model.JETRAX has been developed formilitary air operation support; itincorporates a fixed humidityparameter iza t ion scheme foratmospheric pressure levels below 300hPa. For illustration of the iterative

process needed in generating solution wegive a short review of equations for theexplicit calculation of criticaltemperatures for the formation ofcontrails.Assuming an isobaric mixing process,the temperature increase,

€

dT , of theaffected ambient air due to thecombustion of one mass unit of fuel iscalculated according to:

€

dT =Q

k ⋅ N ⋅ cp(1)

The parameter

€

Q denotes the liberatedheat by the combustion of one mass unitof fuel, and

€

k is the ratio of exhaust gasto the mass of fuel (

€

k≈12 kg kg -1). Thetemperature difference is large at thebeginning of the entrainment of exhaustgases; Jensen (1998) showed typicallapse rates of temperature differenceswith increasing exhaust gas dilutionderived from model studies. The massratio N accounts for the amount ofentrained environmental air to theexhaust gas; thus the product

€

k ⋅ Ncharacterizes the mass of environmentalair that is affected by the combustion ofone mass unit of fuel. The value of Ndepends strongly on the distance of theconsidered mixing parcel behind theaircraft, the combustion efficiency of theengines, and the density and stability ofthe atmosphere controlling the spreadingof the exhaust gases. The mass ratioranges from 0 immediately behind theaircraft engine to infinity. Thetemperature increase

€

dT further dependson the specific heat of air (

€

cp= 1004 J(kg K)-1). Estimating the emission indexfor water vapor as the amount of watervapor produced by the combustion of 1mass unit of fuel,

€

kH2O (≈1.4 kg kg -1),

the increase of mixing ratio

€

drf (g kg -1)in the considered air parcel is derived:

STUEFER ET AL.

3

€

drf =kH2O

⋅1000k ⋅ N

(2)

The combination of equations (1) and (2)results in a relation independent of thestate of mixing (N):

€

drfdT

=kH2O

⋅ cp ⋅1000Q

(3)

The ratio

€

drfdT

is called the contrail factor

(CF), which is characteristic for aircraftengine combustion and thus is notconstant during the different states of aflight. Busen and Schumann (1995)discuss the dependence of combustion tofuel, aircraft engine and flightparameters. Contrail factors wereestimated for certain aircraft and enginetypes; these contrail factors might berepresentative for an airplane duringcruise and level flight. Applemanderived an original value of 0.0336 g (kgK)-1. Schrader (1997) compiled values oftypical contrail factors ranging from0.0300 g (kg K)-1 to 0.039 g (kg K)-1.Busen and Schumann (1995) calculateda minimum contrail factor of 0.028 g (kgK)-1. The maximum contrail factor waspublished with 0.049 g (kg K)-1 (Peters,1993).

Jensen et al. (1998) report that forambient air temperatures between thecritical temperatures for liquid watersaturation and ice saturation no visiblecontrails were found during the ‘Contrailand Cloud Effects Special Study’(SUCCESS). For visible contrails toform, super-saturation with respect towater was observed and a phase changefrom water droplets to ice crystals mightoccur immediately. We use thesaturation of water for criticaltemperature calculations.

The relation of the saturation-mixing

ratio to temperature change (

€

drSdT

) is

compared with the contrail factor toderive a threshold temperature forcontrails formation, which is furtherdenoted as the critical temperature for asaturated environmental atmosphere,Tcrit,100. With p the air pressure, and eSthe saturation-vapor pressure, thesaturation-mixing ratio is calculated. ForeS <<p, a good approximation is:

€

rS =0.622eSp − eS

1000 ≅ 622eSp

(4)

Assuming isobaric processes, equation(4) yields the change of the saturation-mixing ratio with temperature asfunction of the pressure p and the airtemperature T:

€

drSdT

=622p

deSdT

(5)

Downie and Silverman (1957) and Pilieand Jiusto (1958) used the approach ofGoff and Gratch (1946) in order to

specify

€

deSdT

(T). Goff and Gratch

introduced a numerical approximation ofthe Clausius-Clapeyron equation. Thethreshold temperature Tcrit,100 is definedfor:

€

drSdT(Tcrit ,100) =

drfdT

= CF (6)

The critical temperatures Tcrit,h for non-saturated conditions (relative humidity h<100%) are derived from equation (6)according to:

STUEFER ET AL.

