[1897] | 1 | ! Copyright 2013-2015,2017 Université de Reims Champagne-Ardenne |
---|
[1793] | 2 | ! Contributor: J. Burgalat (GSMA, URCA) |
---|
| 3 | ! email of the author : jeremie.burgalat@univ-reims.fr |
---|
| 4 | ! |
---|
| 5 | ! This software is a computer program whose purpose is to compute |
---|
| 6 | ! microphysics processes using a two-moments scheme. |
---|
| 7 | ! |
---|
| 8 | ! This library is governed by the CeCILL-B license under French law and |
---|
| 9 | ! abiding by the rules of distribution of free software. You can use, |
---|
| 10 | ! modify and/ or redistribute the software under the terms of the CeCILL-B |
---|
| 11 | ! license as circulated by CEA, CNRS and INRIA at the following URL |
---|
| 12 | ! "http://www.cecill.info". |
---|
| 13 | ! |
---|
| 14 | ! As a counterpart to the access to the source code and rights to copy, |
---|
| 15 | ! modify and redistribute granted by the license, users are provided only |
---|
| 16 | ! with a limited warranty and the software's author, the holder of the |
---|
| 17 | ! economic rights, and the successive licensors have only limited |
---|
| 18 | ! liability. |
---|
| 19 | ! |
---|
| 20 | ! In this respect, the user's attention is drawn to the risks associated |
---|
| 21 | ! with loading, using, modifying and/or developing or reproducing the |
---|
| 22 | ! software by the user in light of its specific status of free software, |
---|
| 23 | ! that may mean that it is complicated to manipulate, and that also |
---|
| 24 | ! therefore means that it is reserved for developers and experienced |
---|
| 25 | ! professionals having in-depth computer knowledge. Users are therefore |
---|
| 26 | ! encouraged to load and test the software's suitability as regards their |
---|
| 27 | ! requirements in conditions enabling the security of their systems and/or |
---|
| 28 | ! data to be ensured and, more generally, to use and operate it in the |
---|
| 29 | ! same conditions as regards security. |
---|
| 30 | ! |
---|
| 31 | ! The fact that you are presently reading this means that you have had |
---|
| 32 | ! knowledge of the CeCILL-B license and that you accept its terms. |
---|
| 33 | |
---|
| 34 | !! file: mm_clouds.f90 |
---|
| 35 | !! summary: Clouds microphysics module |
---|
| 36 | !! author: J. Burgalat |
---|
[1897] | 37 | !! date: 2013-2015,2017 |
---|
[1793] | 38 | |
---|
| 39 | MODULE MM_CLOUDS |
---|
| 40 | !! Clouds microphysics module. |
---|
| 41 | !! |
---|
| 42 | !! The module contains all definitions of the microphysics processes related to clouds: |
---|
| 43 | !! |
---|
| 44 | !! - [nucleation](page/clouds.html#nucleation) |
---|
| 45 | !! - [condensation](page/clouds.html#condensation) |
---|
| 46 | !! - [sedimentation](page/clouds.html#sedimentation) |
---|
| 47 | !! |
---|
| 48 | !! |
---|
| 49 | !! The interface methods always use the global variables defined in [[mm_globals(module)]] when values |
---|
| 50 | !! (any kind, temperature, pressure, moments...) over the vertical grid are required. |
---|
| 51 | !! Consequently, all these functions only deals with output argument which are most of the time the |
---|
| 52 | !! tendencies of relevant variables on the atmospheric column. |
---|
| 53 | !! |
---|
| 54 | !! @note |
---|
| 55 | !! Tendencies returned by public methods are always defined from __TOP__ of the atmosphere to the |
---|
| 56 | !! __GROUND__. |
---|
| 57 | USE MM_MPREC |
---|
| 58 | USE MM_GLOBALS |
---|
| 59 | USE MM_METHODS |
---|
| 60 | IMPLICIT NONE |
---|
| 61 | |
---|
| 62 | PRIVATE |
---|
| 63 | |
---|
| 64 | PUBLIC :: mm_cloud_microphysics, mm_cloud_sedimentation, mm_cloud_nucond |
---|
| 65 | |
---|
| 66 | CONTAINS |
---|
| 67 | |
---|
| 68 | !============================================================================ |
---|
| 69 | ! CLOUDS MICROPHYSICS INTERFACE SUBROUTINE |
---|
| 70 | !============================================================================ |
---|
| 71 | |
---|
| 72 | SUBROUTINE mm_cloud_microphysics(dm0a,dm3a,dm0n,dm3n,dm3i,dgazs) |
---|
| 73 | !! Get the evolution of moments tracers through clouds microphysics processes. |
---|
| 74 | !! |
---|
| 75 | !! The subroutine is a wrapper to the clouds microphysics methods. It computes the tendencies of moments |
---|
| 76 | !! tracers for nucleation, condensation and sedimentation processes for the atmospheric column. |
---|
| 77 | !! |
---|
| 78 | !! @note |
---|
| 79 | !! Both __dm3i__ and __dgazs__ are 2D-array with the vertical layers in first dimension and the number |
---|
| 80 | !! of ice components in the second. |
---|
| 81 | REAL(kind=mm_wp), DIMENSION(:), INTENT(out) :: dm0a |
---|
| 82 | !! Tendency of the 0th order moment of the aerosols (fractal mode) (\(m^{-3}\)). |
---|
| 83 | REAL(kind=mm_wp), DIMENSION(:), INTENT(out) :: dm3a |
---|
| 84 | !! Tendency of the 3rd order moment of the aerosols distribution (fractal mode) (\(m^{3}.m^{-3}\)) . |
---|
| 85 | REAL(kind=mm_wp), DIMENSION(:), INTENT(out) :: dm0n |
---|
