[1585] | 1 | ! path: $Source: /storm/rc1/cvsroot/rc/rrtmg_lw/src/rrtmg_lw_rtrnmr.f90,v $ |
---|
| 2 | ! author: $Author: mike $ |
---|
| 3 | ! revision: $Revision: 1.7 $ |
---|
| 4 | ! created: $Date: 2009/11/12 20:52:26 $ |
---|
| 5 | ! |
---|
| 6 | module rrtmg_lw_rtrnmr |
---|
| 7 | |
---|
| 8 | ! -------------------------------------------------------------------------- |
---|
| 9 | ! | | |
---|
| 10 | ! | Copyright 2002-2009, Atmospheric & Environmental Research, Inc. (AER). | |
---|
| 11 | ! | This software may be used, copied, or redistributed as long as it is | |
---|
| 12 | ! | not sold and this copyright notice is reproduced on each copy made. | |
---|
| 13 | ! | This model is provided as is without any express or implied warranties. | |
---|
| 14 | ! | (http://www.rtweb.aer.com/) | |
---|
| 15 | ! | | |
---|
| 16 | ! -------------------------------------------------------------------------- |
---|
| 17 | |
---|
| 18 | ! ------- Modules ------- |
---|
| 19 | |
---|
| 20 | use parkind, only : im => kind_im, rb => kind_rb |
---|
| 21 | use parrrtm, only : mg, nbndlw, ngptlw |
---|
| 22 | use rrlw_con, only: fluxfac, heatfac |
---|
| 23 | use rrlw_wvn, only: delwave, ngs |
---|
| 24 | use rrlw_tbl, only: tblint, bpade, tau_tbl, exp_tbl, tfn_tbl |
---|
| 25 | use rrlw_vsn, only: hvrrtx, hnamrtx |
---|
| 26 | |
---|
| 27 | implicit none |
---|
| 28 | |
---|
| 29 | contains |
---|
| 30 | |
---|
| 31 | !----------------------------------------------------------------------------- |
---|
| 32 | subroutine rtrnmr(nlayers, istart, iend, iout, pz, semiss, ncbands, & |
---|
| 33 | cldfrac, taucloud, planklay, planklev, plankbnd, & |
---|
| 34 | pwvcm, fracs, taut, & |
---|
| 35 | totuflux, totdflux, fnet, htr, & |
---|
| 36 | totuclfl, totdclfl, fnetc, htrc, & |
---|
| 37 | idrv, dplankbnd_dt, dtotuflux_dt, dtotuclfl_dt ) |
---|
| 38 | !----------------------------------------------------------------------------- |
---|
| 39 | ! |
---|
| 40 | ! Original version: E. J. Mlawer, et al. RRTM_V3.0 |
---|
| 41 | ! Revision for GCMs: Michael J. Iacono; October, 2002 |
---|
| 42 | ! Revision for F90: Michael J. Iacono; June, 2006 |
---|
| 43 | ! Revision for dFdT option: M. J. Iacono and E. J. Mlawer, November 2009 |
---|
| 44 | ! |
---|
| 45 | ! This program calculates the upward fluxes, downward fluxes, and |
---|
| 46 | ! heating rates for an arbitrary clear or cloudy atmosphere. The input |
---|
| 47 | ! to this program is the atmospheric profile, all Planck function |
---|
| 48 | ! information, and the cloud fraction by layer. A variable diffusivity |
---|
| 49 | ! angle (SECDIFF) is used for the angle integration. Bands 2-3 and 5-9 |
---|
| 50 | ! use a value for SECDIFF that varies from 1.50 to 1.80 as a function of |
---|
| 51 | ! the column water vapor, and other bands use a value of 1.66. The Gaussian |
---|
| 52 | ! weight appropriate to this angle (WTDIFF=0.5) is applied here. Note that |
---|
| 53 | ! use of the emissivity angle for the flux integration can cause errors of |
---|
| 54 | ! 1 to 4 W/m2 within cloudy layers. |
---|
| 55 | ! Clouds are treated with a maximum-random cloud overlap method. |
---|
| 56 | ! This subroutine also provides the optional capability to calculate |
---|
| 57 | ! the derivative of upward flux respect to surface temperature using |
---|
| 58 | ! the pre-tabulated derivative of the Planck function with respect to |
---|
| 59 | ! temperature integrated over each spectral band. |
---|
| 60 | !*************************************************************************** |
---|
| 61 | |
---|
| 62 | ! ------- Declarations ------- |
---|
| 63 | |
---|
| 64 | ! ----- Input ----- |
---|
| 65 | integer(kind=im), intent(in) :: nlayers ! total number of layers |
---|
| 66 | integer(kind=im), intent(in) :: istart ! beginning band of calculation |
---|
| 67 | integer(kind=im), intent(in) :: iend ! ending band of calculation |
---|
| 68 | integer(kind=im), intent(in) :: iout ! output option flag |
---|
| 69 | |
---|
| 70 | ! Atmosphere |
---|
| 71 | real(kind=rb), intent(in) :: pz(0:) ! level (interface) pressures (hPa, mb) |
---|
| 72 | ! Dimensions: (0:nlayers) |
---|
| 73 | real(kind=rb), intent(in) :: pwvcm ! precipitable water vapor (cm) |
---|
| 74 | real(kind=rb), intent(in) :: semiss(:) ! lw surface emissivity |
---|
| 75 | ! Dimensions: (nbndlw) |
---|
| 76 | real(kind=rb), intent(in) :: planklay(:,:) ! |
---|
| 77 | ! Dimensions: (nlayers,nbndlw) |
---|
| 78 | real(kind=rb), intent(in) :: planklev(0:,:) ! |
---|
| 79 | ! Dimensions: (0:nlayers,nbndlw) |
