1 | ! %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% |
---|
2 | ! Copyright (c) 2009, Centre National de la Recherche Scientifique |
---|
3 | ! All rights reserved. |
---|
4 | ! |
---|
5 | ! Redistribution and use in source and binary forms, with or without modification, are |
---|
6 | ! permitted provided that the following conditions are met: |
---|
7 | ! |
---|
8 | ! 1. Redistributions of source code must retain the above copyright notice, this list of |
---|
9 | ! conditions and the following disclaimer. |
---|
10 | ! |
---|
11 | ! 2. Redistributions in binary form must reproduce the above copyright notice, this list |
---|
12 | ! of conditions and the following disclaimer in the documentation and/or other |
---|
13 | ! materials provided with the distribution. |
---|
14 | ! |
---|
15 | ! 3. Neither the name of the copyright holder nor the names of its contributors may be |
---|
16 | ! used to endorse or promote products derived from this software without specific prior |
---|
17 | ! written permission. |
---|
18 | ! |
---|
19 | ! THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY |
---|
20 | ! EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF |
---|
21 | ! MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL |
---|
22 | ! THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, |
---|
23 | ! SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT |
---|
24 | ! OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS |
---|
25 | ! INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
---|
26 | ! LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
---|
27 | ! OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
---|
28 | ! |
---|
29 | ! History |
---|
30 | ! May 2007: ActSim code of M. Chiriaco and H. Chepfer rewritten by S. Bony |
---|
31 | ! |
---|
32 | ! May 2008, H. Chepfer: |
---|
33 | ! - Units of pressure inputs: Pa |
---|
34 | ! - Non Spherical particles : LS Ice NS coefficients, CONV Ice NS coefficients |
---|
35 | ! - New input: ice_type (0=ice-spheres ; 1=ice-non-spherical) |
---|
36 | ! |
---|
37 | ! June 2008, A. Bodas-Salcedo: |
---|
38 | ! - Ported to Fortran 90 and optimisation changes |
---|
39 | ! |
---|
40 | ! August 2008, J-L Dufresne: |
---|
41 | ! - Optimisation changes (sum instructions suppressed) |
---|
42 | ! |
---|
43 | ! October 2008, S. Bony, H. Chepfer and J-L. Dufresne : |
---|
44 | ! - Interface with COSP v2.0: |
---|
45 | ! cloud fraction removed from inputs |
---|
46 | ! in-cloud condensed water now in input (instead of grid-averaged value) |
---|
47 | ! depolarisation diagnostic removed |
---|
48 | ! parasol (polder) reflectances (for 5 different solar zenith angles) added |
---|
49 | ! |
---|
50 | ! December 2008, S. Bony, H. Chepfer and J-L. Dufresne : |
---|
51 | ! - Modification of the integration of the lidar equation. |
---|
52 | ! - change the cloud detection threshold |
---|
53 | ! |
---|
54 | ! April 2008, A. Bodas-Salcedo: |
---|
55 | ! - Bug fix in computation of pmol and pnorm of upper layer |
---|
56 | ! |
---|
57 | ! April 2008, J-L. Dufresne |
---|
58 | ! - Bug fix in computation of pmol and pnorm, thanks to Masaki Satoh: a factor 2 |
---|
59 | ! was missing. This affects the ATB values but not the cloud fraction. |
---|
60 | ! |
---|
61 | ! January 2013, G. Cesana and H. Chepfer: |
---|
62 | ! - Add the perpendicular component of the backscattered signal (pnorm_perp_tot) in the arguments |
---|
63 | ! - Add the temperature for each levels (temp) in the arguments |
---|
64 | ! - Add the computation of the perpendicular component of the backscattered lidar signal |
---|
65 | ! Reference: Cesana G. and H. Chepfer (2013): Evaluation of the cloud water phase |
---|
66 | ! in a climate model using CALIPSO-GOCCP, J. Geophys. Res., doi: 10.1002/jgrd.50376 |
---|
67 | ! |
---|
68 | ! May 2015 - D. Swales - Modified for COSPv2.0 |
---|
69 | ! %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% |
---|
70 | module mod_lidar_simulator |
---|
71 | USE COSP_KINDS, ONLY: wp |
---|
72 | USE MOD_COSP_CONFIG, ONLY: SR_BINS,S_CLD,S_ATT,S_CLD_ATT,R_UNDEF,calipso_histBsct, & |
---|
73 | use_vgrid,vgrid_zl,vgrid_zu |
---|
74 | USE MOD_COSP_STATS, ONLY: COSP_CHANGE_VERTICAL_GRID,hist1d |
---|
75 | implicit none |
---|
76 | |
---|
77 | ! Polynomial coefficients (Alpha, Beta, Gamma) which allow to compute the |
---|
78 | ! ATBperpendicular as a function of the ATB for ice or liquid cloud particles |
---|
79 | ! derived from CALIPSO-GOCCP observations at 120m vertical grid |
---|
80 | ! (Cesana and Chepfer, JGR, 2013). |
---|
81 | ! |
---|
82 | ! Relationship between ATBice and ATBperp,ice for ice particles: |
---|
83 | ! ATBperp,ice = Alpha*ATBice |
---|
84 | ! Relationship between ATBice and ATBperp,ice for liquid particles: |
---|
85 | ! ATBperp,ice = Beta*ATBice^2 + Gamma*ATBice |
---|
86 | real(wp) :: & |
---|
87 | alpha,beta,gamma |
---|
88 | |
---|
89 | contains |
---|
90 | ! ###################################################################################### |
---|
91 | ! SUBROUTINE lidar_subcolumn |
---|
92 | ! Inputs with a vertical dimensions (nlev) should ordered in along the vertical |
---|
93 | ! dimension from TOA-2-SFC, for example: varIN(nlev) is varIN @ SFC. |
---|
94 | ! ###################################################################################### |
---|
95 | subroutine lidar_subcolumn(npoints,ncolumns,nlev,beta_mol,tau_mol,betatot,tautot, & |
---|
96 | betatot_ice,tautot_ice,betatot_liq,tautot_liq, & |
---|
97 | pmol,pnorm,pnorm_perp_tot) |
---|
98 | |
---|
99 | ! INPUTS |
---|
100 | INTEGER,intent(in) :: & |
---|
101 | npoints, & ! Number of gridpoints |
---|
102 | ncolumns, & ! Number of subcolumns |
---|
103 | nlev ! Number of levels |
---|
104 | REAL(WP),intent(in),dimension(npoints,nlev) :: & |
---|
105 | beta_mol, & ! Molecular backscatter coefficient |
---|
106 | tau_mol ! Molecular optical depth |
---|
107 | |
---|
108 | REAL(WP),intent(in),dimension(npoints,ncolumns,nlev) :: & |
---|
109 | betatot, & ! |
---|
110 | tautot, & ! Optical thickess integrated from top |
---|
111 | betatot_ice, & ! Backscatter coefficient for ice particles |
---|
112 | betatot_liq, & ! Backscatter coefficient for liquid particles |
