1 | SUBROUTINE advec_s_bc( sk, sk_char ) |
---|
2 | |
---|
3 | !------------------------------------------------------------------------------! |
---|
4 | ! Current revisions: |
---|
5 | ! ----------------- |
---|
6 | ! |
---|
7 | ! |
---|
8 | ! Former revisions: |
---|
9 | ! ----------------- |
---|
10 | ! $Id: advec_s_bc.f90 392 2009-09-24 10:39:14Z raasch $ |
---|
11 | ! |
---|
12 | ! 247 2009-02-27 14:01:30Z heinze |
---|
13 | ! Output of messages replaced by message handling routine |
---|
14 | ! |
---|
15 | ! 216 2008-11-25 07:12:43Z raasch |
---|
16 | ! Neumann boundary condition at k=nzb is explicitly set for better reading, |
---|
17 | ! although this has been already done in boundary_conds |
---|
18 | ! |
---|
19 | ! 97 2007-06-21 08:23:15Z raasch |
---|
20 | ! Advection of salinity included |
---|
21 | ! Bugfix: Error in boundary condition for TKE removed |
---|
22 | ! |
---|
23 | ! 63 2007-03-13 03:52:49Z raasch |
---|
24 | ! Calculation extended for gridpoint nzt |
---|
25 | ! |
---|
26 | ! RCS Log replace by Id keyword, revision history cleaned up |
---|
27 | ! |
---|
28 | ! Revision 1.22 2006/02/23 09:42:08 raasch |
---|
29 | ! anz renamed ngp |
---|
30 | ! |
---|
31 | ! Revision 1.1 1997/08/29 08:53:46 raasch |
---|
32 | ! Initial revision |
---|
33 | ! |
---|
34 | ! |
---|
35 | ! Description: |
---|
36 | ! ------------ |
---|
37 | ! Advection term for scalar quantities using the Bott-Chlond scheme. |
---|
38 | ! Computation in individual steps for each of the three dimensions. |
---|
39 | ! Limiting assumptions: |
---|
40 | ! So far the scheme has been assuming equidistant grid spacing. As this is not |
---|
41 | ! the case in the stretched portion of the z-direction, there dzw(k) is used as |
---|
42 | ! a substitute for a constant grid length. This certainly causes incorrect |
---|
43 | ! results; however, it is hoped that they are not too apparent for weakly |
---|
44 | ! stretched grids. |
---|
45 | ! NOTE: This is a provisional, non-optimised version! |
---|
46 | !------------------------------------------------------------------------------! |
---|
47 | |
---|
48 | USE advection |
---|
49 | USE arrays_3d |
---|
50 | USE control_parameters |
---|
51 | USE cpulog |
---|
52 | USE grid_variables |
---|
53 | USE indices |
---|
54 | USE interfaces |
---|
55 | USE pegrid |
---|
56 | USE statistics |
---|
57 | |
---|
58 | IMPLICIT NONE |
---|
59 | |
---|
60 | CHARACTER (LEN=*) :: sk_char |
---|
61 | |
---|
62 | INTEGER :: i, ix, j, k, ngp, sr, type_xz_2 |
---|
63 | |
---|
64 | REAL :: cim, cimf, cip, cipf, d_new, ffmax, fminus, fplus, f2, f4, f8, & |
---|
65 | f12, f24, f48, f1920, im, ip, m2, m3, nenner, snenn, sterm, & |
---|
66 | tendenz, t1, t2, zaehler |
---|
67 | REAL :: fmax(2), fmax_l(2) |
---|
68 | REAL, DIMENSION(:,:,:), POINTER :: sk |