---|
| 80 | real(kind=rb), intent(in) :: plankbnd(:) ! |
---|
| 81 | ! Dimensions: (nbndlw) |
---|
| 82 | real(kind=rb), intent(in) :: fracs(:,:) ! |
---|
| 83 | ! Dimensions: (nlayers,ngptw) |
---|
| 84 | real(kind=rb), intent(in) :: taut(:,:) ! gaseous + aerosol optical depths |
---|
| 85 | ! Dimensions: (nlayers,ngptlw) |
---|
| 86 | |
---|
| 87 | ! Clouds |
---|
| 88 | integer(kind=im), intent(in) :: ncbands ! number of cloud spectral bands |
---|
| 89 | real(kind=rb), intent(in) :: cldfrac(:) ! layer cloud fraction |
---|
| 90 | ! Dimensions: (nlayers) |
---|
| 91 | real(kind=rb), intent(in) :: taucloud(:,:) ! layer cloud optical depth |
---|
| 92 | ! Dimensions: (nlayers,nbndlw) |
---|
| 93 | integer(kind=im), intent(in) :: idrv ! flag for calculation of dF/dt from |
---|
| 94 | ! Planck derivative [0=off, 1=on] |
---|
| 95 | real(kind=rb), intent(in) :: dplankbnd_dt(:) ! derivative of Planck function wrt temp |
---|
| 96 | ! Dimensions: (nbndlw) |
---|
| 97 | |
---|
| 98 | ! ----- Output ----- |
---|
| 99 | real(kind=rb), intent(out) :: totuflux(0:) ! upward longwave flux (w/m2) |
---|
| 100 | ! Dimensions: (0:nlayers) |
---|
| 101 | real(kind=rb), intent(out) :: totdflux(0:) ! downward longwave flux (w/m2) |
---|
| 102 | ! Dimensions: (0:nlayers) |
---|
| 103 | real(kind=rb), intent(out) :: fnet(0:) ! net longwave flux (w/m2) |
---|
| 104 | ! Dimensions: (0:nlayers) |
---|
| 105 | real(kind=rb), intent(out) :: htr(0:) ! longwave heating rate (k/day) |
---|
| 106 | ! Dimensions: (0:nlayers) |
---|
| 107 | real(kind=rb), intent(out) :: totuclfl(0:) ! clear sky upward longwave flux (w/m2) |
---|
| 108 | ! Dimensions: (0:nlayers) |
---|
| 109 | real(kind=rb), intent(out) :: totdclfl(0:) ! clear sky downward longwave flux (w/m2) |
---|
| 110 | ! Dimensions: (0:nlayers) |
---|
| 111 | real(kind=rb), intent(out) :: fnetc(0:) ! clear sky net longwave flux (w/m2) |
---|
| 112 | ! Dimensions: (0:nlayers) |
---|
| 113 | real(kind=rb), intent(out) :: htrc(0:) ! clear sky longwave heating rate (k/day) |
---|
| 114 | ! Dimensions: (0:nlayers) |
---|
| 115 | real(kind=rb), intent(out) :: dtotuflux_dt(0:) ! change in upward longwave flux (w/m2/k) |
---|
| 116 | ! with respect to surface temperature |
---|
| 117 | ! Dimensions: (0:nlayers) |
---|
| 118 | real(kind=rb), intent(out) :: dtotuclfl_dt(0:) ! change in upward longwave flux (w/m2/k) |
---|
| 119 | ! with respect to surface temperature |
---|
| 120 | ! Dimensions: (0:nlayers) |
---|
| 121 | |
---|
| 122 | ! ----- Local ----- |
---|
| 123 | ! Declarations for radiative transfer |
---|
| 124 | real(kind=rb) :: abscld(nlayers,nbndlw) |
---|
| 125 | real(kind=rb) :: atot(nlayers) |
---|
| 126 | real(kind=rb) :: atrans(nlayers) |
---|
| 127 | real(kind=rb) :: bbugas(nlayers) |
---|
| 128 | real(kind=rb) :: bbutot(nlayers) |
---|
| 129 | real(kind=rb) :: clrurad(0:nlayers) |
---|
| 130 | real(kind=rb) :: clrdrad(0:nlayers) |
---|
| 131 | real(kind=rb) :: efclfrac(nlayers,nbndlw) |
---|
| 132 | real(kind=rb) :: uflux(0:nlayers) |
---|
| 133 | real(kind=rb) :: dflux(0:nlayers) |
---|
| 134 | real(kind=rb) :: urad(0:nlayers) |
---|
| 135 | real(kind=rb) :: drad(0:nlayers) |
---|
| 136 | real(kind=rb) :: uclfl(0:nlayers) |
---|
| 137 | real(kind=rb) :: dclfl(0:nlayers) |
---|
| 138 | real(kind=rb) :: odcld(nlayers,nbndlw) |
---|
| 139 | |
---|
| 140 | real(kind=rb) :: secdiff(nbndlw) ! secant of diffusivity angle |
---|
| 141 | real(kind=rb) :: a0(nbndlw),a1(nbndlw),a2(nbndlw)! diffusivity angle adjustment coefficients |
---|
| 142 | real(kind=rb) :: wtdiff, rec_6 |
---|
| 143 | real(kind=rb) :: transcld, radld, radclrd, plfrac, blay, dplankup, dplankdn |
---|
| 144 | real(kind=rb) :: odepth, odtot, odepth_rec, odtot_rec, gassrc, ttot |
---|
| 145 | real(kind=rb) :: tblind, tfactot, bbd, bbdtot, tfacgas, transc, tausfac |
---|
| 146 | real(kind=rb) :: rad0, reflect, radlu, radclru |
---|
| 147 | |
---|
| 148 | real(kind=rb) :: duflux_dt(0:nlayers) |
---|
| 149 | real(kind=rb) :: duclfl_dt(0:nlayers) |
---|
| 150 | real(kind=rb) :: d_urad_dt(0:nlayers) |
---|
| 151 | real(kind=rb) :: d_clrurad_dt(0:nlayers) |
---|
| 152 | real(kind=rb) :: d_rad0_dt, d_radlu_dt, d_radclru_dt |
---|
| 153 | |
---|
| 154 | integer(kind=im) :: icldlyr(nlayers) ! flag for cloud in layer |
---|
| 155 | integer(kind=im) :: ibnd, ib, iband, lay, lev, l ! loop indices |
---|
| 156 | integer(kind=im) :: igc ! g-point interval counter |
---|
| 157 | integer(kind=im) :: iclddn ! flag for cloud in down path |
---|
| 158 | integer(kind=im) :: ittot, itgas, itr ! lookup table indices |