---|
113 | tautot_ice, & ! Total optical thickness of ice |
---|
114 | tautot_liq ! Total optical thickness of liq |
---|
115 | |
---|
116 | ! OUTPUTS |
---|
117 | REAL(WP),intent(out),dimension(npoints,nlev) :: & |
---|
118 | pmol ! Molecular attenuated backscatter lidar signal power(m^-1.sr^-1) |
---|
119 | REAL(WP),intent(out),dimension(npoints,ncolumns,nlev) :: & |
---|
120 | pnorm, & ! Molecular backscatter signal power (m^-1.sr^-1) |
---|
121 | pnorm_perp_tot ! Perpendicular lidar backscattered signal power |
---|
122 | |
---|
123 | ! LOCAL VARIABLES |
---|
124 | INTEGER :: k,icol |
---|
125 | REAL(WP),dimension(npoints) :: & |
---|
126 | tautot_lay ! |
---|
127 | REAL(WP),dimension(npoints,ncolumns,nlev) :: & |
---|
128 | pnorm_liq, & ! Lidar backscattered signal power for liquid |
---|
129 | pnorm_ice, & ! Lidar backscattered signal power for ice |
---|
130 | pnorm_perp_ice, & ! Perpendicular lidar backscattered signal power for ice |
---|
131 | pnorm_perp_liq, & ! Perpendicular lidar backscattered signal power for liq |
---|
132 | beta_perp_ice, & ! Perpendicular backscatter coefficient for ice |
---|
133 | beta_perp_liq ! Perpendicular backscatter coefficient for liquid |
---|
134 | |
---|
135 | ! #################################################################################### |
---|
136 | ! *) Molecular signal |
---|
137 | ! #################################################################################### |
---|
138 | call cmp_backsignal(nlev,npoints,beta_mol(1:npoints,1:nlev),& |
---|
139 | tau_mol(1:npoints,1:nlev),pmol(1:npoints,1:nlev)) |
---|
140 | |
---|
141 | ! #################################################################################### |
---|
142 | ! PLANE PARRALLEL FIELDS |
---|
143 | ! #################################################################################### |
---|
144 | do icol=1,ncolumns |
---|
145 | ! ################################################################################# |
---|
146 | ! *) Total Backscatter signal |
---|
147 | ! ################################################################################# |
---|
148 | call cmp_backsignal(nlev,npoints,betatot(1:npoints,icol,1:nlev),& |
---|
149 | tautot(1:npoints,icol,1:nlev),pnorm(1:npoints,icol,1:nlev)) |
---|
150 | ! ################################################################################# |
---|
151 | ! *) Ice/Liq Backscatter signal |
---|
152 | ! ################################################################################# |
---|
153 | ! Computation of the ice and liquid lidar backscattered signal (ATBice and ATBliq) |
---|
154 | ! Ice only |
---|
155 | call cmp_backsignal(nlev,npoints,betatot_ice(1:npoints,icol,1:nlev),& |
---|
156 | tautot_ice(1:npoints,icol,1:nlev),& |
---|
157 | pnorm_ice(1:npoints,icol,1:nlev)) |
---|
158 | ! Liquid only |
---|
159 | call cmp_backsignal(nlev,npoints,betatot_liq(1:npoints,icol,1:nlev),& |
---|
160 | tautot_liq(1:npoints,icol,1:nlev),& |
---|
161 | pnorm_liq(1:npoints,icol,1:nlev)) |
---|
162 | enddo |
---|
163 | |
---|
164 | ! #################################################################################### |
---|
165 | ! PERDENDICULAR FIELDS |
---|
166 | ! #################################################################################### |
---|
167 | do icol=1,ncolumns |
---|
168 | |
---|
169 | ! ################################################################################# |
---|
170 | ! *) Ice/Liq Perpendicular Backscatter signal |
---|
171 | ! ################################################################################# |
---|
172 | ! Computation of ATBperp,ice/liq from ATBice/liq including the multiple scattering |
---|
173 | ! contribution (Cesana and Chepfer 2013, JGR) |
---|
174 | do k=1,nlev |
---|
175 | ! Ice particles |
---|
176 | pnorm_perp_ice(1:npoints,icol,k) = Alpha * pnorm_ice(1:npoints,icol,k) |
---|
177 | |
---|
178 | ! Liquid particles |
---|
179 | pnorm_perp_liq(1:npoints,icol,k) = 1000._wp*Beta*pnorm_liq(1:npoints,icol,k)**2+& |
---|
180 | Gamma*pnorm_liq(1:npoints,icol,k) |
---|
181 | enddo |
---|
182 | |
---|
183 | ! ################################################################################# |
---|
184 | ! *) Computation of beta_perp_ice/liq using the lidar equation |
---|
185 | ! ################################################################################# |
---|
186 | ! Ice only |
---|
187 | call cmp_beta(nlev,npoints,pnorm_perp_ice(1:npoints,icol,1:nlev),& |
---|
188 | tautot_ice(1:npoints,icol,1:nlev),beta_perp_ice(1:npoints,icol,1:nlev)) |
---|
189 | |
---|
190 | ! Liquid only |
---|
191 | call cmp_beta(nlev,npoints,pnorm_perp_liq(1:npoints,icol,1:nlev),& |
---|
192 | tautot_liq(1:npoints,icol,1:nlev),beta_perp_liq(1:npoints,icol,1:nlev)) |
---|
193 | |
---|
194 | ! ################################################################################# |
---|
195 | ! *) Perpendicular Backscatter signal |
---|
196 | ! ################################################################################# |
---|
197 | ! Computation of the total perpendicular lidar signal (ATBperp for liq+ice) |
---|
198 | ! Upper layer |
---|
199 | WHERE(tautot(1:npoints,icol,1) .gt. 0) |
---|
200 | pnorm_perp_tot(1:npoints,icol,1) = (beta_perp_ice(1:npoints,icol,1)+ & |
---|
201 | beta_perp_liq(1:npoints,icol,1)- & |
---|
202 | (beta_mol(1:npoints,1)/(1._wp+1._wp/0.0284_wp))) / & |
---|
203 | (2._wp*tautot(1:npoints,icol,1))* & |
---|
204 | (1._wp-exp(-2._wp*tautot(1:npoints,icol,1))) |
---|
205 | ELSEWHERE |
---|
206 | pnorm_perp_tot(1:npoints,icol,1) = 0._wp |
---|
207 | ENDWHERE |
---|
208 | |
---|
209 | ! Other layers |
---|
210 | do k=2,nlev |
---|
211 | ! Optical thickness of layer k |
---|
212 | tautot_lay(1:npoints) = tautot(1:npoints,icol,k)-tautot(1:npoints,icol,k-1) |
---|
213 | |
---|
214 | ! The perpendicular component of the molecular backscattered signal (Betaperp) |
---|
215 | ! has been taken into account two times (once for liquid and once for ice). |
---|
216 | ! We remove one contribution using |
---|
217 | ! Betaperp=beta_mol(:,k)/(1+1/0.0284)) [bodhaine et al. 1999] in the following |
---|
218 | ! equations: |
---|
219 | WHERE (pnorm(1:npoints,icol,k) .eq. 0) |
---|
220 | pnorm_perp_tot(1:npoints,icol,k)=0._wp |
