1 | SUBROUTINE flow_statistics |
---|
2 | |
---|
3 | !--------------------------------------------------------------------------------! |
---|
4 | ! This file is part of PALM. |
---|
5 | ! |
---|
6 | ! PALM is free software: you can redistribute it and/or modify it under the terms |
---|
7 | ! of the GNU General Public License as published by the Free Software Foundation, |
---|
8 | ! either version 3 of the License, or (at your option) any later version. |
---|
9 | ! |
---|
10 | ! PALM is distributed in the hope that it will be useful, but WITHOUT ANY |
---|
11 | ! WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR |
---|
12 | ! A PARTICULAR PURPOSE. See the GNU General Public License for more details. |
---|
13 | ! |
---|
14 | ! You should have received a copy of the GNU General Public License along with |
---|
15 | ! PALM. If not, see <http://www.gnu.org/licenses/>. |
---|
16 | ! |
---|
17 | ! Copyright 1997-2012 Leibniz University Hannover |
---|
18 | !--------------------------------------------------------------------------------! |
---|
19 | ! |
---|
20 | ! Current revisions: |
---|
21 | ! ----------------- |
---|
22 | ! |
---|
23 | ! Former revisions: |
---|
24 | ! ----------------- |
---|
25 | ! $Id: flow_statistics.f90 1054 2012-11-13 17:30:09Z raasch $ |
---|
26 | ! |
---|
27 | ! 1053 2012-11-13 17:11:03Z hoffmann |
---|
28 | ! necessary additions for two-moment cloud physics scheme: |
---|
29 | ! +nr, qr, qc, prr |
---|
30 | ! |
---|
31 | ! 1036 2012-10-22 13:43:42Z raasch |
---|
32 | ! code put under GPL (PALM 3.9) |
---|
33 | ! |
---|
34 | ! 1007 2012-09-19 14:30:36Z franke |
---|
35 | ! Calculation of buoyancy flux for humidity in case of WS-scheme is now using |
---|
36 | ! turbulent fluxes of WS-scheme |
---|
37 | ! Bugfix: Calculation of subgridscale buoyancy flux for humidity and cloud |
---|
38 | ! droplets at nzb and nzt added |
---|
39 | ! |
---|
40 | ! 801 2012-01-10 17:30:36Z suehring |
---|
41 | ! Calculation of turbulent fluxes in advec_ws is now thread-safe. |
---|
42 | ! |
---|
43 | ! 743 2011-08-18 16:10:16Z suehring |
---|
44 | ! Calculation of turbulent fluxes with WS-scheme only for the whole model |
---|
45 | ! domain, not for user-defined subregions. |
---|
46 | ! |
---|
47 | ! 709 2011-03-30 09:31:40Z raasch |
---|
48 | ! formatting adjustments |
---|
49 | ! |
---|
50 | ! 699 2011-03-22 17:52:22Z suehring |
---|
51 | ! Bugfix in calculation of vertical velocity skewness. The added absolute value |
---|
52 | ! avoid negative values in the root. Negative values of w'w' can occur at the |
---|
53 | ! top or bottom of the model domain due to degrading the order of advection |
---|
54 | ! scheme. Furthermore the calculation will be the same for all advection |
---|
55 | ! schemes. |
---|
56 | ! |
---|
57 | ! 696 2011-03-18 07:03:49Z raasch |
---|
58 | ! Bugfix: Summation of Wicker-Skamarock scheme fluxes and variances for all |
---|
59 | ! threads |
---|
60 | ! |
---|
61 | ! 678 2011-02-02 14:31:56Z raasch |
---|
62 | ! Bugfix in calculation of the divergence of vertical flux of resolved scale |
---|
63 | ! energy, pressure fluctuations, and flux of pressure fluctuation itself |
---|
64 | ! |
---|
65 | ! 673 2011-01-18 16:19:48Z suehring |
---|
66 | ! Top bc for the horizontal velocity variances added for ocean runs. |
---|
67 | ! Setting the corresponding bottom bc moved to advec_ws. |
---|
68 | ! |
---|
69 | ! 667 2010-12-23 12:06:00Z suehring/gryschka |
---|
70 | ! When advection is computed with ws-scheme, turbulent fluxes are already |
---|
71 | ! computed in the respective advection routines and buffered in arrays |
---|
72 | ! sums_xx_ws_l(). This is due to a consistent treatment of statistics with the |
---|
73 | ! numerics and to avoid unphysical kinks near the surface. |
---|
74 | ! So some if requests has to be done to dicern between fluxes from ws-scheme |
---|
75 | ! other advection schemes. |
---|
76 | ! Furthermore the computation of z_i is only done if the heat flux exceeds a |
---|
77 | ! minimum value. This affects only simulations of a neutral boundary layer and |
---|
78 | ! is due to reasons of computations in the advection scheme. |
---|
79 | ! |
---|
80 | ! 624 2010-12-10 11:46:30Z heinze |
---|
81 | ! Calculation of q*2 added |
---|
82 | ! |
---|
83 | ! 622 2010-12-10 08:08:13Z raasch |
---|
84 | ! optional barriers included in order to speed up collective operations |
---|
85 | ! |
---|
86 | ! 388 2009-09-23 09:40:33Z raasch |
---|
87 | ! Vertical profiles of potential density and hydrostatic pressure are |
---|
88 | ! calculated. |
---|
89 | ! Added missing timeseries calculation of w"q"(0), moved timeseries q* to the |
---|
90 | ! end. |
---|
91 | ! Temperature gradient criterion for estimating the boundary layer height |
---|
92 | ! replaced by the gradient criterion of Sullivan et al. (1998). |
---|
93 | ! Output of messages replaced by message handling routine. |
---|
94 | ! |
---|
95 | ! 197 2008-09-16 15:29:03Z raasch |
---|
96 | ! Spline timeseries splptx etc. removed, timeseries w'u', w'v', w'q' (k=0) |
---|
97 | ! added, |
---|
98 | ! bugfix: divide sums(k,8) (e) and sums(k,34) (e*) by ngp_2dh_s_inner(k,sr) |
---|
99 | ! (like other scalars) |
---|
100 | ! |
---|
101 | ! 133 2007-11-20 10:10:53Z letzel |
---|
102 | ! Vertical profiles now based on nzb_s_inner; they are divided by |
---|
103 | ! ngp_2dh_s_inner (scalars, procucts of scalars) and ngp_2dh (staggered |
---|
104 | ! velocity components and their products, procucts of scalars and velocity |
---|
105 | ! components), respectively. |
---|
106 | ! |
---|
107 | ! 106 2007-08-16 14:30:26Z raasch |
---|
108 | ! Prescribed momentum fluxes at the top surface are used, |
---|
109 | ! profiles for w*p* and w"e are calculated |
---|
110 | ! |
---|
111 | ! 97 2007-06-21 08:23:15Z raasch |
---|
112 | ! Statistics for ocean version (salinity, density) added, |
---|
113 | ! calculation of z_i and Deardorff velocity scale adjusted to be used with |
---|
114 | ! the ocean version |
---|
115 | ! |
---|
116 | ! 87 2007-05-22 15:46:47Z raasch |
---|
117 | ! Two more arguments added to user_statistics, which is now also called for |
---|
118 | ! user-defined profiles, |
---|
119 | ! var_hom and var_sum renamed pr_palm |
---|
120 | ! |
---|
121 | ! 82 2007-04-16 15:40:52Z raasch |
---|
122 | ! Cpp-directive lcmuk changed to intel_openmp_bug |
---|
123 | ! |
---|
124 | ! 75 2007-03-22 09:54:05Z raasch |
---|
125 | ! Collection of time series quantities moved from routine flow_statistics to |
---|
126 | ! here, routine user_statistics is called for each statistic region, |
---|
127 | ! moisture renamed humidity |
---|
128 | ! |
---|
129 | ! 19 2007-02-23 04:53:48Z raasch |
---|
130 | ! fluxes at top modified (tswst, qswst) |
---|
131 | ! |
---|
132 | ! RCS Log replace by Id keyword, revision history cleaned up |
---|
133 | ! |
---|
134 | ! Revision 1.41 2006/08/04 14:37:50 raasch |
---|
135 | ! Error removed in non-parallel part (sums_l) |
---|
136 | ! |
---|
137 | ! Revision 1.1 1997/08/11 06:15:17 raasch |