---|
| 159 | integer(kind=im) :: ipat(16,0:2) |
---|
| 160 | |
---|
| 161 | ! Declarations for cloud overlap adjustment |
---|
| 162 | real(kind=rb) :: faccld1(nlayers+1),faccld2(nlayers+1) |
---|
| 163 | real(kind=rb) :: facclr1(nlayers+1),facclr2(nlayers+1) |
---|
| 164 | real(kind=rb) :: faccmb1(nlayers+1),faccmb2(nlayers+1) |
---|
| 165 | real(kind=rb) :: faccld1d(0:nlayers),faccld2d(0:nlayers) |
---|
| 166 | real(kind=rb) :: facclr1d(0:nlayers),facclr2d(0:nlayers) |
---|
| 167 | real(kind=rb) :: faccmb1d(0:nlayers),faccmb2d(0:nlayers) |
---|
| 168 | |
---|
| 169 | real(kind=rb) :: fmax, fmin, rat1, rat2 |
---|
| 170 | real(kind=rb) :: clrradd, cldradd, clrradu, cldradu, oldclr, oldcld |
---|
| 171 | real(kind=rb) :: rad, cldsrc, radmod |
---|
| 172 | |
---|
| 173 | integer(kind=im) :: istcld(nlayers+1),istcldd(0:nlayers) |
---|
| 174 | |
---|
| 175 | ! ------- Definitions ------- |
---|
| 176 | ! input |
---|
| 177 | ! nlayers ! number of model layers |
---|
| 178 | ! ngptlw ! total number of g-point subintervals |
---|
| 179 | ! nbndlw ! number of longwave spectral bands |
---|
| 180 | ! ncbands ! number of spectral bands for clouds |
---|
| 181 | ! secdiff ! diffusivity angle |
---|
| 182 | ! wtdiff ! weight for radiance to flux conversion |
---|
| 183 | ! pavel ! layer pressures (mb) |
---|
| 184 | ! pz ! level (interface) pressures (mb) |
---|
| 185 | ! tavel ! layer temperatures (k) |
---|
| 186 | ! tz ! level (interface) temperatures(mb) |
---|
| 187 | ! tbound ! surface temperature (k) |
---|
| 188 | ! cldfrac ! layer cloud fraction |
---|
| 189 | ! taucloud ! layer cloud optical depth |
---|
| 190 | ! itr ! integer look-up table index |
---|
| 191 | ! icldlyr ! flag for cloudy layers |
---|
| 192 | ! iclddn ! flag for cloud in column at any layer |
---|
| 193 | ! semiss ! surface emissivities for each band |
---|
| 194 | ! reflect ! surface reflectance |
---|
| 195 | ! bpade ! 1/(pade constant) |
---|
| 196 | ! tau_tbl ! clear sky optical depth look-up table |
---|
| 197 | ! exp_tbl ! exponential look-up table for transmittance |
---|
| 198 | ! tfn_tbl ! tau transition function look-up table |
---|
| 199 | |
---|
| 200 | ! local |
---|
| 201 | ! atrans ! gaseous absorptivity |
---|
| 202 | ! abscld ! cloud absorptivity |
---|
| 203 | ! atot ! combined gaseous and cloud absorptivity |
---|
| 204 | ! odclr ! clear sky (gaseous) optical depth |
---|
| 205 | ! odcld ! cloud optical depth |
---|
| 206 | ! odtot ! optical depth of gas and cloud |
---|
| 207 | ! tfacgas ! gas-only pade factor, used for planck fn |
---|
| 208 | ! tfactot ! gas and cloud pade factor, used for planck fn |
---|
| 209 | ! bbdgas ! gas-only planck function for downward rt |
---|
| 210 | ! bbugas ! gas-only planck function for upward rt |
---|
| 211 | ! bbdtot ! gas and cloud planck function for downward rt |
---|
| 212 | ! bbutot ! gas and cloud planck function for upward calc. |
---|
| 213 | ! gassrc ! source radiance due to gas only |
---|
| 214 | ! efclfrac ! effective cloud fraction |
---|
| 215 | ! radlu ! spectrally summed upward radiance |
---|
| 216 | ! radclru ! spectrally summed clear sky upward radiance |
---|
| 217 | ! urad ! upward radiance by layer |
---|
| 218 | ! clrurad ! clear sky upward radiance by layer |
---|
| 219 | ! radld ! spectrally summed downward radiance |
---|
| 220 | ! radclrd ! spectrally summed clear sky downward radiance |
---|
| 221 | ! drad ! downward radiance by layer |
---|
| 222 | ! clrdrad ! clear sky downward radiance by layer |
---|
| 223 | ! d_radlu_dt ! spectrally summed upward radiance |
---|
| 224 | ! d_radclru_dt ! spectrally summed clear sky upward radiance |
---|
| 225 | ! d_urad_dt ! upward radiance by layer |
---|
| 226 | ! d_clrurad_dt ! clear sky upward radiance by layer |
---|
| 227 | |
---|
| 228 | ! output |
---|
| 229 | ! totuflux ! upward longwave flux (w/m2) |
---|
| 230 | ! totdflux ! downward longwave flux (w/m2) |
---|
| 231 | ! fnet ! net longwave flux (w/m2) |
---|
| 232 | ! htr ! longwave heating rate (k/day) |
---|
| 233 | ! totuclfl ! clear sky upward longwave flux (w/m2) |
---|
| 234 | ! totdclfl ! clear sky downward longwave flux (w/m2) |
---|
| 235 | ! fnetc ! clear sky net longwave flux (w/m2) |
---|
| 236 | ! htrc ! clear sky longwave heating rate (k/day) |
---|
| 237 | ! dtotuflux_dt ! change in upward longwave flux (w/m2/k) |
---|
| 238 | ! ! with respect to surface temperature |
---|
| 239 | ! dtotuclfl_dt ! change in clear sky upward longwave flux (w/m2/k) |
---|
| 240 | ! ! with respect to surface temperature |