4

€

Tcrit,h = Tcrit ,100 −(rs Tcrit ,100 − r Tcrit ,h )

CF=

= Tcrit,100 −(rs Tcrit ,100 −

h100

rs Tcrit ,h )

CF

(7)

Both equations (6) and (7) are solvediteratively in order to obtain the criticaltemperatures Tcrit,100 (p) and Tcrit,h (p, h)for a previously estimated contrail factorCF. The determination of Tcrit,h requiresaccurate humidity measurements. Anexample programming code is given byWendler and Stuefer (2002).

4. Source Data

4.1. ObservationWe used contrail observations conductedat the Geophysical Institute of theUniversity of Alaska, Fairbanks(Wendler and Stuefer 2002). Observerscarried out visual contrail identificationoverhead Fairbanks. In addition a digitalcamera photographed the sky everyminute through a fish eye lens pointed tothe zenith (Fig. 1). The aircraft flightinformation, including airplane type,altitude, origin, destination, and speedwas available directly to the observers innear real time using an online FAA datadisplay. In this way aircraftsapproaching Fairbanks were identifiedon the display and an observer wasalerted to check for the presence of acontrail. Observations were carried outcontinuously from June 2001 to March2004 during daylight hours. The contraildatabase was compared withatmospheric measurements fromradiosondes, launched by the NationalWeather Service (NWS) twice daily (0and 12 UTC) from Fairbanks Airport.From our observation database we

FIG. 1: All-sky digital camera imagedisplaying a contrail formed by a Boeing747-400 (Northwest Airline Flight 69 fromDetroit to Osaka/Japan, 31. Jul. 2002).

selected 377 observations, which werewithin a time range of 2 hours before orafter the respective sounding. Thechosen observations were divided intothree groups; 55 negative contrail events(no-contrails), 168 transient (threshold)contrails with lifetimes of few seconds to1 minute, and 154 contrails lastinglonger than 1 minute. Due to drifting andspreading effects, contrails sometimescould not be identified and distinguishedfrom surrounding natural cirrus clouds,therefore the lifetimes in our databaserepresent a lower boundary value. Thelongest clearly identified contrail lasted6 hours.Most frequently observed aircraft typeswere Boeing 747-200 and Boeing 747-400, together comprising 65% of allobservations. Cruising speeds werebetween 400 knots (741 km h-1) and 550knots (1019 km h-1). The cruisingaltitudes varied from 7223 m (23700 ft)to 11277 m (37000 ft).

STUEFER ET AL.

5

FIG. 2: Temperatures versus air-pressureshown at the altitudes of aircraftobservations. Different symbols refer to thepersistence of contrail observations.

4.2. Atmospheric SoundingAtmospheric parameters were obtainedfrom the radiosonde ascents. Linearinterpolation between two respectivedata points was used for obtainingtemperature and dew-point temperatureat flight level and the pressure wasinterpolated with the barometric heightequation. Figure 2 shows the aircraftambient temperatures and pressures forthe three contrail observation classes.No-contrail cases were observed at atemperature range from –56ºC to –36ºC,threshold contrails occurred from –65ºCto –39ºC, and slightly coldertemperatures from –67ºC to –44ºC weremeasured for contrails persisting longerthan one minute. Typical pressure valuesfor all observations were between 330hPa and 200 hPa.Inaccuracies in critical temperaturecalculations might occur as the humidityreported by radiosondes at lowtemperatures is subject to sources oferror (Pratt 1985, Elliott and Gaffen1991). Miloshevich et al. (2001)investigated a strong bias towards dryhumidity values measured with Vaisala

FIG. 3: Errors in critical temperaturesderived due to a dry bias of relativehumidity of 10% for different ambientrelative humidity values. Criticaltemperatures were calculated with p= 250hPa and CF= 0.036 g/kgK.

RS80-A radiosondes at coldtemperatures. Correction factors for therelative humidity measurements of 1.3 attemperatures of –35ºC increasing to 2.4at –70ºC were suggested for thisparticular radiosonde instrument. InFairbanks the NWS used Vaisala RS80-57H radiosondes for upper airmeasurements. As possible correctionfactors depend strongly on sensor type(Miloshevich et al., 2001) and as nosimultaneous hygrometer measurementsto Vaisala RS80-57H measurementswere available for comparison purpose,we used the radiosonde humiditymeasurements without correction.Schrader (1997) reported a littlesensitivity of critical temperatures toambient relative humidity valuesbetween 0% and 70%. Criticaltemperature changes due to a 10% errorin relative humidity are shown in Figure3; the changes are similar for differentpressure levels and contrail factors.Errors in critical temperatures increasewith increasing relative humidity; morethan 1ºC lower critical temperaturesresult from 10% humidity errors at