---|
138 | ! Initial revision |
---|
139 | ! |
---|
140 | ! |
---|
141 | ! Description: |
---|
142 | ! ------------ |
---|
143 | ! Compute average profiles and further average flow quantities for the different |
---|
144 | ! user-defined (sub-)regions. The region indexed 0 is the total model domain. |
---|
145 | ! |
---|
146 | ! NOTE: For simplicity, nzb_s_inner and nzb_diff_s_inner are being used as a |
---|
147 | ! ---- lower vertical index for k-loops for all variables, although strictly |
---|
148 | ! speaking the k-loops would have to be split up according to the staggered |
---|
149 | ! grid. However, this implies no error since staggered velocity components are |
---|
150 | ! zero at the walls and inside buildings. |
---|
151 | !------------------------------------------------------------------------------! |
---|
152 | |
---|
153 | USE arrays_3d |
---|
154 | USE cloud_parameters |
---|
155 | USE control_parameters |
---|
156 | USE cpulog |
---|
157 | USE grid_variables |
---|
158 | USE indices |
---|
159 | USE interfaces |
---|
160 | USE pegrid |
---|
161 | USE statistics |
---|
162 | |
---|
163 | IMPLICIT NONE |
---|
164 | |
---|
165 | INTEGER :: i, j, k, omp_get_thread_num, sr, tn |
---|
166 | LOGICAL :: first |
---|
167 | REAL :: dptdz_threshold, height, pts, sums_l_eper, sums_l_etot, ust, & |
---|
168 | ust2, u2, vst, vst2, v2, w2, z_i(2) |
---|
169 | REAL :: dptdz(nzb+1:nzt+1) |
---|
170 | REAL :: sums_ll(nzb:nzt+1,2) |
---|
171 | |
---|
172 | CALL cpu_log( log_point(10), 'flow_statistics', 'start' ) |
---|
173 | |
---|
174 | ! |
---|
175 | !-- To be on the safe side, check whether flow_statistics has already been |
---|
176 | !-- called once after the current time step |
---|
177 | IF ( flow_statistics_called ) THEN |
---|
178 | |
---|
179 | message_string = 'flow_statistics is called two times within one ' // & |
---|
180 | 'timestep' |
---|
181 | CALL message( 'flow_statistics', 'PA0190', 1, 2, 0, 6, 0 ) |
---|
182 | |
---|
183 | ENDIF |
---|
184 | |
---|
185 | ! |
---|
186 | !-- Compute statistics for each (sub-)region |
---|
187 | DO sr = 0, statistic_regions |
---|
188 | |
---|
189 | ! |
---|
190 | !-- Initialize (local) summation array |
---|
191 | sums_l = 0.0 |
---|
192 | |
---|
193 | ! |
---|
194 | !-- Store sums that have been computed in other subroutines in summation |
---|
195 | !-- array |
---|
196 | sums_l(:,11,:) = sums_l_l(:,sr,:) ! mixing length from diffusivities |
---|
197 | !-- WARNING: next line still has to be adjusted for OpenMP |
---|
198 | sums_l(:,21,0) = sums_wsts_bc_l(:,sr) ! heat flux from advec_s_bc |
---|
199 | sums_l(nzb+9,pr_palm,0) = sums_divold_l(sr) ! old divergence from pres |
---|
200 | sums_l(nzb+10,pr_palm,0) = sums_divnew_l(sr) ! new divergence from pres |
---|
201 | |
---|
202 | ! |
---|
203 | !-- Copy the turbulent quantities, evaluated in the advection routines to |
---|
204 | !-- the local array sums_l() for further computations |
---|
205 | IF ( ws_scheme_mom .AND. sr == 0 ) THEN |
---|
206 | |
---|
207 | ! |
---|
208 | !-- According to the Neumann bc for the horizontal velocity components, |
---|
209 | !-- the corresponding fluxes has to satisfiy the same bc. |
---|
210 | IF ( ocean ) THEN |
---|
211 | sums_us2_ws_l(nzt+1,:) = sums_us2_ws_l(nzt,:) |
---|
212 | sums_vs2_ws_l(nzt+1,:) = sums_vs2_ws_l(nzt,:) |
---|
213 | ENDIF |
---|
214 | |
---|
215 | DO i = 0, threads_per_task-1 |
---|
216 | ! |
---|
217 | !-- Swap the turbulent quantities evaluated in advec_ws. |
---|
218 | sums_l(:,13,i) = sums_wsus_ws_l(:,i) ! w*u* |
---|
219 | sums_l(:,15,i) = sums_wsvs_ws_l(:,i) ! w*v* |
---|
220 | sums_l(:,30,i) = sums_us2_ws_l(:,i) ! u*2 |
---|
221 | sums_l(:,31,i) = sums_vs2_ws_l(:,i) ! v*2 |
---|
222 | sums_l(:,32,i) = sums_ws2_ws_l(:,i) ! w*2 |
---|
223 | sums_l(:,34,i) = sums_l(:,34,i) + 0.5 * & |
---|
224 | ( sums_us2_ws_l(:,i) + sums_vs2_ws_l(:,i) + & |
---|
225 | sums_ws2_ws_l(:,i) ) ! e* |
---|
226 | DO k = nzb, nzt |
---|
227 | sums_l(nzb+5,pr_palm,i) = sums_l(nzb+5,pr_palm,i) + 0.5 * ( & |
---|
228 | sums_us2_ws_l(k,i) + & |
---|
229 | sums_vs2_ws_l(k,i) + & |
---|
230 | sums_ws2_ws_l(k,i) ) |
---|
231 | ENDDO |
---|
232 | ENDDO |
---|
233 | |
---|
234 | ENDIF |
---|
235 | |
---|
236 | IF ( ws_scheme_sca .AND. sr == 0 ) THEN |
---|
237 | |
---|
238 | DO i = 0, threads_per_task-1 |
---|
239 | sums_l(:,17,i) = sums_wspts_ws_l(:,i) ! w*pt* from advec_s_ws |
---|
240 | IF ( ocean ) sums_l(:,66,i) = sums_wssas_ws_l(:,i) ! w*sa* |
---|
241 | IF ( humidity .OR. passive_scalar ) sums_l(:,49,i) = & |
---|
242 | sums_wsqs_ws_l(:,i) !w*q* |
---|
243 | ENDDO |
---|
244 | |
---|
245 | ENDIF |
---|
246 | ! |
---|
247 | !-- Horizontally averaged profiles of horizontal velocities and temperature. |
---|
248 | !-- They must have been computed before, because they are already required |
---|
249 | !-- for other horizontal averages. |
---|
250 | tn = 0 |
---|
251 | |
---|
252 | !$OMP PARALLEL PRIVATE( i, j, k, tn ) |
---|
253 | #if defined( __intel_openmp_bug ) |
---|
254 | tn = omp_get_thread_num() |
---|
255 | #else |
---|
256 | !$ tn = omp_get_thread_num() |
---|
257 | #endif |
---|
258 | |
---|
259 | !$OMP DO |
---|
260 | DO i = nxl, nxr |
---|
261 | DO j = nys, nyn |
---|
262 | DO k = nzb_s_inner(j,i), nzt+1 |
---|
263 | sums_l(k,1,tn) = sums_l(k,1,tn) + u(k,j,i) * rmask(j,i,sr) |
---|
264 | sums_l(k,2,tn) = sums_l(k,2,tn) + v(k,j,i) * rmask(j,i,sr) |
---|
265 | sums_l(k,4,tn) = sums_l(k,4,tn) + pt(k,j,i) * rmask(j,i,sr) |
---|
266 | ENDDO |
---|
267 | ENDDO |
---|
268 | ENDDO |
---|
269 | |
---|
270 | ! |
---|
271 | !-- Horizontally averaged profile of salinity |
---|
272 | IF ( ocean ) THEN |
---|
273 | !$OMP DO |
---|
274 | DO i = nxl, nxr |
---|
275 | DO j = nys, nyn |
---|
276 | DO k = nzb_s_inner(j,i), nzt+1 |
---|
277 | sums_l(k,23,tn) = sums_l(k,23,tn) + & |
---|
278 | sa(k,j,i) * rmask(j,i,sr) |
---|
279 | ENDDO |
---|
280 | ENDDO |
---|
281 | ENDDO |
---|
282 | ENDIF |
---|
283 | |
---|
284 | ! |
---|
285 | !-- Horizontally averaged profiles of virtual potential temperature, |
---|
286 | !-- total water content, specific humidity and liquid water potential |
---|
287 | !-- temperature |
---|
288 | IF ( humidity ) THEN |
---|
289 | !$OMP DO |
---|
290 | DO i = nxl, nxr |
---|
291 | DO j = nys, nyn |
---|
292 | DO k = nzb_s_inner(j,i), nzt+1 |
---|
293 | sums_l(k,44,tn) = sums_l(k,44,tn) + & |
---|
294 | vpt(k,j,i) * rmask(j,i,sr) |
---|
295 | sums_l(k,41,tn) = sums_l(k,41,tn) + & |
---|
296 | q(k,j,i) * rmask(j,i,sr) |
---|
297 | ENDDO |
---|
298 | ENDDO |
---|
299 | ENDDO |
---|
300 | IF ( cloud_physics ) THEN |
---|
301 | !$OMP DO |
---|
302 | DO i = nxl, nxr |
---|
303 | DO j = nys, nyn |
---|
304 | DO k = nzb_s_inner(j,i), nzt+1 |
---|
305 | sums_l(k,42,tn) = sums_l(k,42,tn) + & |
---|
306 | ( q(k,j,i) - ql(k,j,i) ) * rmask(j,i,sr) |
---|
307 | sums_l(k,43,tn) = sums_l(k,43,tn) + ( & |
---|
308 | pt(k,j,i) + l_d_cp*pt_d_t(k) * ql(k,j,i) & |
---|
309 | ) * rmask(j,i,sr) |
---|
310 | ENDDO |
---|
311 | ENDDO |
---|
312 | ENDDO |
---|
313 | ENDIF |