---|
| 241 | |
---|
| 242 | |
---|
| 243 | ! These arrays indicate the spectral 'region' (used in the |
---|
| 244 | ! calculation of ice cloud optical depths) corresponding |
---|
| 245 | ! to each spectral band. See cldprop.f for more details. |
---|
| 246 | data ipat /1,1,1,1,1,1,1,1,1, 1, 1, 1, 1, 1, 1, 1, & |
---|
| 247 | 1,2,3,3,3,4,4,4,5, 5, 5, 5, 5, 5, 5, 5, & |
---|
| 248 | 1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16/ |
---|
| 249 | |
---|
| 250 | ! This secant and weight corresponds to the standard diffusivity |
---|
| 251 | ! angle. This initial value is redefined below for some bands. |
---|
| 252 | data wtdiff /0.5_rb/ |
---|
| 253 | data rec_6 /0.166667_rb/ |
---|
| 254 | |
---|
| 255 | ! Reset diffusivity angle for Bands 2-3 and 5-9 to vary (between 1.50 |
---|
| 256 | ! and 1.80) as a function of total column water vapor. The function |
---|
| 257 | ! has been defined to minimize flux and cooling rate errors in these bands |
---|
| 258 | ! over a wide range of precipitable water values. |
---|
| 259 | data a0 / 1.66_rb, 1.55_rb, 1.58_rb, 1.66_rb, & |
---|
| 260 | 1.54_rb, 1.454_rb, 1.89_rb, 1.33_rb, & |
---|
| 261 | 1.668_rb, 1.66_rb, 1.66_rb, 1.66_rb, & |
---|
| 262 | 1.66_rb, 1.66_rb, 1.66_rb, 1.66_rb / |
---|
| 263 | data a1 / 0.00_rb, 0.25_rb, 0.22_rb, 0.00_rb, & |
---|
| 264 | 0.13_rb, 0.446_rb, -0.10_rb, 0.40_rb, & |
---|
| 265 | -0.006_rb, 0.00_rb, 0.00_rb, 0.00_rb, & |
---|
| 266 | 0.00_rb, 0.00_rb, 0.00_rb, 0.00_rb / |
---|
| 267 | data a2 / 0.00_rb, -12.0_rb, -11.7_rb, 0.00_rb, & |
---|
| 268 | -0.72_rb,-0.243_rb, 0.19_rb,-0.062_rb, & |
---|
| 269 | 0.414_rb, 0.00_rb, 0.00_rb, 0.00_rb, & |
---|
| 270 | 0.00_rb, 0.00_rb, 0.00_rb, 0.00_rb / |
---|
| 271 | |
---|
| 272 | do ibnd = 1,nbndlw |
---|
| 273 | if (ibnd.eq.1 .or. ibnd.eq.4 .or. ibnd.ge.10) then |
---|
| 274 | secdiff(ibnd) = 1.66_rb |
---|
| 275 | else |
---|
| 276 | secdiff(ibnd) = a0(ibnd) + a1(ibnd)*exp(a2(ibnd)*pwvcm) |
---|
| 277 | if (secdiff(ibnd) .gt. 1.80_rb) secdiff(ibnd) = 1.80_rb |
---|
| 278 | if (secdiff(ibnd) .lt. 1.50_rb) secdiff(ibnd) = 1.50_rb |
---|
| 279 | endif |
---|
| 280 | enddo |
---|
| 281 | |
---|
| 282 | hvrrtx = '$Revision: 1.7 $' |
---|
| 283 | |
---|
| 284 | urad(0) = 0.0_rb |
---|
| 285 | drad(0) = 0.0_rb |
---|
| 286 | totuflux(0) = 0.0_rb |
---|
| 287 | totdflux(0) = 0.0_rb |
---|
| 288 | clrurad(0) = 0.0_rb |
---|
| 289 | clrdrad(0) = 0.0_rb |
---|
| 290 | totuclfl(0) = 0.0_rb |
---|
| 291 | totdclfl(0) = 0.0_rb |
---|
| 292 | if (idrv .eq. 1) then |
---|
| 293 | d_urad_dt(0) = 0.0_rb |
---|
| 294 | d_clrurad_dt(0) = 0.0_rb |
---|
| 295 | dtotuflux_dt(0) = 0.0_rb |
---|
| 296 | dtotuclfl_dt(0) = 0.0_rb |
---|
| 297 | endif |
---|
| 298 | |
---|
| 299 | do lay = 1, nlayers |
---|
| 300 | urad(lay) = 0.0_rb |
---|
| 301 | drad(lay) = 0.0_rb |
---|
| 302 | totuflux(lay) = 0.0_rb |
---|
| 303 | totdflux(lay) = 0.0_rb |
---|
| 304 | clrurad(lay) = 0.0_rb |
---|
| 305 | clrdrad(lay) = 0.0_rb |
---|
| 306 | totuclfl(lay) = 0.0_rb |
---|
| 307 | totdclfl(lay) = 0.0_rb |
---|
| 308 | if (idrv .eq. 1) then |
---|
| 309 | d_urad_dt(lay) = 0.0_rb |
---|
| 310 | d_clrurad_dt(lay) = 0.0_rb |
---|
| 311 | dtotuflux_dt(lay) = 0.0_rb |
---|
| 312 | dtotuclfl_dt(lay) = 0.0_rb |
---|
| 313 | endif |
---|
| 314 | |
---|
| 315 | do ib = 1, ncbands |
---|
| 316 | if (cldfrac(lay) .ge. 1.e-6_rb) then |
---|
| 317 | odcld(lay,ib) = secdiff(ib) * taucloud(lay,ib) |
---|
| 318 | icldlyr(lay) = 1 |
---|
| 319 | else |
---|
| 320 | odcld(lay,ib) = 0.0_rb |
---|
| 321 | icldlyr(lay) = 0 |
---|
| 322 | endif |
---|
| 323 | enddo |
---|
| 324 | enddo |
---|
| 325 | |
---|
| 326 | ! Maximum/Random cloud overlap parameter |
---|
| 327 | |
---|
| 328 | istcld(1) = 1 |
---|
| 329 | istcldd(nlayers) = 1 |
---|
| 330 | do lev = 1, nlayers |
---|
| 331 | |
---|
| 332 | if (icldlyr(lev).eq.1) then |
---|
| 333 | ! Maximum/random cloud overlap |
---|
| 334 | istcld(lev+1) = 0 |
---|
| 335 | if (lev .eq. nlayers) then |
---|
| 336 | faccld1(lev+1) = 0._rb |
---|
| 337 | faccld2(lev+1) = 0._rb |
---|
| 338 | facclr1(lev+1) = 0._rb |
---|
| 339 | facclr2(lev+1) = 0._rb |
---|
| 340 | faccmb1(lev+1) = 0._rb |
---|
| 341 | faccmb2(lev+1) = 0._rb |
---|
| 342 | elseif (cldfrac(lev+1) .ge. cldfrac(lev)) then |
---|
| 343 | faccld1(lev+1) = 0._rb |
---|
| 344 | faccld2(lev+1) = 0._rb |
---|
| 345 | if (istcld(lev) .eq. 1) then |
---|
| 346 | facclr1(lev+1) = 0._rb |
---|
| 347 | facclr2(lev+1) = 0._rb |
---|
| 348 | if (cldfrac(lev) .lt. 1._rb) facclr2(lev+1) = & |
---|
