Forecasting dust storms using the CARMA-dust model ?· Environmental Modelling & Software 19 (2004)…

  • Published on
    21-Jun-2018

  • View
    212

  • Download
    0

Embed Size (px)

Transcript

<ul><li><p>Environmental Modelling &amp; Software 19 (2004) 129140www.elsevier.com/locate/envsoft</p><p>Forecasting dust storms using the CARMA-dust model and MM5weather data</p><p>B.H. Barnum a,, N.S. Winstead a, J. Wesely b, A. Hakola b, P.R. Colarco d, O.B. Toon c,P. Ginoux d, G. Brooks b, L. Hasselbarth a, B. Toth a</p><p>a Johns Hopkins University, Applied Physics Laboratory, Laurel, MD 20723, USAb United States Air Force Weather Agency, Offut AFB, NE, USA</p><p>c University of Colorado, PAOS Group, Boulder, CO, USAd NASA Goddard Space Flight Center, Greenbelt, MD, USA</p><p>Received 26 August 2002; received in revised form 15 January 2003; accepted 18 February 2003</p><p>Abstract</p><p>An operational model for the forecast of dust storms in Northern Africa, the Middle East and Southwest Asia has been developedfor the United States Air Force Weather Agency (AFWA). The dust forecast model uses the 5th generation Penn State MesoscaleMeteorology Model (MM5) as input to the University of Colorado CARMA dust transport model. AFWA undertook a 60 dayevaluation of the effectiveness of the dust model to make short, medium and long- range (72 h) forecasts of dust storms. The studyis unique in using satellite and ground observations of dust storms to score the models effectiveness using standard meteorologicalstatistics. Each of the main forecast regions was broken down into smaller areas for more detailed analysis. The study found theforecast model is an effective forecast tool with Probability of Detection of dust storm occurrence exceeding 68 percent overNorthern Africa, with a 16 percent False Alarm Rate. Southwest Asia forecasts had average Probability of Detection values of 61percent with False Alarm Rates averaging 10 percent. 2003 Elsevier Ltd. All rights reserved.</p><p>Keywords: Dust storm forecasting; MM5 weather model; CARMA model; Skill scores; Mineral dust</p><p>Software availabilityName of Software: CARMA-DustContact address: B. H. Barnum, Johns Hopkins Applied</p><p>Physics, Laboratory, Laurel, MD 20723-6099,Tel.:+1-443-778-7082, fax:1+443-778-1899,email: ben.barnum@jhuapl.edu</p><p>Year first available: 2002Hardware required: Sun or Pentium PC with 512 Mb</p><p>RAMSoftware required: Solaris Linux with IDL, Perl and</p><p>F77 compiler with netCDF library.Program lan-guage: Fortran 77, Perl, IDL</p><p>Program size: 4 Mb</p><p> Corresponding author, Tel.: +1-443-778-7082; fax +1-443-778-4130.</p><p>E-mail address: ben.barnum@jhuapl.edu (B.H. Barnum).</p><p>1364-8152/$ - see front matter 2003 Elsevier Ltd. All rights reserved.doi:10.1016/S1364-8152(03)00115-4</p><p>Availability and cost: contact the developer for furtherinformation.</p><p>1. Introduction</p><p>Dust storms throughout Saharan Africa, the MiddleEast and Asia are estimated to place more than 2005000 million metric tons of mineral dust into the earthsatmosphere each year (Tegen and Fung, 1994). Duststorms directly affect visibility and impact daily com-mercial and military operations near desert regions. TheUnited States Air Force Weather Agency (AFWA) hassupported the development of a dust forecast model witha 72 h forecast capability. The dust model called Com-munity Aerosol and Radiation Model for Atmospheres(CARMA), was originally developed by Professor OwenToon and Dr. Peter Colarco at the University of Color-ado, Boulder (Toon et al., 1988; Colarco et al., 2002).</p><p>mailto:ben.barnum@jhuapl.edu</p></li><li><p>130 B.H. Barnum et al. / Environmental Modelling &amp; Software 19 (2004) 129140</p><p>The CARMA model has been modified by Johns Hop-kins Applied Physics Laboratory to make daily dust fore-casts using weather forecast data generated by the UnitedStates Air Force Weather Agency MM5 weather model.</p><p>Several other dust aerosol models are being used forthe daily forecasting of dust storms. These models aresimilar to the CARMA-Dust model, in that they use datafrom standard weather models such as ETA, NOGAPSor MM5. The University of Malta and the University ofAthens uses a modified version of the Eta weather modelto make dust forecasts over Northern Africa and theMediterranean (Nickovic and Dobricic, 1996). TheUnited States Naval Reasearch Laboratory makes dailyforecasts of dust using the Navy Aerosol Analysis andPrediction System (NAAPS). The NAAPS aerosolmodel uses daily weather forecast products from theNavy Operational Global Atmospheric Prediction Sys-tem (NOGAPS) (Hogan and Rosmond, 1991). A dustforecast model developed by Yaping Shao is being usedto forecast dust storms over China and East Asia usingweather data from the National Meteorological Center(CMA) of China (Shao, 2001; Lu and Shao, 2001).</p><p>Our goal in this paper is to evaluate how wellCARMA does in making forecasts of dust storm occur-rence using mesoscale weather forecast data. A directcomparison of the CARMA-Dust model to the other dustforecast models is beyond the scope of this paper. Webelieve, however, that the forecast statistics and capabili-ties of the CARMA model are representative of currentdust models now in use worldwide.