---|
221 | ELSEWHERE |
---|
222 | where(tautot_lay(1:npoints) .gt. 0.) |
---|
223 | pnorm_perp_tot(1:npoints,icol,k) = (beta_perp_ice(1:npoints,icol,k)+ & |
---|
224 | beta_perp_liq(1:npoints,icol,k)-(beta_mol(1:npoints,k)/(1._wp+1._wp/ & |
---|
225 | 0.0284_wp)))*EXP(-2._wp*tautot(1:npoints,icol,k-1))/ & |
---|
226 | (2._wp*tautot_lay(1:npoints))* (1._wp-EXP(-2._wp*tautot_lay(1:npoints))) |
---|
227 | elsewhere |
---|
228 | pnorm_perp_tot(1:npoints,icol,k) = (beta_perp_ice(1:npoints,icol,k)+ & |
---|
229 | beta_perp_liq(1:npoints,icol,k)-(beta_mol(1:npoints,k)/(1._wp+1._wp/ & |
---|
230 | 0.0284_wp)))*EXP(-2._wp*tautot(1:npoints,icol,k-1)) |
---|
231 | endwhere |
---|
232 | ENDWHERE |
---|
233 | END DO |
---|
234 | enddo |
---|
235 | end subroutine lidar_subcolumn |
---|
236 | |
---|
237 | ! ###################################################################################### |
---|
238 | ! SUBROUTINE lidar_column |
---|
239 | ! ###################################################################################### |
---|
240 | subroutine lidar_column(npoints,ncol,nlevels,llm,max_bin,tmp, pnorm, & |
---|
241 | pnorm_perp, pmol, pplay, ok_lidar_cfad, ncat, cfad2, & |
---|
242 | lidarcld, lidarcldphase, cldlayer, zlev, zlev_half, & |
---|
243 | cldlayerphase, lidarcldtmp) |
---|
244 | integer,parameter :: & |
---|
245 | nphase = 6 ! Number of cloud layer phase types |
---|
246 | |
---|
247 | ! Inputs |
---|
248 | integer,intent(in) :: & |
---|
249 | npoints, & ! Number of horizontal grid points |
---|
250 | ncol, & ! Number of subcolumns |
---|
251 | nlevels, & ! Number of vertical layers (OLD grid) |
---|
252 | llm, & ! Number of vertical layers (NEW grid) |
---|
253 | max_bin, & ! Number of bins for SR CFADs |
---|
254 | ncat ! Number of cloud layer types (low,mid,high,total) |
---|
255 | real(wp),intent(in),dimension(npoints,ncol,Nlevels) :: & |
---|
256 | pnorm, & ! Lidar ATB |
---|
257 | pnorm_perp ! Lidar perpendicular ATB |
---|
258 | real(wp),intent(in),dimension(npoints,Nlevels) :: & |
---|
259 | pmol, & ! Molecular ATB |
---|
260 | pplay, & ! Pressure on model levels (Pa) |
---|
261 | tmp ! Temperature at each levels |
---|
262 | logical,intent(in) :: & |
---|
263 | ok_lidar_cfad ! True if lidar CFAD diagnostics need to be computed |
---|
264 | real(wp),intent(in),dimension(npoints,nlevels) :: & |
---|
265 | zlev ! Model full levels |
---|
266 | real(wp),intent(in),dimension(npoints,nlevels+1) :: & |
---|
267 | zlev_half ! Model half levels |
---|
268 | |
---|
269 | ! Outputs |
---|
270 | real(wp),intent(inout),dimension(npoints,llm) :: & |
---|
271 | lidarcld ! 3D "lidar" cloud fraction |
---|
272 | real(wp),intent(inout),dimension(npoints,ncat) :: & |
---|
273 | cldlayer ! "lidar" cloud layer fraction (low, mid, high, total) |
---|
274 | real(wp),intent(inout),dimension(npoints,llm,nphase) :: & |
---|
275 | lidarcldphase ! 3D "lidar" phase cloud fraction |
---|
276 | real(wp),intent(inout),dimension(npoints,40,5) :: & |
---|
277 | lidarcldtmp ! 3D "lidar" phase cloud fraction as a function of temp |
---|
278 | real(wp),intent(inout),dimension(npoints,ncat,nphase) :: & |
---|
279 | cldlayerphase ! "lidar" phase low mid high cloud fraction |
---|
280 | real(wp),intent(inout),dimension(npoints,max_bin,llm) :: & |
---|
281 | cfad2 ! CFADs of SR |
---|
282 | |
---|
283 | ! Local Variables |
---|
284 | integer :: ic,i,j |
---|
285 | real(wp),dimension(npoints,ncol,llm) :: & |
---|
286 | x3d |
---|
287 | real(wp),dimension(npoints,llm) :: & |
---|
288 | x3d_c,pnorm_c |
---|
289 | real(wp) :: & |
---|
290 | xmax |
---|
291 | real(wp),dimension(npoints,1,Nlevels) :: t_in,ph_in,betamol_in |
---|
292 | real(wp),dimension(npoints,ncol,llm) :: pnormFlip,pnorm_perpFlip |
---|
293 | real(wp),dimension(npoints,1,llm) :: tmpFlip,pplayFlip,betamolFlip |
---|
294 | |
---|
295 | ! Vertically regrid input data |
---|
296 | if (use_vgrid) then |
---|
297 | t_in(:,1,:)=tmp(:,nlevels:1:-1) |
---|
298 | call cosp_change_vertical_grid(Npoints,1,Nlevels,zlev(:,nlevels:1:-1),zlev_half(:,nlevels:1:-1),& |
---|
299 | t_in,llm,vgrid_zl(llm:1:-1),vgrid_zu(llm:1:-1),tmpFlip(:,1,llm:1:-1)) |
---|
300 | ph_in(:,1,:) = pplay(:,nlevels:1:-1) |
---|
301 | call cosp_change_vertical_grid(Npoints,1,Nlevels,zlev(:,nlevels:1:-1),zlev_half(:,nlevels:1:-1),& |
---|
302 | ph_in,llm,vgrid_zl(llm:1:-1),vgrid_zu(llm:1:-1),pplayFlip(:,1,llm:1:-1)) |
---|
303 | betamol_in(:,1,:) = pmol(:,nlevels:1:-1) |
---|
304 | call cosp_change_vertical_grid(Npoints,1,Nlevels,zlev(:,nlevels:1:-1),zlev_half(:,nlevels:1:-1),& |
---|
305 | betamol_in,llm,vgrid_zl(llm:1:-1),vgrid_zu(llm:1:-1),betamolFlip(:,1,llm:1:-1)) |
---|
306 | call cosp_change_vertical_grid(Npoints,Ncol,Nlevels,zlev(:,nlevels:1:-1),zlev_half(:,nlevels:1:-1),& |
---|
307 | pnorm(:,:,nlevels:1:-1),llm,vgrid_zl(llm:1:-1),vgrid_zu(llm:1:-1),pnormFlip(:,:,llm:1:-1)) |
---|
308 | call cosp_change_vertical_grid(Npoints,Ncol,Nlevels,zlev(:,nlevels:1:-1),zlev_half(:,nlevels:1:-1),& |
---|
309 | pnorm_perp(:,:,nlevels:1:-1),llm,vgrid_zl(llm:1:-1),vgrid_zu(llm:1:-1),pnorm_perpFlip(:,:,llm:1:-1)) |
---|
310 | endif |
---|
311 | |
---|
312 | ! Initialization (The histogram bins, are set up during initialization and the |
---|
313 | ! maximum value is used as the upper bounds.) |
---|
314 | xmax = maxval(calipso_histBsct) |
---|
315 | |
---|
316 | ! Compute LIDAR scattering ratio |
---|
317 | if (use_vgrid) then |
---|
318 | do ic = 1, ncol |
---|
319 | pnorm_c = pnormFlip(:,ic,:) |
---|
320 | where ((pnorm_c .lt. xmax) .and. (betamolFlip(:,1,:) .lt. xmax) .and. & |
---|
321 | (betamolFlip(:,1,:) .gt. 0.0 )) |
---|
322 | x3d_c = pnorm_c/betamolFlip(:,1,:) |
---|
323 | elsewhere |
---|
324 | x3d_c = R_UNDEF |
---|
325 | end where |
---|
326 | x3d(:,ic,:) = x3d_c |
---|
327 | enddo |
---|
328 | ! Diagnose cloud fractions for subcolumn lidar scattering ratios |
---|
329 | CALL COSP_CLDFRAC(npoints,ncol,llm,ncat,nphase,tmpFlip,x3d,pnormFlip, & |
---|
330 | pnorm_perpFlip,pplayFlip,S_att,S_cld,S_cld_att,R_UNDEF, & |
---|
331 | lidarcld,cldlayer,lidarcldphase,cldlayerphase,lidarcldtmp) |
---|
332 | else |
---|
333 | do ic = 1, ncol |
---|
334 | pnorm_c = pnorm(:,ic,:) |
---|