---|
314 | ENDIF |
---|
315 | |
---|
316 | ! |
---|
317 | !-- Horizontally averaged profiles of passive scalar |
---|
318 | IF ( passive_scalar ) THEN |
---|
319 | !$OMP DO |
---|
320 | DO i = nxl, nxr |
---|
321 | DO j = nys, nyn |
---|
322 | DO k = nzb_s_inner(j,i), nzt+1 |
---|
323 | sums_l(k,41,tn) = sums_l(k,41,tn) + q(k,j,i) * rmask(j,i,sr) |
---|
324 | ENDDO |
---|
325 | ENDDO |
---|
326 | ENDDO |
---|
327 | ENDIF |
---|
328 | !$OMP END PARALLEL |
---|
329 | ! |
---|
330 | !-- Summation of thread sums |
---|
331 | IF ( threads_per_task > 1 ) THEN |
---|
332 | DO i = 1, threads_per_task-1 |
---|
333 | sums_l(:,1,0) = sums_l(:,1,0) + sums_l(:,1,i) |
---|
334 | sums_l(:,2,0) = sums_l(:,2,0) + sums_l(:,2,i) |
---|
335 | sums_l(:,4,0) = sums_l(:,4,0) + sums_l(:,4,i) |
---|
336 | IF ( ocean ) THEN |
---|
337 | sums_l(:,23,0) = sums_l(:,23,0) + sums_l(:,23,i) |
---|
338 | ENDIF |
---|
339 | IF ( humidity ) THEN |
---|
340 | sums_l(:,41,0) = sums_l(:,41,0) + sums_l(:,41,i) |
---|
341 | sums_l(:,44,0) = sums_l(:,44,0) + sums_l(:,44,i) |
---|
342 | IF ( cloud_physics ) THEN |
---|
343 | sums_l(:,42,0) = sums_l(:,42,0) + sums_l(:,42,i) |
---|
344 | sums_l(:,43,0) = sums_l(:,43,0) + sums_l(:,43,i) |
---|
345 | ENDIF |
---|
346 | ENDIF |
---|
347 | IF ( passive_scalar ) THEN |
---|
348 | sums_l(:,41,0) = sums_l(:,41,0) + sums_l(:,41,i) |
---|
349 | ENDIF |
---|
350 | ENDDO |
---|
351 | ENDIF |
---|
352 | |
---|
353 | #if defined( __parallel ) |
---|
354 | ! |
---|
355 | !-- Compute total sum from local sums |
---|
356 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
357 | CALL MPI_ALLREDUCE( sums_l(nzb,1,0), sums(nzb,1), nzt+2-nzb, MPI_REAL, & |
---|
358 | MPI_SUM, comm2d, ierr ) |
---|
359 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
360 | CALL MPI_ALLREDUCE( sums_l(nzb,2,0), sums(nzb,2), nzt+2-nzb, MPI_REAL, & |
---|
361 | MPI_SUM, comm2d, ierr ) |
---|
362 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
363 | CALL MPI_ALLREDUCE( sums_l(nzb,4,0), sums(nzb,4), nzt+2-nzb, MPI_REAL, & |
---|
364 | MPI_SUM, comm2d, ierr ) |
---|
365 | IF ( ocean ) THEN |
---|
366 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
367 | CALL MPI_ALLREDUCE( sums_l(nzb,23,0), sums(nzb,23), nzt+2-nzb, & |
---|
368 | MPI_REAL, MPI_SUM, comm2d, ierr ) |
---|
369 | ENDIF |
---|
370 | IF ( humidity ) THEN |
---|
371 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
372 | CALL MPI_ALLREDUCE( sums_l(nzb,44,0), sums(nzb,44), nzt+2-nzb, & |
---|
373 | MPI_REAL, MPI_SUM, comm2d, ierr ) |
---|
374 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
375 | CALL MPI_ALLREDUCE( sums_l(nzb,41,0), sums(nzb,41), nzt+2-nzb, & |
---|
376 | MPI_REAL, MPI_SUM, comm2d, ierr ) |
---|
377 | IF ( cloud_physics ) THEN |
---|
378 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
379 | CALL MPI_ALLREDUCE( sums_l(nzb,42,0), sums(nzb,42), nzt+2-nzb, & |
---|
380 | MPI_REAL, MPI_SUM, comm2d, ierr ) |
---|
381 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
382 | CALL MPI_ALLREDUCE( sums_l(nzb,43,0), sums(nzb,43), nzt+2-nzb, & |
---|
383 | MPI_REAL, MPI_SUM, comm2d, ierr ) |
---|
384 | ENDIF |
---|
385 | ENDIF |
---|
386 | |
---|
387 | IF ( passive_scalar ) THEN |
---|
388 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
389 | CALL MPI_ALLREDUCE( sums_l(nzb,41,0), sums(nzb,41), nzt+2-nzb, & |
---|
390 | MPI_REAL, MPI_SUM, comm2d, ierr ) |
---|
391 | ENDIF |
---|
392 | #else |
---|
393 | sums(:,1) = sums_l(:,1,0) |
---|
394 | sums(:,2) = sums_l(:,2,0) |
---|
395 | sums(:,4) = sums_l(:,4,0) |
---|
396 | IF ( ocean ) sums(:,23) = sums_l(:,23,0) |
---|
397 | IF ( humidity ) THEN |
---|
398 | sums(:,44) = sums_l(:,44,0) |
---|
399 | sums(:,41) = sums_l(:,41,0) |
---|
400 | IF ( cloud_physics ) THEN |
---|
401 | sums(:,42) = sums_l(:,42,0) |
---|
402 | sums(:,43) = sums_l(:,43,0) |
---|
403 | ENDIF |
---|
404 | ENDIF |
---|
405 | IF ( passive_scalar ) sums(:,41) = sums_l(:,41,0) |
---|
406 | #endif |
---|
407 | |
---|
408 | ! |
---|
409 | !-- Final values are obtained by division by the total number of grid points |
---|
410 | !-- used for summation. After that store profiles. |
---|
411 | sums(:,1) = sums(:,1) / ngp_2dh(sr) |
---|
412 | sums(:,2) = sums(:,2) / ngp_2dh(sr) |
---|
413 | sums(:,4) = sums(:,4) / ngp_2dh_s_inner(:,sr) |
---|
414 | hom(:,1,1,sr) = sums(:,1) ! u |
---|
415 | hom(:,1,2,sr) = sums(:,2) ! v |
---|
416 | hom(:,1,4,sr) = sums(:,4) ! pt |
---|
417 | |
---|
418 | |
---|
419 | ! |
---|
420 | !-- Salinity |
---|
421 | IF ( ocean ) THEN |
---|
422 | sums(:,23) = sums(:,23) / ngp_2dh_s_inner(:,sr) |
---|
423 | hom(:,1,23,sr) = sums(:,23) ! sa |
---|
424 | ENDIF |
---|
425 | |
---|
426 | ! |
---|
427 | !-- Humidity and cloud parameters |
---|
428 | IF ( humidity ) THEN |
---|
429 | sums(:,44) = sums(:,44) / ngp_2dh_s_inner(:,sr) |
---|
430 | sums(:,41) = sums(:,41) / ngp_2dh_s_inner(:,sr) |
---|
431 | hom(:,1,44,sr) = sums(:,44) ! vpt |
---|
432 | hom(:,1,41,sr) = sums(:,41) ! qv (q) |
---|
433 | IF ( cloud_physics ) THEN |
---|
434 | sums(:,42) = sums(:,42) / ngp_2dh_s_inner(:,sr) |
---|
435 | sums(:,43) = sums(:,43) / ngp_2dh_s_inner(:,sr) |
---|
436 | hom(:,1,42,sr) = sums(:,42) ! qv |
---|
437 | hom(:,1,43,sr) = sums(:,43) ! pt |
---|
438 | ENDIF |
---|
439 | ENDIF |
---|
440 | |
---|
441 | ! |
---|
442 | !-- Passive scalar |
---|
443 | IF ( passive_scalar ) hom(:,1,41,sr) = sums(:,41) / & |
---|
444 | ngp_2dh_s_inner(:,sr) ! s (q) |
---|
445 | |
---|
446 | ! |
---|
447 | !-- Horizontally averaged profiles of the remaining prognostic variables, |
---|
448 | !-- variances, the total and the perturbation energy (single values in last |
---|
449 | !-- column of sums_l) and some diagnostic quantities. |
---|
450 | !-- NOTE: for simplicity, nzb_s_inner is used below, although strictly |
---|
451 | !-- ---- speaking the following k-loop would have to be split up and |
---|
452 | !-- rearranged according to the staggered grid. |
---|
453 | !-- However, this implies no error since staggered velocity components |
---|
454 | !-- are zero at the walls and inside buildings. |
---|
455 | tn = 0 |
---|
456 | #if defined( __intel_openmp_bug ) |
---|
457 | !$OMP PARALLEL PRIVATE( i, j, k, pts, sums_ll, sums_l_eper, sums_l_etot, & |
---|
458 | !$OMP tn, ust, ust2, u2, vst, vst2, v2, w2 ) |
---|
459 | tn = omp_get_thread_num() |
---|
460 | #else |
---|
461 | !$OMP PARALLEL PRIVATE( i, j, k, pts, sums_ll, sums_l_eper, sums_l_etot, tn, ust, ust2, u2, vst, vst2, v2, w2 ) |
---|
462 | !$ tn = omp_get_thread_num() |
---|
463 | #endif |
---|
464 | !$OMP DO |
---|
465 | DO i = nxl, nxr |
---|
466 | DO j = nys, nyn |
---|
467 | sums_l_etot = 0.0 |
---|
468 | DO k = nzb_s_inner(j,i), nzt+1 |
---|
469 | ! |
---|
470 | !-- Prognostic and diagnostic variables |
---|
471 | sums_l(k,3,tn) = sums_l(k,3,tn) + w(k,j,i) * rmask(j,i,sr) |
---|
472 | sums_l(k,8,tn) = sums_l(k,8,tn) + e(k,j,i) * rmask(j,i,sr) |
---|
473 | sums_l(k,9,tn) = sums_l(k,9,tn) + km(k,j,i) * rmask(j,i,sr) |
---|
474 | sums_l(k,10,tn) = sums_l(k,10,tn) + kh(k,j,i) * rmask(j,i,sr) |