| 349 | (cldfrac(lev+1)-cldfrac(lev))/(1._rb-cldfrac(lev)) |
---|
| 350 | facclr2(lev) = 0._rb |
---|
| 351 | faccld2(lev) = 0._rb |
---|
| 352 | else |
---|
| 353 | fmax = max(cldfrac(lev),cldfrac(lev-1)) |
---|
| 354 | if (cldfrac(lev+1) .gt. fmax) then |
---|
| 355 | facclr1(lev+1) = rat2 |
---|
| 356 | facclr2(lev+1) = (cldfrac(lev+1)-fmax)/(1._rb-fmax) |
---|
| 357 | elseif (cldfrac(lev+1) .lt. fmax) then |
---|
| 358 | facclr1(lev+1) = (cldfrac(lev+1)-cldfrac(lev))/ & |
---|
| 359 | (cldfrac(lev-1)-cldfrac(lev)) |
---|
| 360 | facclr2(lev+1) = 0._rb |
---|
| 361 | else |
---|
| 362 | facclr1(lev+1) = rat2 |
---|
| 363 | facclr2(lev+1) = 0._rb |
---|
| 364 | endif |
---|
| 365 | endif |
---|
| 366 | if (facclr1(lev+1).gt.0._rb .or. facclr2(lev+1).gt.0._rb) then |
---|
| 367 | rat1 = 1._rb |
---|
| 368 | rat2 = 0._rb |
---|
| 369 | else |
---|
| 370 | rat1 = 0._rb |
---|
| 371 | rat2 = 0._rb |
---|
| 372 | endif |
---|
| 373 | else |
---|
| 374 | facclr1(lev+1) = 0._rb |
---|
| 375 | facclr2(lev+1) = 0._rb |
---|
| 376 | if (istcld(lev) .eq. 1) then |
---|
| 377 | faccld1(lev+1) = 0._rb |
---|
| 378 | faccld2(lev+1) = (cldfrac(lev)-cldfrac(lev+1))/cldfrac(lev) |
---|
| 379 | |
---|
| 380 | facclr2(lev) = 0._rb |
---|
| 381 | faccld2(lev) = 0._rb |
---|
| 382 | else |
---|
| 383 | fmin = min(cldfrac(lev),cldfrac(lev-1)) |
---|
| 384 | if (cldfrac(lev+1) .le. fmin) then |
---|
| 385 | faccld1(lev+1) = rat1 |
---|
| 386 | faccld2(lev+1) = (fmin-cldfrac(lev+1))/fmin |
---|
| 387 | else |
---|
| 388 | faccld1(lev+1) = (cldfrac(lev)-cldfrac(lev+1))/(cldfrac(lev)-fmin) |
---|
| 389 | faccld2(lev+1) = 0._rb |
---|
| 390 | endif |
---|
| 391 | endif |
---|
| 392 | if (faccld1(lev+1).gt.0._rb .or. faccld2(lev+1).gt.0._rb) then |
---|
| 393 | rat1 = 0._rb |
---|
| 394 | rat2 = 1._rb |
---|
| 395 | else |
---|
| 396 | rat1 = 0._rb |
---|
| 397 | rat2 = 0._rb |
---|
| 398 | endif |
---|
| 399 | endif |
---|
| 400 | faccmb1(lev+1) = facclr1(lev+1) * faccld2(lev) * cldfrac(lev-1) |
---|
| 401 | faccmb2(lev+1) = faccld1(lev+1) * facclr2(lev) * (1._rb - cldfrac(lev-1)) |
---|
| 402 | else |
---|
| 403 | istcld(lev+1) = 1 |
---|
| 404 | endif |
---|
| 405 | enddo |
---|
| 406 | |
---|
| 407 | do lev = nlayers, 1, -1 |
---|
| 408 | if (icldlyr(lev).eq.1) then |
---|
| 409 | istcldd(lev-1) = 0 |
---|
| 410 | if (lev .eq. 1) then |
---|
| 411 | faccld1d(lev-1) = 0._rb |
---|
| 412 | faccld2d(lev-1) = 0._rb |
---|
| 413 | facclr1d(lev-1) = 0._rb |
---|
| 414 | facclr2d(lev-1) = 0._rb |
---|
| 415 | faccmb1d(lev-1) = 0._rb |
---|
| 416 | faccmb2d(lev-1) = 0._rb |
---|
| 417 | elseif (cldfrac(lev-1) .ge. cldfrac(lev)) then |
---|
| 418 | faccld1d(lev-1) = 0._rb |
---|
| 419 | faccld2d(lev-1) = 0._rb |
---|
| 420 | if (istcldd(lev) .eq. 1) then |
---|
| 421 | facclr1d(lev-1) = 0._rb |
---|
| 422 | facclr2d(lev-1) = 0._rb |
---|
| 423 | if (cldfrac(lev) .lt. 1._rb) facclr2d(lev-1) = & |
---|
| 424 | (cldfrac(lev-1)-cldfrac(lev))/(1._rb-cldfrac(lev)) |
---|
| 425 | facclr2d(lev) = 0._rb |
---|
| 426 | faccld2d(lev) = 0._rb |
---|
| 427 | else |
---|
| 428 | fmax = max(cldfrac(lev),cldfrac(lev+1)) |
---|
| 429 | if (cldfrac(lev-1) .gt. fmax) then |
---|
| 430 | facclr1d(lev-1) = rat2 |
---|
| 431 | facclr2d(lev-1) = (cldfrac(lev-1)-fmax)/(1._rb-fmax) |
---|
| 432 | elseif (cldfrac(lev-1) .lt. fmax) then |
---|
| 433 | facclr1d(lev-1) = (cldfrac(lev-1)-cldfrac(lev))/ & |
---|
| 434 | (cldfrac(lev+1)-cldfrac(lev)) |
---|
| 435 | facclr2d(lev-1) = 0. |
---|
| 436 | else |
---|
| 437 | facclr1d(lev-1) = rat2 |
---|
| 438 | facclr2d(lev-1) = 0._rb |
---|
| 439 | endif |
---|
| 440 | endif |
---|
| 441 | if (facclr1d(lev-1).gt.0._rb .or. facclr2d(lev-1).gt.0._rb)then |
---|
| 442 | rat1 = 1._rb |
---|
| 443 | rat2 = 0._rb |
---|
| 444 | else |
---|
| 445 | rat1 = 0._rb |
---|
| 446 | rat2 = 0._rb |
---|
| 447 | endif |
---|
| 448 | else |
---|
| 449 | facclr1d(lev-1) = 0._rb |
---|
| 450 | facclr2d(lev-1) = 0._rb |
---|
| 451 | if (istcldd(lev) .eq. 1) then |
---|
| 452 | faccld1d(lev-1) = 0._rb |
---|
| 453 | faccld2d(lev-1) = (cldfrac(lev)-cldfrac(lev-1))/cldfrac(lev) |
---|
| 454 | facclr2d(lev) = 0._rb |
---|
| 455 | faccld2d(lev) = 0._rb |
---|
| 456 | else |
---|
| 457 | fmin = min(cldfrac(lev),cldfrac(lev+1)) |
---|
| 458 | if (cldfrac(lev-1) .le. fmin) then |
---|
| 459 | faccld1d(lev-1) = rat1 |
---|
| 460 | faccld2d(lev-1) = (fmin-cldfrac(lev-1))/fmin |
---|
| 461 | else |
---|
| 462 | faccld1d(lev-1) = (cldfrac(lev)-cldfrac(lev-1))/(cldfrac(lev)-fmin) |
---|
| 463 | faccld2d(lev-1) = 0._rb |
---|
| 464 | endif |