CF x y w z POD POD(th) POD(>1min) PODnil FAR HR PODm0.025 183 139 54 1 0.568 0.365 0.792 0.982 0.005 0.629 0.7750.026 201 121 54 1 0.624 0.461 0.805 0.982 0.005 0.676 0.8030.027 210 112 53 2 0.652 0.491 0.831 0.964 0.009 0.698 0.8080.028 228 94 52 3 0.708 0.575 0.857 0.945 0.013 0.743 0.8270.029 232 90 52 3 0.720 0.587 0.870 0.945 0.013 0.753 0.8330.03 246 76 52 3 0.764 0.641 0.903 0.945 0.012 0.790 0.855

0.031 255 67 52 3 0.792 0.677 0.922 0.945 0.012 0.814 0.8690.032 259 63 52 3 0.804 0.689 0.935 0.945 0.011 0.825 0.8750.033 264 58 52 3 0.820 0.707 0.948 0.945 0.011 0.838 0.8830.034 270 52 52 3 0.839 0.743 0.948 0.945 0.011 0.854 0.8920.035 273 49 52 3 0.848 0.754 0.955 0.945 0.011 0.862 0.8970.036 276 46 52 3 0.857 0.772 0.955 0.945 0.011 0.870 0.9010.037 280 42 51 4 0.870 0.796 0.955 0.927 0.014 0.878 0.8980.038 283 39 49 6 0.879 0.814 0.955 0.891 0.021 0.881 0.8850.039 290 32 47 8 0.901 0.850 0.961 0.855 0.027 0.894 0.8780.04 293 29 46 9 0.910 0.868 0.961 0.836 0.030 0.899 0.873

TAB. 1: Contingency table with the number of forecasted and observed contrails. X: number ofobserved and forecasted contrails, y: false- observed, but not forecasted, w: not observed and notforecasted, z: false- not observed, but forecasted. The total probability of detection is given for allobserved contrails (POD), for threshold contrails (POD(th), 0<lifetime≤1minute), and forcontrails persisting longer than 1 minute. PODnil: POD for no-contrail events. FAR; false alarmrate; HR: hit rate defined as the number of correct contrail/no-contrail forecasts related to the totalnumber of observation; PODm: mean value of POD and PODnil.

ambient relative humidity values above80%.

4.3. Forecast ModelsFor contrail layer forecasts and forecastverification we used two differentversions of the non-hydrostatic, fifth-generation mesoscale model M M 5 ,which was originally developed by theNational Center for AtmosphericResearch (NCAR) and the PennsylvaniaState University (Dudhia 1993, Grell etal. 1994). A coupled modeling system,referred to as the Arctic MM5 wasdeveloped by the ‘Mesoscale Modelingand Applications Group’ at theGeophysical Institute of the Universityof Alaska, Fairbanks (Tilley et al. 1999,Zhang et al. 2004). The Arctic MM5treats radiative processes andmicrophysical cloud and precipitationphysics according to the Polar MM5

(Bromwich et al. 2001, Cassano et al.2001). Hack et al. (1993) described thelongwave and shortwave radiationscheme, which is also used in the NCARCommunity Climate Model, version2(CCM2).Experimental Arctic MM5 runs wereexecuted three times daily by nudgingmeteorological fields with analysis ofactual observations. The initializationtimes were at 6, 12, and 18 UTC, theforecast range was from 0 to 48 hours.Most forecast products were saved formodel verification purpose; forecastsfrom mid-December 2002 until end ofFebruary 2004 were available from theArctic MM5 for contrail layerverification. Due to the designatedatmospheric sounding times at 00 and 12UTC, near real- time (0 hour) forecastverification with radiosonde data wasrestricted to Arctic MM5 initialized at 12

STUEFER ET AL.

7

UTC. The different model initializationtimes and forecast ranges had to beconsidered for verification of MM5forecasts with atmospheric soundingsand actual observations of contrails.Starting June 2003, we received, inaddition to Arctic MM5, Air ForceWeather Agency (AFWA) MM5initialized at 00 and 12 UTC. For AFWAMM5 characteristics we refer the readerto the following URL address:http://meted.ucar.edu/nwp/pcu2/afintro.htm. AFWA MM5 produced 0 to 72 hoursforecasts with time intervals of 3 hours.The forecasts covering the Alaskatheater (region) were obtained afteractual model runs via the AppliedPhysics Laboratory of Johns HopkinsUniversity. Due to the amount of datatransfer (max. ~253 Mbyte, compressed*.tar format), gaps in forecast dataacquisition occurred.Both the AFWA MM5 and the ArcticMM5 use a 45 km horizontal grid. Thecloud and precipitation physics areequivalent in both models and follow theReisner scheme (Reisner et al. 1998).The vertical resolution of AFWA MM5consists of 24 levels, forecasts areperformed every 50 hPa for pressurelevels above 850hPa. Arctic MM5 ischaracterized by a higher verticalresolution with 41 terrain following σlevels (Zhang et al. 2004). Contraillayers calculated from MM5 forecastsfor the nearest model grid point toFairbanks were verified with radiosondegenerated contrail layers, to which werefer as reference layers in the followingsection.

