1 | SUBROUTINE flow_statistics |
---|
2 | |
---|
3 | !------------------------------------------------------------------------------! |
---|
4 | ! Actual revisions: |
---|
5 | ! ----------------- |
---|
6 | ! fluxes at top modified (tswst, qswst) |
---|
7 | ! |
---|
8 | ! Former revisions: |
---|
9 | ! ----------------- |
---|
10 | ! $Id: flow_statistics.f90 19 2007-02-23 04:53:48Z raasch $ |
---|
11 | ! RCS Log replace by Id keyword, revision history cleaned up |
---|
12 | ! |
---|
13 | ! Revision 1.41 2006/08/04 14:37:50 raasch |
---|
14 | ! Error removed in non-parallel part (sums_l) |
---|
15 | ! |
---|
16 | ! Revision 1.1 1997/08/11 06:15:17 raasch |
---|
17 | ! Initial revision |
---|
18 | ! |
---|
19 | ! |
---|
20 | ! Description: |
---|
21 | ! ------------ |
---|
22 | ! Compute average profiles and further average flow quantities for the different |
---|
23 | ! user-defined (sub-)regions. The region indexed 0 is the total model domain. |
---|
24 | ! |
---|
25 | ! NOTE: For simplicity, nzb_s_outer and nzb_diff_s_outer are being used as a |
---|
26 | ! ---- lower vertical index for k-loops for all variables so that regardless |
---|
27 | ! of the variable and its respective staggered grid always the same number of |
---|
28 | ! grid points is used for 2D averages. The disadvantage: depending on the |
---|
29 | ! variable, up to one grid layer adjacent to the (vertical walls of the) |
---|
30 | ! topography is missed out by this simplification. |
---|
31 | !------------------------------------------------------------------------------! |
---|
32 | |
---|
33 | USE arrays_3d |
---|
34 | USE cloud_parameters |
---|
35 | USE cpulog |
---|
36 | USE grid_variables |
---|
37 | USE indices |
---|
38 | USE interfaces |
---|
39 | USE pegrid |
---|
40 | USE statistics |
---|
41 | USE control_parameters |
---|
42 | |
---|
43 | IMPLICIT NONE |
---|
44 | |
---|
45 | INTEGER :: i, j, k, omp_get_thread_num, sr, tn |
---|
46 | LOGICAL :: first |
---|
47 | REAL :: height, pts, sums_l_eper, sums_l_etot, ust, ust2, u2, vst, & |
---|
48 | vst2, v2, w2, z_i(2) |
---|
49 | REAL :: sums_ll(nzb:nzt+1,2) |
---|
50 | |
---|
51 | |
---|
52 | CALL cpu_log( log_point(10), 'flow_statistics', 'start' ) |
---|
53 | |
---|
54 | ! |
---|
55 | !-- To be on the safe side, check whether flow_statistics has already been |
---|
56 | !-- called once after the current time step |
---|
57 | IF ( flow_statistics_called ) THEN |
---|
58 | IF ( myid == 0 ) PRINT*, '+++ WARNING: flow_statistics is called two', & |
---|
59 | ' times within one timestep' |
---|
60 | CALL local_stop |
---|
61 | ENDIF |
---|
62 | |
---|
63 | ! |
---|
64 | !-- Compute statistics for each (sub-)region |
---|
65 | DO sr = 0, statistic_regions |
---|
66 | |
---|
67 | ! |
---|
68 | !-- Initialize (local) summation array |
---|
69 | sums_l = 0.0 |
---|
70 | |
---|
71 | ! |
---|
72 | !-- Store sums that have been computed in other subroutines in summation |
---|
73 | !-- array |
---|
74 | sums_l(:,11,:) = sums_l_l(:,sr,:) ! mixing length from diffusivities |
---|
75 | !-- WARNING: next line still has to be adjusted for OpenMP |
---|
76 | sums_l(:,21,0) = sums_wsts_bc_l(:,sr) ! heat flux from advec_s_bc |
---|
77 | sums_l(nzb+9,var_sum,0) = sums_divold_l(sr) ! old divergence from pres |
---|
78 | sums_l(nzb+10,var_sum,0) = sums_divnew_l(sr) ! new divergence from pres |
---|
79 | !-- WARNING: next four lines still may have to be adjusted for OpenMP |
---|
80 | sums_l(nzb:nzb+2,var_sum-1,0) = sums_up_fraction_l(1,1:3,sr)! upstream |
---|
81 | sums_l(nzb+3:nzb+5,var_sum-1,0) = sums_up_fraction_l(2,1:3,sr)! parts |
---|
82 | sums_l(nzb+6:nzb+8,var_sum-1,0) = sums_up_fraction_l(3,1:3,sr)! from |
---|
83 | sums_l(nzb+9:nzb+11,var_sum-1,0) = sums_up_fraction_l(4,1:3,sr)! spline |
---|
84 | |
---|
85 | ! |
---|
86 | !-- Horizontally averaged profiles of horizontal velocities and temperature. |
---|
87 | !-- They must have been computed before, because they are already required |
---|
88 | !-- for other horizontal averages. |
---|
89 | tn = 0 |
---|
90 | !$OMP PARALLEL PRIVATE( i, j, k, tn ) |
---|
91 | #if defined( __lcmuk ) |
---|
92 | tn = omp_get_thread_num() |
---|
93 | #else |
---|
94 | !$ tn = omp_get_thread_num() |
---|
95 | #endif |
---|
96 | |
---|
97 | !$OMP DO |
---|
98 | DO i = nxl, nxr |
---|
99 | DO j = nys, nyn |