---|
| 465 | endif |
---|
| 466 | if (faccld1d(lev-1).gt.0._rb .or. faccld2d(lev-1).gt.0._rb)then |
---|
| 467 | rat1 = 0._rb |
---|
| 468 | rat2 = 1._rb |
---|
| 469 | else |
---|
| 470 | rat1 = 0._rb |
---|
| 471 | rat2 = 0._rb |
---|
| 472 | endif |
---|
| 473 | endif |
---|
| 474 | faccmb1d(lev-1) = facclr1d(lev-1) * faccld2d(lev) * cldfrac(lev+1) |
---|
| 475 | faccmb2d(lev-1) = faccld1d(lev-1) * facclr2d(lev) * (1._rb - cldfrac(lev+1)) |
---|
| 476 | else |
---|
| 477 | istcldd(lev-1) = 1 |
---|
| 478 | endif |
---|
| 479 | enddo |
---|
| 480 | |
---|
| 481 | igc = 1 |
---|
| 482 | ! Loop over frequency bands. |
---|
| 483 | do iband = istart, iend |
---|
| 484 | |
---|
| 485 | ! Reinitialize g-point counter for each band if output for each band is requested. |
---|
| 486 | if (iout.gt.0.and.iband.ge.2) igc = ngs(iband-1)+1 |
---|
| 487 | if (ncbands .eq. 1) then |
---|
| 488 | ib = ipat(iband,0) |
---|
| 489 | elseif (ncbands .eq. 5) then |
---|
| 490 | ib = ipat(iband,1) |
---|
| 491 | elseif (ncbands .eq. 16) then |
---|
| 492 | ib = ipat(iband,2) |
---|
| 493 | endif |
---|
| 494 | |
---|
| 495 | ! Loop over g-channels. |
---|
| 496 | 1000 continue |
---|
| 497 | |
---|
| 498 | ! Radiative transfer starts here. |
---|
| 499 | radld = 0._rb |
---|
| 500 | radclrd = 0._rb |
---|
| 501 | iclddn = 0 |
---|
| 502 | |
---|
| 503 | ! Downward radiative transfer loop. |
---|
| 504 | |
---|
| 505 | do lev = nlayers, 1, -1 |
---|
| 506 | plfrac = fracs(lev,igc) |
---|
| 507 | blay = planklay(lev,iband) |
---|
| 508 | dplankup = planklev(lev,iband) - blay |
---|
| 509 | dplankdn = planklev(lev-1,iband) - blay |
---|
| 510 | odepth = secdiff(iband) * taut(lev,igc) |
---|
| 511 | |
---|
| 512 | if (odepth .lt. 0.0_rb) odepth = 0.0_rb |
---|
| 513 | ! Cloudy layer |
---|
| 514 | if (icldlyr(lev).eq.1) then |
---|
| 515 | iclddn = 1 |
---|
| 516 | odtot = odepth + odcld(lev,ib) |
---|
| 517 | if (odtot .lt. 0.06_rb) then |
---|
| 518 | atrans(lev) = odepth - 0.5_rb*odepth*odepth |
---|
| 519 | odepth_rec = rec_6*odepth |
---|
| 520 | gassrc = plfrac*(blay+dplankdn*odepth_rec)*atrans(lev) |
---|
| 521 | |
---|
| 522 | atot(lev) = odtot - 0.5_rb*odtot*odtot |
---|
| 523 | odtot_rec = rec_6*odtot |
---|
| 524 | bbdtot = plfrac * (blay+dplankdn*odtot_rec) |
---|
| 525 | bbd = plfrac*(blay+dplankdn*odepth_rec) |
---|
| 526 | |
---|
| 527 | bbugas(lev) = plfrac * (blay+dplankup*odepth_rec) |
---|
| 528 | bbutot(lev) = plfrac * (blay+dplankup*odtot_rec) |
---|
| 529 | elseif (odepth .le. 0.06_rb) then |
---|
| 530 | atrans(lev) = odepth - 0.5_rb*odepth*odepth |
---|
| 531 | odepth_rec = rec_6*odepth |
---|
| 532 | gassrc = plfrac*(blay+dplankdn*odepth_rec)*atrans(lev) |
---|
| 533 | |
---|
| 534 | odtot = odepth + odcld(lev,ib) |
---|
| 535 | tblind = odtot/(bpade+odtot) |
---|
| 536 | ittot = tblint*tblind + 0.5_rb |
---|
| 537 | tfactot = tfn_tbl(ittot) |
---|
| 538 | bbdtot = plfrac * (blay + tfactot*dplankdn) |
---|
| 539 | bbd = plfrac*(blay+dplankdn*odepth_rec) |
---|
| 540 | atot(lev) = 1._rb - exp_tbl(ittot) |
---|
| 541 | |
---|
| 542 | bbugas(lev) = plfrac * (blay + dplankup*odepth_rec) |
---|
| 543 | bbutot(lev) = plfrac * (blay + tfactot * dplankup) |
---|
| 544 | else |
---|
| 545 | tblind = odepth/(bpade+odepth) |
---|
| 546 | itgas = tblint*tblind+0.5_rb |
---|
| 547 | odepth = tau_tbl(itgas) |
---|
| 548 | atrans(lev) = 1._rb - exp_tbl(itgas) |
---|
| 549 | tfacgas = tfn_tbl(itgas) |
---|
| 550 | gassrc = atrans(lev) * plfrac * (blay + tfacgas*dplankdn) |
---|
| 551 | |
---|
| 552 | odtot = odepth + odcld(lev,ib) |
---|
| 553 | tblind = odtot/(bpade+odtot) |
---|
| 554 | ittot = tblint*tblind + 0.5_rb |
---|
| 555 | tfactot = tfn_tbl(ittot) |
---|
| 556 | bbdtot = plfrac * (blay + tfactot*dplankdn) |
---|
| 557 | bbd = plfrac*(blay+tfacgas*dplankdn) |
---|
| 558 | atot(lev) = 1._rb - exp_tbl(ittot) |
---|
| 559 | |
---|
| 560 | bbugas(lev) = plfrac * (blay + tfacgas * dplankup) |
---|
| 561 | bbutot(lev) = plfrac * (blay + tfactot * dplankup) |
---|
| 562 | endif |
---|
| 563 | |
---|
| 564 | if (istcldd(lev) .eq. 1) then |
---|
| 565 | cldradd = cldfrac(lev) * radld |
---|
| 566 | clrradd = radld - cldradd |
---|
| 567 | oldcld = cldradd |
---|
| 568 | oldclr = clrradd |
---|
| 569 | rad = 0._rb |
---|
| 570 | endif |
---|
| 571 | ttot = 1._rb - atot(lev) |
---|
| 572 | cldsrc = bbdtot * atot(lev) |
---|
| 573 | cldradd = cldradd * ttot + cldfrac(lev) * cldsrc |
---|
| 574 | clrradd = clrradd * (1._rb-atrans(lev)) + (1._rb - cldfrac(lev)) * gassrc |
---|