</p><p>The latest version of the CARMA MM5 dust modelcan make 72 h forecasts of surface and airborne dustconcentrations in 3 different mesoscale theaters coveringSaharan Africa and the Middle East, Southwest Asia andChina. A new global dust source database developed byDr. Paul Ginoux et al. (2001) is used in the CARMAmodel. The dust source model is based on topographicalfeatures associated with dust sources and has beenfurther developed using TOMS satellite data (Prosperoet al., 2002; Herman et al., 1997).</p><p>The forecast ability of the dust model was evaluatedover a 60 day period, beginning February 15th, 2002,for two of the AFWA MM5 forecast theaters, SaharanAfrica and Southwest Asia. The Middle East has beengrouped with Southwest Asia for this evaluation. Themodel forecasts were compared with Defense Meteoro-logical Satellite Program (DMSP) satellite imagery andground observations. Each theater was broken into sub-regions for detailed evaluation of the short (612 h), mid(3036 h) and long-term (5460 h) forecast ability ofthe model. Results of the study show the dust model hasgood skill in forecasting dust conditions for short,medium and long range forecast periods.</p><p>2. CARMA MM5 dust forecasting</p><p>The Community Aerosol and Radiation Model forAtmospheres (CARMA) was originally developed by theUniversity of Colorado and NASA Ames to be a scalableaerosol model to study a variety of atmospheric pro-cesses, such as cloud formation, smoke and dust aerosols(Toon et al., 1988). The version of CARMA developedfor daily forecasting of dust has been modified to assimi-late meteorological forecast data from the Penn State 5thgeneration Mesoscale Meteorology Model (MM5)(Anthes and Warner, 1978). The model also incorporatesthe global dust source database developed by Ginoux etal. (2001). The model uses 10 particle size bins whichcover dust particles with radii from 0.5 m to 10.0 m.Following the model initialization, the MM5 72 h fore-cast data for winds, pressure, and temperature, at the sur-face and at 22 selected sigma pressure levels are inputinto CARMA. The dust model outputs a set of dust con-centration maps and vertical concentration profiles foreach 3 h time period during the 72 h forecast.</p><p>The MM5 weather forecast data is run by the UnitedStates Air Force Weather Agency (AFWA) for theatersworldwide on a daily basis (Fig. 1). The MM5 data isobtained directly from AFWA for the mesoscale theaterscovering Saharan Africa and Middle East (T09a) andSouthwest Asia (T04a). The MM5 model is run with41 vertical sigma pressure coordinate levels with 45 kmhorizontal grid spacing.</p><p>The CARMA dust model reads in a subset of theMM5 data, using 22 vertical sigma pressure levels anda 90 km horizontal latitude, longitude grid spacing. Thisgrid scheme was chosen to have approximately the sameresolution as the 1 1 (111 km) dust source databaseand to reduce the run time for daily forecasting. Thevertical levels were chosen to optimize vertical resol-ution in the boundary layer, with 18 vertical levels usedbetween the surface and the 500 mb pressure level. Ver-tical winds are calculated internally in CARMA for eachgrid location based on the divergence of the MM5 press-ure fields at each sigma vertical pressure level using themethod of Jacobson (1999). In the model, dust aerosolsare lofted by vertical advection and diffusion. The verti-cal diffusion is calculated in CARMA using the MM5input meteorology. The model calculates the verticalpotential temperature, sensible heat flux, MoninObu-khov length and friction velocity using the MM5meteorological profile at each grid cell location. Themodel then calculates the vertical diffusion for each ver-tical level following the method developed by Zhang andAnthes (1982).</p><p>The dust model forecast is initialized by running themodel for a simulated 2 day (48 h) spin-up period.The spin-up uses the first 24 h of each daily MM5 fore-cast during the spin-up portion of the model run. Thedata from the end of the spin-up period is used as the</p></li><li><p>131B.H. Barnum et al. / Environmental Modelling &amp; Software 19 (2004) 129140</p><p>Fig. 1. Weather forecast data is run daily by the USAF Weather Agency for the theaters shown using MM5. Input meteorology used in CARMAis run with 45 km grid resolution for Africa (t09a) and Southwest Asia (T04a).</p><p>initial dust concentration condition at the beginning ofthe 72 h CARMA forecast. During model development,we compared 2, 5 and 10 day spin-up cycles for duststorm prediction. The use of 5 or 10 day forecasts werefound to be better in a few cases over Saharan Africafor the prediction of total dust loading; however, the 2day spin-up cycle was able to capture all of the mainfeatures required for dust forecasting. Since the modelwas to be used for daily operational forecasting atAFWA, the 2 day spin-up version was implemented.