---|
100 | DO k = nzb_s_outer(j,i), nzt+1 |
---|
101 | sums_l(k,1,tn) = sums_l(k,1,tn) + u(k,j,i) * rmask(j,i,sr) |
---|
102 | sums_l(k,2,tn) = sums_l(k,2,tn) + v(k,j,i) * rmask(j,i,sr) |
---|
103 | sums_l(k,4,tn) = sums_l(k,4,tn) + pt(k,j,i) * rmask(j,i,sr) |
---|
104 | ENDDO |
---|
105 | ENDDO |
---|
106 | ENDDO |
---|
107 | |
---|
108 | ! |
---|
109 | !-- Horizontally averaged profiles of virtual potential temperature, |
---|
110 | !-- total water content, specific humidity and liquid water potential |
---|
111 | !-- temperature |
---|
112 | IF ( moisture ) THEN |
---|
113 | !$OMP DO |
---|
114 | DO i = nxl, nxr |
---|
115 | DO j = nys, nyn |
---|
116 | DO k = nzb_s_outer(j,i), nzt+1 |
---|
117 | sums_l(k,44,tn) = sums_l(k,44,tn) + & |
---|
118 | vpt(k,j,i) * rmask(j,i,sr) |
---|
119 | sums_l(k,41,tn) = sums_l(k,41,tn) + & |
---|
120 | q(k,j,i) * rmask(j,i,sr) |
---|
121 | ENDDO |
---|
122 | ENDDO |
---|
123 | ENDDO |
---|
124 | IF ( cloud_physics ) THEN |
---|
125 | !$OMP DO |
---|
126 | DO i = nxl, nxr |
---|
127 | DO j = nys, nyn |
---|
128 | DO k = nzb_s_outer(j,i), nzt+1 |
---|
129 | sums_l(k,42,tn) = sums_l(k,42,tn) + & |
---|
130 | ( q(k,j,i) - ql(k,j,i) ) * rmask(j,i,sr) |
---|
131 | sums_l(k,43,tn) = sums_l(k,43,tn) + ( & |
---|
132 | pt(k,j,i) + l_d_cp*pt_d_t(k) * ql(k,j,i) & |
---|
133 | ) * rmask(j,i,sr) |
---|
134 | ENDDO |
---|
135 | ENDDO |
---|
136 | ENDDO |
---|
137 | ENDIF |
---|
138 | ENDIF |
---|
139 | |
---|
140 | ! |
---|
141 | !-- Horizontally averaged profiles of passive scalar |
---|
142 | IF ( passive_scalar ) THEN |
---|
143 | !$OMP DO |
---|
144 | DO i = nxl, nxr |
---|
145 | DO j = nys, nyn |
---|
146 | DO k = nzb_s_outer(j,i), nzt+1 |
---|
147 | sums_l(k,41,tn) = sums_l(k,41,tn) + q(k,j,i) * rmask(j,i,sr) |
---|
148 | ENDDO |
---|
149 | ENDDO |
---|
150 | ENDDO |
---|
151 | ENDIF |
---|
152 | !$OMP END PARALLEL |
---|
153 | |
---|
154 | ! |
---|
155 | !-- Summation of thread sums |
---|
156 | IF ( threads_per_task > 1 ) THEN |
---|
157 | DO i = 1, threads_per_task-1 |
---|
158 | sums_l(:,1,0) = sums_l(:,1,0) + sums_l(:,1,i) |
---|
159 | sums_l(:,2,0) = sums_l(:,2,0) + sums_l(:,2,i) |
---|
160 | sums_l(:,4,0) = sums_l(:,4,0) + sums_l(:,4,i) |
---|
161 | IF ( moisture ) THEN |
---|
162 | sums_l(:,41,0) = sums_l(:,41,0) + sums_l(:,41,i) |
---|
163 | sums_l(:,44,0) = sums_l(:,44,0) + sums_l(:,44,i) |
---|
164 | IF ( cloud_physics ) THEN |
---|
165 | sums_l(:,42,0) = sums_l(:,42,0) + sums_l(:,42,i) |
---|
166 | sums_l(:,43,0) = sums_l(:,43,0) + sums_l(:,43,i) |
---|
167 | ENDIF |
---|
168 | ENDIF |
---|
169 | IF ( passive_scalar ) THEN |
---|
170 | sums_l(:,41,0) = sums_l(:,41,0) + sums_l(:,41,i) |
---|
171 | ENDIF |
---|
172 | ENDDO |
---|
173 | ENDIF |
---|
174 | |
---|
175 | #if defined( __parallel ) |
---|
176 | ! |
---|
177 | !-- Compute total sum from local sums |
---|
178 | CALL MPI_ALLREDUCE( sums_l(nzb,1,0), sums(nzb,1), nzt+2-nzb, MPI_REAL, & |
---|
179 | MPI_SUM, comm2d, ierr ) |
---|
180 | CALL MPI_ALLREDUCE( sums_l(nzb,2,0), sums(nzb,2), nzt+2-nzb, MPI_REAL, & |
---|
181 | MPI_SUM, comm2d, ierr ) |
---|
182 | CALL MPI_ALLREDUCE( sums_l(nzb,4,0), sums(nzb,4), nzt+2-nzb, MPI_REAL, & |
---|
183 | MPI_SUM, comm2d, ierr ) |
---|
184 | IF ( moisture ) THEN |
---|
185 | CALL MPI_ALLREDUCE( sums_l(nzb,44,0), sums(nzb,44), nzt+2-nzb, & |
---|
186 | MPI_REAL, MPI_SUM, comm2d, ierr ) |
---|
187 | CALL MPI_ALLREDUCE( sums_l(nzb,41,0), sums(nzb,41), nzt+2-nzb, & |
---|
188 | MPI_REAL, MPI_SUM, comm2d, ierr ) |
---|
189 | IF ( cloud_physics ) THEN |
---|
190 | CALL MPI_ALLREDUCE( sums_l(nzb,42,0), sums(nzb,42), nzt+2-nzb, & |
---|
191 | MPI_REAL, MPI_SUM, comm2d, ierr ) |
---|
192 | CALL MPI_ALLREDUCE( sums_l(nzb,43,0), sums(nzb,43), nzt+2-nzb, & |
---|
193 | MPI_REAL, MPI_SUM, comm2d, ierr ) |
---|
194 | ENDIF |
---|
195 | ENDIF |
---|
196 | |
---|
197 | IF ( passive_scalar ) THEN |
---|
198 | CALL MPI_ALLREDUCE( sums_l(nzb,41,0), sums(nzb,41), nzt+2-nzb, & |
---|
199 | MPI_REAL, MPI_SUM, comm2d, ierr ) |
---|
200 | ENDIF |
---|
201 | #else |
---|
202 | sums(:,1) = sums_l(:,1,0) |
---|
203 | sums(:,2) = sums_l(:,2,0) |
---|
204 | sums(:,4) = sums_l(:,4,0) |
---|
205 | IF ( moisture ) THEN |
---|
206 | sums(:,44) = sums_l(:,44,0) |
---|
207 | sums(:,41) = sums_l(:,41,0) |
---|
208 | IF ( cloud_physics ) THEN |