| 575 | radld = cldradd + clrradd |
---|
| 576 | drad(lev-1) = drad(lev-1) + radld |
---|
| 577 | |
---|
| 578 | radmod = rad * & |
---|
| 579 | (facclr1d(lev-1) * (1.-atrans(lev)) + & |
---|
| 580 | faccld1d(lev-1) * ttot) - & |
---|
| 581 | faccmb1d(lev-1) * gassrc + & |
---|
| 582 | faccmb2d(lev-1) * cldsrc |
---|
| 583 | |
---|
| 584 | oldcld = cldradd - radmod |
---|
| 585 | oldclr = clrradd + radmod |
---|
| 586 | rad = -radmod + facclr2d(lev-1)*oldclr - faccld2d(lev-1)*oldcld |
---|
| 587 | cldradd = cldradd + rad |
---|
| 588 | clrradd = clrradd - rad |
---|
| 589 | ! Clear layer |
---|
| 590 | else |
---|
| 591 | if (odepth .le. 0.06_rb) then |
---|
| 592 | atrans(lev) = odepth-0.5_rb*odepth*odepth |
---|
| 593 | odepth = rec_6*odepth |
---|
| 594 | bbd = plfrac*(blay+dplankdn*odepth) |
---|
| 595 | bbugas(lev) = plfrac*(blay+dplankup*odepth) |
---|
| 596 | else |
---|
| 597 | tblind = odepth/(bpade+odepth) |
---|
| 598 | itr = tblint*tblind+0.5_rb |
---|
| 599 | transc = exp_tbl(itr) |
---|
| 600 | atrans(lev) = 1._rb-transc |
---|
| 601 | tausfac = tfn_tbl(itr) |
---|
| 602 | bbd = plfrac*(blay+tausfac*dplankdn) |
---|
| 603 | bbugas(lev) = plfrac * (blay + tausfac * dplankup) |
---|
| 604 | endif |
---|
| 605 | radld = radld + (bbd-radld)*atrans(lev) |
---|
| 606 | drad(lev-1) = drad(lev-1) + radld |
---|
| 607 | endif |
---|
| 608 | ! Set clear sky stream to total sky stream as long as layers |
---|
| 609 | ! remain clear. Streams diverge when a cloud is reached (iclddn=1), |
---|
| 610 | ! and clear sky stream must be computed separately from that point. |
---|
| 611 | if (iclddn.eq.1) then |
---|
| 612 | radclrd = radclrd + (bbd-radclrd) * atrans(lev) |
---|
| 613 | clrdrad(lev-1) = clrdrad(lev-1) + radclrd |
---|
| 614 | else |
---|
| 615 | radclrd = radld |
---|
| 616 | clrdrad(lev-1) = drad(lev-1) |
---|
| 617 | endif |
---|
| 618 | enddo |
---|
| 619 | |
---|
| 620 | ! Spectral emissivity & reflectance |
---|
| 621 | ! Include the contribution of spectrally varying longwave emissivity |
---|
| 622 | ! and reflection from the surface to the upward radiative transfer. |
---|
| 623 | ! Note: Spectral and Lambertian reflection are identical for the |
---|
| 624 | ! diffusivity angle flux integration used here. |
---|
| 625 | ! Note: The emissivity is applied to plankbnd and dplankbnd_dt when |
---|
| 626 | ! they are defined in subroutine setcoef. |
---|
| 627 | |
---|
| 628 | rad0 = fracs(1,igc) * plankbnd(iband) |
---|
| 629 | if (idrv .eq. 1) then |
---|
| 630 | d_rad0_dt = fracs(1,igc) * dplankbnd_dt(iband) |
---|
| 631 | endif |
---|
| 632 | |
---|
| 633 | ! Add in reflection of surface downward radiance. |
---|
| 634 | reflect = 1._rb - semiss(iband) |
---|
| 635 | radlu = rad0 + reflect * radld |
---|
| 636 | radclru = rad0 + reflect * radclrd |
---|
| 637 | |
---|
| 638 | ! Upward radiative transfer loop. |
---|
| 639 | |
---|
| 640 | urad(0) = urad(0) + radlu |
---|
| 641 | clrurad(0) = clrurad(0) + radclru |
---|
| 642 | if (idrv .eq. 1) then |
---|
| 643 | d_radlu_dt = d_rad0_dt |
---|
| 644 | d_urad_dt(0) = d_urad_dt(0) + d_radlu_dt |
---|
| 645 | d_radclru_dt = d_rad0_dt |
---|
| 646 | d_clrurad_dt(0) = d_clrurad_dt(0) + d_radclru_dt |
---|
| 647 | endif |
---|
| 648 | |
---|
| 649 | do lev = 1, nlayers |
---|
| 650 | ! Cloudy layer |
---|
| 651 | if (icldlyr(lev) .eq. 1) then |
---|
| 652 | gassrc = bbugas(lev) * atrans(lev) |
---|
| 653 | if (istcld(lev) .eq. 1) then |
---|
| 654 | cldradu = cldfrac(lev) * radlu |
---|
| 655 | clrradu = radlu - cldradu |
---|
| 656 | oldcld = cldradu |
---|
| 657 | oldclr = clrradu |
---|
| 658 | rad = 0._rb |
---|
| 659 | endif |
---|
| 660 | ttot = 1._rb - atot(lev) |
---|
| 661 | cldsrc = bbutot(lev) * atot(lev) |
---|
| 662 | cldradu = cldradu * ttot + cldfrac(lev) * cldsrc |
---|
| 663 | clrradu = clrradu * (1.0_rb-atrans(lev)) + (1._rb - cldfrac(lev)) * gassrc |
---|
| 664 | ! Total sky radiance |
---|
| 665 | radlu = cldradu + clrradu |
---|
| 666 | urad(lev) = urad(lev) + radlu |
---|
| 667 | radmod = rad * & |
---|
| 668 | (facclr1(lev+1)*(1.0_rb-atrans(lev))+ & |
---|
| 669 | faccld1(lev+1) * ttot) - & |
---|
| 670 | faccmb1(lev+1) * gassrc + & |
---|
| 671 | faccmb2(lev+1) * cldsrc |
---|
| 672 | oldcld = cldradu - radmod |
---|
| 673 | oldclr = clrradu + radmod |
---|
| 674 | rad = -radmod + facclr2(lev+1)*oldclr - faccld2(lev+1)*oldcld |
---|
| 675 | cldradu = cldradu + rad |
---|
| 676 | clrradu = clrradu - rad |
---|
| 677 | if (idrv .eq. 1) then |
---|
| 678 | d_radlu_dt = d_radlu_dt * cldfrac(lev) * (1.0_rb - atot(lev)) + & |