---|
69 | |
---|
70 | REAL, DIMENSION(:,:), ALLOCATABLE :: a0, a1, a12, a2, a22, immb, imme, & |
---|
71 | impb, impe, ipmb, ipme, ippb, ippe |
---|
72 | REAL, DIMENSION(:,:,:), ALLOCATABLE :: sk_p |
---|
73 | |
---|
74 | #if defined( __nec ) |
---|
75 | REAL (kind=4) :: m1n, m1z !Wichtig: Division |
---|
76 | REAL (kind=4), DIMENSION(:,:), ALLOCATABLE :: m1, sw |
---|
77 | #else |
---|
78 | REAL :: m1n, m1z |
---|
79 | REAL, DIMENSION(:,:), ALLOCATABLE :: m1, sw |
---|
80 | #endif |
---|
81 | |
---|
82 | |
---|
83 | ! |
---|
84 | !-- Array sk_p requires 2 extra elements for each dimension |
---|
85 | ALLOCATE( sk_p(nzb-2:nzt+3,nys-3:nyn+3,nxl-3:nxr+3) ) |
---|
86 | sk_p = 0.0 |
---|
87 | |
---|
88 | ! |
---|
89 | !-- Assign reciprocal values in order to avoid divisions later |
---|
90 | f2 = 0.5 |
---|
91 | f4 = 0.25 |
---|
92 | f8 = 0.125 |
---|
93 | f12 = 0.8333333333333333E-01 |
---|
94 | f24 = 0.4166666666666666E-01 |
---|
95 | f48 = 0.2083333333333333E-01 |
---|
96 | f1920 = 0.5208333333333333E-03 |
---|
97 | |
---|
98 | ! |
---|
99 | !-- Advection in x-direction: |
---|
100 | |
---|
101 | ! |
---|
102 | !-- Save the quantity to be advected in a local array |
---|
103 | !-- add an enlarged boundary in x-direction |
---|
104 | DO i = nxl-1, nxr+1 |
---|
105 | DO j = nys, nyn |
---|
106 | DO k = nzb, nzt+1 |
---|
107 | sk_p(k,j,i) = sk(k,j,i) |
---|
108 | ENDDO |
---|
109 | ENDDO |
---|
110 | ENDDO |
---|
111 | #if defined( __parallel ) |
---|
112 | ngp = 2 * ( nzt - nzb + 6 ) * ( nyn - nys + 7 ) |
---|
113 | CALL cpu_log( log_point_s(11), 'advec_s_bc:sendrecv', 'start' ) |
---|
114 | ! |
---|
115 | !-- Send left boundary, receive right boundary |
---|
116 | CALL MPI_SENDRECV( sk_p(nzb-2,nys-3,nxl+1), ngp, MPI_REAL, pleft, 0, & |
---|
117 | sk_p(nzb-2,nys-3,nxr+2), ngp, MPI_REAL, pright, 0, & |
---|
118 | comm2d, status, ierr ) |
---|
119 | ! |
---|
120 | !-- Send right boundary, receive left boundary |
---|
121 | CALL MPI_SENDRECV( sk_p(nzb-2,nys-3,nxr-2), ngp, MPI_REAL, pright, 1, & |
---|
122 | sk_p(nzb-2,nys-3,nxl-3), ngp, MPI_REAL, pleft, 1, & |
---|
123 | comm2d, status, ierr ) |
---|
124 | CALL cpu_log( log_point_s(11), 'advec_s_bc:sendrecv', 'pause' ) |
---|
125 | #else |
---|
126 | |
---|
127 | ! |
---|
128 | !-- Cyclic boundary conditions |
---|
129 | sk_p(:,nys:nyn,nxl-3) = sk_p(:,nys:nyn,nxr-2) |
---|
130 | sk_p(:,nys:nyn,nxl-2) = sk_p(:,nys:nyn,nxr-1) |
---|
131 | sk_p(:,nys:nyn,nxr+2) = sk_p(:,nys:nyn,nxl+1) |
---|
132 | sk_p(:,nys:nyn,nxr+3) = sk_p(:,nys:nyn,nxl+2) |
---|
133 | #endif |
---|
134 | |
---|
135 | ! |
---|
136 | !-- In case of a sloping surface, the additional gridpoints in x-direction |
---|
137 | !-- of the temperature field at the left and right boundary of the total |
---|