---|
209 | sums(:,42) = sums_l(:,42,0) |
---|
210 | sums(:,43) = sums_l(:,43,0) |
---|
211 | ENDIF |
---|
212 | ENDIF |
---|
213 | IF ( passive_scalar ) sums(:,41) = sums_l(:,41,0) |
---|
214 | #endif |
---|
215 | |
---|
216 | ! |
---|
217 | !-- Final values are obtained by division by the total number of grid points |
---|
218 | !-- used for summation. After that store profiles. |
---|
219 | sums(:,1) = sums(:,1) / ngp_2dh_outer(:,sr) |
---|
220 | sums(:,2) = sums(:,2) / ngp_2dh_outer(:,sr) |
---|
221 | sums(:,4) = sums(:,4) / ngp_2dh_outer(:,sr) |
---|
222 | hom(:,1,1,sr) = sums(:,1) ! u |
---|
223 | hom(:,1,2,sr) = sums(:,2) ! v |
---|
224 | hom(:,1,4,sr) = sums(:,4) ! pt |
---|
225 | |
---|
226 | ! |
---|
227 | !-- Humidity and cloud parameters |
---|
228 | IF ( moisture ) THEN |
---|
229 | sums(:,44) = sums(:,44) / ngp_2dh_outer(:,sr) |
---|
230 | sums(:,41) = sums(:,41) / ngp_2dh_outer(:,sr) |
---|
231 | hom(:,1,44,sr) = sums(:,44) ! vpt |
---|
232 | hom(:,1,41,sr) = sums(:,41) ! qv (q) |
---|
233 | IF ( cloud_physics ) THEN |
---|
234 | sums(:,42) = sums(:,42) / ngp_2dh_outer(:,sr) |
---|
235 | sums(:,43) = sums(:,43) / ngp_2dh_outer(:,sr) |
---|
236 | hom(:,1,42,sr) = sums(:,42) ! qv |
---|
237 | hom(:,1,43,sr) = sums(:,43) ! pt |
---|
238 | ENDIF |
---|
239 | ENDIF |
---|
240 | |
---|
241 | ! |
---|
242 | !-- Passive scalar |
---|
243 | IF ( passive_scalar ) hom(:,1,41,sr) = sums(:,41) / ngp_2dh_outer(:,sr) |
---|
244 | |
---|
245 | ! |
---|
246 | !-- Horizontally averaged profiles of the remaining prognostic variables, |
---|
247 | !-- variances, the total and the perturbation energy (single values in last |
---|
248 | !-- column of sums_l) and some diagnostic quantities. |
---|
249 | !-- NOTE: for simplicity, nzb_s_outer is used below, although strictly |
---|
250 | !-- ---- speaking the following k-loop would have to be split up and |
---|
251 | !-- rearranged according to the staggered grid. |
---|
252 | tn = 0 |
---|
253 | #if defined( __lcmuk ) |
---|
254 | !$OMP PARALLEL PRIVATE( i, j, k, pts, sums_ll, sums_l_eper, sums_l_etot, & |
---|
255 | !$OMP tn, ust, ust2, u2, vst, vst2, v2, w2 ) |
---|
256 | tn = omp_get_thread_num() |
---|
257 | #else |
---|
258 | !$OMP PARALLEL PRIVATE( i, j, k, pts, sums_ll, sums_l_eper, sums_l_etot, tn, ust, ust2, u2, vst, vst2, v2, w2 ) |
---|
259 | !$ tn = omp_get_thread_num() |
---|
260 | #endif |
---|
261 | !$OMP DO |
---|
262 | DO i = nxl, nxr |
---|
263 | DO j = nys, nyn |
---|
264 | sums_l_etot = 0.0 |
---|
265 | sums_l_eper = 0.0 |
---|
266 | DO k = nzb_s_outer(j,i), nzt+1 |
---|
267 | u2 = u(k,j,i)**2 |
---|
268 | v2 = v(k,j,i)**2 |
---|
269 | w2 = w(k,j,i)**2 |
---|
270 | ust2 = ( u(k,j,i) - hom(k,1,1,sr) )**2 |
---|
271 | vst2 = ( v(k,j,i) - hom(k,1,2,sr) )**2 |
---|
272 | ! |
---|
273 | !-- Prognostic and diagnostic variables |
---|
274 | sums_l(k,3,tn) = sums_l(k,3,tn) + w(k,j,i) * rmask(j,i,sr) |
---|
275 | sums_l(k,8,tn) = sums_l(k,8,tn) + e(k,j,i) * rmask(j,i,sr) |
---|
276 | sums_l(k,9,tn) = sums_l(k,9,tn) + km(k,j,i) * rmask(j,i,sr) |
---|
277 | sums_l(k,10,tn) = sums_l(k,10,tn) + kh(k,j,i) * rmask(j,i,sr) |
---|
278 | sums_l(k,40,tn) = sums_l(k,40,tn) + p(k,j,i) |
---|
279 | |
---|
280 | ! |
---|
281 | !-- Variances |
---|
282 | sums_l(k,30,tn) = sums_l(k,30,tn) + ust2 * rmask(j,i,sr) |
---|
283 | sums_l(k,31,tn) = sums_l(k,31,tn) + vst2 * rmask(j,i,sr) |
---|
284 | sums_l(k,32,tn) = sums_l(k,32,tn) + w2 * rmask(j,i,sr) |
---|
285 | sums_l(k,33,tn) = sums_l(k,33,tn) + & |
---|
286 | ( pt(k,j,i)-hom(k,1,4,sr) )**2 * rmask(j,i,sr) |
---|
287 | ! |
---|
288 | !-- Higher moments |
---|
289 | !-- (Computation of the skewness of w further below) |
---|
290 | sums_l(k,38,tn) = sums_l(k,38,tn) + w(k,j,i) * w2 * & |
---|
291 | rmask(j,i,sr) |
---|
292 | ! |
---|
293 | !-- Perturbation energy |
---|
294 | sums_l(k,34,tn) = sums_l(k,34,tn) + 0.5 * ( ust2 + vst2 + w2 ) & |
---|
295 | * rmask(j,i,sr) |
---|
296 | sums_l_etot = sums_l_etot + & |
---|
297 | 0.5 * ( u2 + v2 + w2 ) * rmask(j,i,sr) |
---|
298 | sums_l_eper = sums_l_eper + & |
---|
299 | 0.5 * ( ust2+vst2+w2 ) * rmask(j,i,sr) |
---|
300 | ENDDO |
---|
301 | ! |
---|
302 | !-- Total and perturbation energy for the total domain (being |
---|
303 | !-- collected in the last column of sums_l). Summation of these |
---|