---|
| 679 | d_radlu_dt * (1.0_rb - cldfrac(lev)) * (1.0_rb - atrans(lev)) |
---|
| 680 | d_urad_dt(lev) = d_urad_dt(lev) + d_radlu_dt |
---|
| 681 | endif |
---|
| 682 | ! Clear layer |
---|
| 683 | else |
---|
| 684 | radlu = radlu + (bbugas(lev)-radlu)*atrans(lev) |
---|
| 685 | urad(lev) = urad(lev) + radlu |
---|
| 686 | if (idrv .eq. 1) then |
---|
| 687 | d_radlu_dt = d_radlu_dt * (1.0_rb - atrans(lev)) |
---|
| 688 | d_urad_dt(lev) = d_urad_dt(lev) + d_radlu_dt |
---|
| 689 | endif |
---|
| 690 | endif |
---|
| 691 | ! Set clear sky stream to total sky stream as long as all layers |
---|
| 692 | ! are clear (iclddn=0). Streams must be calculated separately at |
---|
| 693 | ! all layers when a cloud is present (iclddn=1), because surface |
---|
| 694 | ! reflectance is different for each stream. |
---|
| 695 | if (iclddn.eq.1) then |
---|
| 696 | radclru = radclru + (bbugas(lev)-radclru)*atrans(lev) |
---|
| 697 | clrurad(lev) = clrurad(lev) + radclru |
---|
| 698 | else |
---|
| 699 | radclru = radlu |
---|
| 700 | clrurad(lev) = urad(lev) |
---|
| 701 | endif |
---|
| 702 | if (idrv .eq. 1) then |
---|
| 703 | if (iclddn.eq.1) then |
---|
| 704 | d_radclru_dt = d_radclru_dt * (1.0_rb - atrans(lev)) |
---|
| 705 | d_clrurad_dt(lev) = d_clrurad_dt(lev) + d_radclru_dt |
---|
| 706 | else |
---|
| 707 | d_radclru_dt = d_radlu_dt |
---|
| 708 | d_clrurad_dt(lev) = d_urad_dt(lev) |
---|
| 709 | endif |
---|
| 710 | endif |
---|
| 711 | enddo |
---|
| 712 | |
---|
| 713 | ! Increment g-point counter |
---|
| 714 | igc = igc + 1 |
---|
| 715 | ! Return to continue radiative transfer for all g-channels in present band |
---|
| 716 | if (igc .le. ngs(iband)) go to 1000 |
---|
| 717 | |
---|
| 718 | ! Process longwave output from band. |
---|
| 719 | ! Calculate upward, downward, and net flux. |
---|
| 720 | do lev = nlayers, 0, -1 |
---|
| 721 | uflux(lev) = urad(lev)*wtdiff |
---|
| 722 | dflux(lev) = drad(lev)*wtdiff |
---|
| 723 | urad(lev) = 0.0_rb |
---|
| 724 | drad(lev) = 0.0_rb |
---|
| 725 | totuflux(lev) = totuflux(lev) + uflux(lev) * delwave(iband) |
---|
| 726 | totdflux(lev) = totdflux(lev) + dflux(lev) * delwave(iband) |
---|
| 727 | uclfl(lev) = clrurad(lev)*wtdiff |
---|
| 728 | dclfl(lev) = clrdrad(lev)*wtdiff |
---|
| 729 | clrurad(lev) = 0.0_rb |
---|
| 730 | clrdrad(lev) = 0.0_rb |
---|
| 731 | totuclfl(lev) = totuclfl(lev) + uclfl(lev) * delwave(iband) |
---|
| 732 | totdclfl(lev) = totdclfl(lev) + dclfl(lev) * delwave(iband) |
---|
| 733 | enddo |
---|
| 734 | |
---|
| 735 | ! Calculate total change in upward flux wrt surface temperature |
---|
| 736 | if (idrv .eq. 1) then |
---|
| 737 | do lev = nlayers, 0, -1 |
---|
| 738 | duflux_dt(lev) = d_urad_dt(lev) * wtdiff |
---|
| 739 | d_urad_dt(lev) = 0.0_rb |
---|
| 740 | dtotuflux_dt(lev) = dtotuflux_dt(lev) + duflux_dt(lev) * delwave(iband) * fluxfac |
---|
| 741 | |
---|
| 742 | duclfl_dt(lev) = d_clrurad_dt(lev) * wtdiff |
---|
| 743 | d_clrurad_dt(lev) = 0.0_rb |
---|
| 744 | dtotuclfl_dt(lev) = dtotuclfl_dt(lev) + duclfl_dt(lev) * delwave(iband) * fluxfac |
---|
| 745 | enddo |
---|
| 746 | endif |
---|
| 747 | |
---|
| 748 | ! End spectral band loop |
---|
| 749 | enddo |
---|
| 750 | |
---|
| 751 | ! Calculate fluxes at surface |
---|
| 752 | totuflux(0) = totuflux(0) * fluxfac |
---|
| 753 | totdflux(0) = totdflux(0) * fluxfac |
---|
| 754 | fnet(0) = totuflux(0) - totdflux(0) |
---|
| 755 | |
---|
| 756 | totuclfl(0) = totuclfl(0) * fluxfac |
---|
| 757 | totdclfl(0) = totdclfl(0) * fluxfac |
---|
| 758 | fnetc(0) = totuclfl(0) - totdclfl(0) |
---|
| 759 | |
---|
| 760 | ! Calculate fluxes at model levels |
---|
| 761 | do lev = 1, nlayers |
---|
| 762 | totuflux(lev) = totuflux(lev) * fluxfac |
---|
| 763 | totdflux(lev) = totdflux(lev) * fluxfac |
---|
| 764 | fnet(lev) = totuflux(lev) - totdflux(lev) |
---|
| 765 | totuclfl(lev) = totuclfl(lev) * fluxfac |
---|
| 766 | totdclfl(lev) = totdclfl(lev) * fluxfac |
---|
| 767 | fnetc(lev) = totuclfl(lev) - totdclfl(lev) |
---|
| 768 | l = lev - 1 |
---|
| 769 | |
---|
| 770 | ! Calculate heating rates at model layers |
---|
| 771 | htr(l)=heatfac*(fnet(l)-fnet(lev))/(pz(l)-pz(lev)) |
---|
| 772 | htrc(l)=heatfac*(fnetc(l)-fnetc(lev))/(pz(l)-pz(lev)) |
---|
| 773 | enddo |
---|
| 774 | |
---|
| 775 | ! Set heating rate to zero in top layer |
---|
| 776 | htr(nlayers) = 0.0_rb |
---|
| 777 | htrc(nlayers) = 0.0_rb |
---|
| 778 | |
---|
| 779 | end subroutine rtrnmr |
---|
| 780 | |
---|
| 781 | end module rrtmg_lw_rtrnmr |
---|
| 782 | |
---|