---|
475 | sums_l(k,40,tn) = sums_l(k,40,tn) + p(k,j,i) |
---|
476 | |
---|
477 | sums_l(k,33,tn) = sums_l(k,33,tn) + & |
---|
478 | ( pt(k,j,i)-hom(k,1,4,sr) )**2 * rmask(j,i,sr) |
---|
479 | |
---|
480 | IF ( humidity ) THEN |
---|
481 | sums_l(k,70,tn) = sums_l(k,70,tn) + & |
---|
482 | ( q(k,j,i)-hom(k,1,41,sr) )**2 * rmask(j,i,sr) |
---|
483 | ENDIF |
---|
484 | |
---|
485 | ! |
---|
486 | !-- Higher moments |
---|
487 | !-- (Computation of the skewness of w further below) |
---|
488 | sums_l(k,38,tn) = sums_l(k,38,tn) + w(k,j,i)**3 * rmask(j,i,sr) |
---|
489 | |
---|
490 | sums_l_etot = sums_l_etot + & |
---|
491 | 0.5 * ( u(k,j,i)**2 + v(k,j,i)**2 + & |
---|
492 | w(k,j,i)**2 ) * rmask(j,i,sr) |
---|
493 | ENDDO |
---|
494 | ! |
---|
495 | !-- Total and perturbation energy for the total domain (being |
---|
496 | !-- collected in the last column of sums_l). Summation of these |
---|
497 | !-- quantities is seperated from the previous loop in order to |
---|
498 | !-- allow vectorization of that loop. |
---|
499 | sums_l(nzb+4,pr_palm,tn) = sums_l(nzb+4,pr_palm,tn) + sums_l_etot |
---|
500 | ! |
---|
501 | !-- 2D-arrays (being collected in the last column of sums_l) |
---|
502 | sums_l(nzb,pr_palm,tn) = sums_l(nzb,pr_palm,tn) + & |
---|
503 | us(j,i) * rmask(j,i,sr) |
---|
504 | sums_l(nzb+1,pr_palm,tn) = sums_l(nzb+1,pr_palm,tn) + & |
---|
505 | usws(j,i) * rmask(j,i,sr) |
---|
506 | sums_l(nzb+2,pr_palm,tn) = sums_l(nzb+2,pr_palm,tn) + & |
---|
507 | vsws(j,i) * rmask(j,i,sr) |
---|
508 | sums_l(nzb+3,pr_palm,tn) = sums_l(nzb+3,pr_palm,tn) + & |
---|
509 | ts(j,i) * rmask(j,i,sr) |
---|
510 | IF ( humidity ) THEN |
---|
511 | sums_l(nzb+12,pr_palm,tn) = sums_l(nzb+12,pr_palm,tn) + & |
---|
512 | qs(j,i) * rmask(j,i,sr) |
---|
513 | ENDIF |
---|
514 | ENDDO |
---|
515 | ENDDO |
---|
516 | |
---|
517 | ! |
---|
518 | !-- Computation of statistics when ws-scheme is not used. Else these |
---|
519 | !-- quantities are evaluated in the advection routines. |
---|
520 | IF ( .NOT. ws_scheme_mom .OR. sr /= 0 ) THEN |
---|
521 | !$OMP DO |
---|
522 | DO i = nxl, nxr |
---|
523 | DO j = nys, nyn |
---|
524 | sums_l_eper = 0.0 |
---|
525 | DO k = nzb_s_inner(j,i), nzt+1 |
---|
526 | u2 = u(k,j,i)**2 |
---|
527 | v2 = v(k,j,i)**2 |
---|
528 | w2 = w(k,j,i)**2 |
---|
529 | ust2 = ( u(k,j,i) - hom(k,1,1,sr) )**2 |
---|
530 | vst2 = ( v(k,j,i) - hom(k,1,2,sr) )**2 |
---|
531 | |
---|
532 | sums_l(k,30,tn) = sums_l(k,30,tn) + ust2 * rmask(j,i,sr) |
---|
533 | sums_l(k,31,tn) = sums_l(k,31,tn) + vst2 * rmask(j,i,sr) |
---|
534 | sums_l(k,32,tn) = sums_l(k,32,tn) + w2 * rmask(j,i,sr) |
---|
535 | ! |
---|
536 | !-- Perturbation energy |
---|
537 | |
---|
538 | sums_l(k,34,tn) = sums_l(k,34,tn) + 0.5 * & |
---|
539 | ( ust2 + vst2 + w2 ) * rmask(j,i,sr) |
---|
540 | sums_l_eper = sums_l_eper + & |
---|
541 | 0.5 * ( ust2+vst2+w2 ) * rmask(j,i,sr) |
---|
542 | |
---|
543 | ENDDO |
---|
544 | sums_l(nzb+5,pr_palm,tn) = sums_l(nzb+5,pr_palm,tn) & |
---|
545 | + sums_l_eper |
---|
546 | ENDDO |
---|
547 | ENDDO |
---|
548 | ENDIF |
---|
549 | ! |
---|
550 | !-- Horizontally averaged profiles of the vertical fluxes |
---|
551 | |
---|
552 | !$OMP DO |
---|
553 | DO i = nxl, nxr |
---|
554 | DO j = nys, nyn |
---|
555 | ! |
---|
556 | !-- Subgridscale fluxes (without Prandtl layer from k=nzb, |
---|
557 | !-- oterwise from k=nzb+1) |
---|
558 | !-- NOTE: for simplicity, nzb_diff_s_inner is used below, although |
---|
559 | !-- ---- strictly speaking the following k-loop would have to be |
---|
560 | !-- split up according to the staggered grid. |
---|
561 | !-- However, this implies no error since staggered velocity |
---|
562 | !-- components are zero at the walls and inside buildings. |
---|
563 | |
---|
564 | DO k = nzb_diff_s_inner(j,i)-1, nzt_diff |
---|
565 | ! |
---|
566 | !-- Momentum flux w"u" |
---|
567 | sums_l(k,12,tn) = sums_l(k,12,tn) - 0.25 * ( & |
---|
568 | km(k,j,i)+km(k+1,j,i)+km(k,j,i-1)+km(k+1,j,i-1) & |
---|
569 | ) * ( & |
---|
570 | ( u(k+1,j,i) - u(k,j,i) ) * ddzu(k+1) & |
---|
571 | + ( w(k,j,i) - w(k,j,i-1) ) * ddx & |
---|
572 | ) * rmask(j,i,sr) |
---|
573 | ! |
---|
574 | !-- Momentum flux w"v" |
---|
575 | sums_l(k,14,tn) = sums_l(k,14,tn) - 0.25 * ( & |
---|
576 | km(k,j,i)+km(k+1,j,i)+km(k,j-1,i)+km(k+1,j-1,i) & |
---|
577 | ) * ( & |
---|
578 | ( v(k+1,j,i) - v(k,j,i) ) * ddzu(k+1) & |
---|
579 | + ( w(k,j,i) - w(k,j-1,i) ) * ddy & |
---|
580 | ) * rmask(j,i,sr) |
---|
581 | ! |
---|
582 | !-- Heat flux w"pt" |
---|
583 | sums_l(k,16,tn) = sums_l(k,16,tn) & |
---|
584 | - 0.5 * ( kh(k,j,i) + kh(k+1,j,i) ) & |
---|
585 | * ( pt(k+1,j,i) - pt(k,j,i) ) & |
---|
586 | * ddzu(k+1) * rmask(j,i,sr) |
---|
587 | |
---|
588 | |
---|
589 | ! |
---|
590 | !-- Salinity flux w"sa" |
---|
591 | IF ( ocean ) THEN |
---|
592 | sums_l(k,65,tn) = sums_l(k,65,tn) & |
---|
593 | - 0.5 * ( kh(k,j,i) + kh(k+1,j,i) ) & |
---|
594 | * ( sa(k+1,j,i) - sa(k,j,i) ) & |
---|
595 | * ddzu(k+1) * rmask(j,i,sr) |
---|
596 | ENDIF |
---|
597 | |
---|
598 | ! |
---|
599 | !-- Buoyancy flux, water flux (humidity flux) w"q" |
---|
600 | IF ( humidity ) THEN |
---|
601 | sums_l(k,45,tn) = sums_l(k,45,tn) & |
---|
602 | - 0.5 * ( kh(k,j,i) + kh(k+1,j,i) ) & |
---|
603 | * ( vpt(k+1,j,i) - vpt(k,j,i) ) & |
---|
604 | * ddzu(k+1) * rmask(j,i,sr) |
---|
605 | sums_l(k,48,tn) = sums_l(k,48,tn) & |
---|
606 | - 0.5 * ( kh(k,j,i) + kh(k+1,j,i) ) & |
---|
607 | * ( q(k+1,j,i) - q(k,j,i) ) & |
---|
608 | * ddzu(k+1) * rmask(j,i,sr) |
---|
609 | |
---|
610 | IF ( cloud_physics ) THEN |
---|
611 | sums_l(k,51,tn) = sums_l(k,51,tn) & |
---|
612 | - 0.5 * ( kh(k,j,i) + kh(k+1,j,i) ) & |
---|
613 | * ( ( q(k+1,j,i) - ql(k+1,j,i) )& |
---|
614 | - ( q(k,j,i) - ql(k,j,i) ) ) & |
---|
615 | * ddzu(k+1) * rmask(j,i,sr) |
---|
616 | ENDIF |
---|
617 | ENDIF |
---|
618 | |
---|
619 | ! |
---|
620 | !-- Passive scalar flux |
---|
621 | IF ( passive_scalar ) THEN |
---|
622 | sums_l(k,48,tn) = sums_l(k,48,tn) & |
---|
623 | - 0.5 * ( kh(k,j,i) + kh(k+1,j,i) ) & |
---|
624 | * ( q(k+1,j,i) - q(k,j,i) ) & |
---|
625 | * ddzu(k+1) * rmask(j,i,sr) |
---|
626 | ENDIF |
---|
627 | |
---|
628 | ENDDO |
---|
629 | |
---|
630 | ! |
---|
631 | !-- Subgridscale fluxes in the Prandtl layer |
---|
632 | IF ( use_surface_fluxes ) THEN |
---|
633 | sums_l(nzb,12,tn) = sums_l(nzb,12,tn) + & |
---|
634 | usws(j,i) * rmask(j,i,sr) ! w"u" |
---|
635 | sums_l(nzb,14,tn) = sums_l(nzb,14,tn) + & |
---|
636 | vsws(j,i) * rmask(j,i,sr) ! w"v" |
---|
637 | sums_l(nzb,16,tn) = sums_l(nzb,16,tn) + & |
---|
638 | shf(j,i) * rmask(j,i,sr) ! w"pt" |
---|
639 | sums_l(nzb,58,tn) = sums_l(nzb,58,tn) + & |
---|
640 | 0.0 * rmask(j,i,sr) ! u"pt" |
---|
641 | sums_l(nzb,61,tn) = sums_l(nzb,61,tn) + & |
---|
642 | 0.0 * rmask(j,i,sr) ! v"pt" |
---|
643 | IF ( ocean ) THEN |
---|
644 | sums_l(nzb,65,tn) = sums_l(nzb,65,tn) + & |
---|
645 | saswsb(j,i) * rmask(j,i,sr) ! w"sa" |
---|
646 | ENDIF |
---|
647 | IF ( humidity ) THEN |
---|
648 | sums_l(nzb,48,tn) = sums_l(nzb,48,tn) + & |