304 | !-- quantities is seperated from the previous loop in order to |
---|
305 | !-- allow vectorization of that loop. |
---|
306 | sums_l(nzb+4,var_sum,tn) = sums_l(nzb+4,var_sum,tn) + sums_l_etot |
---|
307 | sums_l(nzb+5,var_sum,tn) = sums_l(nzb+5,var_sum,tn) + sums_l_eper |
---|
308 | ! |
---|
309 | !-- 2D-arrays (being collected in the last column of sums_l) |
---|
310 | sums_l(nzb,var_sum,tn) = sums_l(nzb,var_sum,tn) + & |
---|
311 | us(j,i) * rmask(j,i,sr) |
---|
312 | sums_l(nzb+1,var_sum,tn) = sums_l(nzb+1,var_sum,tn) + & |
---|
313 | usws(j,i) * rmask(j,i,sr) |
---|
314 | sums_l(nzb+2,var_sum,tn) = sums_l(nzb+2,var_sum,tn) + & |
---|
315 | vsws(j,i) * rmask(j,i,sr) |
---|
316 | sums_l(nzb+3,var_sum,tn) = sums_l(nzb+3,var_sum,tn) + & |
---|
317 | ts(j,i) * rmask(j,i,sr) |
---|
318 | ENDDO |
---|
319 | ENDDO |
---|
320 | |
---|
321 | ! |
---|
322 | !-- Horizontally averaged profiles of the vertical fluxes |
---|
323 | !$OMP DO |
---|
324 | DO i = nxl, nxr |
---|
325 | DO j = nys, nyn |
---|
326 | ! |
---|
327 | !-- Subgridscale fluxes (without Prandtl layer from k=nzb, |
---|
328 | !-- oterwise from k=nzb+1) |
---|
329 | !-- NOTE: for simplicity, nzb_diff_s_outer is used below, although |
---|
330 | !-- ---- strictly speaking the following k-loop would have to be |
---|
331 | !-- split up according to the staggered grid. |
---|
332 | DO k = nzb_diff_s_outer(j,i)-1, nzt |
---|
333 | ! |
---|
334 | !-- Momentum flux w"u" |
---|
335 | sums_l(k,12,tn) = sums_l(k,12,tn) - 0.25 * ( & |
---|
336 | km(k,j,i)+km(k+1,j,i)+km(k,j,i-1)+km(k+1,j,i-1) & |
---|
337 | ) * ( & |
---|
338 | ( u(k+1,j,i) - u(k,j,i) ) * ddzu(k+1) & |
---|
339 | + ( w(k,j,i) - w(k,j,i-1) ) * ddx & |
---|
340 | ) * rmask(j,i,sr) |
---|
341 | ! |
---|
342 | !-- Momentum flux w"v" |
---|
343 | sums_l(k,14,tn) = sums_l(k,14,tn) - 0.25 * ( & |
---|
344 | km(k,j,i)+km(k+1,j,i)+km(k,j-1,i)+km(k+1,j-1,i) & |
---|
345 | ) * ( & |
---|
346 | ( v(k+1,j,i) - v(k,j,i) ) * ddzu(k+1) & |
---|
347 | + ( w(k,j,i) - w(k,j-1,i) ) * ddy & |
---|
348 | ) * rmask(j,i,sr) |
---|
349 | ENDDO |
---|
350 | |
---|
351 | DO k = nzb_diff_s_outer(j,i)-1, nzt_diff |
---|
352 | ! |
---|
353 | !-- Heat flux w"pt" |
---|
354 | sums_l(k,16,tn) = sums_l(k,16,tn) & |
---|
355 | - 0.5 * ( kh(k,j,i) + kh(k+1,j,i) ) & |
---|
356 | * ( pt(k+1,j,i) - pt(k,j,i) ) & |
---|
357 | * ddzu(k+1) * rmask(j,i,sr) |
---|
358 | |
---|
359 | |
---|
360 | ! |
---|
361 | !-- Buoyancy flux, water flux (humidity flux) w"q" |
---|
362 | IF ( moisture ) THEN |
---|
363 | sums_l(k,45,tn) = sums_l(k,45,tn) & |
---|
364 | - 0.5 * ( kh(k,j,i) + kh(k+1,j,i) ) & |
---|
365 | * ( vpt(k+1,j,i) - vpt(k,j,i) ) & |
---|
366 | * ddzu(k+1) * rmask(j,i,sr) |
---|
367 | sums_l(k,48,tn) = sums_l(k,48,tn) & |
---|
368 | - 0.5 * ( kh(k,j,i) + kh(k+1,j,i) ) & |
---|
369 | * ( q(k+1,j,i) - q(k,j,i) ) & |
---|
370 | * ddzu(k+1) * rmask(j,i,sr) |
---|
371 | IF ( cloud_physics ) THEN |
---|
372 | sums_l(k,51,tn) = sums_l(k,51,tn) & |
---|
373 | - 0.5 * ( kh(k,j,i) + kh(k+1,j,i) ) & |
---|
374 | * ( ( q(k+1,j,i) - ql(k+1,j,i) )& |
---|
375 | - ( q(k,j,i) - ql(k,j,i) ) ) & |
---|
376 | * ddzu(k+1) * rmask(j,i,sr) |
---|
377 | ENDIF |
---|
378 | ENDIF |
---|
379 | |
---|
380 | ! |
---|
381 | !-- Passive scalar flux |
---|
382 | IF ( passive_scalar ) THEN |
---|
383 | sums_l(k,48,tn) = sums_l(k,48,tn) & |
---|
384 | - 0.5 * ( kh(k,j,i) + kh(k+1,j,i) ) & |
---|
385 | * ( q(k+1,j,i) - q(k,j,i) ) & |
---|
386 | * ddzu(k+1) * rmask(j,i,sr) |
---|
387 | ENDIF |
---|
388 | |
---|
389 | ENDDO |
---|
390 | |
---|
391 | ! |
---|
392 | !-- Subgridscale fluxes in the Prandtl layer |
---|
393 | IF ( use_surface_fluxes ) THEN |
---|
394 | sums_l(nzb,12,tn) = sums_l(nzb,12,tn) + & |
---|
395 | usws(j,i) * rmask(j,i,sr) ! w"u" |
---|
396 | sums_l(nzb,14,tn) = sums_l(nzb,14,tn) + & |
---|
397 | vsws(j,i) * rmask(j,i,sr) ! w"v" |
---|
398 | sums_l(nzb,16,tn) = sums_l(nzb,16,tn) + & |
---|
399 | shf(j,i) * rmask(j,i,sr) ! w"pt" |
---|
400 | sums_l(nzb,58,tn) = sums_l(nzb,58,tn) + & |
---|
401 | 0.0 * rmask(j,i,sr) ! u"pt" |
---|
402 | sums_l(nzb,61,tn) = sums_l(nzb,61,tn) + & |
---|
403 | 0.0 * rmask(j,i,sr) ! v"pt" |
---|
404 | IF ( moisture ) THEN |
---|
405 | sums_l(nzb,48,tn) = sums_l(nzb,48,tn) + & |
---|
406 | qsws(j,i) * rmask(j,i,sr) ! w"q" (w"qv") |
---|