| 86 | !! Tendency of the 0th order moment of the aerosols distribution (fractal mode) (\(m^{-3}\)). |
---|
| 87 | REAL(kind=mm_wp), DIMENSION(:), INTENT(out) :: dm3n |
---|
| 88 | !! Tendency of the 3rd order moment of the ccn distribution (\(m^{3}.m^{-3}\)). |
---|
| 89 | REAL(kind=mm_wp), DIMENSION(:,:), INTENT(out) :: dm3i |
---|
| 90 | !! Tendencies of the 3rd order moments of each ice components (\(m^{3}.m^{-3}\)). |
---|
| 91 | REAL(kind=mm_wp), DIMENSION(:,:), INTENT(out) :: dgazs |
---|
| 92 | !! Tendencies of each condensible gaz species (\(mol.mol^{-1}\)). |
---|
| 93 | REAL(kind=mm_wp), DIMENSION(:), ALLOCATABLE :: zdm0n,zdm3n |
---|
| 94 | REAL(kind=mm_wp), DIMENSION(:,:), ALLOCATABLE :: zdm3i |
---|
| 95 | INTEGER :: i |
---|
| 96 | dm0a = 0._mm_wp ; dm3a = 0._mm_wp |
---|
| 97 | dm0n = 0._mm_wp ; dm3n = 0._mm_wp |
---|
| 98 | dm3i = 0._mm_wp ; dgazs = 0._mm_wp |
---|
| 99 | |
---|
| 100 | IF (mm_w_cloud_nucond) THEN |
---|
| 101 | ! Calls condensation/nucleation (and update saturation ratio diagnostic) |
---|
| 102 | call mm_cloud_nucond(dm0a,dm3a,dm0n,dm3n,dm3i,dgazs,mm_gazs_sat) |
---|
| 103 | ENDIF |
---|
| 104 | |
---|
| 105 | IF (mm_w_cloud_sed) THEN |
---|
| 106 | ! Calls sedimentation |
---|
| 107 | ALLOCATE(zdm0n(mm_nla),zdm3n(mm_nla),zdm3i(mm_nla,mm_nesp)) |
---|
| 108 | call mm_cloud_sedimentation(zdm0n,zdm3n,zdm3i) |
---|
| 109 | |
---|
| 110 | ! computes precipitations / ice fluxes |
---|
| 111 | mm_ccn_prec = SUM(zdm3n*mm_dzlev) |
---|
| 112 | mm_ccn_flux(:) = get_mass_flux(mm_rhoaer,mm_m3ccn(:)) |
---|
| 113 | |
---|
| 114 | DO i=1, mm_nesp |
---|
| 115 | mm_ice_prec(i) = SUM(zdm3i(:,i)*mm_dzlev) |
---|
| 116 | mm_ice_fluxes(:,i) = get_mass_flux(mm_xESPS(i)%rho,mm_m3ice(:,i)) |
---|
| 117 | ENDDO |
---|
| 118 | ! updates tendencies |
---|
| 119 | dm0n = dm0n + zdm0n |
---|
| 120 | dm3n = dm3n + zdm3n |
---|
| 121 | dm3i = dm3i + zdm3i |
---|
| 122 | ENDIF |
---|
| 123 | |
---|
| 124 | END SUBROUTINE mm_cloud_microphysics |
---|
| 125 | |
---|
| 126 | !----------------------------------------------------------------------------- |
---|
| 127 | ! NUCLEATION/CONDENSATION PROCESS RELATED METHODS |
---|
| 128 | !----------------------------------------------------------------------------- |
---|
| 129 | |
---|
| 130 | SUBROUTINE mm_cloud_nucond(dm0a,dm3a,dm0n,dm3n,dm3i,dgazs,gazsat) |
---|
| 131 | !! Get moments tendencies through nucleation/condensation/evaporation. |
---|
| 132 | !! |
---|
| 133 | !! The method is a wrapper of [[mm_clouds(module):nc_esp(subroutine)]] which computes the |
---|
| 134 | !! tendencies of tracers for all the condensible species given in the vector __xESPS__. |
---|
| 135 | !! |
---|
| 136 | !! @warning |
---|
| 137 | !! __xESPS__, __m3i__ and __gazes__ must share the same indexing. For example if __xESPS(IDX)__ |
---|
| 138 | !! corresponds to \(CH_{4}\) properties then __m3i(IDX)__ must be the total volume of solid |
---|
| 139 | !! \(CH_{4}\) (ice) and __gazs(IDX)__ its vapor mole fraction. |
---|
| 140 | REAL(kind=mm_wp), DIMENSION(:), INTENT(out) :: dm0a |
---|
| 141 | !! Tendency of the 0th order moment of the aerosols (fractal mode) (\(m^{-3}\)). |
---|
| 142 | REAL(kind=mm_wp), DIMENSION(:), INTENT(out) :: dm3a |
---|
| 143 | !! Tendency of the 3rd order moment of the aerosols distribution (fractal mode) (\(m^{3}.m^{-3}\)). |
---|
| 144 | REAL(kind=mm_wp), DIMENSION(:), INTENT(out) :: dm0n |
---|
| 145 | !! Tendency of the 0th order moment of the aerosols distribution (fractal mode) (\(m^{-3}\)). |
---|
| 146 | REAL(kind=mm_wp), DIMENSION(:), INTENT(out) :: dm3n |
---|
| 147 | !! Tendency of the 3rd order moment of the ccn distribution (\(m^{3}.m^{-3}\)). |
---|
| 148 | REAL(kind=mm_wp), DIMENSION(:,:), INTENT(out) :: dm3i |
---|
| 149 | !! Tendencies of the 3rd order moments of each ice components (\(m^{3}.m^{-3}\)). |
---|
| 150 | REAL(kind=mm_wp), DIMENSION(:,:), INTENT(out) :: dgazs |
---|
| 151 | !! Tendencies of each condensible gaz species (\(mol.mol^{-1}\)) . |
---|
| 152 | REAL(kind=mm_wp), DIMENSION(:,:), INTENT(out) :: gazsat |
---|
| 153 | !! Saturation ratio of each condensible specie. |
---|
| 154 | INTEGER :: i,idx,ng |
---|
| 155 | TYPE(mm_esp) :: xESP |
---|
| 156 | REAL(kind=mm_wp), DIMENSION(:,:), ALLOCATABLE :: zdm0a,zdm3a,zdm0n,zdm3n |
---|
| 157 | ALLOCATE(zdm0a(mm_nla,mm_nesp),zdm3a(mm_nla,mm_nesp), & |
---|
| 158 | zdm0n(mm_nla,mm_nesp),zdm3n(mm_nla,mm_nesp)) |
---|
| 159 | zdm0a(:,:) = 0._mm_wp ; zdm3a(:,:) = 0._mm_wp |
---|
| 160 | zdm0n(:,:) = 0._mm_wp ; zdm3n(:,:) = 0._mm_wp |
---|
| 161 | DO i = 1, mm_nesp |
---|
| 162 | call nc_esp(mm_xESPS(i),mm_gazs(:,i),mm_m3ice(:,i),dgazs(:,i),dm3i(:,i), & |
---|
| 163 | zdm0a(:,i),zdm3a(:,i),zdm0n(:,i),zdm3n(:,i),gazsat(:,i)) |
---|
| 164 | ENDDO |
---|
| 165 | |
---|
| 166 | ! Computes balance : |
---|
| 167 | ! Each ice components has been treated independently from the others, and |
---|
| 168 | ! their tendencies are returned as is (just converted in input units). |
---|
| 169 | ! |
---|
| 170 | ! Aerosols and CCN distribution evolution depends on the ice components: |