5. VerificationCritical temperatures were calculatedwith atmospheric data derived fromFairbanks radiosonde ascents and fromMM5 forecasts for Fairbanks. Contrail

formation was predicted for those layers,where critical temperatures exceeded themeasured or forecasted temperatures.

5.1. Verification of contrailobservations with predicted contrailoccurrence based on radiosondemeasurementsIn order to verify the algorithm withradiosonde data we used various contrailfactors ranging from 0.025 to 0.05g/kgK. The results of dichotomouscontrail forecasts for selected contrailfactors are shown in Table 1. Thecontingency of forecasted hits (x) ,misses (y, not forecasted but observed),the number of cases when a contrail wasforecasted but not observed (z, falsealarm) and the correct negative forecasts(w) were calculated for no contrails,threshold contrails (0 < lifetime 1minute) and contrails persisting longerthan 1 minute. In addition the probabilityof detection (POD = x/(x+y)), theprobability of detection for no-contrailevents (PODnil = w/(w+z)), the falsealarm rate (FAR=z/(x+z)), the hit rate(HR = (x+w)/(x+y+w+z)) and a meanp r o b a b i l i t y o f d e t e c t i o n(PODm=(POD+PODnil)/2) are given inTable 1. A steady increase of the hit rate(HR) combined with low false alarmrates (FAR) were obtained due to thelarger number of contrail incidencecompared to no-contrails. Theprobability of detection showed majordifferences between threshold contrails(PODth) and contrails lasting longer thanone minute (POD>1). The mean valuePODm is considered appropriate for thederivation of a threshold betweencontrails and no-contrails. Theradiosonde prediction provided the bestresults (PODm = 90%) with a contrailfactor CF = 0.036 g/kgK. This contrailfactor represents an average value for

STUEFER ET AL.

8

different aircraft types with differentcombustion characteristics for cruisingcommercial flights overhead ofFairbanks. Contrail layers were definedas the range with critical temperatureswarmer than ambient temperatures.PODnil values decrease with increasingcontrail factors due to a thickening of theforecasted contrail layers.

5.2. Verification of contrail layers fromradiosonde measurements with contraillayers derived from MM5For the verification of MM5 we derivedall contrail layers from criticaltemperature calculations using a contrailfactor of 0.036 g/kgK. Temperatures andcritical temperatures were linearlyinterpolated between those successivereference (or model) levels where thesign of the difference betweentemperature and critical temperaturechanged. The altitudes where the criticaltemperatures equaled the measured(forecasted) temperatures correspond tothe boundaries of contrail layers. 870radiosonde profiles were available forthe derivation of reference contraillayers for the verification period fromDecember 2002 until February 2004.Most reference profiles resulted in onecontrail layer; in 6% of the cases wefound two layers, and another 6% of theprofiles showed no contrail probabilitywith critical temperatures continuouslybelow the measured temperature. Mainlydue to different vertical resolution of themodels the number of multiple contraillayers in forecasted profiles is less thanin reference profiles. For simplicity weomitted cases with multiple layeroccurrence. Table 2 shows the numberof contrail layers, which were derivedfrom MM5 forecast outputs and were

compared with corresponding referencelayers. The mean altitudes of referencecontrail layers during winter monthswere typically between 8000 and 9000 ma.s.l.; a minimum value of 5940 m wasderived for the period of verification.The altitudes raised in summerfrequently above 10000 m a.s.l.. Thecomparison of mean altitudes from thedifferent models and forecast times withcorresponding reference data resultedmostly in increasing deviation withincreasing forecast time. Figure 4 showsexample plots of mean layer altitudes for0 and 72 hour forecasts derived fromAFWA MM5 data. Significant spreadingof the data and a reduced correlationcoefficient was found for the 72 hourforecast period. Mean values of theabsolute deviation of forecasted layeraltitudes from reference altitudes areshown in Figure 5 for all forecasts. Thealtitudes derived from the 0 hourforecasts for all models are on theaverage about 200 m above or below thecontrail layer altitudes derived fromreference sounding. For 24 hourforecasts we obtained differencesbetween 270 m and 395 m; thedifference increased to about 500 m forthe 72 hour forecasts. Steadily increasingaltitude differences with increasingforecast time also resulted for the 0 to 48hour Arctic MM5 forecasts. AFWAMM5 showed for both initializationtimes partly irregular performancecharacteristics, however generalincreasing altitude deviations indicated asignificant decrease of contrail layerforecast accuracy.In order to estimate effects of contrailforecast errors, layer thickness data haveto be considered besides altitudes.