407 | IF ( cloud_physics ) THEN |
---|
408 | sums_l(nzb,45,tn) = sums_l(nzb,45,tn) + ( & |
---|
409 | ( 1.0 + 0.61 * q(nzb,j,i) ) * & |
---|
410 | shf(j,i) + 0.61 * pt(nzb,j,i) * & |
---|
411 | qsws(j,i) & |
---|
412 | ) |
---|
413 | ! |
---|
414 | !-- Formula does not work if ql(nzb) /= 0.0 |
---|
415 | sums_l(nzb,51,tn) = sums_l(nzb,51,tn) + & ! w"q" (w"qv") |
---|
416 | qsws(j,i) * rmask(j,i,sr) |
---|
417 | ENDIF |
---|
418 | ENDIF |
---|
419 | IF ( passive_scalar ) THEN |
---|
420 | sums_l(nzb,48,tn) = sums_l(nzb,48,tn) + & |
---|
421 | qsws(j,i) * rmask(j,i,sr) ! w"q" (w"qv") |
---|
422 | ENDIF |
---|
423 | ENDIF |
---|
424 | |
---|
425 | ! |
---|
426 | !-- Subgridscale fluxes at the top surface |
---|
427 | IF ( use_top_fluxes ) THEN |
---|
428 | sums_l(nzt,16,tn) = sums_l(nzt,16,tn) + & |
---|
429 | tswst(j,i) * rmask(j,i,sr) ! w"pt" |
---|
430 | sums_l(nzt,58,tn) = sums_l(nzt,58,tn) + & |
---|
431 | 0.0 * rmask(j,i,sr) ! u"pt" |
---|
432 | sums_l(nzt,61,tn) = sums_l(nzt,61,tn) + & |
---|
433 | 0.0 * rmask(j,i,sr) ! v"pt" |
---|
434 | IF ( moisture ) THEN |
---|
435 | sums_l(nzt,48,tn) = sums_l(nzt,48,tn) + & |
---|
436 | qswst(j,i) * rmask(j,i,sr) ! w"q" (w"qv") |
---|
437 | IF ( cloud_physics ) THEN |
---|
438 | sums_l(nzt,45,tn) = sums_l(nzt,45,tn) + ( & |
---|
439 | ( 1.0 + 0.61 * q(nzt,j,i) ) * & |
---|
440 | tswst(j,i) + 0.61 * pt(nzt,j,i) * & |
---|
441 | qsws(j,i) & |
---|
442 | ) |
---|
443 | ! |
---|
444 | !-- Formula does not work if ql(nzb) /= 0.0 |
---|
445 | sums_l(nzt,51,tn) = sums_l(nzt,51,tn) + & ! w"q" (w"qv") |
---|
446 | qswst(j,i) * rmask(j,i,sr) |
---|
447 | ENDIF |
---|
448 | ENDIF |
---|
449 | IF ( passive_scalar ) THEN |
---|
450 | sums_l(nzt,48,tn) = sums_l(nzt,48,tn) + & |
---|
451 | qswst(j,i) * rmask(j,i,sr) ! w"q" (w"qv") |
---|
452 | ENDIF |
---|
453 | ENDIF |
---|
454 | |
---|
455 | ! |
---|
456 | !-- Resolved fluxes (can be computed for all horizontal points) |
---|
457 | !-- NOTE: for simplicity, nzb_s_outer is used below, although strictly |
---|
458 | !-- ---- speaking the following k-loop would have to be split up and |
---|
459 | !-- rearranged according to the staggered grid. |
---|
460 | DO k = nzb_s_outer(j,i), nzt |
---|
461 | ust = 0.5 * ( u(k,j,i) - hom(k,1,1,sr) + & |
---|
462 | u(k+1,j,i) - hom(k+1,1,1,sr) ) |
---|
463 | vst = 0.5 * ( v(k,j,i) - hom(k,1,2,sr) + & |
---|
464 | v(k+1,j,i) - hom(k+1,1,2,sr) ) |
---|
465 | pts = 0.5 * ( pt(k,j,i) - hom(k,1,4,sr) + & |
---|
466 | pt(k+1,j,i) - hom(k+1,1,4,sr) ) |
---|
467 | ! |
---|
468 | !-- Momentum flux w*u* |
---|
469 | sums_l(k,13,tn) = sums_l(k,13,tn) + 0.5 * & |
---|
470 | ( w(k,j,i-1) + w(k,j,i) ) & |
---|
471 | * ust * rmask(j,i,sr) |
---|
472 | ! |
---|
473 | !-- Momentum flux w*v* |
---|
474 | sums_l(k,15,tn) = sums_l(k,15,tn) + 0.5 * & |
---|
475 | ( w(k,j-1,i) + w(k,j,i) ) & |
---|
476 | * vst * rmask(j,i,sr) |
---|
477 | ! |
---|
478 | !-- Heat flux w*pt* |
---|
479 | !-- The following formula (comment line, not executed) does not |
---|
480 | !-- work if applied to subregions |
---|
481 | ! sums_l(k,17,tn) = sums_l(k,17,tn) + 0.5 * & |
---|
482 | ! ( pt(k,j,i)+pt(k+1,j,i) ) & |
---|
483 | ! * w(k,j,i) * rmask(j,i,sr) |
---|
484 | sums_l(k,17,tn) = sums_l(k,17,tn) + pts * w(k,j,i) * & |
---|
485 | rmask(j,i,sr) |
---|
486 | ! |
---|
487 | !-- Higher moments |
---|
488 | sums_l(k,35,tn) = sums_l(k,35,tn) + pts * w(k,j,i)**2 * & |
---|
489 | rmask(j,i,sr) |
---|
490 | sums_l(k,36,tn) = sums_l(k,36,tn) + pts**2 * w(k,j,i) * & |
---|
491 | rmask(j,i,sr) |
---|
492 | |
---|
493 | ! |
---|
494 | !-- Buoyancy flux, water flux, humidity flux and liquid water |
---|
495 | !-- content |
---|
496 | IF ( moisture ) THEN |
---|
497 | pts = 0.5 * ( vpt(k,j,i) - hom(k,1,44,sr) + & |
---|
498 | vpt(k+1,j,i) - hom(k+1,1,44,sr) ) |
---|
499 | sums_l(k,46,tn) = sums_l(k,46,tn) + pts * w(k,j,i) * & |
---|
500 | rmask(j,i,sr) |
---|
501 | pts = 0.5 * ( q(k,j,i) - hom(k,1,41,sr) + & |
---|
502 | q(k+1,j,i) - hom(k+1,1,41,sr) ) |
---|
503 | sums_l(k,49,tn) = sums_l(k,49,tn) + pts * w(k,j,i) * & |
---|
504 | rmask(j,i,sr) |
---|
505 | IF ( cloud_physics .OR. cloud_droplets ) THEN |
---|
506 | pts = 0.5 * & |
---|
507 | ( ( q(k,j,i) - ql(k,j,i) ) - hom(k,1,42,sr) & |
---|
508 | + ( q(k+1,j,i) - ql(k+1,j,i) ) - hom(k+1,1,42,sr) ) |
---|
509 | sums_l(k,52,tn) = sums_l(k,52,tn) + pts * w(k,j,i) * & |