---|
649 | qsws(j,i) * rmask(j,i,sr) ! w"q" (w"qv") |
---|
650 | sums_l(nzb,45,tn) = sums_l(nzb,45,tn) + ( & |
---|
651 | ( 1.0 + 0.61 * q(nzb,j,i) ) * & |
---|
652 | shf(j,i) + 0.61 * pt(nzb,j,i) * & |
---|
653 | qsws(j,i) ) |
---|
654 | IF ( cloud_droplets ) THEN |
---|
655 | sums_l(nzb,45,tn) = sums_l(nzb,45,tn) + ( & |
---|
656 | ( 1.0 + 0.61 * q(nzb,j,i) - & |
---|
657 | ql(nzb,j,i) ) * shf(j,i) + & |
---|
658 | 0.61 * pt(nzb,j,i) * qsws(j,i) ) |
---|
659 | ENDIF |
---|
660 | IF ( cloud_physics ) THEN |
---|
661 | ! |
---|
662 | !-- Formula does not work if ql(nzb) /= 0.0 |
---|
663 | sums_l(nzb,51,tn) = sums_l(nzb,51,tn) + & ! w"q" (w"qv") |
---|
664 | qsws(j,i) * rmask(j,i,sr) |
---|
665 | ENDIF |
---|
666 | ENDIF |
---|
667 | IF ( passive_scalar ) THEN |
---|
668 | sums_l(nzb,48,tn) = sums_l(nzb,48,tn) + & |
---|
669 | qsws(j,i) * rmask(j,i,sr) ! w"q" (w"qv") |
---|
670 | ENDIF |
---|
671 | ENDIF |
---|
672 | |
---|
673 | ! |
---|
674 | !-- Subgridscale fluxes at the top surface |
---|
675 | IF ( use_top_fluxes ) THEN |
---|
676 | sums_l(nzt:nzt+1,12,tn) = sums_l(nzt:nzt+1,12,tn) + & |
---|
677 | uswst(j,i) * rmask(j,i,sr) ! w"u" |
---|
678 | sums_l(nzt:nzt+1,14,tn) = sums_l(nzt:nzt+1,14,tn) + & |
---|
679 | vswst(j,i) * rmask(j,i,sr) ! w"v" |
---|
680 | sums_l(nzt:nzt+1,16,tn) = sums_l(nzt:nzt+1,16,tn) + & |
---|
681 | tswst(j,i) * rmask(j,i,sr) ! w"pt" |
---|
682 | sums_l(nzt:nzt+1,58,tn) = sums_l(nzt:nzt+1,58,tn) + & |
---|
683 | 0.0 * rmask(j,i,sr) ! u"pt" |
---|
684 | sums_l(nzt:nzt+1,61,tn) = sums_l(nzt:nzt+1,61,tn) + & |
---|
685 | 0.0 * rmask(j,i,sr) ! v"pt" |
---|
686 | |
---|
687 | IF ( ocean ) THEN |
---|
688 | sums_l(nzt,65,tn) = sums_l(nzt,65,tn) + & |
---|
689 | saswst(j,i) * rmask(j,i,sr) ! w"sa" |
---|
690 | ENDIF |
---|
691 | IF ( humidity ) THEN |
---|
692 | sums_l(nzt,48,tn) = sums_l(nzt,48,tn) + & |
---|
693 | qswst(j,i) * rmask(j,i,sr) ! w"q" (w"qv") |
---|
694 | sums_l(nzt,45,tn) = sums_l(nzt,45,tn) + ( & |
---|
695 | ( 1.0 + 0.61 * q(nzt,j,i) ) * & |
---|
696 | tswst(j,i) + 0.61 * pt(nzt,j,i) * & |
---|
697 | qswst(j,i) ) |
---|
698 | IF ( cloud_droplets ) THEN |
---|
699 | sums_l(nzt,45,tn) = sums_l(nzt,45,tn) + ( & |
---|
700 | ( 1.0 + 0.61 * q(nzt,j,i) - & |
---|
701 | ql(nzt,j,i) ) * tswst(j,i) + & |
---|
702 | 0.61 * pt(nzt,j,i) * qswst(j,i) ) |
---|
703 | ENDIF |
---|
704 | IF ( cloud_physics ) THEN |
---|
705 | ! |
---|
706 | !-- Formula does not work if ql(nzb) /= 0.0 |
---|
707 | sums_l(nzt,51,tn) = sums_l(nzt,51,tn) + & ! w"q" (w"qv") |
---|
708 | qswst(j,i) * rmask(j,i,sr) |
---|
709 | ENDIF |
---|
710 | ENDIF |
---|
711 | IF ( passive_scalar ) THEN |
---|
712 | sums_l(nzt,48,tn) = sums_l(nzt,48,tn) + & |
---|
713 | qswst(j,i) * rmask(j,i,sr) ! w"q" (w"qv") |
---|
714 | ENDIF |
---|
715 | ENDIF |
---|
716 | |
---|
717 | ! |
---|
718 | !-- Resolved fluxes (can be computed for all horizontal points) |
---|
719 | !-- NOTE: for simplicity, nzb_s_inner is used below, although strictly |
---|
720 | !-- ---- speaking the following k-loop would have to be split up and |
---|
721 | !-- rearranged according to the staggered grid. |
---|
722 | DO k = nzb_s_inner(j,i), nzt |
---|
723 | ust = 0.5 * ( u(k,j,i) - hom(k,1,1,sr) + & |
---|
724 | u(k+1,j,i) - hom(k+1,1,1,sr) ) |
---|
725 | vst = 0.5 * ( v(k,j,i) - hom(k,1,2,sr) + & |
---|
726 | v(k+1,j,i) - hom(k+1,1,2,sr) ) |
---|
727 | pts = 0.5 * ( pt(k,j,i) - hom(k,1,4,sr) + & |
---|
728 | pt(k+1,j,i) - hom(k+1,1,4,sr) ) |
---|
729 | |
---|
730 | !-- Higher moments |
---|
731 | sums_l(k,35,tn) = sums_l(k,35,tn) + pts * w(k,j,i)**2 * & |
---|
732 | rmask(j,i,sr) |
---|
733 | sums_l(k,36,tn) = sums_l(k,36,tn) + pts**2 * w(k,j,i) * & |
---|
734 | rmask(j,i,sr) |
---|
735 | |
---|
736 | ! |
---|
737 | !-- Salinity flux and density (density does not belong to here, |
---|
738 | !-- but so far there is no other suitable place to calculate) |
---|
739 | IF ( ocean ) THEN |
---|
740 | IF( .NOT. ws_scheme_sca .OR. sr /= 0 ) THEN |
---|
741 | pts = 0.5 * ( sa(k,j,i) - hom(k,1,23,sr) + & |
---|
742 | sa(k+1,j,i) - hom(k+1,1,23,sr) ) |
---|
743 | sums_l(k,66,tn) = sums_l(k,66,tn) + pts * w(k,j,i) * & |
---|
744 | rmask(j,i,sr) |
---|
745 | ENDIF |
---|
746 | sums_l(k,64,tn) = sums_l(k,64,tn) + rho(k,j,i) * & |
---|
747 | rmask(j,i,sr) |
---|
748 | sums_l(k,71,tn) = sums_l(k,71,tn) + prho(k,j,i) * & |
---|
749 | rmask(j,i,sr) |
---|
750 | ENDIF |
---|
751 | |
---|
752 | ! |
---|
753 | !-- Buoyancy flux, water flux, humidity flux, liquid water |
---|
754 | !-- content, rain drop concentration and rain water content |
---|
755 | IF ( humidity ) THEN |
---|
756 | IF ( cloud_physics .OR. cloud_droplets ) THEN |
---|
757 | pts = 0.5 * ( vpt(k,j,i) - hom(k,1,44,sr) + & |
---|
758 | vpt(k+1,j,i) - hom(k+1,1,44,sr) ) |
---|
759 | sums_l(k,46,tn) = sums_l(k,46,tn) + pts * w(k,j,i) * & |
---|
760 | rmask(j,i,sr) |
---|
761 | IF ( .NOT. cloud_droplets ) THEN |
---|
762 | pts = 0.5 * & |
---|
763 | ( ( q(k,j,i) - ql(k,j,i) ) - & |
---|
764 | hom(k,1,42,sr) + & |
---|
765 | ( q(k+1,j,i) - ql(k+1,j,i) ) - & |
---|
766 | hom(k+1,1,42,sr) ) |
---|
767 | sums_l(k,52,tn) = sums_l(k,52,tn) + pts * w(k,j,i) * & |
---|
768 | rmask(j,i,sr) |
---|
769 | IF ( icloud_scheme == 0 ) THEN |
---|
770 | sums_l(k,54,tn) = sums_l(k,54,tn) + ( ql(k,j,i) + & |
---|
771 | qr(k,j,i) ) * & |
---|
772 | rmask(j,i,sr) |
---|
773 | sums_l(k,73,tn) = sums_l(k,73,tn) + nr(k,j,i) * & |
---|
774 | rmask(j,i,sr) |
---|
775 | sums_l(k,74,tn) = sums_l(k,74,tn) + qr(k,j,i) * & |
---|
776 | rmask(j,i,sr) |
---|
777 | sums_l(k,75,tn) = sums_l(k,75,tn) + ql(k,j,i) * & |
---|
778 | rmask(j,i,sr) |
---|
779 | sums_l(k,76,tn) = sums_l(k,76,tn) + prr(k,j,i) * & |
---|
780 | rmask(j,i,sr) |
---|
781 | ELSE |
---|
782 | sums_l(k,54,tn) = sums_l(k,54,tn) + ql(k,j,i) * & |
---|
783 | rmask(j,i,sr) |
---|
784 | ENDIF |
---|
785 | ELSE |
---|
786 | sums_l(k,54,tn) = sums_l(k,54,tn) + ql(k,j,i) * & |
---|
787 | rmask(j,i,sr) |
---|
788 | ENDIF |
---|
789 | ELSE |
---|
790 | IF( .NOT. ws_scheme_sca .OR. sr /= 0 ) THEN |
---|
791 | pts = 0.5 * ( vpt(k,j,i) - hom(k,1,44,sr) + & |
---|
792 | vpt(k+1,j,i) - hom(k+1,1,44,sr) ) |
---|
793 | sums_l(k,46,tn) = sums_l(k,46,tn) + pts * w(k,j,i) * & |
---|
794 | rmask(j,i,sr) |
---|
795 | ELSE IF ( ws_scheme_sca .AND. sr == 0 ) THEN |
---|
796 | sums_l(k,46,tn) = ( 1.0 + 0.61 * hom(k,1,41,sr) ) * & |
---|
797 | sums_l(k,17,tn) + & |
---|
798 | 0.61 * hom(k,1,4,sr) * sums_l(k,49,tn) |
---|
799 | END IF |
---|
800 | END IF |
---|
801 | ENDIF |
---|
802 | ! |
---|
803 | !-- Passive scalar flux |
---|
804 | IF ( passive_scalar .AND. ( .NOT. ws_scheme_sca & |
---|
805 | .OR. sr /= 0 ) ) THEN |
---|
806 | pts = 0.5 * ( q(k,j,i) - hom(k,1,41,sr) + & |
---|
807 | q(k+1,j,i) - hom(k+1,1,41,sr) ) |
---|
808 | sums_l(k,49,tn) = sums_l(k,49,tn) + pts * w(k,j,i) * & |
---|
809 | rmask(j,i,sr) |
---|
810 | ENDIF |
---|
811 | |
---|
812 | ! |
---|
813 | !-- Energy flux w*e* |
---|
814 | !-- has to be adjusted |
---|
815 | sums_l(k,37,tn) = sums_l(k,37,tn) + w(k,j,i) * 0.5 * & |