---|
| 171 | ! - For nucleation only creation of CCN can occur. |
---|
| 172 | ! - For condensation only loss of CCN can occur. |
---|
| 173 | ! We use the simple following rule : |
---|
| 174 | ! The global variation of CCN (and thus aerosols) is determined from the |
---|
| 175 | ! most intense activity of the ice components. |
---|
| 176 | ! that is the maximum value of the CCN tendencies regardless of its sign. |
---|
| 177 | DO i=1, mm_nla |
---|
| 178 | idx = MAXLOC(zdm0n(i,:),DIM=1) ! WARNING this is not the definition above (should be in ABS() func) |
---|
| 179 | dm0n(i) = zdm0n(i,idx) |
---|
| 180 | dm3n(i) = zdm3n(i,idx) |
---|
| 181 | dm0a(i) = zdm0a(i,idx) |
---|
| 182 | dm3a(i) = zdm3a(i,idx) |
---|
| 183 | ! all ice are returned but we must convert their units |
---|
| 184 | dm3i(i,:) = dm3i(i,:) |
---|
| 185 | ENDDO |
---|
| 186 | END SUBROUTINE mm_cloud_nucond |
---|
| 187 | |
---|
| 188 | SUBROUTINE nc_esp(xESP,vapX,m3iX,dvapX,dm3iX,dm0aer,dm3aer,dm0ccn,dm3ccn,Xsat) |
---|
| 189 | !! Get moments tendencies through nucleation/condensation/evaporation of a given condensible specie. |
---|
| 190 | !! |
---|
| 191 | !! The method computes the global tendencies of the aerosols, ccn and "ice" moments through cloud |
---|
| 192 | !! microphysics processes (nucleation & condensation). |
---|
| 193 | !! |
---|
| 194 | !! @warning |
---|
| 195 | !! Input quantities __m3iX__,__m3iO__, __m0aer__,__m3aer__, __m0ccn__,__m3ccn__ are assumed to be in |
---|
| 196 | !! \(X.kg^{-1}\) (where X is the unit of the moment that is, a number for M0 and a volume - \(m^3\) |
---|
| 197 | !! for M3) ; __vapX__ must be expressed in term of molar fraction. |
---|
| 198 | TYPE(mm_esp), INTENT(in) :: xESP |
---|
| 199 | !! Condensate specie properties. |
---|
| 200 | REAL(kind=mm_wp),INTENT(in), DIMENSION(:) :: vapX |
---|
| 201 | !! Gas specie molar fraction on the vertical grid from __TOP__ to __GROUND__ (\(mol.mol^{-1}\)). |
---|
| 202 | REAL(kind=mm_wp),INTENT(in), DIMENSION(:) :: m3iX |
---|
| 203 | !! 3rd order moment of the ice component (\(m^{3}.m^{-3}\)). |
---|
| 204 | REAL(kind=mm_wp),INTENT(out), DIMENSION(:) :: dvapX |
---|
| 205 | !! Tendency of gas specie (\(mol.mol^{-1}\)). |
---|
| 206 | REAL(kind=mm_wp),INTENT(out), DIMENSION(:) :: dm3iX |
---|
| 207 | !! Tendency of the 3rd order moment of the ice component (\(m^{3}.m^{-3}\)). |
---|
| 208 | REAL(kind=mm_wp),INTENT(out), DIMENSION(:) :: dm0aer |
---|
| 209 | !! Tendency of the 0th order moment of the fractal mode distribution (\(m^{-3}\)). |
---|
| 210 | REAL(kind=mm_wp),INTENT(out), DIMENSION(:) :: dm3aer |
---|
| 211 | !! Tendency of the 3rd order moment of the fractal mode distribution (\(m^{3}.m^{-3}\)). |
---|
| 212 | REAL(kind=mm_wp),INTENT(out), DIMENSION(:) :: dm0ccn |
---|
| 213 | !! Tendency of the 0th order moment of the ccn distribution (\(m^{-3}\)). |
---|
| 214 | REAL(kind=mm_wp),INTENT(out), DIMENSION(:) :: dm3ccn |
---|
| 215 | !! Tendency of the 3rd order moment of the ccn distribution (\(m^{3}.m^{-3}\)). |
---|
| 216 | REAL(kind=mm_wp),INTENT(out), DIMENSION(:) :: Xsat |
---|
| 217 | !! Saturation ratio values on the vertical grid (--). |
---|
| 218 | INTEGER :: i |
---|
| 219 | REAL(kind=mm_wp) :: bef,aft |
---|
| 220 | REAL(kind=mm_wp), DIMENSION(SIZE(vapX)) :: sm0a,sm3a,sm0n,sm3n,sm3iX |
---|
| 221 | REAL(kind=mm_wp), DIMENSION(SIZE(vapX)) :: zm0a,zm3a,zm0n,zm3n,zm3iX,zvapX |
---|
| 222 | REAL(kind=mm_wp), DIMENSION(SIZE(vapX)) :: pX,sig,qsat,lheat,seq,up,down, & |
---|
| 223 | ctot,newvap,nucr,grate,cm0,cm3 |
---|
| 224 | ! Initialization : |
---|
| 225 | ! Copy input argument and convert units X.m-3 -> X.kg-1) |
---|
| 226 | ! sXXX is the initial converted value saved |
---|
| 227 | sm3iX = m3iX/mm_rhoair |
---|
| 228 | sm0a = mm_m0aer_f/mm_rhoair ; sm3a = mm_m3aer_f/mm_rhoair |
---|
| 229 | sm0n = mm_m0ccn/mm_rhoair ; sm3n = mm_m3ccn/mm_rhoair |
---|
| 230 | ! zXXX is our working copy |
---|
| 231 | zm3ix = sm3ix ; zm0a = sm0a ; zm3a = sm3a ; zm0n = sm0n ; zm3n = sm3n |
---|
| 232 | |
---|
| 233 | ! Molar fraction of X specie is set in mass mixing ratio |
---|
| 234 | zvapX = vapX * xESP%fmol2fmas |
---|
| 235 | ! Surface tension |
---|
| 236 | sig = mm_sigX(mm_temp,xESP) |
---|
| 237 | ! X specie mass mixing ratio at saturation |
---|
| 238 | qsat = mm_qsatX(mm_temp,mm_play,xESP) |
---|
| 239 | ! partial pressure of X specie |
---|
| 240 | pX = vapX * mm_play |
---|
| 241 | ! Saturation ratio |
---|
| 242 | Xsat = zvapX / qsat |
---|
| 243 | ! Equilibrium saturation near the drop |
---|
| 244 | seq = dexp(2._mm_wp*sig*xESP%masmol/(xESP%rho*mm_rgas*mm_temp*mm_drad)) |
---|
| 245 | ! Latent heat released |
---|
| 246 | lheat = mm_lheatX(mm_temp,xESP) |
---|
| 247 | ! Gets nucleation rate (ccn radius is the monomer !) |
---|
| 248 | call nuc_rate((/(mm_rm, i=1,mm_nla)/),mm_temp,xESP,pX,Xsat,nucr) |
---|
| 249 | ! Gets growth rate |
---|
| 250 | call growth_rate(mm_temp,mm_play,pX/Xsat,xESP,seq,mm_drad,grate) |
---|
| 251 | ctot = zvapx + xESP%rho * m3iX |