---|
510 | rmask(j,i,sr) |
---|
511 | sums_l(k,54,tn) = sums_l(k,54,tn) + ql(k,j,i) * & |
---|
512 | rmask(j,i,sr) |
---|
513 | ENDIF |
---|
514 | ENDIF |
---|
515 | |
---|
516 | ! |
---|
517 | !-- Passive scalar flux |
---|
518 | IF ( passive_scalar ) THEN |
---|
519 | pts = 0.5 * ( q(k,j,i) - hom(k,1,41,sr) + & |
---|
520 | q(k+1,j,i) - hom(k+1,1,41,sr) ) |
---|
521 | sums_l(k,49,tn) = sums_l(k,49,tn) + pts * w(k,j,i) * & |
---|
522 | rmask(j,i,sr) |
---|
523 | ENDIF |
---|
524 | |
---|
525 | ! |
---|
526 | !-- Energy flux w*e* |
---|
527 | sums_l(k,37,tn) = sums_l(k,37,tn) + w(k,j,i) * 0.5 * & |
---|
528 | ( ust**2 + vst**2 + w(k,j,i)**2 )& |
---|
529 | * rmask(j,i,sr) |
---|
530 | |
---|
531 | ENDDO |
---|
532 | ENDDO |
---|
533 | ENDDO |
---|
534 | |
---|
535 | ! |
---|
536 | !-- Divergence of vertical flux of resolved scale energy and pressure |
---|
537 | !-- fluctuations. First calculate the products, then the divergence. |
---|
538 | !-- Calculation is time consuming. Do it only, if profiles shall be plotted. |
---|
539 | IF ( hom(nzb+1,2,55,0) /= 0.0 ) THEN |
---|
540 | |
---|
541 | sums_ll = 0.0 ! local array |
---|
542 | |
---|
543 | !$OMP DO |
---|
544 | DO i = nxl, nxr |
---|
545 | DO j = nys, nyn |
---|
546 | DO k = nzb_s_outer(j,i)+1, nzt |
---|
547 | |
---|
548 | sums_ll(k,1) = sums_ll(k,1) + 0.5 * w(k,j,i) * ( & |
---|
549 | ( 0.25 * ( u(k,j,i)+u(k+1,j,i)+u(k,j,i+1)+u(k+1,j,i+1) & |
---|
550 | - 2.0 * ( hom(k,1,1,sr) + hom(k+1,1,1,sr) ) & |
---|
551 | ) )**2 & |
---|
552 | + ( 0.25 * ( v(k,j,i)+v(k+1,j,i)+v(k,j+1,i)+v(k+1,j+1,i) & |
---|
553 | - 2.0 * ( hom(k,1,2,sr) + hom(k+1,1,2,sr) ) & |
---|
554 | ) )**2 & |
---|
555 | + w(k,j,i)**2 ) |
---|
556 | |
---|
557 | sums_ll(k,2) = sums_ll(k,2) + 0.5 * w(k,j,i) & |
---|
558 | * ( p(k,j,i) + p(k+1,j,i) ) |
---|
559 | |
---|
560 | ENDDO |
---|
561 | ENDDO |
---|
562 | ENDDO |
---|
563 | sums_ll(0,1) = 0.0 ! because w is zero at the bottom |
---|
564 | sums_ll(nzt+1,1) = 0.0 |
---|
565 | sums_ll(0,2) = 0.0 |
---|
566 | sums_ll(nzt+1,2) = 0.0 |
---|
567 | |
---|
568 | DO k = nzb_s_outer(j,i)+1, nzt |
---|
569 | sums_l(k,55,tn) = ( sums_ll(k,1) - sums_ll(k-1,1) ) * ddzw(k) |
---|
570 | sums_l(k,56,tn) = ( sums_ll(k,2) - sums_ll(k-1,2) ) * ddzw(k) |
---|
571 | ENDDO |
---|
572 | sums_l(nzb,55,tn) = sums_l(nzb+1,55,tn) |
---|
573 | sums_l(nzb,56,tn) = sums_l(nzb+1,56,tn) |
---|
574 | |
---|
575 | ENDIF |
---|
576 | |
---|
577 | ! |
---|
578 | !-- Divergence of vertical flux of SGS TKE |
---|
579 | IF ( hom(nzb+1,2,57,0) /= 0.0 ) THEN |
---|
580 | |
---|
581 | !$OMP DO |
---|
582 | DO i = nxl, nxr |
---|
583 | DO j = nys, nyn |
---|
584 | DO k = nzb_s_outer(j,i)+1, nzt |
---|
585 | |
---|
586 | sums_l(k,57,tn) = sums_l(k,57,tn) + ( & |
---|
587 | (km(k,j,i)+km(k+1,j,i)) * (e(k+1,j,i)-e(k,j,i)) * ddzu(k+1) & |
---|
588 | - (km(k-1,j,i)+km(k,j,i)) * (e(k,j,i)-e(k-1,j,i)) * ddzu(k) & |
---|
589 | ) * ddzw(k) |
---|
590 | |
---|
591 | ENDDO |
---|
592 | ENDDO |
---|
593 | ENDDO |
---|
594 | sums_l(nzb,57,tn) = sums_l(nzb+1,57,tn) |
---|
595 | |
---|
596 | ENDIF |
---|
597 | |
---|
598 | ! |
---|
599 | !-- Horizontal heat fluxes (subgrid, resolved, total). |
---|
600 | !-- Do it only, if profiles shall be plotted. |
---|
601 | IF ( hom(nzb+1,2,58,0) /= 0.0 ) THEN |
---|
602 | |
---|
603 | !$OMP DO |
---|
604 | DO i = nxl, nxr |
---|
605 | DO j = nys, nyn |
---|
606 | DO k = nzb_s_outer(j,i)+1, nzt |
---|
607 | ! |
---|
608 | !-- Subgrid horizontal heat fluxes u"pt", v"pt" |
---|
609 | sums_l(k,58,tn) = sums_l(k,58,tn) - 0.5 * & |
---|
610 | ( kh(k,j,i) + kh(k,j,i-1) ) & |
---|
611 | * ( pt(k,j,i-1) - pt(k,j,i) ) & |
---|
612 | * ddx * rmask(j,i,sr) |
---|
613 | sums_l(k,61,tn) = sums_l(k,61,tn) - 0.5 * & |
---|
614 | ( kh(k,j,i) + kh(k,j-1,i) ) & |
---|
615 | * ( pt(k,j-1,i) - pt(k,j,i) ) & |
---|
616 | * ddy * rmask(j,i,sr) |
---|
617 | ! |
---|
618 | !-- Resolved horizontal heat fluxes u*pt*, v*pt* |
---|
619 | sums_l(k,59,tn) = sums_l(k,59,tn) + & |
---|
620 | ( u(k,j,i) - hom(k,1,1,sr) ) & |
---|
621 | * 0.5 * ( pt(k,j,i-1) - hom(k,1,4,sr) + & |
---|
622 | pt(k,j,i) - hom(k,1,4,sr) ) |
---|
623 | pts = 0.5 * ( pt(k,j-1,i) - hom(k,1,4,sr) + & |
---|
624 | pt(k,j,i) - hom(k,1,4,sr) ) |
---|
625 | sums_l(k,62,tn) = sums_l(k,62,tn) + & |
---|