335 | where ((pnorm_c.lt.xmax) .and. (pmol.lt.xmax) .and. (pmol.gt. 0.0 )) |
---|
336 | x3d_c = pnorm_c/pmol |
---|
337 | elsewhere |
---|
338 | x3d_c = R_UNDEF |
---|
339 | end where |
---|
340 | x3d(:,ic,:) = x3d_c |
---|
341 | enddo |
---|
342 | ! Diagnose cloud fractions for subcolumn lidar scattering ratios |
---|
343 | CALL COSP_CLDFRAC(npoints,ncol,nlevels,ncat,nphase,tmp,x3d,pnorm,pnorm_perp,pplay,& |
---|
344 | S_att,S_cld,S_cld_att,R_UNDEF,lidarcld,cldlayer,lidarcldphase, & |
---|
345 | cldlayerphase,lidarcldtmp) |
---|
346 | endif |
---|
347 | |
---|
348 | ! CFADs |
---|
349 | if (ok_lidar_cfad) then |
---|
350 | ! CFADs of subgrid-scale lidar scattering ratios |
---|
351 | do i=1,Npoints |
---|
352 | do j=1,llm |
---|
353 | cfad2(i,:,j) = hist1D(ncol,x3d(i,:,j),SR_BINS,calipso_histBsct) |
---|
354 | enddo |
---|
355 | enddo |
---|
356 | where(cfad2 .ne. R_UNDEF) cfad2=cfad2/ncol |
---|
357 | |
---|
358 | endif |
---|
359 | |
---|
360 | ! Unit conversions |
---|
361 | where(lidarcld /= R_UNDEF) lidarcld = lidarcld*100._wp |
---|
362 | where(cldlayer /= R_UNDEF) cldlayer = cldlayer*100._wp |
---|
363 | where(cldlayerphase /= R_UNDEF) cldlayerphase = cldlayerphase*100._wp |
---|
364 | where(lidarcldphase /= R_UNDEF) lidarcldphase = lidarcldphase*100._wp |
---|
365 | where(lidarcldtmp /= R_UNDEF) lidarcldtmp = lidarcldtmp*100._wp |
---|
366 | |
---|
367 | end subroutine lidar_column |
---|
368 | |
---|
369 | ! ###################################################################################### |
---|
370 | ! The subroutines below compute the attenuated backscatter signal and the lidar |
---|
371 | ! backscatter coefficients using eq (1) from doi:0094-8276/08/2008GL034207 |
---|
372 | ! ###################################################################################### |
---|
373 | subroutine cmp_backsignal(nlev,npoints,beta,tau,pnorm) |
---|
374 | ! INPUTS |
---|
375 | integer, intent(in) :: nlev,npoints |
---|
376 | real(wp),intent(in),dimension(npoints,nlev) :: beta,tau |
---|
377 | |
---|
378 | ! OUTPUTS |
---|
379 | real(wp),intent(out),dimension(npoints,nlev) :: pnorm |
---|
380 | |
---|
381 | ! Internal Variables |
---|
382 | real(wp), dimension(npoints) :: tautot_lay |
---|
383 | integer :: k |
---|
384 | |
---|
385 | ! Uppermost layer |
---|
386 | pnorm(:,1) = beta(:,1) / (2._wp*tau(:,1)) * (1._wp-exp(-2._wp*tau(:,1))) |
---|
387 | |
---|
388 | ! Other layers |
---|
389 | do k=2,nlev |
---|
390 | tautot_lay(:) = tau(:,k)-tau(:,k-1) |
---|
391 | WHERE ( EXP(-2._wp*tau(:,k-1)) .gt. 0. ) |
---|
392 | WHERE (tautot_lay(:) .gt. 0.) |
---|
393 | pnorm(:,k) = beta(:,k)*EXP(-2._wp*tau(:,k-1)) /& |
---|
394 | (2._wp*tautot_lay(:))*(1._wp-EXP(-2._wp*tautot_lay(:))) |
---|
395 | ELSEWHERE |
---|
396 | ! This must never happen, but just in case, to avoid div. by 0 |
---|
397 | pnorm(:,k) = beta(:,k) * EXP(-2._wp*tau(:,k-1)) |
---|
398 | END WHERE |
---|
399 | ELSEWHERE |
---|
400 | pnorm(:,k) = 0._wp!beta(:,k) |
---|
401 | END WHERE |
---|
402 | END DO |
---|
403 | end subroutine cmp_backsignal |
---|
404 | |
---|
405 | subroutine cmp_beta(nlev,npoints,pnorm,tau,beta) |
---|
406 | ! INPUTS |
---|
407 | integer, intent(in) :: nlev,npoints |
---|
408 | real(wp),intent(in),dimension(npoints,nlev) :: pnorm,tau |
---|
409 | |
---|
410 | ! OUTPUTS |
---|
411 | real(wp),intent(out),dimension(npoints,nlev) :: beta |
---|
412 | |
---|
413 | ! Internal Variables |
---|
414 | real(wp), dimension(npoints) :: tautot_lay |
---|
415 | integer :: k |
---|
416 | |
---|
417 | beta(:,1) = pnorm(:,1) * (2._wp*tau(:,1))/(1._wp-exp(-2._wp*tau(:,1))) |
---|
418 | do k=2,nlev |
---|
419 | tautot_lay(:) = tau(:,k)-tau(:,k-1) |
---|
420 | WHERE ( EXP(-2._wp*tau(:,k-1)) .gt. 0. ) |
---|
421 | WHERE (tautot_lay(:) .gt. 0.) |
---|
422 | beta(:,k) = pnorm(:,k)/ EXP(-2._wp*tau(:,k-1))* & |
---|
423 | (2._wp*tautot_lay(:))/(1._wp-exp(-2._wp*tautot_lay(:))) |
---|
424 | ELSEWHERE |
---|
425 | beta(:,k)=pnorm(:,k)/EXP(-2._wp*tau(:,k-1)) |
---|
426 | END WHERE |
---|
427 | ELSEWHERE |
---|
428 | beta(:,k)=pnorm(:,k) |
---|
429 | END WHERE |
---|
430 | ENDDO |
---|
431 | |
---|
432 | end subroutine cmp_beta |
---|
433 | ! #################################################################################### |
---|
434 | ! SUBROUTINE cosp_cldfrac |
---|
435 | ! Conventions: Ncat must be equal to 4 |
---|
436 | ! #################################################################################### |
---|
437 | SUBROUTINE COSP_CLDFRAC(Npoints,Ncolumns,Nlevels,Ncat,Nphase,tmp,x,ATB,ATBperp, & |
---|
438 | pplay,S_att,S_cld,S_cld_att,undef,lidarcld,cldlayer, & |
---|
439 | lidarcldphase,cldlayerphase,lidarcldtemp) |
---|
440 | ! Parameters |
---|
441 | integer,parameter :: Ntemp=40 ! indice of the temperature vector |
---|
442 | real(wp),parameter,dimension(Ntemp+1) :: & |
---|
443 | tempmod = [0.0, 183.15,186.15,189.15,192.15,195.15,198.15,201.15,204.15,207.15, & |
---|
444 | 210.15,213.15,216.15,219.15,222.15,225.15,228.15,231.15,234.15,237.15, & |
---|
445 | 240.15,243.15,246.15,249.15,252.15,255.15,258.15,261.15,264.15,267.15, & |
---|
446 | 270.15,273.15,276.15,279.15,282.15,285.15,288.15,291.15,294.15,297.15, & |
---|
447 | 473.15] |
---|
448 | |
---|
449 | ! Polynomial coefficient of the phase discrimination line used to separate liquid from ice |
---|
450 | ! (Cesana and Chepfer, JGR, 2013) |
---|
451 | ! ATBperp = ATB^5*alpha50 + ATB^4*beta50 + ATB^3*gamma50 + ATB^2*delta50 + ATB*epsilon50 + zeta50 |
---|
452 | real(wp),parameter :: & |
---|
453 | alpha50 = 9.0322e+15_wp, & ! |
---|
454 | beta50 = -2.1358e+12_wp, & ! |
---|
455 | gamma50 = 173.3963e06_wp, & ! |
---|
456 | delta50 = -3.9514e03_wp, & ! |
---|
457 | epsilon50 = 0.2559_wp, & ! |
---|
458 | zeta50 = -9.4776e-07_wp ! |
---|
459 | |
---|
460 | ! Inputs |
---|
461 | integer,intent(in) :: & |
---|
462 | Npoints, & ! Number of gridpoints |
---|
463 | Ncolumns, & ! Number of subcolumns |
---|
464 | Nlevels, & ! Number of vertical levels |
---|
465 | Ncat, & ! Number of cloud layer types |
---|
466 | Nphase ! Number of cloud layer phase types |
---|
467 | ! [ice,liquid,undefined,false ice,false liquid,Percent of ice] |
---|
468 | real(wp),intent(in) :: & |
---|
469 | S_att, & ! |
---|