---|
| 252 | up = vapx + mm_dt * grate * 4._mm_wp * mm_pi * xESP%rho * mm_drad * seq * zm0n |
---|
| 253 | down = 1._mm_wp + mm_dt * grate * 4._mm_wp * mm_pi * xESP%rho * mm_drad / qsat * zm0n |
---|
| 254 | ! gets new vapor X specie mass mixing ratio : cannot be greater than the |
---|
| 255 | ! total gas + ice and lower than nothing :) |
---|
| 256 | newvap = max(min(up/down,ctot),0._mm_wp) |
---|
| 257 | ! gets "true" growth rate |
---|
| 258 | grate = grate * (newvap/qsat - seq) |
---|
| 259 | |
---|
| 260 | ! computes tendencies through condensation |
---|
| 261 | ! 1) check for the specific case : NO ICE and SUBLIMATION |
---|
| 262 | WHERE (zm3iX <= 0._mm_wp .AND. grate <= 0._mm_wp) |
---|
| 263 | ! no ice and sublimation : reset ice to 0 |
---|
| 264 | zm3iX = 0._mm_wp |
---|
| 265 | ELSEWHERE |
---|
| 266 | ! update ice volume ... |
---|
| 267 | zm3iX = zm3iX + mm_dt*grate*4._mm_wp*mm_pi*mm_drad*zm0n |
---|
| 268 | ! ... and check if there ice left in the ccn |
---|
| 269 | WHERE (zm3ix <= 0._mm_wp) |
---|
| 270 | zm3ix = 0._mm_wp |
---|
| 271 | zm0a = zm0a + zm0n ; zm0n = 0._mm_wp |
---|
| 272 | zm3a = zm3a + zm3n ; zm3n = 0._mm_wp |
---|
| 273 | ENDWHERE |
---|
| 274 | ENDWHERE |
---|
| 275 | |
---|
| 276 | ! computes tendencies |
---|
| 277 | ! all of these tendencies are in X.kg-1 |
---|
| 278 | dm0aer = zm0a - sm0a |
---|
| 279 | dm3aer = zm3a - sm3a |
---|
| 280 | dm0ccn = zm0n - sm0n |
---|
| 281 | dm3ccn = zm3n - sm3n |
---|
| 282 | dm3ix = zm3ix - sm3ix |
---|
| 283 | ! and this one is in mol.mol-1 |
---|
| 284 | dvapx = -xESP%rho * dm3ix / xESP%fmol2fmas |
---|
| 285 | |
---|
| 286 | ! reset temporary arrays to initial values |
---|
| 287 | zm3ix = sm3ix ; zm0a = sm0a ; zm3a = sm3a ; zm0n = sm0n ; zm3n = sm3n |
---|
| 288 | |
---|
| 289 | ! Computes global tendencies (nucleation && condensation) |
---|
| 290 | ! |
---|
| 291 | ! for nucleation we have the following equations: |
---|
| 292 | ! dMaer_(k)/dt = - dMccn_(k)/dt (conservation of aerosols+ccn) (1) |
---|
| 293 | ! dMaer_(k)/dt = - 4*PI*nucr/rm * Maer_(k+3) (2) |
---|
| 294 | ! = - 4*PI*nucr/rm * alpha(k+3)/alpha(k) * Maer_(k) |
---|
| 295 | ! where alpha(k+3) = Maer_(k+3)/Maer_(0) / rc**(k+3) |
---|
| 296 | ! We solve (implicit scheme) : |
---|
| 297 | ! CONST_M(k) = 4*PI*nucr/rm * alpha(k+3)/alpha(k)*rc**3 * dt |
---|
| 298 | ! Maer_(k)[t+dt] = 1/(1+CONST_M(k)) * Maer_(k)[t] (3) |
---|
| 299 | ! Then, from eq. 2: |
---|
| 300 | ! Mccn_(k)[t+dt] = Mccn_(k)[t] + CONST_M(k)/(1+CONST_M(k))*Maer_(k)[t] (4) |
---|
| 301 | ! |
---|
| 302 | cm0 = 4._mm_wp*mm_pi*nucr/mm_rm*mm_alpha_f(3._mm_wp)*mm_rcf**3*mm_dt |
---|
| 303 | cm3 = 4._mm_wp*mm_pi*nucr/mm_rm*mm_alpha_f(6._mm_wp)/mm_alpha_f(3._mm_wp)*mm_rcf**3*mm_dt |
---|
| 304 | zm0a = 1._mm_wp/(1._mm_wp+cm0) * zm0a |
---|
| 305 | zm3a = 1._mm_wp/(1._mm_wp+cm0) * zm3a |
---|
| 306 | WHERE (zm0a <= 0._mm_wp .OR. zm3a <= 0._mm_wp) |
---|
| 307 | zm0a=0._mm_wp |
---|
| 308 | zm3a=0._mm_wp |
---|
| 309 | zm0n = zm0n + sm0a |
---|
| 310 | zm3n = zm3n + sm3a |
---|
| 311 | ELSEWHERE |
---|
| 312 | zm0n = zm0n + cm0/(1.+cm0)*zm0a |
---|
| 313 | zm3n = zm3n + cm3/(1.+cm3)*zm3a |
---|
| 314 | ENDWHERE |
---|
| 315 | |
---|
| 316 | ! Adds condensation tendencies |
---|
| 317 | zm0a = zm0a + dm0aer |
---|
| 318 | zm3a = zm3a + dm3aer |
---|
| 319 | zm0n = zm0n + dm0ccn |
---|
| 320 | zm3n = zm3n + dm3ccn |
---|
| 321 | |
---|
| 322 | ! Computes balance |
---|
| 323 | IF (mm_debug) THEN |
---|
| 324 | WRITE(*,'(a)') "Condensation/nucleation balance :" |
---|
| 325 | DO i=1,mm_nla |
---|
| 326 | bef = sm0a(i) + sm0n(i) |
---|
| 327 | aft = zm0a(i) + zm0n(i) |
---|
| 328 | IF (ABS(bef-aft)/bef > 1e-10_mm_wp) WRITE(*,'((a),I2.2,(a))') & |
---|
| 329 | "[WARNING] nc_esp speaking: Total number not conserved (z=",i,")" |
---|
| 330 | bef = sm3a(i) + sm3n(i) |
---|
| 331 | aft = zm3a(i) + zm3n(i) |
---|
| 332 | IF (ABS(bef-aft)/bef > 1e-10_mm_wp) WRITE(*,'((a),I2.2,(a))') & |
---|
| 333 | "[WARNING] nc_esp speaking: Total volume not conserved (z=",i,")" |
---|
| 334 | ENDDO |
---|
| 335 | ENDIF |
---|
| 336 | |
---|
| 337 | ! Now updates tendencies |
---|
| 338 | dm0aer = (zm0a-sm0a)*mm_rhoair |
---|
| 339 | dm3aer = (zm3a-sm3a)*mm_rhoair |
---|
| 340 | dm0ccn = (zm0n-sm0n)*mm_rhoair |
---|
| 341 | dm3ccn = (zm3n-sm3n)*mm_rhoair |
---|
| 342 | |
---|
| 343 | END SUBROUTINE nc_esp |
---|
| 344 | |
---|
| 345 | SUBROUTINE nuc_rate(rccn,temp,xESP,pvp,sat,rate) |
---|
| 346 | !! Get nucleation rate. |
---|
| 347 | !! |
---|
| 348 | !! The method computes the heterogeneous nucleation rate for the given specie on a fractal particle |
---|
| 349 | !! of size __rccn__. |
---|
| 350 | !! Except __xESP__, all arguments are vectors of the same size (vertical grid). |
---|
| 351 | REAL(kind=mm_wp), INTENT(in), DIMENSION(:) :: rccn !! Radius of the cloud condensation nuclei (m). |
---|
| 352 | REAL(kind=mm_wp), INTENT(in), DIMENSION(:) :: temp !! Temperature (K). |
---|
| 353 | REAL(kind=mm_wp), INTENT(in), DIMENSION(:) :: pvp !! Partial vapor pressure of X specie (Pa). |