---|
816 | ( ust**2 + vst**2 + w(k,j,i)**2 )& |
---|
817 | * rmask(j,i,sr) |
---|
818 | ENDDO |
---|
819 | ENDDO |
---|
820 | ENDDO |
---|
821 | ! |
---|
822 | !-- For speed optimization fluxes which have been computed in part directly |
---|
823 | !-- inside the WS advection routines are treated seperatly |
---|
824 | !-- Momentum fluxes first: |
---|
825 | IF ( .NOT. ws_scheme_mom .OR. sr /= 0 ) THEN |
---|
826 | !$OMP DO |
---|
827 | DO i = nxl, nxr |
---|
828 | DO j = nys, nyn |
---|
829 | DO k = nzb_diff_s_inner(j,i)-1, nzt_diff |
---|
830 | ust = 0.5 * ( u(k,j,i) - hom(k,1,1,sr) + & |
---|
831 | u(k+1,j,i) - hom(k+1,1,1,sr) ) |
---|
832 | vst = 0.5 * ( v(k,j,i) - hom(k,1,2,sr) + & |
---|
833 | v(k+1,j,i) - hom(k+1,1,2,sr) ) |
---|
834 | ! |
---|
835 | !-- Momentum flux w*u* |
---|
836 | sums_l(k,13,tn) = sums_l(k,13,tn) + 0.5 * & |
---|
837 | ( w(k,j,i-1) + w(k,j,i) ) & |
---|
838 | * ust * rmask(j,i,sr) |
---|
839 | ! |
---|
840 | !-- Momentum flux w*v* |
---|
841 | sums_l(k,15,tn) = sums_l(k,15,tn) + 0.5 * & |
---|
842 | ( w(k,j-1,i) + w(k,j,i) ) & |
---|
843 | * vst * rmask(j,i,sr) |
---|
844 | ENDDO |
---|
845 | ENDDO |
---|
846 | ENDDO |
---|
847 | |
---|
848 | ENDIF |
---|
849 | IF ( .NOT. ws_scheme_sca .OR. sr /= 0 ) THEN |
---|
850 | !$OMP DO |
---|
851 | DO i = nxl, nxr |
---|
852 | DO j = nys, nyn |
---|
853 | DO k = nzb_diff_s_inner(j,i)-1, nzt_diff |
---|
854 | ! |
---|
855 | !-- Vertical heat flux |
---|
856 | sums_l(k,17,tn) = sums_l(k,17,tn) + 0.5 * & |
---|
857 | ( pt(k,j,i) - hom(k,1,4,sr) + & |
---|
858 | pt(k+1,j,i) - hom(k+1,1,4,sr) ) & |
---|
859 | * w(k,j,i) * rmask(j,i,sr) |
---|
860 | IF ( humidity ) THEN |
---|
861 | pts = 0.5 * ( q(k,j,i) - hom(k,1,41,sr) + & |
---|
862 | q(k+1,j,i) - hom(k+1,1,41,sr) ) |
---|
863 | sums_l(k,49,tn) = sums_l(k,49,tn) + pts * w(k,j,i) * & |
---|
864 | rmask(j,i,sr) |
---|
865 | ENDIF |
---|
866 | ENDDO |
---|
867 | ENDDO |
---|
868 | ENDDO |
---|
869 | |
---|
870 | ENDIF |
---|
871 | |
---|
872 | |
---|
873 | ! |
---|
874 | !-- Density at top follows Neumann condition |
---|
875 | IF ( ocean ) THEN |
---|
876 | sums_l(nzt+1,64,tn) = sums_l(nzt,64,tn) |
---|
877 | sums_l(nzt+1,71,tn) = sums_l(nzt,71,tn) |
---|
878 | ENDIF |
---|
879 | |
---|
880 | ! |
---|
881 | !-- Divergence of vertical flux of resolved scale energy and pressure |
---|
882 | !-- fluctuations as well as flux of pressure fluctuation itself (68). |
---|
883 | !-- First calculate the products, then the divergence. |
---|
884 | !-- Calculation is time consuming. Do it only, if profiles shall be plotted. |
---|
885 | IF ( hom(nzb+1,2,55,0) /= 0.0 .OR. hom(nzb+1,2,68,0) /= 0.0 ) THEN |
---|
886 | |
---|
887 | sums_ll = 0.0 ! local array |
---|
888 | |
---|
889 | !$OMP DO |
---|
890 | DO i = nxl, nxr |
---|
891 | DO j = nys, nyn |
---|
892 | DO k = nzb_s_inner(j,i)+1, nzt |
---|
893 | |
---|
894 | sums_ll(k,1) = sums_ll(k,1) + 0.5 * w(k,j,i) * ( & |
---|
895 | ( 0.25 * ( u(k,j,i)+u(k+1,j,i)+u(k,j,i+1)+u(k+1,j,i+1) & |
---|
896 | - 0.5 * ( hom(k,1,1,sr) + hom(k+1,1,1,sr) ) & |
---|
897 | ) )**2 & |
---|
898 | + ( 0.25 * ( v(k,j,i)+v(k+1,j,i)+v(k,j+1,i)+v(k+1,j+1,i) & |
---|
899 | - 0.5 * ( hom(k,1,2,sr) + hom(k+1,1,2,sr) ) & |
---|
900 | ) )**2 & |
---|
901 | + w(k,j,i)**2 ) |
---|
902 | |
---|
903 | sums_ll(k,2) = sums_ll(k,2) + 0.5 * w(k,j,i) & |
---|
904 | * ( p(k,j,i) + p(k+1,j,i) ) |
---|
905 | |
---|
906 | ENDDO |
---|
907 | ENDDO |
---|
908 | ENDDO |
---|
909 | sums_ll(0,1) = 0.0 ! because w is zero at the bottom |
---|
910 | sums_ll(nzt+1,1) = 0.0 |
---|
911 | sums_ll(0,2) = 0.0 |
---|
912 | sums_ll(nzt+1,2) = 0.0 |
---|
913 | |
---|
914 | DO k = nzb+1, nzt |
---|
915 | sums_l(k,55,tn) = ( sums_ll(k,1) - sums_ll(k-1,1) ) * ddzw(k) |
---|
916 | sums_l(k,56,tn) = ( sums_ll(k,2) - sums_ll(k-1,2) ) * ddzw(k) |
---|
917 | sums_l(k,68,tn) = sums_ll(k,2) |
---|
918 | ENDDO |
---|
919 | sums_l(nzb,55,tn) = sums_l(nzb+1,55,tn) |
---|
920 | sums_l(nzb,56,tn) = sums_l(nzb+1,56,tn) |
---|
921 | sums_l(nzb,68,tn) = 0.0 ! because w* = 0 at nzb |
---|
922 | |
---|
923 | ENDIF |
---|
924 | |
---|
925 | ! |
---|
926 | !-- Divergence of vertical flux of SGS TKE and the flux itself (69) |
---|
927 | IF ( hom(nzb+1,2,57,0) /= 0.0 .OR. hom(nzb+1,2,69,0) /= 0.0 ) THEN |
---|
928 | |
---|
929 | !$OMP DO |
---|
930 | DO i = nxl, nxr |
---|
931 | DO j = nys, nyn |
---|
932 | DO k = nzb_s_inner(j,i)+1, nzt |
---|
933 | |
---|
934 | sums_l(k,57,tn) = sums_l(k,57,tn) - 0.5 * ( & |
---|
935 | (km(k,j,i)+km(k+1,j,i)) * (e(k+1,j,i)-e(k,j,i)) * ddzu(k+1) & |
---|
936 | - (km(k-1,j,i)+km(k,j,i)) * (e(k,j,i)-e(k-1,j,i)) * ddzu(k) & |
---|
937 | ) * ddzw(k) |
---|
938 | |
---|
939 | sums_l(k,69,tn) = sums_l(k,69,tn) - 0.5 * ( & |
---|
940 | (km(k,j,i)+km(k+1,j,i)) * (e(k+1,j,i)-e(k,j,i)) * ddzu(k+1) & |
---|
941 | ) |
---|
942 | |
---|
943 | ENDDO |
---|
944 | ENDDO |
---|
945 | ENDDO |
---|
946 | sums_l(nzb,57,tn) = sums_l(nzb+1,57,tn) |
---|
947 | sums_l(nzb,69,tn) = sums_l(nzb+1,69,tn) |
---|
948 | |
---|
949 | ENDIF |
---|
950 | |
---|
951 | ! |
---|
952 | !-- Horizontal heat fluxes (subgrid, resolved, total). |
---|
953 | !-- Do it only, if profiles shall be plotted. |
---|
954 | IF ( hom(nzb+1,2,58,0) /= 0.0 ) THEN |
---|
955 | |
---|
956 | !$OMP DO |
---|
957 | DO i = nxl, nxr |
---|
958 | DO j = nys, nyn |
---|
959 | DO k = nzb_s_inner(j,i)+1, nzt |
---|
960 | ! |
---|
961 | !-- Subgrid horizontal heat fluxes u"pt", v"pt" |
---|
962 | sums_l(k,58,tn) = sums_l(k,58,tn) - 0.5 * & |
---|
963 | ( kh(k,j,i) + kh(k,j,i-1) ) & |
---|
964 | * ( pt(k,j,i-1) - pt(k,j,i) ) & |
---|
965 | * ddx * rmask(j,i,sr) |
---|
966 | sums_l(k,61,tn) = sums_l(k,61,tn) - 0.5 * & |
---|
967 | ( kh(k,j,i) + kh(k,j-1,i) ) & |
---|
968 | * ( pt(k,j-1,i) - pt(k,j,i) ) & |
---|
969 | * ddy * rmask(j,i,sr) |
---|
970 | ! |
---|
971 | !-- Resolved horizontal heat fluxes u*pt*, v*pt* |
---|
972 | sums_l(k,59,tn) = sums_l(k,59,tn) + & |
---|
973 | ( u(k,j,i) - hom(k,1,1,sr) ) & |
---|
974 | * 0.5 * ( pt(k,j,i-1) - hom(k,1,4,sr) + & |
---|
975 | pt(k,j,i) - hom(k,1,4,sr) ) |
---|
976 | pts = 0.5 * ( pt(k,j-1,i) - hom(k,1,4,sr) + & |
---|
977 | pt(k,j,i) - hom(k,1,4,sr) ) |
---|
978 | sums_l(k,62,tn) = sums_l(k,62,tn) + & |
---|
979 | ( v(k,j,i) - hom(k,1,2,sr) ) & |
---|
980 | * 0.5 * ( pt(k,j-1,i) - hom(k,1,4,sr) + & |
---|
981 | pt(k,j,i) - hom(k,1,4,sr) ) |
---|
982 | ENDDO |
---|
983 | ENDDO |
---|
984 | ENDDO |
---|
985 | ! |
---|
986 | !-- Fluxes at the surface must be zero (e.g. due to the Prandtl-layer) |
---|
987 | sums_l(nzb,58,tn) = 0.0 |
---|
988 | sums_l(nzb,59,tn) = 0.0 |
---|
989 | sums_l(nzb,60,tn) = 0.0 |
---|
990 | sums_l(nzb,61,tn) = 0.0 |
---|
991 | sums_l(nzb,62,tn) = 0.0 |
---|
992 | sums_l(nzb,63,tn) = 0.0 |
---|
993 | |
---|
994 | ENDIF |
---|
995 | |
---|
996 | ! |
---|
997 | !-- Calculate the user-defined profiles |
---|
998 | CALL user_statistics( 'profiles', sr, tn ) |
---|
999 | !$OMP END PARALLEL |
---|
1000 | |
---|
1001 | ! |
---|
1002 | !-- Summation of thread sums |
---|
1003 | IF ( threads_per_task > 1 ) THEN |