FIG. 4: Mean altitudes of contrail layers from reference sounding data versus the forecasted meanaltitudes. The 0 hour (left) and 72 hour (right) forecast comparisons are shown.

Thickness depends strongly on thecontrail factor; for a contrail factor of0.036 g/kgK the contrail layer thicknessderived from Fairbanks radiosondemeasurements is 1700 m on the averageduring the summer months June, Julyand August. For winter months fromDecember to February we found morethan twice as thick layers with anaverage value of 3700 m. Modelverification differences might result dueto pronounced seasonal thicknessdifferences. The thickness differenceswere almost coincident using 6 and 18UTC initialization Arctic MM5 data.Figure 6 shows slight differences for the12 UTC Arctic model, which is mostlikely a result of the limited number ofdata available (Table 2). For near-realtime 0 hour model forecasts the modeledthickness was on the average about 400m thinner or thicker than the referencethickness. For 48 hour forecasts thethickness errors increased to 690 m(Arctic MM5) and 800 m (AFWAMM5). Strong coincident thicknessdifferences were obtained from AFWA

MM5 for forecast periods from 48 to 72hours. Errors of more than 30% of thereference thickness could be expected onthe average for 72 hour forecasts.

ArcticMM5

ArcticMM5

ArcticMM5

AFWAMM5

AFWAMM5

Forecast 06 UTC12 UTC18 UTC00 UTC12 UTC0 227 85 926 362 369 12 220 84 8918 346 345 24 213 176 17530 329 336 36 211 162 15742 179 185 48 183 147 14160 143 13572 130 136

TAB. 1: Numbers of contrail layersavailable for comparison with layerscalculated from atmospheric soundings atFairbanks. The 00, 06, 12 and 18 UTCcolumns correspond to the different modelinitialization times.

A mean overlap (MO) was calculated forverification of the altitude and thethickness of contrail layers. The MO hasbeen introduced as a forecast skill

STUEFER ET AL.

10

parameter, which relates overlappingcontrail layers derived from MM5forecasts and from atmospheric soundingmeasurements; values of MO may rangefrom 0 to 100%. Defining dr as thethickness of the reference layer from thesounding profile, df as the thickness ofthe forecasted layer, and db as the sectionof the overlapping contrail layer fromboth profiles, we calculated the meanoverlap in percent as MO =((db/dr)+(db/df))/2*100 (Fig.7). For aperfect forecast showing coincidingcontrail layers the value of MO would be100%. A small MO may result from avertical shift in the model and referencelayers even at perfectly forecastedthickness. Figure 8 shows the resultingdecrease of the MO with an increasingforecast hour. Arctic MM5 verificationresults coincided well with AFWAMM5; all 0 hour model runs showedmean overlaps of about 90%. A steadydecrease of 4% per 24 hours on theaverage of all forecast datasets wasobserved. MO values between 82% and83% were found for 48 hour forecasts.

FIG. 5: Mean absolute deviation of meanaltitudes of contrail layers derived fromMM5 in reference to altitudes derived fromsounding measurements. Absolutedeviations are shown for Arctic- and AFWAMM5 for 0-72 hour forecasts.

FIG. 6: Mean absolute deviation of contraillayer thickness derived from MM5 inreference to layer thickness derived fromsounding measurements. Absolutedeviations are shown for Arctic- and AFWAMM5 for 0-72 hour forecasts.

5.3. Contrail observations verified withforecasted contrail occurrence based onMM5The algorithm for the derivation ofcritical temperatures as threshold forcontrail formation (equation 7) wastested with actual observations ofcontrails overhead Fairbanks and MM5forecasts. Only daytime traffic within atime range of 0 UTC +/- 2 hours wasconsidered for verification, thereforeinitialization times with forecasts for 0UTC were used. The number of

FIG. 7: Example of contrail layer thicknessderived from reference sounding (left) andfrom AFWA MM5 forecast (right).