470 | S_cld, & ! |
---|
471 | S_cld_att,& ! New threshold for undefine cloud phase detection |
---|
472 | undef ! Undefined value |
---|
473 | real(wp),intent(in),dimension(Npoints,Ncolumns,Nlevels) :: & |
---|
474 | x, & ! |
---|
475 | ATB, & ! 3D attenuated backscatter |
---|
476 | ATBperp ! 3D attenuated backscatter (perpendicular) |
---|
477 | real(wp),intent(in),dimension(Npoints,Nlevels) :: & |
---|
478 | tmp, & ! Temperature |
---|
479 | pplay ! Pressure |
---|
480 | |
---|
481 | ! Outputs |
---|
482 | real(wp),intent(out),dimension(Npoints,Ntemp,5) :: & |
---|
483 | lidarcldtemp ! 3D Temperature 1=tot,2=ice,3=liq,4=undef,5=ice/ice+liq |
---|
484 | real(wp),intent(out),dimension(Npoints,Nlevels,Nphase) :: & |
---|
485 | lidarcldphase ! 3D cloud phase fraction |
---|
486 | real(wp),intent(out),dimension(Npoints,Nlevels) :: & |
---|
487 | lidarcld ! 3D cloud fraction |
---|
488 | real(wp),intent(out),dimension(Npoints,Ncat) :: & |
---|
489 | cldlayer ! Low, middle, high, total cloud fractions |
---|
490 | real(wp),intent(out),dimension(Npoints,Ncat,Nphase) :: & |
---|
491 | cldlayerphase ! Low, middle, high, total cloud fractions for ice liquid and undefine phase |
---|
492 | |
---|
493 | ! Local variables |
---|
494 | integer :: & |
---|
495 | ip, k, iz, ic, ncol, nlev, i, itemp, toplvlsat |
---|
496 | real(wp) :: & |
---|
497 | p1,checktemp, ATBperp_tmp,checkcldlayerphase, checkcldlayerphase2 |
---|
498 | real(wp),dimension(Npoints,Nlevels) :: & |
---|
499 | nsub,lidarcldphasetmp |
---|
500 | real(wp),dimension(Npoints,Ntemp) :: & |
---|
501 | sumlidarcldtemp,lidarcldtempind |
---|
502 | real(wp),dimension(Npoints,Ncolumns,Ncat) :: & |
---|
503 | cldlay,nsublay |
---|
504 | real(wp),dimension(Npoints,Ncat) :: & |
---|
505 | nsublayer,cldlayerphasetmp,cldlayerphasesum |
---|
506 | real(wp),dimension(Npoints,Ncolumns,Nlevels) :: & |
---|
507 | tmpi, & ! Temperature of ice cld |
---|
508 | tmpl, & ! Temperature of liquid cld |
---|
509 | tmpu, & ! Temperature of undef cld |
---|
510 | cldy, & ! |
---|
511 | srok ! |
---|
512 | real(wp),dimension(Npoints,Ncolumns,Ncat,Nphase) :: & |
---|
513 | cldlayphase ! subgrided low mid high phase cloud fraction |
---|
514 | |
---|
515 | ! #################################################################################### |
---|
516 | ! 1) Initialize |
---|
517 | ! #################################################################################### |
---|
518 | lidarcld = 0._wp |
---|
519 | nsub = 0._wp |
---|
520 | cldlay = 0._wp |
---|
521 | nsublay = 0._wp |
---|
522 | ATBperp_tmp = 0._wp |
---|
523 | lidarcldphase(:,:,:) = 0._wp |
---|
524 | cldlayphase(:,:,:,:) = 0._wp |
---|
525 | cldlayerphase(:,:,:) = 0._wp |
---|
526 | tmpi(:,:,:) = 0._wp |
---|
527 | tmpl(:,:,:) = 0._wp |
---|
528 | tmpu(:,:,:) = 0._wp |
---|
529 | cldlayerphasesum(:,:) = 0._wp |
---|
530 | lidarcldtemp(:,:,:) = 0._wp |
---|
531 | lidarcldtempind(:,:) = 0._wp |
---|
532 | sumlidarcldtemp(:,:) = 0._wp |
---|
533 | lidarcldphasetmp(:,:) = 0._wp |
---|
534 | toplvlsat = 0 |
---|
535 | |
---|
536 | ! #################################################################################### |
---|
537 | ! 2) Cloud detection |
---|
538 | ! #################################################################################### |
---|
539 | do k=1,Nlevels |
---|
540 | ! Cloud detection at subgrid-scale: |
---|
541 | where ((x(:,:,k) .gt. S_cld) .and. (x(:,:,k) .ne. undef) ) |
---|
542 | cldy(:,:,k)=1._wp |
---|
543 | elsewhere |
---|
544 | cldy(:,:,k)=0._wp |
---|
545 | endwhere |
---|
546 | |
---|
547 | ! Number of usefull sub-columns: |
---|
548 | where ((x(:,:,k) .gt. S_att) .and. (x(:,:,k) .ne. undef) ) |
---|
549 | srok(:,:,k)=1._wp |
---|
550 | elsewhere |
---|
551 | srok(:,:,k)=0._wp |
---|
552 | endwhere |
---|
553 | enddo |
---|
554 | |
---|
555 | ! #################################################################################### |
---|
556 | ! 3) Grid-box 3D cloud fraction and layered cloud fractions(ISCCP pressure categories) |
---|
557 | ! #################################################################################### |
---|
558 | lidarcld = 0._wp |
---|
559 | nsub = 0._wp |
---|
560 | cldlay = 0._wp |
---|
561 | nsublay = 0._wp |
---|
562 | do k=1,Nlevels |
---|
563 | do ic = 1, Ncolumns |
---|
564 | do ip = 1, Npoints |
---|
565 | |
---|
566 | ! Computation of the cloud fraction as a function of the temperature instead |
---|
567 | ! of height, for ice,liquid and all clouds |
---|
568 | if(srok(ip,ic,k).gt.0.)then |
---|
569 | do itemp=1,Ntemp |
---|
570 | if( (tmp(ip,k).ge.tempmod(itemp)).and.(tmp(ip,k).lt.tempmod(itemp+1)) )then |
---|
571 | lidarcldtempind(ip,itemp)=lidarcldtempind(ip,itemp)+1._wp |
---|
572 | endif |
---|
573 | enddo |
---|
574 | endif |
---|
575 | |
---|
576 | if(cldy(ip,ic,k).eq.1.)then |
---|
577 | do itemp=1,Ntemp |
---|
578 | if( (tmp(ip,k) .ge. tempmod(itemp)).and.(tmp(ip,k) .lt. tempmod(itemp+1)) )then |
---|
579 | lidarcldtemp(ip,itemp,1)=lidarcldtemp(ip,itemp,1)+1._wp |
---|
580 | endif |
---|
581 | enddo |
---|
582 | endif |
---|
583 | |
---|
584 | iz=1 |
---|
585 | p1 = pplay(ip,k) |
---|
586 | if ( p1.gt.0. .and. p1.lt.(440._wp*100._wp)) then ! high clouds |
---|
587 | iz=3 |
---|
588 | else if(p1.ge.(440._wp*100._wp) .and. p1.lt.(680._wp*100._wp)) then ! mid clouds |
---|
589 | iz=2 |
---|
590 | endif |
---|
591 | |
---|
592 | cldlay(ip,ic,iz) = MAX(cldlay(ip,ic,iz),cldy(ip,ic,k)) |
---|
593 | cldlay(ip,ic,4) = MAX(cldlay(ip,ic,4),cldy(ip,ic,k)) |
---|
594 | lidarcld(ip,k) = lidarcld(ip,k) + cldy(ip,ic,k) |
---|
595 | |
---|
596 | nsublay(ip,ic,iz) = MAX(nsublay(ip,ic,iz),srok(ip,ic,k)) |
---|
597 | nsublay(ip,ic,4) = MAX(nsublay(ip,ic,4),srok(ip,ic,k)) |
---|
598 | nsub(ip,k) = nsub(ip,k) + srok(ip,ic,k) |
---|
599 | |
---|
600 | enddo |
---|
601 | enddo |
---|
602 | enddo |
---|
603 | |
---|
604 | ! Grid-box 3D cloud fraction |
---|
605 | where ( nsub(:,:).gt.0.0 ) |
---|
606 | lidarcld(:,:) = lidarcld(:,:)/nsub(:,:) |
---|
607 | elsewhere |
---|
608 | lidarcld(:,:) = undef |
---|
609 | endwhere |
---|
610 | |
---|
611 | ! Layered cloud fractions |
---|
612 | cldlayer = 0._wp |
---|
613 | nsublayer = 0._wp |
---|
614 | do iz = 1, Ncat |
---|