---|
| 354 | REAL(kind=mm_wp), INTENT(in), DIMENSION(:) :: sat !! Saturation ratio of given specie (--). |
---|
| 355 | TYPE(mm_esp), INTENT(in) :: xESP !! X specie properties (--). |
---|
| 356 | REAL(kind=mm_wp), INTENT(out), DIMENSION(:) :: rate !! The nucleation rate (\(m^-{2}.s^{-1}\)). |
---|
| 357 | INTEGER :: nv |
---|
| 358 | REAL(kind=mm_wp), DIMENSION(:), ALLOCATABLE :: nX,rstar,gstar,x,zeldov,deltaf, & |
---|
| 359 | fsh,fstar,sig |
---|
| 360 | nv = SIZE(rccn) |
---|
| 361 | ALLOCATE(nX(nv), rstar(nv), gstar(nv), x(nv), zeldov(nv), & |
---|
| 362 | deltaf(nv), fsh(nv), fstar(nv)) |
---|
| 363 | ! Activation condition |
---|
| 364 | WHERE (sat > 1._mm_wp) |
---|
| 365 | sig = mm_sigX(temp,xESP) |
---|
| 366 | nX = pvp/mm_kboltz/temp |
---|
| 367 | rstar = 2._mm_wp*sig*xESP%vol/(mm_kboltz*temp*dlog(sat)) |
---|
| 368 | ! curvature radius |
---|
| 369 | x = rccn/rstar |
---|
| 370 | fsh = mm_fshape(xESP%mteta,x) |
---|
| 371 | fstar = (4._mm_wp/3._mm_wp*mm_pi)*sig*(rstar**2.)*fsh |
---|
| 372 | deltaf=MIN(MAX((2.*mm_fdes-mm_fdif-fstar)/(mm_kboltz*temp),-100._mm_wp),100._mm_wp) |
---|
| 373 | WHERE (deltaf == -100._mm_wp) |
---|
| 374 | rate = 0._mm_wp |
---|
| 375 | ELSEWHERE |
---|
| 376 | gstar = 4._mm_wp*mm_pi*(rstar**3)/(3._mm_wp*xESP%vol) |
---|
| 377 | zeldov = dsqrt(fstar/(3._mm_wp*mm_pi*mm_kboltz*temp*(gstar**2))) |
---|
| 378 | rate = zeldov*mm_kboltz*temp*(nX*rstar)**2._mm_wp*dexp(deltaf)/ & |
---|
| 379 | (fsh*mm_nus*xESP%mas) |
---|
| 380 | ENDWHERE |
---|
| 381 | ELSEWHERE |
---|
| 382 | rate = 0._mm_wp |
---|
| 383 | ENDWHERE |
---|
| 384 | DEALLOCATE(nX,rstar,gstar,x,zeldov,deltaf,fsh,fstar) |
---|
| 385 | RETURN |
---|
| 386 | END SUBROUTINE nuc_rate |
---|
| 387 | |
---|
| 388 | SUBROUTINE growth_rate(temp,pres,pXsat,xESP,seq,drad,rate) |
---|
| 389 | !! Get growth rate through condensation/evaporation process. |
---|
| 390 | !! |
---|
| 391 | !! The method computes the growth rate a drop through condensation/evaporation processes: |
---|
| 392 | !! |
---|
| 393 | !! $$ r \times \frac{dr}{dt} = g_{rate} \times (S - S_{eq}) $$ |
---|
| 394 | !! |
---|
| 395 | !! Except __xESP__ which is a scalar, all arguments are vectors of the same size (vertical grid). |
---|
| 396 | REAL(kind=mm_wp), INTENT(in), DIMENSION(:) :: temp !! Temperature (K). |
---|
| 397 | REAL(kind=mm_wp), INTENT(in), DIMENSION(:) :: pres !! Pressure level (Pa). |
---|
| 398 | REAL(kind=mm_wp), INTENT(in), DIMENSION(:) :: seq !! Saturation vapor pressure of specie (Pa). |
---|
| 399 | REAL(kind=mm_wp), INTENT(in), DIMENSION(:) :: drad !! Specie properties. |
---|
| 400 | REAL(kind=mm_wp), INTENT(in), DIMENSION(:) :: pXsat !! Equilibrium saturation near the drop. |
---|
| 401 | TYPE(mm_esp), INTENT(in) :: xESP !! Drop radius (m). |
---|
| 402 | REAL(kind=mm_wp), INTENT(out), DIMENSION(:) :: rate !! Growth rate (\(m^{2}.s^{-1}\)). |
---|
| 403 | REAL(kind=mm_wp), DIMENSION(:), ALLOCATABLE :: k,knu,slf,rkc,rdc,l,dv |
---|
| 404 | INTEGER :: n |
---|
| 405 | n = SIZE(temp) |
---|
| 406 | ALLOCATE(k(n),knu(n),slf(n),rkc(n),rdc(n),l(n),dv(n)) |
---|
| 407 | |
---|
| 408 | ! N2 (air) Thermal conductivity (where does it come from ?) |
---|
| 409 | k(:) = ( 2.857e-2_mm_wp*temp-0.5428_mm_wp)*4.184e-3_mm_wp |
---|
| 410 | ! Gas mean free path |
---|
| 411 | l(:) = mm_lambda_g(temp,pres) |
---|
| 412 | ! Diffusion coefficient of X gas |
---|
| 413 | Dv(:) = 1._mm_wp/3._mm_wp*dsqrt(8._mm_wp*mm_rgas*temp(:)/(mm_pi*xESP%masmol))*mm_kboltz*temp(:) / & |
---|
| 414 | (mm_pi*pres(:)*(mm_air_rad+xESP%ray)**2* dsqrt(1._mm_wp+xESP%fmol2fmas)) |
---|
| 415 | knu(:) = l(:)/drad(:) ! The knudsen number of the drop |
---|
| 416 | slf(:) = (1.333_mm_wp+0.71_mm_wp/knu(:))/(1._mm_wp+1._mm_wp/knu(:)) ! Slip flow correction |
---|
| 417 | Dv(:) = Dv(:)/(1._mm_wp+slf(:)*knu(:)) |
---|
| 418 | ! latent heat resistance coefficient |
---|
| 419 | rkc(:) = mm_lheatX(temp(:),xESP)**2 * xESP%rho * xESP%masmol / (k(:)*mm_rgas*temp(:)**2) |
---|
| 420 | ! Diffusion resistance coefficient |
---|
| 421 | rdc(:) = mm_rgas * temp(:) * xESP%rho / (Dv(:)*pXsat(:)*xESP%masmol) |
---|
| 422 | ! Growth rate: rdr/dt = rate * (S-Seq) ; rate is returned |
---|
| 423 | rate(:) = 1._mm_wp / (seq(:)*rkc(:)+rdc(:)) |
---|
| 424 | RETURN |
---|
| 425 | END SUBROUTINE growth_rate |
---|
| 426 | |
---|
| 427 | |
---|
| 428 | !----------------------------------------------------------------------------- |
---|
| 429 | ! SEDIMENTATION PROCESS RELATED METHODS |
---|
| 430 | !----------------------------------------------------------------------------- |
---|
| 431 | |
---|
| 432 | SUBROUTINE mm_cloud_sedimentation(dm0n,dm3n,dm3i) |
---|
| 433 | !! Compute the tendency of _clouds_ related moments through sedimentation process. |
---|
| 434 | !! |
---|
| 435 | !! The method computes the tendencies of moments related to cloud microphysics through |
---|
| 436 | !! sedimentation process. The algorithm used here differs from |
---|
| 437 | !! [[mm_haze(module):mm_haze_sedimentation(subroutine)]] as all moments settle with the same |