Model Fc.(h) all x y w z x(th) y(th) POD FAR PODnil PODm HR

Arctic MM5 18 6 162 108 17 37 0 54 16 0.86 0 1 0.93 0.90Arctic MM5 12 12 104 72 17 15 0 36 17 0.81 0 1 0.90 0.84Arctic MM5 06 18 164 104 23 34 3 50 22 0.82 0.03 0.92 0.87 0.84Arctic MM5 18 30 167 101 29 37 0 49 25 0.78 0 1 0.89 0.83Arctic MM5 12 36 106 71 19 16 0 34 19 0.79 0 1 0.89 0.82Arctic MM5 06 42 95 62 21 12 0 28 20 0.75 0 1 0.87 0.78AFWA MM5 00 0 45 28 10 7 0 16 9 0.74 0 1 0.87 0.78AFWA MM5 12 12 49 26 14 9 0 12 13 0.65 0 1 0.83 0.71AFWA MM5 00 24 96 58 24 14 0 30 21 0.71 0 1 0.85 0.75AFWA MM5 12 36 94 46 34 14 0 20 30 0.58 0 1 0.79 0.64AFWA MM5 00 48 98 44 38 16 0 21 30 0.54 0 1 0.77 0.61AFWA MM5 12 60 97 42 40 14 1 21 30 0.51 0.02 0.93 0.72 0.58AFWA MM5 00 72 90 43 32 14 1 23 25 0.57 0.02 0.93 0.75 0.63

TAB. 3: Contingency table with the number of MM5-forecasted and observed contrails. Themodel with initialization time and respective forecast hour (Fc) is given. Numbers x(th) and y(th)refer to observed threshold cases with lifetimes less or equal 1 minute. Abbreviations areconsistent with table 1: POD: probability of detection; FAR; false alarm rate; HR: hit rate;PODm: mean value of POD and PODnil.

observation to forecast pairs varied withavailable model and initialization timeswithin the period from mid-December2002 until end of February 2004, thusresults have to be compared with somereservations (Table 3). Nevertheless, theoverall observation counts between 95and 167 cases for Arctic MM5 resultedin high mean probability of detection(≥87%) and hit rates (≥78%) for allforecasts (6 to 42 hours). No-contrailevents were forecasted correctly exceptfor the 18 hour forecast, when 3observed no-contrail cases wereforecasted as contrails due to a smallmargin between critical temperaturesand forecasted temperatures. Errors inforecasts of observed contrails weremostly caused by threshold contrailswith short lifetimes; a ratio y(th)/y higherthan 86% was observed.The performance of the AFWA MM5was slightly less than that of the ArcticMM5 (Table3). Similar to Arctic MM5,

most forecast errors originated fromthreshold contrails; almost no errorswere obtained for no-contrail cases.Mean probability of detection values ofless than 80% were found after aforecast period of only 36 hours.

FIG. 8: The MO (mean overlap) indicatingthe agreement of modeled layers and layersfrom radiosonde measurements. Arctic(solid lines) and AFWA (dashed lines) MM5overlap parameters are indicated for thedifferent model runs.

STUEFER ET AL.

12

We related the differences in modelperformance mostly to the highervertical resolution of the Arctic MM5.Contrails with short lifetimes were oftencharacterized by little differencesTcrit–T (Jensen et al. 1998); smalltemperature differences occurred mostlyin marginal zones of contrail layers or inthin layers, which might be inaccurate,or missed, by the AFWA verticalresolution of 50 hPa.

6. Summary and discussionBoth Arctic MM5 and AFWA MM5performed similar in forecasting thealtitude of contrail layers; contrails wereobserved in Fairbanks typically ataltitudes higher than 8000 m a.s.l..Slightly better results were obtained forlayer thickness forecasts with ArcticMM5 than with AFWA MM5. Thisdifference in performance might resultfrom the different vertical resolution ofthe available forecast data, as thedifferences in the physics of AFWA- andArctic MM5 mainly affect processes inthe boundary layer and lowert roposphere . Different modelperformance in the considered altitudesof contrail formation may not result frommodel initial and boundary conditions,which are similar for Arctic- and AFWAMM5 models; differences may occur atlower altitudes over the Arctic Ocean,where Arctic MM5 uses a different seaice parameterization scheme.A mean overlap (MO) was defined as anoverall check of the forecast contraillayer. For short-term forecasts MOs ofbetween 85 and 90% imply goodagreement of forecasted contrail layeraltitude and thickness with referencedata. The MO was consistent for allmodel runs, a steady decrease ofperformance with increasing forecastperiod was found. The MO for 48 hour