626 | ( v(k,j,i) - hom(k,1,2,sr) ) & |
---|
627 | * 0.5 * ( pt(k,j-1,i) - hom(k,1,4,sr) + & |
---|
628 | pt(k,j,i) - hom(k,1,4,sr) ) |
---|
629 | ENDDO |
---|
630 | ENDDO |
---|
631 | ENDDO |
---|
632 | ! |
---|
633 | !-- Fluxes at the surface must be zero (e.g. due to the Prandtl-layer) |
---|
634 | sums(nzb,58) = 0.0 |
---|
635 | sums(nzb,59) = 0.0 |
---|
636 | sums(nzb,60) = 0.0 |
---|
637 | sums(nzb,61) = 0.0 |
---|
638 | sums(nzb,62) = 0.0 |
---|
639 | sums(nzb,63) = 0.0 |
---|
640 | |
---|
641 | ENDIF |
---|
642 | !$OMP END PARALLEL |
---|
643 | |
---|
644 | ! |
---|
645 | !-- Summation of thread sums |
---|
646 | IF ( threads_per_task > 1 ) THEN |
---|
647 | DO i = 1, threads_per_task-1 |
---|
648 | sums_l(:,3,0) = sums_l(:,3,0) + sums_l(:,3,i) |
---|
649 | sums_l(:,4:40,0) = sums_l(:,4:40,0) + sums_l(:,4:40,i) |
---|
650 | sums_l(:,45:var_sum,0) = sums_l(:,45:var_sum,0) + & |
---|
651 | sums_l(:,45:var_sum,i) |
---|
652 | ENDDO |
---|
653 | ENDIF |
---|
654 | |
---|
655 | #if defined( __parallel ) |
---|
656 | ! |
---|
657 | !-- Compute total sum from local sums |
---|
658 | CALL MPI_ALLREDUCE( sums_l(nzb,1,0), sums(nzb,1), ngp_sums, MPI_REAL, & |
---|
659 | MPI_SUM, comm2d, ierr ) |
---|
660 | #else |
---|
661 | sums = sums_l(:,:,0) |
---|
662 | #endif |
---|
663 | |
---|
664 | ! |
---|
665 | !-- Final values are obtained by division by the total number of grid points |
---|
666 | !-- used for summation. After that store profiles. |
---|
667 | !-- Profiles: |
---|
668 | DO k = nzb, nzt+1 |
---|
669 | sums(k,:var_sum-2) = sums(k,:var_sum-2) / ngp_2dh_outer(k,sr) |
---|
670 | ENDDO |
---|
671 | !-- Upstream-parts |
---|
672 | sums(nzb:nzb+11,var_sum-1) = sums(nzb:nzb+11,var_sum-1) / ngp_3d(sr) |
---|
673 | !-- u* and so on |
---|
674 | !-- As sums(nzb:nzb+3,var_sum) are full 2D arrays (us, usws, vsws, ts) whose |
---|
675 | !-- size is always ( nx + 1 ) * ( ny + 1 ), defined at the first grid layer |
---|
676 | !-- above the topography, they are being divided by ngp_2dh(sr) |
---|
677 | sums(nzb:nzb+3,var_sum) = sums(nzb:nzb+3,var_sum) / & |
---|
678 | ngp_2dh(sr) |
---|
679 | !-- eges, e* |
---|
680 | sums(nzb+4:nzb+5,var_sum) = sums(nzb+4:nzb+5,var_sum) / & |
---|
681 | ngp_3d_inner(sr) |
---|
682 | !-- Old and new divergence |
---|
683 | sums(nzb+9:nzb+10,var_sum) = sums(nzb+9:nzb+10,var_sum) / & |
---|
684 | ngp_3d_inner(sr) |
---|
685 | |
---|
686 | ! |
---|
687 | !-- Collect horizontal average in hom. |
---|
688 | !-- Compute deduced averages (e.g. total heat flux) |
---|
689 | hom(:,1,3,sr) = sums(:,3) ! w |
---|
690 | hom(:,1,8,sr) = sums(:,8) ! e profiles 5-7 are initial profiles |
---|
691 | hom(:,1,9,sr) = sums(:,9) ! km |
---|
692 | hom(:,1,10,sr) = sums(:,10) ! kh |
---|
693 | hom(:,1,11,sr) = sums(:,11) ! l |
---|
694 | hom(:,1,12,sr) = sums(:,12) ! w"u" |
---|
695 | hom(:,1,13,sr) = sums(:,13) ! w*u* |
---|
696 | hom(:,1,14,sr) = sums(:,14) ! w"v" |
---|
697 | hom(:,1,15,sr) = sums(:,15) ! w*v* |
---|
698 | hom(:,1,16,sr) = sums(:,16) ! w"pt" |
---|
699 | hom(:,1,17,sr) = sums(:,17) ! w*pt* |
---|
700 | hom(:,1,18,sr) = sums(:,16) + sums(:,17) ! wpt |
---|
701 | hom(:,1,19,sr) = sums(:,12) + sums(:,13) ! wu |
---|
702 | hom(:,1,20,sr) = sums(:,14) + sums(:,15) ! wv |
---|
703 | hom(:,1,21,sr) = sums(:,21) ! w*pt*BC |
---|
704 | hom(:,1,22,sr) = sums(:,16) + sums(:,21) ! wptBC |
---|
705 | ! profiles 23-29 left empty for initial |
---|
706 | ! profiles |
---|
707 | hom(:,1,30,sr) = sums(:,30) ! u*2 |
---|
708 | hom(:,1,31,sr) = sums(:,31) ! v*2 |
---|
709 | hom(:,1,32,sr) = sums(:,32) ! w*2 |
---|
710 | hom(:,1,33,sr) = sums(:,33) ! pt*2 |
---|
711 | hom(:,1,34,sr) = sums(:,34) ! e* |
---|
712 | hom(:,1,35,sr) = sums(:,35) ! w*2pt* |
---|
713 | hom(:,1,36,sr) = sums(:,36) ! w*pt*2 |
---|
714 | hom(:,1,37,sr) = sums(:,37) ! w*e* |
---|
715 | hom(:,1,38,sr) = sums(:,38) ! w*3 |
---|
716 | hom(:,1,39,sr) = sums(:,38) / ( sums(:,32) + 1E-20 )**1.5 ! Sw |
---|
717 | hom(:,1,40,sr) = sums(:,40) ! p |
---|
718 | hom(:,1,45,sr) = sums(:,45) ! w"q" |
---|
719 | hom(:,1,46,sr) = sums(:,46) ! w*vpt* |
---|
720 | hom(:,1,47,sr) = sums(:,45) + sums(:,46) ! wvpt |
---|
721 | hom(:,1,48,sr) = sums(:,48) ! w"q" (w"qv") |
---|
722 | hom(:,1,49,sr) = sums(:,49) ! w*q* (w*qv*) |
---|
723 | hom(:,1,50,sr) = sums(:,48) + sums(:,49) ! wq (wqv) |
---|