---|
| 438 | !! terminal velocity which is computed with the average drop radius of the size distribution. |
---|
| 439 | !! We simply compute an _exchange matrix_ that stores the new positions of each cells through |
---|
| 440 | !! sedimentation process and then computes the matrix |
---|
| 441 | !! product with input moments values to get final tendencies. |
---|
| 442 | REAL(kind=mm_wp), INTENT(out), DIMENSION(:) :: dm0n |
---|
| 443 | !! Tendency of the 0th order moment of the ccn distribution (\(m^{-3}\)). |
---|
| 444 | REAL(kind=mm_wp), INTENT(out), DIMENSION(:) :: dm3n |
---|
| 445 | !! Tendency of the 3rd order moment of the ccn distribution (\(m^{3}.m^{-3}\)). |
---|
| 446 | REAL(kind=mm_wp), INTENT(out), DIMENSION(:,:) :: dm3i |
---|
| 447 | !! Tendencies of the 3rd order moment of each ice component of the cloud (\(m^{3}m^{-3}\)). |
---|
| 448 | INTEGER :: im,nm |
---|
| 449 | REAL(kind=mm_wp), DIMENSION(:,:), ALLOCATABLE :: moms, momsf,chg_matrix |
---|
| 450 | nm = 2 + mm_nesp |
---|
| 451 | ALLOCATE(moms(mm_nla,nm),momsf(mm_nla,nm),chg_matrix(mm_nla,mm_nla)) |
---|
| 452 | ! Initializes moms |
---|
| 453 | moms(:,1) = mm_m0ccn * mm_dzlev |
---|
| 454 | moms(:,2) = mm_m3ccn * mm_dzlev |
---|
| 455 | DO im=1,mm_nesp |
---|
| 456 | moms(:,2+im) = mm_m3ice(:,im) * mm_dzlev |
---|
| 457 | ENDDO |
---|
| 458 | ! Computes exchange matrix |
---|
| 459 | CALL exchange(mm_drad,mm_drho,mm_dt,chg_matrix) |
---|
| 460 | ! Computes final moments values |
---|
| 461 | DO im=1,nm ; momsf(:,im) = MATMUL(chg_matrix,moms(:,im)) ; ENDDO |
---|
| 462 | ! Computes tendencies (converted in X.m-3) |
---|
| 463 | dm0n = (momsf(:,1)-moms(:,1))/mm_dzlev |
---|
| 464 | dm3n = (momsf(:,2)-moms(:,2))/mm_dzlev |
---|
| 465 | DO im=1,mm_nesp |
---|
| 466 | dm3i(:,im) = (momsf(:,2+im)-moms(:,2+im))/mm_dzlev |
---|
| 467 | ENDDO |
---|
| 468 | RETURN |
---|
| 469 | END SUBROUTINE mm_cloud_sedimentation |
---|
| 470 | |
---|
| 471 | SUBROUTINE exchange(rad,rhog,dt,matrix) |
---|
| 472 | !! Compute the exchange matrix. |
---|
| 473 | !! |
---|
| 474 | !! The subroutine computes the matrix exchange used by |
---|
| 475 | !! [[mm_clouds(module):mm_cloud_sedimentation(subroutine)]] to compute moments tendencies |
---|
| 476 | !! through sedimentation process. Both __rad__ and __rhog__ must be vector with relevant |
---|
| 477 | !! values over the atmospheric vertical structure. __matrix__ is square 2D-array with same |
---|
| 478 | !! dimension size than __rad__. |
---|
| 479 | REAL(kind=mm_wp), INTENT(in), DIMENSION(:) :: rad |
---|
| 480 | !! Cloud drop radius over the atmospheric vertical structure (m). |
---|
| 481 | REAL(kind=mm_wp), INTENT(in), DIMENSION(:) :: rhog |
---|
| 482 | !! Cloud drop density over the atmospheric vertical structure (\(kg.m^{-3}\)). |
---|
| 483 | REAL(kind=mm_wp), INTENT(in) :: dt |
---|
| 484 | !! Timestep (s). |
---|
| 485 | REAL(kind=mm_wp), INTENT(out) :: matrix(:,:) |
---|
| 486 | !! The output _exchange matrix_. |
---|
| 487 | INTEGER :: nz,i,j,jj,jinf,jsup |
---|
| 488 | REAL(kind=mm_wp) :: zni,znip1,xf,xft,xcnt |
---|
| 489 | REAL(kind=mm_wp), DIMENSION(:), ALLOCATABLE :: puit |
---|
| 490 | REAL(kind=mm_wp) :: cpte,cpte2 |
---|
| 491 | INTEGER, PARAMETER :: ichx = 1 |
---|
| 492 | matrix = 0._mm_wp ; nz = SIZE(rad) ; ALLOCATE(puit(nz)) |
---|
| 493 | ! compute exchange matrix |
---|
| 494 | DO i=1,nz |
---|
| 495 | puit(i) = 0._mm_wp |
---|
| 496 | xcnt = 0._mm_wp |
---|
| 497 | ! computes drop move (i.e. its new positions) |
---|
| 498 | CALL getnzs(ichx,i,rad,rhog,dt,zni,znip1) |
---|
| 499 | |
---|
| 500 | ! Peculiar case : Ground level precipitation [znip1<0 && (zni<0 || zni>0)] |
---|
| 501 | ! Complete precipitation [ znip1 <= 0 && zni <= 0 ] : |
---|
| 502 | IF(zni<=0._mm_wp.and.znip1<=0._mm_wp) THEN |
---|
| 503 | xft=0._mm_wp |
---|
| 504 | xf=1._mm_wp |
---|
| 505 | xcnt=xcnt+xf |
---|
| 506 | puit(i)=puit(i)+xf |
---|
| 507 | ENDIF |
---|
| 508 | ! partial precipitation [ znip1 <= 0 && zni > 0 ] : |
---|
| 509 | IF (zni>0._mm_wp .and. znip1 <= 0._mm_wp) THEN |
---|
| 510 | xft=zni/(zni-znip1) |
---|
| 511 | xf=(1.-xft) |
---|
| 512 | xcnt=xcnt+xf |
---|
| 513 | puit(i)=puit(i)+xf |
---|
| 514 | ENDIF |
---|
| 515 | ! General case: no ground precipitation [ znip1 > 0 && zni > 0 ] |
---|
| 516 | IF (zni>0._mm_wp.and.znip1>0._mm_wp) THEN |
---|
| 517 | xft = 1._mm_wp ! on a la totalite de la case |
---|
| 518 | xf = 0._mm_wp |
---|
| 519 | xcnt=xcnt+xf |
---|
| 520 | puit(i)=puit(i)+xf |
---|
| 521 | ENDIF |
---|
| 522 | ! Fix minimum level to the ground |
---|
| 523 | znip1 = MAX(znip1,0.) |
---|
| 524 | zni = MAX(zni,0.) |
---|
| 525 | ! Locate new "drop" position in the verical grid |
---|
| 526 | jsup=nz+1 |
---|
| 527 | jinf=nz+1 |
---|
| 528 | DO j=1,nz |
---|
| 529 | IF (zni<=mm_zlev(j).and.zni>=mm_zlev(j+1)) jsup=j |
---|
| 530 | IF (znip1<=mm_zlev(j).and.znip1>=mm_zlev(j+1)) jinf=j |
---|
| 531 | ENDDO |
---|
| 532 | ! Volume is out of range: (all drops have touched the ground!) |