forecasts is above 80%. Due to adecrease in the MO, forecast errors areexpected to occur mostly in marginalregions of contrail layers with small(Tcrit–T) values. These cases wereobserved mostly as threshold contrailswith lifetimes of less than one minute.The forecast verification of visualaircraft observations from the groundconfirmed that main errors originatefrom threshold contrails; for someforecast model runs all cases of observedcontrails, which were wrongly classifiedas no-contrails, were contrails withlifetimes less or equal one minute.Though transient (fast dissolving)contrails can be considered to have littlepotential to affect directly the radiationbudget at the surface, they are of greatinterest for military when producingcontrail forecasts. Transient contrailsmay have importance for climate studiesdue to the added water vapor in theatmosphere and an increased potentialfor heterogeneous nucleation, whichfacilitates the formation of persistentcontrails by succeeding aircrafts (Rind etal. 2000).The AFWA JETRAX contrail forecastmodel uses parameterized constanthumidity values depending on thealtitude in reference to the tropopause(Shull 1998). Humidity parameterizationavoids inaccurate contrail forecasts dueto a possible poor quality of humiditymeasurements and forecasts. Shull(1998) showed in his validation study ofJETRAX hit rates of 84.4% as a meanvalue of 18, 24, and 30 hours AFWAMM5 forecasts; his findings were basedon 397 aircraft observations. TheJETRAX hit rates are comparable to ourderived hit rates using Arctic MM5 andusing forecasted humidity values. Ourcomparison of contrail observations withAFWA MM5 source data showed less

STUEFER ET AL.

13

agreement; reasons for the lack offorecast quality might be subject to theconsiderable number of thresholdcontrails in our observation database,and the reduced number of cases, andhence less significant statistics. Thesmaller vertical resolution of AFWAMM5 reduces the probability ofdetection of threshold contrails. A smallvertical resolution may also hamper thedetection of accurate tropopausea l t i t udes r educ ing JETRAXperformance.The obtained quality of contrail forecastsin Fairbanks provides confidence ofusing MM5 especially for short- termforecasts of contrail layers. Contrailforecasts could be incorporated for flightplanning purposes. Even for 60 hourforecasts MOs of forecasted contraillayers of 80% suggest good modelperformance at least for the detection ofthose atmospheric layers wherepersistent contrails are expected to form.Arctic MM5 performed slightly betterwhen compared to AFWA MM5 due toan enhanced vertical resolution andhence a more accurate detection ofcontrail layers.A graphical interpretation of thediscussed MM5 contrail layer forecastsfor Fairbanks and the whole Alaskaregion is available on our website at:http://contrail.gi.alaska.edu/ (Stuefer etal. 2004).

AcknowledgementsThe University Partnering forOperational Support (UPOS) supportedthis research in collaboration with theApplied Physics Laboratory, JohnsHopkins University, by a grant fromDoD. We thank B. Moore, M. Shulski,B. Hartmann and M. Robb, whoparticipated in observation and analysisof contrails. Special thanks go to the

anonymous reviewers of this paper. JeffTilley and the University of AlaskaFairbanks Mesoscale Modeling andApplications Group provided ArcticMM5 source data.

REFERENCESAppleman, H. S., 1953: The Formation

of Exhaust Condensation Trails byJet Aircraft. Bull. Amer. Meteor.Soc., 34, 14-20.

Bromwich, D. H., J. J. Cassano, T.Klein, G. Heinemann, K. M.Hines, K. Steffen, and J. E. Box,2001: Mesoscale modeling ofkatabatic winds over Greenlandwith the Polar MM5. Mon. Wea.Rev., 129 , 2290-2309.

Busen, R. and U. Schumann, 1995:Visible contrail formation fromfuels with different sulfur content.Geophys. Res. Let., 22, 1357-1360.

Carleton, A. and P. Lamb, 1986: Jetcontrails and cirrus clouds: afeasibility study employing highresolution satellite imagery. Bull.Amer. Meteor. Soc., 67, 301-309.

Cassano, J. J., J. E. Box, D. H.Bromwich, L. Li, and K. Steffen,2001: Evaluation of Polar MM5simulations of Greenland'satmospheric circulation. J .Geophys. Res., Special Issue onthe PARCA (Program for ArcticRegional Climate Assessment),106, 33,867-33,890.

Downie, C. S., and B. A. Silverman,1957: Jet aircraft condensationtrails. Handbook of Geophysics forAir Force Designers, GeophysicsResearch Directorate, Air ForceCambridge Research Laboratories,Air Research and DevelopmentCommand, United States AirForces, 19-1 - 19-9.

STUEFER ET AL.

14

Dudhia, J., 1993: A nonhydrostaticversion of the Penn State / NCARmesoscale model: Validation testsand simulation of an Atlanticcyclone and cold front. Mon. Wea.Rev. 121 1493-1513.

Elliott, W. P. and D. J. Gaffen, 1991: Onthe utility of radiosonde humidityarchives for climate studies. Bull.Amer. Meteor. Soc., 72 (10), 1507-1520.