---|
1004 | DO i = 1, threads_per_task-1 |
---|
1005 | sums_l(:,3,0) = sums_l(:,3,0) + sums_l(:,3,i) |
---|
1006 | sums_l(:,4:40,0) = sums_l(:,4:40,0) + sums_l(:,4:40,i) |
---|
1007 | sums_l(:,45:pr_palm,0) = sums_l(:,45:pr_palm,0) + & |
---|
1008 | sums_l(:,45:pr_palm,i) |
---|
1009 | IF ( max_pr_user > 0 ) THEN |
---|
1010 | sums_l(:,pr_palm+1:pr_palm+max_pr_user,0) = & |
---|
1011 | sums_l(:,pr_palm+1:pr_palm+max_pr_user,0) + & |
---|
1012 | sums_l(:,pr_palm+1:pr_palm+max_pr_user,i) |
---|
1013 | ENDIF |
---|
1014 | ENDDO |
---|
1015 | ENDIF |
---|
1016 | |
---|
1017 | #if defined( __parallel ) |
---|
1018 | |
---|
1019 | ! |
---|
1020 | !-- Compute total sum from local sums |
---|
1021 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
---|
1022 | CALL MPI_ALLREDUCE( sums_l(nzb,1,0), sums(nzb,1), ngp_sums, MPI_REAL, & |
---|
1023 | MPI_SUM, comm2d, ierr ) |
---|
1024 | #else |
---|
1025 | sums = sums_l(:,:,0) |
---|
1026 | #endif |
---|
1027 | |
---|
1028 | ! |
---|
1029 | !-- Final values are obtained by division by the total number of grid points |
---|
1030 | !-- used for summation. After that store profiles. |
---|
1031 | !-- Profiles: |
---|
1032 | DO k = nzb, nzt+1 |
---|
1033 | sums(k,3) = sums(k,3) / ngp_2dh(sr) |
---|
1034 | sums(k,8:11) = sums(k,8:11) / ngp_2dh_s_inner(k,sr) |
---|
1035 | sums(k,12:22) = sums(k,12:22) / ngp_2dh(sr) |
---|
1036 | sums(k,23:29) = sums(k,23:29) / ngp_2dh_s_inner(k,sr) |
---|
1037 | sums(k,30:32) = sums(k,30:32) / ngp_2dh(sr) |
---|
1038 | sums(k,33:34) = sums(k,33:34) / ngp_2dh_s_inner(k,sr) |
---|
1039 | sums(k,35:39) = sums(k,35:39) / ngp_2dh(sr) |
---|
1040 | sums(k,40) = sums(k,40) / ngp_2dh_s_inner(k,sr) |
---|
1041 | sums(k,45:53) = sums(k,45:53) / ngp_2dh(sr) |
---|
1042 | sums(k,54) = sums(k,54) / ngp_2dh_s_inner(k,sr) |
---|
1043 | sums(k,55:63) = sums(k,55:63) / ngp_2dh(sr) |
---|
1044 | sums(k,64) = sums(k,64) / ngp_2dh_s_inner(k,sr) |
---|
1045 | sums(k,65:69) = sums(k,65:69) / ngp_2dh(sr) |
---|
1046 | sums(k,70:pr_palm-2) = sums(k,70:pr_palm-2)/ ngp_2dh_s_inner(k,sr) |
---|
1047 | ENDDO |
---|
1048 | |
---|
1049 | !-- Upstream-parts |
---|
1050 | sums(nzb:nzb+11,pr_palm-1) = sums(nzb:nzb+11,pr_palm-1) / ngp_3d(sr) |
---|
1051 | !-- u* and so on |
---|
1052 | !-- As sums(nzb:nzb+3,pr_palm) are full 2D arrays (us, usws, vsws, ts) whose |
---|
1053 | !-- size is always ( nx + 1 ) * ( ny + 1 ), defined at the first grid layer |
---|
1054 | !-- above the topography, they are being divided by ngp_2dh(sr) |
---|
1055 | sums(nzb:nzb+3,pr_palm) = sums(nzb:nzb+3,pr_palm) / & |
---|
1056 | ngp_2dh(sr) |
---|
1057 | sums(nzb+12,pr_palm) = sums(nzb+12,pr_palm) / & ! qs |
---|
1058 | ngp_2dh(sr) |
---|
1059 | !-- eges, e* |
---|
1060 | sums(nzb+4:nzb+5,pr_palm) = sums(nzb+4:nzb+5,pr_palm) / & |
---|
1061 | ngp_3d(sr) |
---|
1062 | !-- Old and new divergence |
---|
1063 | sums(nzb+9:nzb+10,pr_palm) = sums(nzb+9:nzb+10,pr_palm) / & |
---|
1064 | ngp_3d_inner(sr) |
---|
1065 | |
---|
1066 | !-- User-defined profiles |
---|
1067 | IF ( max_pr_user > 0 ) THEN |
---|
1068 | DO k = nzb, nzt+1 |
---|
1069 | sums(k,pr_palm+1:pr_palm+max_pr_user) = & |
---|
1070 | sums(k,pr_palm+1:pr_palm+max_pr_user) / & |
---|
1071 | ngp_2dh_s_inner(k,sr) |
---|
1072 | ENDDO |
---|
1073 | ENDIF |
---|
1074 | |
---|
1075 | ! |
---|
1076 | !-- Collect horizontal average in hom. |
---|
1077 | !-- Compute deduced averages (e.g. total heat flux) |
---|
1078 | hom(:,1,3,sr) = sums(:,3) ! w |
---|
1079 | hom(:,1,8,sr) = sums(:,8) ! e profiles 5-7 are initial profiles |
---|
1080 | hom(:,1,9,sr) = sums(:,9) ! km |
---|
1081 | hom(:,1,10,sr) = sums(:,10) ! kh |
---|
1082 | hom(:,1,11,sr) = sums(:,11) ! l |
---|
1083 | hom(:,1,12,sr) = sums(:,12) ! w"u" |
---|
1084 | hom(:,1,13,sr) = sums(:,13) ! w*u* |
---|
1085 | hom(:,1,14,sr) = sums(:,14) ! w"v" |
---|
1086 | hom(:,1,15,sr) = sums(:,15) ! w*v* |
---|
1087 | hom(:,1,16,sr) = sums(:,16) ! w"pt" |
---|
1088 | hom(:,1,17,sr) = sums(:,17) ! w*pt* |
---|
1089 | hom(:,1,18,sr) = sums(:,16) + sums(:,17) ! wpt |
---|
1090 | hom(:,1,19,sr) = sums(:,12) + sums(:,13) ! wu |
---|
1091 | hom(:,1,20,sr) = sums(:,14) + sums(:,15) ! wv |
---|
1092 | hom(:,1,21,sr) = sums(:,21) ! w*pt*BC |
---|
1093 | hom(:,1,22,sr) = sums(:,16) + sums(:,21) ! wptBC |
---|
1094 | ! profile 24 is initial profile (sa) |
---|
1095 | ! profiles 25-29 left empty for initial |
---|
1096 | ! profiles |
---|
1097 | hom(:,1,30,sr) = sums(:,30) ! u*2 |
---|
1098 | hom(:,1,31,sr) = sums(:,31) ! v*2 |
---|
1099 | hom(:,1,32,sr) = sums(:,32) ! w*2 |
---|
1100 | hom(:,1,33,sr) = sums(:,33) ! pt*2 |
---|
1101 | hom(:,1,34,sr) = sums(:,34) ! e* |
---|
1102 | hom(:,1,35,sr) = sums(:,35) ! w*2pt* |
---|
1103 | hom(:,1,36,sr) = sums(:,36) ! w*pt*2 |
---|
1104 | hom(:,1,37,sr) = sums(:,37) ! w*e* |
---|
1105 | hom(:,1,38,sr) = sums(:,38) ! w*3 |
---|
1106 | hom(:,1,39,sr) = sums(:,38) / ( abs( sums(:,32) ) + 1E-20 )**1.5 ! Sw |
---|
1107 | hom(:,1,40,sr) = sums(:,40) ! p |
---|
1108 | hom(:,1,45,sr) = sums(:,45) ! w"vpt" |
---|
1109 | hom(:,1,46,sr) = sums(:,46) ! w*vpt* |
---|
1110 | hom(:,1,47,sr) = sums(:,45) + sums(:,46) ! wvpt |
---|
1111 | hom(:,1,48,sr) = sums(:,48) ! w"q" (w"qv") |
---|
1112 | hom(:,1,49,sr) = sums(:,49) ! w*q* (w*qv*) |
---|
1113 | hom(:,1,50,sr) = sums(:,48) + sums(:,49) ! wq (wqv) |
---|
1114 | hom(:,1,51,sr) = sums(:,51) ! w"qv" |
---|
1115 | hom(:,1,52,sr) = sums(:,52) ! w*qv* |
---|
1116 | hom(:,1,53,sr) = sums(:,52) + sums(:,51) ! wq (wqv) |
---|
1117 | hom(:,1,54,sr) = sums(:,54) ! ql |
---|
1118 | hom(:,1,55,sr) = sums(:,55) ! w*u*u*/dz |
---|
1119 | hom(:,1,56,sr) = sums(:,56) ! w*p*/dz |
---|
1120 | hom(:,1,57,sr) = sums(:,57) ! ( w"e + w"p"/rho )/dz |
---|
1121 | hom(:,1,58,sr) = sums(:,58) ! u"pt" |
---|
1122 | hom(:,1,59,sr) = sums(:,59) ! u*pt* |
---|
1123 | hom(:,1,60,sr) = sums(:,58) + sums(:,59) ! upt_t |
---|
1124 | hom(:,1,61,sr) = sums(:,61) ! v"pt" |
---|
1125 | hom(:,1,62,sr) = sums(:,62) ! v*pt* |
---|
1126 | hom(:,1,63,sr) = sums(:,61) + sums(:,62) ! vpt_t |
---|
1127 | hom(:,1,64,sr) = sums(:,64) ! rho |
---|
1128 | hom(:,1,65,sr) = sums(:,65) ! w"sa" |
---|
1129 | hom(:,1,66,sr) = sums(:,66) ! w*sa* |
---|
1130 | hom(:,1,67,sr) = sums(:,65) + sums(:,66) ! wsa |
---|
1131 | hom(:,1,68,sr) = sums(:,68) ! w*p* |
---|
1132 | hom(:,1,69,sr) = sums(:,69) ! w"e + w"p"/rho |
---|
1133 | hom(:,1,70,sr) = sums(:,70) ! q*2 |
---|
1134 | hom(:,1,71,sr) = sums(:,71) ! prho |
---|
1135 | hom(:,1,72,sr) = hyp * 1E-4 ! hyp in dbar |
---|
1136 | hom(:,1,73,sr) = sums(:,73) ! nr |
---|
1137 | hom(:,1,74,sr) = sums(:,74) ! qr |
---|
1138 | hom(:,1,75,sr) = sums(:,75) ! qc |
---|
1139 | hom(:,1,76,sr) = sums(:,76) ! prr (precipitation rate) |
---|
1140 | |
---|
1141 | hom(:,1,pr_palm-1,sr) = sums(:,pr_palm-1) |
---|
1142 | ! upstream-parts u_x, u_y, u_z, v_x, |
---|
1143 | ! v_y, usw. (in last but one profile) |
---|
1144 | hom(:,1,pr_palm,sr) = sums(:,pr_palm) |
---|
1145 | ! u*, w'u', w'v', t* (in last profile) |
---|
1146 | |
---|