---|
| 533 | ! Note: can happen here,it has been treated previously :) |
---|
| 534 | IF (jsup>=nz+1.and.jinf==jsup) THEN |
---|
| 535 | WRITE(*,'(a)') "[EXCHANGE] speaking: The impossible happened !" |
---|
| 536 | call EXIT(666) |
---|
| 537 | ENDIF |
---|
| 538 | ! Volume inside a single level |
---|
| 539 | IF (jsup==jinf.and.jsup<=nz) THEN |
---|
| 540 | xf=1._mm_wp |
---|
| 541 | xcnt=xcnt+xft*xf |
---|
| 542 | matrix(jinf,i)=matrix(jinf,i)+xft*xf |
---|
| 543 | ENDIF |
---|
| 544 | |
---|
| 545 | ! Volume over 2 levels |
---|
| 546 | IF (jinf==jsup+1) THEN |
---|
| 547 | xf=(zni-mm_zlev(jinf))/(zni-znip1) |
---|
| 548 | xcnt=xcnt+xf*xft |
---|
| 549 | IF(jsup<=nz) THEN |
---|
| 550 | matrix(jsup,i)=matrix(jsup,i)+xft*xf |
---|
| 551 | ENDIF |
---|
| 552 | xf=(mm_zlev(jinf)-znip1)/(zni-znip1) |
---|
| 553 | xcnt=xcnt+xf*xft |
---|
| 554 | IF (jinf<=nz) THEN |
---|
| 555 | matrix(jinf,i)=matrix(jinf,i)+xft*xf |
---|
| 556 | ENDIF |
---|
| 557 | ENDIF |
---|
| 558 | |
---|
| 559 | ! Volume over 3 or more levels |
---|
| 560 | IF (jinf > jsup+1) THEN |
---|
| 561 | ! first cell |
---|
| 562 | xf=(zni-mm_zlev(jsup+1))/(zni-znip1) |
---|
| 563 | xcnt=xcnt+xf*xft |
---|
| 564 | matrix(jsup,i)=matrix(jsup,i)+xft*xf |
---|
| 565 | ! last cell |
---|
| 566 | xf=(mm_zlev(jinf)-znip1)/(zni-znip1) |
---|
| 567 | xcnt=xcnt+xf*xft |
---|
| 568 | matrix(jinf,i)=matrix(jinf,i)+xft*xf |
---|
| 569 | ! other :) |
---|
| 570 | DO jj=jsup+1,jinf-1 |
---|
| 571 | xf=(mm_zlev(jj)-mm_zlev(jj+1))/(zni-znip1) |
---|
| 572 | xcnt=xcnt+xf*xft |
---|
| 573 | matrix(jj,i)=matrix(jj,i)+xft*xf |
---|
| 574 | ENDDO |
---|
| 575 | ENDIF |
---|
| 576 | ENDDO |
---|
| 577 | ! checking if everything is alright if debug enabled... |
---|
| 578 | IF (mm_debug) THEN |
---|
| 579 | cpte=0._mm_wp ; cpte2=0._mm_wp |
---|
| 580 | DO j=1,nz |
---|
| 581 | DO jj=1,nz |
---|
| 582 | cpte=cpte+matrix(jj,j) |
---|
| 583 | ENDDO |
---|
| 584 | cpte2=cpte+puit(j) |
---|
| 585 | ENDDO |
---|
| 586 | IF (abs(cpte2-nz)>1.e-4_mm_wp) THEN |
---|
| 587 | WRITE(*,'(a)')"[EXCHANGE] speaking :" |
---|
| 588 | WRITE(*,'("tx expl (/nz):",2(2X,ES10.3))') cpte,cpte2 |
---|
| 589 | ENDIF |
---|
| 590 | ENDIF |
---|
| 591 | RETURN |
---|
| 592 | END SUBROUTINE exchange |
---|
| 593 | |
---|
| 594 | SUBROUTINE getnzs(ichx,idx,rad,rho,dt,zni,zns) |
---|
| 595 | !! Compute displacement of a cell under sedimentation process. |
---|
| 596 | !! |
---|
| 597 | !! The method computes the new position of a _drop cell_ through sedimentation process as |
---|
| 598 | !! descibed in the following scheme: |
---|
| 599 | !! |
---|
| 600 | !! ![Cloud sedimentation scheme](|media|/cloud_sed_scheme.svg) |
---|
| 601 | !! |
---|
| 602 | !! New positions are returned in __zni__ and __zns__ ouptut arguments. |
---|
| 603 | !! |
---|
| 604 | !! @note |
---|
| 605 | !! The method uses directly [[mm_globals(module):mm_play(variable)]], [[mm_globals(module):mm_plev(variable)]], |
---|
| 606 | !! [[mm_globals(module):mm_temp(variable)]],[[mm_globals(module):mm_btemp(variable)]], |
---|
| 607 | !! [[mm_globals(module):mm_zlay(variable)]] and [[mm_globals(module):mm_zlev(variable)]] and uses __idx__ to |
---|
| 608 | !! get the relevant value to use on the vertical grid. |
---|
| 609 | INTEGER, INTENT(in) :: ichx |
---|
| 610 | !! Velocity extrapolation control flag (0 for linear, 1 for exponential -preferred -). |
---|
| 611 | INTEGER, INTENT(in) :: idx |
---|
| 612 | !! Initial position of the drop (subscript of vertical layers vectors). |
---|
| 613 | REAL(kind=mm_wp), INTENT(in), DIMENSION(:) :: rad |
---|
| 614 | !! Cloud drop radius over the atmospheric vertical structure (m). |
---|
| 615 | REAL(kind=mm_wp), INTENT(in), DIMENSION(:) :: rho |
---|
| 616 | !! Cloud drop density over the atmospheric vertical structure (\(kg.m^{-3}\)). |
---|
| 617 | REAL(kind=mm_wp), INTENT(in) :: dt |
---|
| 618 | !! Timestep (s). |
---|
| 619 | REAL(kind=mm_wp), INTENT(out) :: zni |
---|
| 620 | !! Final layer center position (m). |
---|
| 621 | REAL(kind=mm_wp), INTENT(out) :: zns |
---|
| 622 | !! Final layer upper boundary position (m). |
---|
| 623 | REAL(kind=mm_wp) :: ws,wi,w,zi,zs |
---|
| 624 | REAL(kind=mm_wp) :: alpha,argexp,v0,arg1,arg2 |
---|
| 625 | INTEGER :: i,nz |
---|
| 626 | REAL(kind=mm_wp), PARAMETER :: es = 30._mm_wp |
---|
| 627 | nz = SIZE(rad) |
---|
| 628 | ! Linear extrapolation of velocity |
---|
| 629 | IF (ichx==0) THEN |
---|
| 630 | ! velocity upper interface |
---|
| 631 | ws = wsettle(mm_plev(idx),mm_btemp(idx),mm_zlev(idx),rho(idx),rad(idx)) |
---|
| 632 | IF (idx==nz) THEN |
---|
| 633 | ! veloctity center layer |
---|
| 634 | wi = wsettle(mm_play(idx),mm_temp(idx),mm_zlay(idx),rho(idx),rad(idx)) |
---|
| 635 | ELSEIF(idx<nz) THEN |
---|
| 636 | ! velocity lower interface |
---|
| 637 | wi = wsettle(mm_plev(idx+1),mm_btemp(idx+1),mm_zlev(idx+1), & |
---|
| 638 | rho(idx+1),rad(idx+1)) |
---|
| 639 | ELSE |
---|
| 640 | WRITE(*,'(a)') "[getnzs] speaking:" |
---|
| 641 | WRITE(*,'(a)') "This is the fatal error..." |