Goff, J. A. and S. Gratch, 1946: Lowpressure properties of water from -160 to 212 F. Trans. Amer. Soc.Heat. Vent. Eng., 52, 95 pp.

Grell, G. A., J. Dudhia and D. R.Stauffer, 1994: A description ofthe fifth-generation Penn State /NCAR mesoscale model (MM5).NCAR Tech. Note, NCAR/TN-398+STR, National Center forAtmospheric Research, Boulder,CO, 138 pp.

Hack, J. J., B. A. Boville, B. P. Briegleb,J. T. Kiehl, P. J. Rasch, and D. L.Williamson, 1993: Description ofthe NCAR Community ClimateModel (CCM2). NCAR Tech.Note , NCAR/TN-382+STR ,National Center for AtmosphericResearch, Boulder, CO, 108 pp.

Jensen, E.J., et al., 1998: Environmentalconditions required for contrailformation and persistence. J.Geophys. Res., 103, No. D4, 3929-3936.

Meerkötter, R., U. Schumann, D. R.Doelling, P. Minnis, T. Nakajimaand Y. Tsushima, 1999: Radiativeforcing by contrails. AnnalesGeophysicae, 17, 1070-1084.

Miloshevich, L. M., H. Voemel, A.Paukkunen, A. J. Heymsfield, andS . J . O l tmans , 2000 :Characterization and correction ofrelative humidity measurements

from Vaisala RS80-A radiosondesat cold temperatures. J. Atmos.Oceanic Technol., 18, 135-156.

Minnis, P., U. Schumann, D. R.Doelling, K. M. Gierens, and D.W.Fahey, 1999: Global distributionof contrail radiative forcing.Geoph. Res. Let., 26 , No. 13,1853-1856.

Peters, J. L., 1993: New techniques forcontrail forecasting. AWS/TR-93/001, 26 pp. [Available from HQAir Weather Service, Scott AFB,IL 62225.]

Pilie, R. J., and J. E. Jiusto, 1958: Alaboratory study of contrails. J.Appl. Meteor., 15, 149-154

Pratt, R. W., 1985: Review ofradiosonde humidi ty andtemperature errors. J. Atmos.Ocean. Technol., 2 (3), 404-407.

Rind, D., P. Lonergan, and K. Shah,2000: Modeled impact of cirruscloud increases along aircraftflight paths. J. Geophys. Res., 105,No. D15, 19927-19940

Sausen, R., K. Gierens, M. Ponater, andU. Schumann, 1998: A DiagnosticStudy of the Global Distribution ofContrails Part I: Present DayClimate. Theor. Appl. Climatol.,61, 127-141.

Schmidt, E., 1941: Die Entstehung vonEisnebel aus den Auspuffgasenvon Flugmotoren. Schriften derDeutschen Akademie derLuftfahrtforschung, Verlag R.Oldenbourg, Muenchen undBerlin, Heft 44, 1-15.

Schrader, M. L. 1997: Calculations ofaircraft contrail formation criticaltempera tures . Notes andcorrespondence, J. Appl. Meteor. ,36, 1725-1729.

STUEFER ET AL.

15

Schumann, U., 1996: On conditions forcontrail formation from aircraftexhausts. Meteor. Z., 5, 4-23.

Shull, J. D., 1998: A validation study ofthe Air Force Weather Agency(AFWA) JETRAX contrailforecast algorithm. M. S. thesis,Dept. of Meteorology, AirforceInstitute of Technology, 120 pp.[Available from Airforce Instituteof Technology; Wright PattersonAFB, OH 45433.]

Stuefer, M. and G. Wendler, 2004:Contrail studies and forecasts inthe subarctic atmosphere aboveFairbanks, Alaska. Preprints, 11thConference on Aviation, Range,and Aerospace Meteorology,Hyannis, MA, Amer. Meteor. Soc.,P8.13.

Tilley, J. S., A. H. Lynch, W.L.Chapman, W. Wu and G. Weller,1999: On the Coupling of Land

Surface Packages within aRegional Climate Model of theArctic. Proceedings of a Workshopon Arctic Regional ClimateModels, H. Cattle, ed. WorldMeteorological Organization/Tech.Document No. 981, pp. C12.1-12.5.

Wendler, G. and M. Stuefer, 2002.Improved Contrail ForecastingTechniques for the SubarcticSetting of Fairbanks, Alaska.Geophysical Institute of theUniversity of Alaska Report (UAGR-329), 35 pp.

Zhang, J. and X. Zhang, 2004: ModelingStudy of Arctic Storm with theCoupled MM5-Sea Ice-OceanModel. Preprints, 5th WRF/ 14th

MM5 User ’s Workshop ,Boulder,CO.