</p><p>2.1. Dust source model</p><p>The CARMA MM5 model uses a global dust sourcedatabase originally described by Ginoux et al. (2001).The dust database was developed using topography anddust sources regions identified using satellite data fromthe Total Ozone Mapping Spectrometer (TOMS). TheTOMS instrument measures the amount of ultravioletabsorption by dust aerosols by taking the ratio of 331nm and 360 nm measured radiance to the calculated radi-ances based on a model Rayleigh scattering atmosphere(Herman et al., 1997). The database uses TOMSobserved sources that are associated topographicaldepressions where sediments accumulate, such as theLake Chad Basin. The source areas are assigned a sourcestrength value between 0 and 1.0. The data is given ona global 1 1 grid, shown in Fig. 1, and is re-interp-olated to the MM5 grid used in the CARMA dust model.</p><p>The current implementation of the CARMA MM5model uses 10 particle size bins, which cover particlesizes from 0.1 to 10 m. Each of the bins are sized sothat the individual particle mass in each succeeding binhas a mass ratio of 2.71 times the mass of a particle</p><p>in the preceding bin size, as listed in Table 1 (Toon etal., 1988).</p><p>The model uses 3 dust particle size ranges or classesto describe soil fractional components consisting ofclays, silts and sand. Each class is assigned a componentfraction, which is 0.1 for clay, 0.33 for silt and 0.33 forsand. The clay component is any particle radius rangingfrom 0.1 m to 1.0 m, silts are 1.0 m to 10 m inparticle radius and sand is any particle larger than10.0 m.</p><p>Dust mobilization normally begins when the surfacewind velocity exceeds a threshold wind speed. At thethreshold speed, larger particles, which are not embed-ded in the soil matrix, are blown along the surface insaltation where they collide and liberate smaller particlesfrom the soil by sandblasting (Gillette, 1980). The thres-hold wind speed calculated in CARMA follows themethod developed by Iversen and White (1982). Themodel mobilizes larger sand particles at lower windspeeds to simulate the sandblasting process. Fig. 3 showsthe threshold wind speed versus particle size used inthe model.</p><p>The surface dust flux in CARMA is calculated usingthe net MM5 wind velocity at 10 m above ground (agl).The flux equation follows the formulation based on Gil-lette and Passi (1988). The dust source model first calcu-lates the mobilization threshold wind velocity at eachgrid location for each particle bin size. Where there ismeasurable accumulated precipitation in a 24 h period,the threshold wind velocity is set so that no dust flux isgenerated at the location.</p><p>The surface dust flux is then calculated for each par-ticle size bin using the MM5 forecast 10 m windspeed using:</p></li><li><p>132 B.H. Barnum et al. / Environmental Modelling &amp; Software 19 (2004) 129140</p><p>Fig. 2. (a) Dust source regions over North Africa and the Middle Easton a 01.0 scale, plotted with 0 (white) being a non-source region and0.6 (yellow) representing the most significant source regions. Themesoscale area is divided and grouped into distinct regions that areused for the computation of skill scores. (b) Dust source regions overthe Middle East and Central Asia with white being a non-source regionand yellow representing the most significant of source regions. Thearea is divided and grouped into distinct regions that are used for thecomputation of skill scores.</p><p>F(i,j,r) CS(i,j,r))w10m(i,j)ut(i,j,r)w10m(i,j)2.</p><p>where C is a model constant equal to 2.34 E-17 gs2m5, used to control the total amount of dust flux emis-sion. C depends on the particular weather model and gridscale used. F(i,j,r) is the surface dust flux in gm/m2-s, ateach of the i,j, grid locations and particle bin number r,S(i,j,r) is the Ginoux database source strength for eachparticle class size, w10m(i,j) is the MM5 wind speed at10 m agl, and ut(ijr) is the calculated threshold wind speedfor each grid location and particle bin size (Ginoux etal., 2001; Chin et al., 2002).</p><p>Table 1CARMA-Dust model particle bin sizes and estimated particle fall velo-cities at sea level. The 10 particle bins are sized such that the massof a particle in the i + 1 bin is 2.71 times the mass of the particle inthe preceding ith bin</p><p>Particle radius (in Particle mass (kg) Particle fall velocity at seam) level (m/s)</p><p>0.50 1.391015 0.00010.71 3.771015 0.00020.97 1.021014 0.00031.36 2.781014 0.00061.89 7.531014 0.00122.64 2.041013 0.00233.68 5.561013 0.00445.14 1.511012 0.00847.17 4.091012 0.016310.00 1.111011 0.0316</p><p>Fig. 3. Dust threshold surface wind velocity calculated in CARMAusing the method described by Iverson and White (1982). Notice thatsmaller size dust grains require higher surface wind speeds to mobilizesince they are embedded in the soil matrix until liberated by larger par-ticles.</p><p>2.2. Dust deposition and advection</p><p>Dust deposition in CARMA is calculated using a 2layer method described by Shao (2000). The particle ver-tical deposition velocity combines the effects of bound-ary layer turbulent motion, molecular diffusion and sedi-mentation. In this way, the particle deposition in thelowest model layer is controlled by the boundary layermeteorological conditions forecast by MM5. The particlesedimentation velocity is calculated at each model layerand particle...</p></li></ul>