---|
| 642 | WRITE(*,'(a)') "index is higher than number of levels" |
---|
| 643 | call EXIT(111) |
---|
| 644 | ENDIF |
---|
| 645 | w = (ws+wi)/2._mm_wp |
---|
| 646 | zni = mm_zlev(idx)-w*dt |
---|
| 647 | zns = mm_zlev(idx)-mm_dzlev(idx)-w*dt |
---|
| 648 | RETURN |
---|
| 649 | ! Exponential extrapolation of velocity |
---|
| 650 | ELSEIF(ichx==1) THEN |
---|
| 651 | ws = wsettle(mm_plev(idx),mm_btemp(idx),mm_zlev(idx),rho(idx),rad(idx)) |
---|
| 652 | zs = mm_zlev(idx) |
---|
| 653 | wi = wsettle(mm_play(idx),mm_temp(idx),mm_zlay(idx),rho(idx),rad(idx)) |
---|
| 654 | zi=mm_zlay(idx) |
---|
| 655 | ! ws & wi must be different ! |
---|
| 656 | IF(dabs(wi-ws)/wi <= 1.e-3_mm_wp) wi=ws/1.001_mm_wp |
---|
| 657 | IF (wi /= 0._mm_wp) alpha = dlog(ws/wi)/(zs-zi) ! alpha < 0 if wi > ws |
---|
| 658 | ! -es < argexp < es |
---|
| 659 | argexp=MAX(MIN(alpha*zs,es),-es) |
---|
| 660 | v0 = ws/dexp(argexp) |
---|
| 661 | arg1=1._mm_wp+v0*alpha*dexp(argexp)*dt |
---|
| 662 | argexp=MAX(MIN(alpha*(mm_zlev(idx)-mm_dzlev(idx)),es),-es) |
---|
| 663 | arg2=1._mm_wp+v0*alpha*dexp(argexp)*dt |
---|
| 664 | IF (arg1<=0._mm_wp.OR.arg2<=0._mm_wp) THEN |
---|
| 665 | ! correct velocity |
---|
| 666 | ! divides the velocity argument in arg1 and arg2 : |
---|
| 667 | ! argX=1+alpha*v0*exp(alpha*z)*dt <==> argX-1=alpha*v0*exp(alpha*z)*dt |
---|
| 668 | ! ==> argX' = 1 + (argX-1)/2 <==> argX' = (1+argX)/2. |
---|
| 669 | DO i=1,25 |
---|
| 670 | IF (arg1<=0._mm_wp.OR.arg2<=0._mm_wp) THEN |
---|
| 671 | IF (mm_debug) & |
---|
| 672 | WRITE(*,'((a),I2.2,(a))') "[getnzs] must adjust velocity (",i,"/25)" |
---|
| 673 | arg1=(arg1+1._mm_wp)/2._mm_wp ; arg2=(arg2+1._mm_wp)/2._mm_wp |
---|
| 674 | ELSE |
---|
| 675 | EXIT |
---|
| 676 | ENDIF |
---|
| 677 | ENDDO |
---|
| 678 | ! sh** we have to stop |
---|
| 679 | IF (i>25) THEN |
---|
| 680 | WRITE(*,'(a)')"[getnzs] speaking:" |
---|
| 681 | WRITE(*,'(a)') "Cannot adjust velocities" |
---|
| 682 | call EXIT(111) |
---|
| 683 | ENDIF |
---|
| 684 | ENDIF |
---|
| 685 | zni = mm_zlev(idx)-dlog(arg1)/alpha |
---|
| 686 | zns = mm_zlev(idx)-mm_dzlev(idx)-dlog(arg2)/alpha |
---|
| 687 | RETURN |
---|
| 688 | ENDIF |
---|
| 689 | END SUBROUTINE getnzs |
---|
| 690 | |
---|
| 691 | ELEMENTAL FUNCTION wsettle(p,t,z,rho,rad) RESULT(w) |
---|
| 692 | !! Compute the settling velocity of a spherical particle. |
---|
| 693 | !! |
---|
| 694 | !! The method computes the effective settling velocity of spherical particle of |
---|
| 695 | !! radius __rad__. It accounts for the slip-flow transition (no approximation). |
---|
| 696 | REAL(kind=mm_wp), INTENT(in) :: p !! The pressure level (Pa). |
---|
| 697 | REAL(kind=mm_wp), INTENT(in) :: t !! The temperature (K). |
---|
| 698 | REAL(kind=mm_wp), INTENT(in) :: z !! The altitude level (m). |
---|
| 699 | REAL(kind=mm_wp), INTENT(in) :: rho !! Density of the particle (\(kg.m^{-3}\)). |
---|
| 700 | REAL(kind=mm_wp), INTENT(in) :: rad !! Radius of the particle (m). |
---|
| 701 | REAL(kind=mm_wp) :: w !! Settling velocity (\(m.s^{-1}\)). |
---|
| 702 | REAL(kind=mm_wp) :: g,a,kn,nu |
---|
| 703 | REAL(kind=mm_wp), PARAMETER :: ra = 1.75e-10_mm_wp, nu0 = 1.74e-4_mm_wp, c = 109._mm_wp |
---|
| 704 | ! Computes corrected gravity |
---|
| 705 | g = mm_effg(z) |
---|
| 706 | ! Knudsen number |
---|
| 707 | kn = mm_kboltz*t/(p*4._mm_wp*sqrt(2._mm_wp)*mm_pi*ra**2)/rad |
---|
| 708 | ! Air viscosity |
---|
| 709 | nu=nu0*sqrt(t/293._mm_wp)*(1._mm_wp+c/293._mm_wp)/(1._mm_wp+c/t) |
---|
| 710 | ! Computes settling velocity |
---|
| 711 | w = 2._mm_wp/9._mm_wp * rad**2*g*rho/nu |
---|
| 712 | ! apply slip-flow correction |
---|
| 713 | w = w*(1._mm_wp+1.2517_mm_wp*kn+0.4_mm_wp*kn*dexp(-1.1_mm_wp/kn)) |
---|
| 714 | END FUNCTION wsettle |
---|
| 715 | |
---|
| 716 | FUNCTION get_mass_flux(rho,m3) RESULT(flx) |
---|
| 717 | !> Get the mass flux of (clouds related) moment through sedimention. |
---|
| 718 | !! |
---|
| 719 | !! @warning |
---|
| 720 | !! The method is __only__ valid for cloud moments (i.e. ice or ccn). It calls |
---|
| 721 | !! [[mm_clouds(module):wsettle(function)]] that compute the _mean_ settling velocity of a |
---|
| 722 | !! cloud drop. |
---|
| 723 | !! |
---|
| 724 | !! @note |
---|
| 725 | !! The computed flux is always positive. |
---|
| 726 | REAL(kind=mm_wp), INTENT(in) :: rho |
---|
| 727 | !! Tracer density (\(kg.m^{-3}\)). |
---|
| 728 | REAL(kind=mm_wp), INTENT(in), DIMENSION(:) :: m3 |
---|
| 729 | !! Vertical profile of the total volume of tracer (i.e. M3) from __TOP__ to __GROUND__ (\(m^{3}.m^{-3}\)). |
---|
| 730 | REAL(kind=mm_wp), DIMENSION(SIZE(m3)) :: flx |
---|
| 731 | !! Mass sedimentation fluxes at each layer from __TOP__ to __GROUND__ (\(kg.m^{-2}.s^{-1}\)). |
---|
| 732 | REAL(kind=mm_wp), SAVE :: fac = 4._mm_wp/3._mm_wp * mm_pi |
---|
| 733 | flx = fac * rho * m3 * wsettle(mm_play,mm_temp,mm_zlay,mm_drho,mm_drad) |
---|
| 734 | RETURN |
---|
| 735 | END FUNCTION get_mass_flux |
---|
| 736 | |
---|
| 737 | END MODULE MM_CLOUDS |
---|