615 | do ic = 1, Ncolumns |
---|
616 | cldlayer(:,iz) = cldlayer(:,iz) + cldlay(:,ic,iz) |
---|
617 | nsublayer(:,iz) = nsublayer(:,iz) + nsublay(:,ic,iz) |
---|
618 | enddo |
---|
619 | enddo |
---|
620 | where (nsublayer(:,:) .gt. 0.0) |
---|
621 | cldlayer(:,:) = cldlayer(:,:)/nsublayer(:,:) |
---|
622 | elsewhere |
---|
623 | cldlayer(:,:) = undef |
---|
624 | endwhere |
---|
625 | |
---|
626 | ! #################################################################################### |
---|
627 | ! 4) Grid-box 3D cloud Phase |
---|
628 | ! #################################################################################### |
---|
629 | |
---|
630 | ! #################################################################################### |
---|
631 | ! 4.1) For Cloudy pixels with 8.16km < z < 19.2km |
---|
632 | ! #################################################################################### |
---|
633 | do ncol=1,Ncolumns |
---|
634 | do i=1,Npoints |
---|
635 | do nlev=1,23 ! from 19.2km until 8.16km |
---|
636 | p1 = pplay(1,nlev) |
---|
637 | |
---|
638 | ! Avoid zero values |
---|
639 | if( (cldy(i,ncol,nlev).eq.1.) .and. (ATBperp(i,ncol,nlev).gt.0.) )then |
---|
640 | ! Computation of the ATBperp along the phase discrimination line |
---|
641 | ATBperp_tmp = (ATB(i,ncol,nlev)**5)*alpha50 + (ATB(i,ncol,nlev)**4)*beta50 + & |
---|
642 | (ATB(i,ncol,nlev)**3)*gamma50 + (ATB(i,ncol,nlev)**2)*delta50 + & |
---|
643 | ATB(i,ncol,nlev)*epsilon50 + zeta50 |
---|
644 | ! ######################################################################## |
---|
645 | ! 4.1.a) Ice: ATBperp above the phase discrimination line |
---|
646 | ! ######################################################################## |
---|
647 | if((ATBperp(i,ncol,nlev)-ATBperp_tmp) .ge. 0.)then ! Ice clouds |
---|
648 | |
---|
649 | ! ICE with temperature above 273,15°K = Liquid (false ice) |
---|
650 | if(tmp(i,nlev) .gt. 273.15) then ! Temperature above 273,15 K |
---|
651 | ! Liquid: False ice corrected by the temperature to Liquid |
---|
652 | lidarcldphase(i,nlev,2) = lidarcldphase(i,nlev,2)+1._wp ! False ice detection ==> added to Liquid |
---|
653 | |
---|
654 | tmpl(i,ncol,nlev) = tmp(i,nlev) |
---|
655 | lidarcldphase(i,nlev,5) = lidarcldphase(i,nlev,5)+1._wp ! Keep the information "temperature criterium used" |
---|
656 | ! to classify the phase cloud |
---|
657 | cldlayphase(i,ncol,4,2) = 1. ! tot cloud |
---|
658 | if (p1 .gt. 0. .and. p1.lt.(440._wp*100._wp)) then ! high cloud |
---|
659 | cldlayphase(i,ncol,3,2) = 1._wp |
---|
660 | else if(p1 .ge. (440._wp*100._wp) .and. p1 .lt. (680._wp*100._wp)) then ! mid cloud |
---|
661 | cldlayphase(i,ncol,2,2) = 1._wp |
---|
662 | else ! low cloud |
---|
663 | cldlayphase(i,ncol,1,2) = 1._wp |
---|
664 | endif |
---|
665 | cldlayphase(i,ncol,4,5) = 1._wp ! tot cloud |
---|
666 | ! High cloud |
---|
667 | if (p1 .gt. 0. .and. p1 .lt. (440._wp*100._wp)) then |
---|
668 | cldlayphase(i,ncol,3,5) = 1._wp |
---|
669 | ! Middle cloud |
---|
670 | else if(p1 .ge. (440._wp*100._wp) .and. p1 .lt. (680._wp*100._wp)) then |
---|
671 | cldlayphase(i,ncol,2,5) = 1._wp |
---|
672 | ! Low cloud |
---|
673 | else |
---|
674 | cldlayphase(i,ncol,1,5) = 1._wp |
---|
675 | endif |
---|
676 | else |
---|
677 | ! ICE with temperature below 273,15°K |
---|
678 | lidarcldphase(i,nlev,1) = lidarcldphase(i,nlev,1)+1._wp |
---|
679 | tmpi(i,ncol,nlev) = tmp(i,nlev) |
---|
680 | cldlayphase(i,ncol,4,1) = 1._wp ! tot cloud |
---|
681 | ! High cloud |
---|
682 | if (p1 .gt. 0. .and. p1 .lt. (440._wp*100._wp)) then |
---|
683 | cldlayphase(i,ncol,3,1) = 1._wp |
---|
684 | ! Middle cloud |
---|
685 | else if(p1 .ge. (440._wp*100._wp) .and. p1 .lt. (680._wp*100._wp)) then |
---|
686 | cldlayphase(i,ncol,2,1) = 1._wp |
---|
687 | ! Low cloud |
---|
688 | else |
---|
689 | cldlayphase(i,ncol,1,1) = 1._wp |
---|
690 | endif |
---|
691 | endif |
---|
692 | ! ######################################################################## |
---|
693 | ! 4.1.b) Liquid: ATBperp below the phase discrimination line |
---|
694 | ! ######################################################################## |
---|
695 | else |
---|
696 | ! Liquid with temperature above 231,15°K |
---|
697 | if(tmp(i,nlev) .gt. 231.15_wp) then |
---|
698 | lidarcldphase(i,nlev,2) = lidarcldphase(i,nlev,2)+1._wp |
---|
699 | tmpl(i,ncol,nlev) = tmp(i,nlev) |
---|
700 | cldlayphase(i,ncol,4,2) = 1._wp ! tot cloud |
---|
701 | ! High cloud |
---|
702 | if (p1 .gt. 0. .and. p1 .lt. (440._wp*100._wp)) then |
---|
703 | cldlayphase(i,ncol,3,2) = 1._wp |
---|
704 | ! Middle cloud |
---|
705 | else if(p1 .ge. (440._wp*100._wp) .and. p1 .lt. (680._wp*100._wp)) then |
---|
706 | cldlayphase(i,ncol,2,2) = 1._wp |
---|
707 | ! Low cloud |
---|
708 | else |
---|
709 | cldlayphase(i,ncol,1,2) = 1._wp |
---|
710 | endif |
---|
711 | else |
---|
712 | ! Liquid with temperature below 231,15°K = Ice (false liquid) |
---|
713 | tmpi(i,ncol,nlev) = tmp(i,nlev) |
---|
714 | lidarcldphase(i,nlev,1) = lidarcldphase(i,nlev,1)+1._wp ! false liquid detection ==> added to ice |
---|
715 | lidarcldphase(i,nlev,4) = lidarcldphase(i,nlev,4)+1._wp |
---|
716 | cldlayphase(i,ncol,4,4) = 1._wp ! tot cloud |
---|
717 | ! High cloud |
---|
718 | if (p1 .gt. 0. .and. p1 .lt. (440._wp*100._wp)) then |
---|
719 | cldlayphase(i,ncol,3,4) = 1._wp |
---|
720 | ! Middle cloud |
---|
721 | else if(p1 .ge. (440._wp*100._wp) .and. p1 .lt. (680._wp*100._wp)) then |
---|
722 | cldlayphase(i,ncol,2,4) = 1._wp |
---|
723 | ! Low cloud |
---|
724 | else |
---|
725 | cldlayphase(i,ncol,1,4) = 1._wp |
---|
726 | endif |
---|
727 | cldlayphase(i,ncol,4,1) = 1._wp ! tot cloud |
---|
728 | ! High cloud |
---|
729 | if (p1 .gt. 0. .and. p1 .lt. (440._wp*100._wp)) then |
---|
730 | cldlayphase(i,ncol,3,1) = 1._wp |
---|
731 | ! Middle cloud |
---|
732 | else if(p1 .ge. (440._wp*100._wp) .and. p1 .lt. (680._wp*100._wp)) then |
---|
733 | cldlayphase(i,ncol,2,1) = 1._wp |
---|
734 | ! Low cloud |
---|
735 | else |
---|
736 | cldlayphase(i,ncol,1,1) = 1._wp |
---|
737 | endif |
---|
738 | endif |
---|
739 | endif ! end of discrimination condition |
---|
740 | endif ! end of cloud condition |
---|
741 | enddo ! end of altitude loop |