724 | hom(:,1,51,sr) = sums(:,51) ! w"qv" |
---|
725 | hom(:,1,52,sr) = sums(:,52) ! w*qv* |
---|
726 | hom(:,1,53,sr) = sums(:,52) + sums(:,51) ! wq (wqv) |
---|
727 | hom(:,1,54,sr) = sums(:,54) ! ql |
---|
728 | hom(:,1,55,sr) = sums(:,55) ! w*u*u*/dz |
---|
729 | hom(:,1,56,sr) = sums(:,56) ! w*p*/dz |
---|
730 | hom(:,1,57,sr) = sums(:,57) ! w"e/dz |
---|
731 | hom(:,1,58,sr) = sums(:,58) ! u"pt" |
---|
732 | hom(:,1,59,sr) = sums(:,59) ! u*pt* |
---|
733 | hom(:,1,60,sr) = sums(:,58) + sums(:,59) ! upt_t |
---|
734 | hom(:,1,61,sr) = sums(:,61) ! v"pt" |
---|
735 | hom(:,1,62,sr) = sums(:,62) ! v*pt* |
---|
736 | hom(:,1,63,sr) = sums(:,61) + sums(:,62) ! vpt_t |
---|
737 | |
---|
738 | hom(:,1,var_hom-1,sr) = sums(:,var_sum-1) |
---|
739 | ! upstream-parts u_x, u_y, u_z, v_x, |
---|
740 | ! v_y, usw. (in last but one profile) |
---|
741 | hom(:,1,var_hom,sr) = sums(:,var_sum) |
---|
742 | ! u*, w'u', w'v', t* (in last profile) |
---|
743 | |
---|
744 | ! |
---|
745 | !-- Determine the boundary layer height using two different schemes. |
---|
746 | !-- First scheme: Starting from the Earth's surface, look for the first |
---|
747 | !-- relative minimum of the total heat flux. The corresponding height is |
---|
748 | !-- accepted as the boundary layer height, if it is less than 1.5 times the |
---|
749 | !-- height where the heat flux becomes negative for the first time. |
---|
750 | !-- NOTE: This criterion is still capable of improving! |
---|
751 | z_i(1) = 0.0 |
---|
752 | first = .TRUE. |
---|
753 | DO k = nzb, nzt-1 |
---|
754 | IF ( first .AND. hom(k,1,18,sr) < 0.0 ) THEN |
---|
755 | first = .FALSE. |
---|
756 | height = zw(k) |
---|
757 | ENDIF |
---|
758 | IF ( hom(k,1,18,sr) < 0.0 .AND. & |
---|
759 | hom(k+1,1,18,sr) > hom(k,1,18,sr) ) THEN |
---|
760 | IF ( zw(k) < 1.5 * height ) THEN |
---|
761 | z_i(1) = zw(k) |
---|
762 | ELSE |
---|
763 | z_i(1) = height |
---|
764 | ENDIF |
---|
765 | EXIT |
---|
766 | ENDIF |
---|
767 | ENDDO |
---|
768 | |
---|
769 | ! |
---|
770 | !-- Second scheme: Starting from the top model boundary, look for the first |
---|
771 | !-- characteristic kink in the temperature profile, where the originally |
---|
772 | !-- stable stratification notably weakens. |
---|
773 | z_i(2) = 0.0 |
---|
774 | DO k = nzt-1, nzb+1, -1 |
---|
775 | IF ( ( hom(k+1,1,4,sr) - hom(k,1,4,sr) ) > & |
---|
776 | 2.0 * ( hom(k,1,4,sr) - hom(k-1,1,4,sr) ) ) THEN |
---|
777 | z_i(2) = zu(k) |
---|
778 | EXIT |
---|
779 | ENDIF |
---|
780 | ENDDO |
---|
781 | |
---|
782 | hom(nzb+6,1,var_hom,sr) = z_i(1) |
---|
783 | hom(nzb+7,1,var_hom,sr) = z_i(2) |
---|
784 | |
---|
785 | ! |
---|
786 | !-- Computation of both the characteristic vertical velocity and |
---|
787 | !-- the characteristic convective boundary layer temperature. |
---|
788 | !-- The horizontal average at nzb+1 is input for the average temperature. |
---|
789 | IF ( hom(nzb,1,18,sr) > 0.0 .AND. z_i(1) /= 0.0 ) THEN |
---|
790 | hom(nzb+8,1,var_hom,sr) = ( g / hom(nzb+1,1,4,sr) * & |
---|
791 | hom(nzb,1,18,sr) * z_i(1) )**0.333333333 |
---|
792 | !-- so far this only works if Prandtl layer is used |
---|
793 | hom(nzb+11,1,var_hom,sr) = hom(nzb,1,16,sr) / hom(nzb+8,1,var_hom,sr) |
---|
794 | ELSE |
---|
795 | hom(nzb+8,1,var_hom,sr) = 0.0 |
---|
796 | hom(nzb+11,1,var_hom,sr) = 0.0 |
---|
797 | ENDIF |
---|
798 | |
---|
799 | ENDDO ! loop of the subregions |
---|
800 | |
---|
801 | ! |
---|
802 | !-- Calculate additional statistics provided by the user interface |
---|
803 | CALL user_statistics |
---|
804 | |
---|
805 | ! |
---|
806 | !-- If required, sum up horizontal averages for subsequent time averaging |
---|
807 | IF ( do_sum ) THEN |
---|
808 | IF ( average_count_pr == 0 ) hom_sum = 0.0 |
---|
809 | hom_sum = hom_sum + hom(:,1,:,:) |
---|
810 | average_count_pr = average_count_pr + 1 |
---|
811 | do_sum = .FALSE. |
---|
812 | ENDIF |
---|
813 | |
---|
814 | ! |
---|
815 | !-- Set flag for other UPs (e.g. output routines, but also buoyancy). |
---|
816 | !-- This flag is reset after each time step in time_integration. |
---|
817 | flow_statistics_called = .TRUE. |
---|
818 | |
---|
819 | CALL cpu_log( log_point(10), 'flow_statistics', 'stop' ) |
---|
820 | |
---|
821 | |
---|
822 | END SUBROUTINE flow_statistics |
---|
823 | |
---|
824 | |
---|
825 | |
---|