1147 | IF ( max_pr_user > 0 ) THEN ! user-defined profiles |
---|
1148 | hom(:,1,pr_palm+1:pr_palm+max_pr_user,sr) = & |
---|
1149 | sums(:,pr_palm+1:pr_palm+max_pr_user) |
---|
1150 | ENDIF |
---|
1151 | |
---|
1152 | ! |
---|
1153 | !-- Determine the boundary layer height using two different schemes. |
---|
1154 | !-- First scheme: Starting from the Earth's (Ocean's) surface, look for the |
---|
1155 | !-- first relative minimum (maximum) of the total heat flux. |
---|
1156 | !-- The corresponding height is assumed as the boundary layer height, if it |
---|
1157 | !-- is less than 1.5 times the height where the heat flux becomes negative |
---|
1158 | !-- (positive) for the first time. |
---|
1159 | z_i(1) = 0.0 |
---|
1160 | first = .TRUE. |
---|
1161 | |
---|
1162 | IF ( ocean ) THEN |
---|
1163 | DO k = nzt, nzb+1, -1 |
---|
1164 | IF ( first .AND. hom(k,1,18,sr) < 0.0 & |
---|
1165 | .AND. abs(hom(k,1,18,sr)) > 1.0E-8) THEN |
---|
1166 | first = .FALSE. |
---|
1167 | height = zw(k) |
---|
1168 | ENDIF |
---|
1169 | IF ( hom(k,1,18,sr) < 0.0 .AND. & |
---|
1170 | abs(hom(k,1,18,sr)) > 1.0E-8 .AND. & |
---|
1171 | hom(k-1,1,18,sr) > hom(k,1,18,sr) ) THEN |
---|
1172 | IF ( zw(k) < 1.5 * height ) THEN |
---|
1173 | z_i(1) = zw(k) |
---|
1174 | ELSE |
---|
1175 | z_i(1) = height |
---|
1176 | ENDIF |
---|
1177 | EXIT |
---|
1178 | ENDIF |
---|
1179 | ENDDO |
---|
1180 | ELSE |
---|
1181 | DO k = nzb, nzt-1 |
---|
1182 | IF ( first .AND. hom(k,1,18,sr) < 0.0 & |
---|
1183 | .AND. abs(hom(k,1,18,sr)) > 1.0E-8 ) THEN |
---|
1184 | first = .FALSE. |
---|
1185 | height = zw(k) |
---|
1186 | ENDIF |
---|
1187 | IF ( hom(k,1,18,sr) < 0.0 .AND. & |
---|
1188 | abs(hom(k,1,18,sr)) > 1.0E-8 .AND. & |
---|
1189 | hom(k+1,1,18,sr) > hom(k,1,18,sr) ) THEN |
---|
1190 | IF ( zw(k) < 1.5 * height ) THEN |
---|
1191 | z_i(1) = zw(k) |
---|
1192 | ELSE |
---|
1193 | z_i(1) = height |
---|
1194 | ENDIF |
---|
1195 | EXIT |
---|
1196 | ENDIF |
---|
1197 | ENDDO |
---|
1198 | ENDIF |
---|
1199 | |
---|
1200 | ! |
---|
1201 | !-- Second scheme: Gradient scheme from Sullivan et al. (1998), modified |
---|
1202 | !-- by Uhlenbrock(2006). The boundary layer height is the height with the |
---|
1203 | !-- maximal local temperature gradient: starting from the second (the last |
---|
1204 | !-- but one) vertical gridpoint, the local gradient must be at least |
---|
1205 | !-- 0.2K/100m and greater than the next four gradients. |
---|
1206 | !-- WARNING: The threshold value of 0.2K/100m must be adjusted for the |
---|
1207 | !-- ocean case! |
---|
1208 | z_i(2) = 0.0 |
---|
1209 | DO k = nzb+1, nzt+1 |
---|
1210 | dptdz(k) = ( hom(k,1,4,sr) - hom(k-1,1,4,sr) ) * ddzu(k) |
---|
1211 | ENDDO |
---|
1212 | dptdz_threshold = 0.2 / 100.0 |
---|
1213 | |
---|
1214 | IF ( ocean ) THEN |
---|
1215 | DO k = nzt+1, nzb+5, -1 |
---|
1216 | IF ( dptdz(k) > dptdz_threshold .AND. & |
---|
1217 | dptdz(k) > dptdz(k-1) .AND. dptdz(k) > dptdz(k-2) .AND. & |
---|
1218 | dptdz(k) > dptdz(k-3) .AND. dptdz(k) > dptdz(k-4) ) THEN |
---|
1219 | z_i(2) = zw(k-1) |
---|
1220 | EXIT |
---|
1221 | ENDIF |
---|
1222 | ENDDO |
---|
1223 | ELSE |
---|
1224 | DO k = nzb+1, nzt-3 |
---|
1225 | IF ( dptdz(k) > dptdz_threshold .AND. & |
---|
1226 | dptdz(k) > dptdz(k+1) .AND. dptdz(k) > dptdz(k+2) .AND. & |
---|
1227 | dptdz(k) > dptdz(k+3) .AND. dptdz(k) > dptdz(k+4) ) THEN |
---|
1228 | z_i(2) = zw(k-1) |
---|
1229 | EXIT |
---|
1230 | ENDIF |
---|
1231 | ENDDO |
---|
1232 | ENDIF |
---|
1233 | |
---|
1234 | hom(nzb+6,1,pr_palm,sr) = z_i(1) |
---|
1235 | hom(nzb+7,1,pr_palm,sr) = z_i(2) |
---|
1236 | |
---|
1237 | ! |
---|
1238 | !-- Computation of both the characteristic vertical velocity and |
---|
1239 | !-- the characteristic convective boundary layer temperature. |
---|
1240 | !-- The horizontal average at nzb+1 is input for the average temperature. |
---|
1241 | IF ( hom(nzb,1,18,sr) > 0.0 .AND. abs(hom(nzb,1,18,sr)) > 1.0E-8 & |
---|
1242 | .AND. z_i(1) /= 0.0 ) THEN |
---|
1243 | hom(nzb+8,1,pr_palm,sr) = ( g / hom(nzb+1,1,4,sr) * & |
---|
1244 | hom(nzb,1,18,sr) * & |
---|
1245 | ABS( z_i(1) ) )**0.333333333 |
---|
1246 | !-- so far this only works if Prandtl layer is used |
---|
1247 | hom(nzb+11,1,pr_palm,sr) = hom(nzb,1,16,sr) / hom(nzb+8,1,pr_palm,sr) |
---|
1248 | ELSE |
---|
1249 | hom(nzb+8,1,pr_palm,sr) = 0.0 |
---|
1250 | hom(nzb+11,1,pr_palm,sr) = 0.0 |
---|
1251 | ENDIF |
---|
1252 | |
---|
1253 | ! |
---|
1254 | !-- Collect the time series quantities |
---|
1255 | ts_value(1,sr) = hom(nzb+4,1,pr_palm,sr) ! E |
---|
1256 | ts_value(2,sr) = hom(nzb+5,1,pr_palm,sr) ! E* |
---|
1257 | ts_value(3,sr) = dt_3d |
---|
1258 | ts_value(4,sr) = hom(nzb,1,pr_palm,sr) ! u* |
---|
1259 | ts_value(5,sr) = hom(nzb+3,1,pr_palm,sr) ! th* |
---|
1260 | ts_value(6,sr) = u_max |
---|
1261 | ts_value(7,sr) = v_max |
---|
1262 | ts_value(8,sr) = w_max |
---|
1263 | ts_value(9,sr) = hom(nzb+10,1,pr_palm,sr) ! new divergence |
---|
1264 | ts_value(10,sr) = hom(nzb+9,1,pr_palm,sr) ! old Divergence |
---|
1265 | ts_value(11,sr) = hom(nzb+6,1,pr_palm,sr) ! z_i(1) |
---|
1266 | ts_value(12,sr) = hom(nzb+7,1,pr_palm,sr) ! z_i(2) |
---|
1267 | ts_value(13,sr) = hom(nzb+8,1,pr_palm,sr) ! w* |
---|
1268 | ts_value(14,sr) = hom(nzb,1,16,sr) ! w'pt' at k=0 |
---|
1269 | ts_value(15,sr) = hom(nzb+1,1,16,sr) ! w'pt' at k=1 |
---|
1270 | ts_value(16,sr) = hom(nzb+1,1,18,sr) ! wpt at k=1 |
---|
1271 | ts_value(17,sr) = hom(nzb,1,4,sr) ! pt(0) |
---|
1272 | ts_value(18,sr) = hom(nzb+1,1,4,sr) ! pt(zp) |
---|
1273 | ts_value(19,sr) = hom(nzb+1,1,pr_palm,sr) ! u'w' at k=0 |
---|
1274 | ts_value(20,sr) = hom(nzb+2,1,pr_palm,sr) ! v'w' at k=0 |
---|
1275 | ts_value(21,sr) = hom(nzb,1,48,sr) ! w"q" at k=0 |
---|
1276 | |
---|
1277 | IF ( ts_value(5,sr) /= 0.0 ) THEN |
---|
1278 | ts_value(22,sr) = ts_value(4,sr)**2 / & |
---|
1279 | ( kappa * g * ts_value(5,sr) / ts_value(18,sr) ) ! L |
---|
1280 | ELSE |
---|
1281 | ts_value(22,sr) = 10000.0 |
---|
1282 | ENDIF |
---|
1283 | |
---|
1284 | ts_value(23,sr) = hom(nzb+12,1,pr_palm,sr) ! q* |
---|
1285 | ! |
---|
1286 | !-- Calculate additional statistics provided by the user interface |
---|
1287 | CALL user_statistics( 'time_series', sr, 0 ) |
---|
1288 | |
---|
1289 | ENDDO ! loop of the subregions |
---|
1290 | |
---|
1291 | ! |
---|
1292 | !-- If required, sum up horizontal averages for subsequent time averaging |
---|
1293 | IF ( do_sum ) THEN |
---|
1294 | IF ( average_count_pr == 0 ) hom_sum = 0.0 |
---|
1295 | hom_sum = hom_sum + hom(:,1,:,:) |
---|
1296 | average_count_pr = average_count_pr + 1 |
---|
1297 | do_sum = .FALSE. |
---|
1298 | ENDIF |
---|
1299 | |
---|
1300 | ! |
---|
1301 | !-- Set flag for other UPs (e.g. output routines, but also buoyancy). |
---|
1302 | !-- This flag is reset after each time step in time_integration. |
---|
1303 | flow_statistics_called = .TRUE. |
---|
1304 | |
---|
1305 | CALL cpu_log( log_point(10), 'flow_statistics', 'stop' ) |
---|
1306 | |
---|
1307 | |
---|
1308 | END SUBROUTINE flow_statistics |
---|