---|
742 | |
---|
743 | ! ############################################################################## |
---|
744 | ! 4.2) For Cloudy pixels with 0km < z < 8.16km |
---|
745 | ! ############################################################################## |
---|
746 | toplvlsat = 0 |
---|
747 | do nlev=24,Nlevels! from 8.16km until 0km |
---|
748 | p1 = pplay(i,nlev) |
---|
749 | |
---|
750 | if((cldy(i,ncol,nlev) .eq. 1.) .and. (ATBperp(i,ncol,nlev) .gt. 0.) )then |
---|
751 | ! Computation of the ATBperp of the phase discrimination line |
---|
752 | ATBperp_tmp = (ATB(i,ncol,nlev)**5)*alpha50 + (ATB(i,ncol,nlev)**4)*beta50 + & |
---|
753 | (ATB(i,ncol,nlev)**3)*gamma50 + (ATB(i,ncol,nlev)**2)*delta50 + & |
---|
754 | ATB(i,ncol,nlev)*epsilon50 + zeta50 |
---|
755 | ! ######################################################################## |
---|
756 | ! 4.2.a) Ice: ATBperp above the phase discrimination line |
---|
757 | ! ######################################################################## |
---|
758 | ! ICE with temperature above 273,15°K = Liquid (false ice) |
---|
759 | if((ATBperp(i,ncol,nlev)-ATBperp_tmp) .ge. 0.)then ! Ice clouds |
---|
760 | if(tmp(i,nlev) .gt. 273.15)then |
---|
761 | lidarcldphase(i,nlev,2) = lidarcldphase(i,nlev,2)+1._wp ! false ice ==> liq |
---|
762 | tmpl(i,ncol,nlev) = tmp(i,nlev) |
---|
763 | lidarcldphase(i,nlev,5) = lidarcldphase(i,nlev,5)+1._wp |
---|
764 | cldlayphase(i,ncol,4,2) = 1._wp ! tot cloud |
---|
765 | ! High cloud |
---|
766 | if (p1 .gt. 0. .and. p1 .lt. (440._wp*100._wp)) then |
---|
767 | cldlayphase(i,ncol,3,2) = 1._wp |
---|
768 | ! Middle cloud |
---|
769 | else if(p1 .ge. (440._wp*100._wp) .and. p1 .lt. (680._wp*100._wp)) then |
---|
770 | cldlayphase(i,ncol,2,2) = 1._wp |
---|
771 | ! Low cloud |
---|
772 | else |
---|
773 | cldlayphase(i,ncol,1,2) = 1._wp |
---|
774 | endif |
---|
775 | |
---|
776 | cldlayphase(i,ncol,4,5) = 1. ! tot cloud |
---|
777 | ! High cloud |
---|
778 | if (p1 .gt. 0. .and. p1 .lt. (440._wp*100._wp)) then |
---|
779 | cldlayphase(i,ncol,3,5) = 1._wp |
---|
780 | ! Middle cloud |
---|
781 | else if(p1 .ge. (440._wp*100._wp) .and. p1 .lt. (680._wp*100._wp)) then |
---|
782 | cldlayphase(i,ncol,2,5) = 1._wp |
---|
783 | ! Low cloud |
---|
784 | else |
---|
785 | cldlayphase(i,ncol,1,5) = 1._wp |
---|
786 | endif |
---|
787 | else |
---|
788 | ! ICE with temperature below 273,15°K |
---|
789 | lidarcldphase(i,nlev,1) = lidarcldphase(i,nlev,1)+1._wp |
---|
790 | tmpi(i,ncol,nlev) = tmp(i,nlev) |
---|
791 | cldlayphase(i,ncol,4,1) = 1._wp ! tot cloud |
---|
792 | ! High cloud |
---|
793 | if (p1 .gt. 0. .and. p1 .lt. (440._wp*100._wp)) then |
---|
794 | cldlayphase(i,ncol,3,1) = 1._wp |
---|
795 | ! Middle cloud |
---|
796 | else if(p1 .ge. (440._wp*100._wp) .and. p1 .lt.(680._wp*100._wp)) then |
---|
797 | cldlayphase(i,ncol,2,1) = 1._wp |
---|
798 | ! Low cloud |
---|
799 | else |
---|
800 | cldlayphase(i,ncol,1,1) = 1._wp |
---|
801 | endif |
---|
802 | endif |
---|
803 | |
---|
804 | ! ######################################################################## |
---|
805 | ! 4.2.b) Liquid: ATBperp below the phase discrimination line |
---|
806 | ! ######################################################################## |
---|
807 | else |
---|
808 | ! Liquid with temperature above 231,15°K |
---|
809 | if(tmp(i,nlev) .gt. 231.15)then |
---|
810 | lidarcldphase(i,nlev,2) = lidarcldphase(i,nlev,2)+1._wp |
---|
811 | tmpl(i,ncol,nlev) = tmp(i,nlev) |
---|
812 | cldlayphase(i,ncol,4,2) = 1._wp ! tot cloud |
---|
813 | ! High cloud |
---|
814 | if (p1 .gt. 0. .and. p1 .lt. (440._wp*100._wp)) then |
---|
815 | cldlayphase(i,ncol,3,2) = 1._wp |
---|
816 | ! Middle cloud |
---|
817 | else if(p1 .ge. (440._wp*100._wp) .and. p1 .lt. (680._wp*100._wp)) then |
---|
818 | cldlayphase(i,ncol,2,2) = 1._wp |
---|
819 | ! Low cloud |
---|
820 | else |
---|
821 | cldlayphase(i,ncol,1,2) = 1._wp |
---|
822 | endif |
---|
823 | else |
---|
824 | ! Liquid with temperature below 231,15°K = Ice (false liquid) |
---|
825 | tmpi(i,ncol,nlev) = tmp(i,nlev) |
---|
826 | lidarcldphase(i,nlev,1) = lidarcldphase(i,nlev,1)+1._wp ! false liq ==> ice |
---|
827 | lidarcldphase(i,nlev,4) = lidarcldphase(i,nlev,4)+1._wp ! false liq ==> ice |
---|
828 | cldlayphase(i,ncol,4,4) = 1._wp ! tot cloud |
---|
829 | ! High cloud |
---|
830 | if (p1 .gt. 0. .and. p1 .lt. (440._wp*100._wp)) then |
---|
831 | cldlayphase(i,ncol,3,4) = 1._wp |
---|
832 | ! Middle |
---|
833 | else if(p1 .ge. (440._wp*100._wp) .and. p1 .lt. (680._wp*100._wp)) then |
---|
834 | cldlayphase(i,ncol,2,4) = 1._wp |
---|
835 | ! Low cloud |
---|
836 | else |
---|
837 | cldlayphase(i,ncol,1,4) = 1._wp |
---|
838 | endif |
---|
839 | |
---|
840 | cldlayphase(i,ncol,4,1) = 1._wp ! tot cloud |
---|
841 | ! High cloud |
---|
842 | if (p1 .gt. 0. .and. p1 .lt. (440._wp*100._wp)) then |
---|
843 | cldlayphase(i,ncol,3,1) = 1._wp |
---|
844 | ! Middle cloud |
---|
845 | else if(p1 .ge. (440._wp*100._wp) .and. p1 .lt. (680._wp*100._wp)) then |
---|
846 | cldlayphase(i,ncol,2,1) = 1._wp |
---|
847 | ! Low cloud |
---|
848 | else |
---|
849 | cldlayphase(i,ncol,1,1) = 1._wp |
---|
850 | endif |
---|
851 | endif |
---|
852 | endif ! end of discrimination condition |
---|
853 | |
---|
854 | toplvlsat=0 |
---|
855 | |
---|
856 | ! Find the level of the highest cloud with SR>30 |
---|
857 | if(x(i,ncol,nlev) .gt. S_cld_att) then ! SR > 30. |
---|
858 | toplvlsat = nlev+1 |
---|
859 | goto 99 |
---|
860 | endif |
---|
861 | endif ! end of cloud condition |
---|
862 | enddo ! end of altitude loop |
---|
863 | 99 continue |
---|
864 | |
---|
865 | ! ############################################################################## |
---|
866 | ! Undefined phase: For a cloud located below another cloud with SR>30 |
---|
867 | ! see Cesana and Chepfer 2013 Sect.III.2 |
---|
868 | ! ############################################################################## |
---|
869 | if(toplvlsat.ne.0) then |
---|
870 | do nlev = toplvlsat,Nlevels |
---|
871 | p1 = pplay(i,nlev) |
---|
872 | if(cldy(i,ncol,nlev).eq.1.)then |
---|
873 | lidarcldphase(i,nlev,3) = lidarcldphase(i,nlev,3)+1._wp |
---|
874 | tmpu(i,ncol,nlev) = tmp(i,nlev) |
---|
875 | cldlayphase(i,ncol,4,3) = 1._wp ! tot cloud |
---|
876 | ! High cloud |
---|
877 | if (p1 .gt. 0. .and. p1 .lt. (440._wp*100._wp)) then |
---|
878 | cldlayphase(i,ncol,3,3) = 1._wp |
---|
879 | ! Middle cloud |
---|
880 | else if(p1 .ge. (440._wp*100._wp) .and. p1 .lt. (680._wp*100._wp)) then |
---|
881 | cldlayphase(i,ncol,2,3) = 1._wp |
---|
882 | ! Low cloud |
---|
883 | else |
---|
884 | cldlayphase(i,ncol,1,3) = 1._wp |
---|
885 | endif |
---|
886 | endif |
---|
887 | enddo |
---|
888 | endif |
---|
889 | toplvlsat=0 |
---|
890 | enddo |
---|
891 | enddo |
---|
892 | |
---|
893 | ! #################################################################################### |
---|
894 | ! Computation of final cloud phase diagnosis |
---|
895 | ! #################################################################################### |
---|
896 | |
---|
897 | ! Compute the Ice percentage in cloud = ice/(ice+liq) as a function of the occurrences |
---|
898 | lidarcldphasetmp(:,:) = lidarcldphase(:,:,1)+lidarcldphase(:,:,2); |
---|
899 | WHERE (lidarcldphasetmp(:,:) .gt. 0.) |
---|
900 | lidarcldphase(:,:,6)=lidarcldphase(:,:,1)/lidarcldphasetmp(:,:) |
---|
901 | ELSEWHERE |
---|
902 | lidarcldphase(:,:,6) = undef |
---|
903 | ENDWHERE |
---|
904 | |
---|
905 | ! Compute Phase 3D Cloud Fraction |
---|
906 | !WHERE (nsub(:,Nlevels:1:-1) .gt. 0.0 ) |
---|
907 | WHERE (nsub(:,:) .gt. 0.0 ) |
---|
908 | lidarcldphase(:,:,1)=lidarcldphase(:,:,1)/nsub(:,:) |
---|
909 | lidarcldphase(:,:,2)=lidarcldphase(:,:,2)/nsub(:,:) |
---|
910 | lidarcldphase(:,:,3)=lidarcldphase(:,:,3)/nsub(:,:) |
---|
911 | lidarcldphase(:,:,4)=lidarcldphase(:,:,4)/nsub(:,:) |
---|
912 | lidarcldphase(:,:,5)=lidarcldphase(:,:,5)/nsub(:,:) |
---|
913 | ELSEWHERE |
---|
914 | lidarcldphase(:,:,1) = undef |
---|
915 | lidarcldphase(:,:,2) = undef |
---|
916 | lidarcldphase(:,:,3) = undef |
---|
917 | lidarcldphase(:,:,4) = undef |
---|
918 | lidarcldphase(:,:,5) = undef |
---|
919 | ENDWHERE |
---|
920 | |
---|
921 | ! Compute Phase low mid high cloud fractions |
---|
922 | do iz = 1, Ncat |
---|
923 | do i=1,Nphase-3 |
---|
924 | do ic = 1, Ncolumns |
---|
925 | cldlayerphase(:,iz,i) = cldlayerphase(:,iz,i) + cldlayphase(:,ic,iz,i) |
---|
926 | cldlayerphasesum(:,iz) = cldlayerphasesum(:,iz) + cldlayphase(:,ic,iz,i) |
---|
927 | enddo |
---|
928 | enddo |
---|
929 | enddo |
---|
930 | do iz = 1, Ncat |
---|
931 | do i=4,5 |
---|
932 | do ic = 1, Ncolumns |
---|
933 | cldlayerphase(:,iz,i) = cldlayerphase(:,iz,i) + cldlayphase(:,ic,iz,i) |
---|
934 | enddo |
---|
935 | enddo |
---|
936 | enddo |
---|
937 | |
---|
938 | ! Compute the Ice percentage in cloud = ice/(ice+liq) |
---|
939 | cldlayerphasetmp(:,:)=cldlayerphase(:,:,1)+cldlayerphase(:,:,2) |
---|
940 | WHERE (cldlayerphasetmp(:,:).gt. 0.) |
---|
941 | cldlayerphase(:,:,6)=cldlayerphase(:,:,1)/cldlayerphasetmp(:,:) |
---|
942 | ELSEWHERE |
---|
943 | cldlayerphase(:,:,6) = undef |
---|
944 | ENDWHERE |
---|
945 | |
---|
946 | do i=1,Nphase-1 |
---|
947 | WHERE ( cldlayerphasesum(:,:).gt.0.0 ) |
---|
948 | cldlayerphase(:,:,i) = (cldlayerphase(:,:,i)/cldlayerphasesum(:,:)) * cldlayer(:,:) |
---|
949 | ENDWHERE |
---|
950 | enddo |
---|
951 | |
---|
952 | do i=1,Npoints |
---|
953 | do iz=1,Ncat |
---|
954 | checkcldlayerphase=0. |
---|
955 | checkcldlayerphase2=0. |
---|
956 | if (cldlayerphasesum(i,iz) .gt. 0.0 )then |
---|
957 | do ic=1,Nphase-3 |
---|
958 | checkcldlayerphase = checkcldlayerphase+cldlayerphase(i,iz,ic) |
---|
959 | enddo |
---|
960 | checkcldlayerphase2 = cldlayer(i,iz)-checkcldlayerphase |
---|
961 | if((checkcldlayerphase2 .gt. 0.01) .or. (checkcldlayerphase2 .lt. -0.01) ) print *, checkcldlayerphase,cldlayer(i,iz) |
---|
962 | endif |
---|
963 | enddo |
---|
964 | enddo |
---|
965 | |
---|
966 | do i=1,Nphase-1 |
---|
967 | WHERE (nsublayer(:,:) .eq. 0.0) |
---|
968 | cldlayerphase(:,:,i) = undef |
---|
969 | ENDWHERE |
---|
970 | enddo |
---|
971 | |
---|
972 | ! Compute Phase 3D as a function of temperature |
---|
973 | do nlev=1,Nlevels |
---|
974 | do ncol=1,Ncolumns |
---|
975 | do i=1,Npoints |
---|
976 | do itemp=1,Ntemp |
---|
977 | if(tmpi(i,ncol,nlev).gt.0.)then |
---|
978 | if((tmpi(i,ncol,nlev) .ge. tempmod(itemp)) .and. (tmpi(i,ncol,nlev) .lt. tempmod(itemp+1)) )then |
---|
979 | lidarcldtemp(i,itemp,2)=lidarcldtemp(i,itemp,2)+1._wp |
---|
980 | endif |
---|
981 | elseif(tmpl(i,ncol,nlev) .gt. 0.)then |
---|
982 | if((tmpl(i,ncol,nlev) .ge. tempmod(itemp)) .and. (tmpl(i,ncol,nlev) .lt. tempmod(itemp+1)) )then |
---|
983 | lidarcldtemp(i,itemp,3)=lidarcldtemp(i,itemp,3)+1._wp |
---|
984 | endif |
---|
985 | elseif(tmpu(i,ncol,nlev) .gt. 0.)then |
---|
986 | if((tmpu(i,ncol,nlev) .ge. tempmod(itemp)) .and. (tmpu(i,ncol,nlev) .lt. tempmod(itemp+1)) )then |
---|
987 | lidarcldtemp(i,itemp,4)=lidarcldtemp(i,itemp,4)+1._wp |
---|
988 | endif |
---|
989 | endif |
---|
990 | enddo |
---|
991 | enddo |
---|
992 | enddo |
---|
993 | enddo |
---|
994 | |
---|
995 | ! Check temperature cloud fraction |
---|
996 | do i=1,Npoints |
---|
997 | do itemp=1,Ntemp |
---|
998 | checktemp=lidarcldtemp(i,itemp,2)+lidarcldtemp(i,itemp,3)+lidarcldtemp(i,itemp,4) |
---|
999 | !if(checktemp .NE. lidarcldtemp(i,itemp,1))then |
---|
1000 | ! print *, i,itemp |
---|
1001 | ! print *, lidarcldtemp(i,itemp,1:4) |
---|
1002 | !endif |
---|
1003 | |
---|
1004 | enddo |
---|
1005 | enddo |
---|
1006 | |
---|
1007 | ! Compute the Ice percentage in cloud = ice/(ice+liq) |
---|
1008 | sumlidarcldtemp(:,:)=lidarcldtemp(:,:,2)+lidarcldtemp(:,:,3) |
---|
1009 | WHERE(sumlidarcldtemp(:,:) .gt. 0.) |
---|
1010 | lidarcldtemp(:,:,5)=lidarcldtemp(:,:,2)/sumlidarcldtemp(:,:) |
---|
1011 | ELSEWHERE |
---|
1012 | lidarcldtemp(:,:,5)=undef |
---|
1013 | ENDWHERE |
---|
1014 | |
---|
1015 | do i=1,4 |
---|
1016 | WHERE(lidarcldtempind(:,:) .gt. 0.) |
---|
1017 | lidarcldtemp(:,:,i) = lidarcldtemp(:,:,i)/lidarcldtempind(:,:) |
---|
1018 | ELSEWHERE |
---|
1019 | lidarcldtemp(:,:,i) = undef |
---|
1020 | ENDWHERE |
---|
1021 | enddo |
---|
1022 | |
---|
1023 | RETURN |
---|
1024 | END SUBROUTINE COSP_CLDFRAC |
---|
1025 | |
---|
1026 | end module mod_lidar_simulator |
---|