138 | !-- domain must be adjusted by the temperature difference between this distance |
---|
139 | IF ( sloping_surface .AND. sk_char == 'pt' ) THEN |
---|
140 | IF ( nxl == 0 ) THEN |
---|
141 | sk_p(:,nys:nyn,nxl-3) = sk_p(:,nys:nyn,nxl-3) - pt_slope_offset |
---|
142 | sk_p(:,nys:nyn,nxl-2) = sk_p(:,nys:nyn,nxl-2) - pt_slope_offset |
---|
143 | ENDIF |
---|
144 | IF ( nxr == nx ) THEN |
---|
145 | sk_p(:,nys:nyn,nxr+2) = sk_p(:,nys:nyn,nxr+2) + pt_slope_offset |
---|
146 | sk_p(:,nys:nyn,nxr+3) = sk_p(:,nys:nyn,nxr+3) + pt_slope_offset |
---|
147 | ENDIF |
---|
148 | ENDIF |
---|
149 | |
---|
150 | ! |
---|
151 | !-- Initialise control density |
---|
152 | d = 0.0 |
---|
153 | |
---|
154 | ! |
---|
155 | !-- Determine maxima of the first and second derivative in x-direction |
---|
156 | fmax_l = 0.0 |
---|
157 | DO i = nxl, nxr |
---|
158 | DO j = nys, nyn |
---|
159 | DO k = nzb+1, nzt |
---|
160 | zaehler = ABS( sk_p(k,j,i+1) - 2.0 * sk_p(k,j,i) + sk_p(k,j,i-1) ) |
---|
161 | nenner = ABS( sk_p(k,j,i+1) - sk_p(k,j,i-1) ) |
---|
162 | fmax_l(1) = MAX( fmax_l(1) , zaehler ) |
---|
163 | fmax_l(2) = MAX( fmax_l(2) , nenner ) |
---|
164 | ENDDO |
---|
165 | ENDDO |
---|
166 | ENDDO |
---|
167 | #if defined( __parallel ) |
---|
168 | CALL MPI_ALLREDUCE( fmax_l, fmax, 2, MPI_REAL, MPI_MAX, comm2d, ierr ) |
---|
169 | #else |
---|
170 | fmax = fmax_l |
---|
171 | #endif |
---|
172 | |
---|
173 | fmax = 0.04 * fmax |
---|
174 | |
---|
175 | ! |
---|
176 | !-- Allocate temporary arrays |
---|
177 | ALLOCATE( a0(nzb+1:nzt,nxl-1:nxr+1), a1(nzb+1:nzt,nxl-1:nxr+1), & |
---|
178 | a2(nzb+1:nzt,nxl-1:nxr+1), a12(nzb+1:nzt,nxl-1:nxr+1), & |
---|
179 | a22(nzb+1:nzt,nxl-1:nxr+1), immb(nzb+1:nzt,nxl-1:nxr+1), & |
---|
180 | imme(nzb+1:nzt,nxl-1:nxr+1), impb(nzb+1:nzt,nxl-1:nxr+1), & |
---|
181 | impe(nzb+1:nzt,nxl-1:nxr+1), ipmb(nzb+1:nzt,nxl-1:nxr+1), & |
---|
182 | ipme(nzb+1:nzt,nxl-1:nxr+1), ippb(nzb+1:nzt,nxl-1:nxr+1), & |
---|
183 | ippe(nzb+1:nzt,nxl-1:nxr+1), m1(nzb+1:nzt,nxl-2:nxr+2), & |
---|
184 | sw(nzb+1:nzt,nxl-1:nxr+1) & |
---|
185 | ) |
---|
186 | imme = 0.0; impe = 0.0; ipme = 0.0; ippe = 0.0 |
---|
187 | |
---|
188 | ! |
---|
189 | !-- Initialise point of time measuring of the exponential portion (this would |
---|
190 | !-- not work if done locally within the loop) |
---|
191 | CALL cpu_log( log_point_s(12), 'advec_s_bc:exp', 'start' ) |
---|
192 | CALL cpu_log( log_point_s(12), 'advec_s_bc:exp', 'pause' ) |
---|
193 | |
---|
194 | ! |
---|
195 | !-- Outer loop of all j |
---|
196 | DO j = nys, nyn |
---|
197 | |
---|
198 | ! |
---|
199 | !-- Compute polynomial coefficients |
---|
200 | DO i = nxl-1, nxr+1 |
---|
201 | DO k = nzb+1, nzt |
---|
202 | a12(k,i) = 0.5 * ( sk_p(k,j,i+1) - sk_p(k,j,i-1) ) |
---|
203 | a22(k,i) = 0.5 * ( sk_p(k,j,i+1) - 2.0 * sk_p(k,j,i) & |
---|
204 | + sk_p(k,j,i-1) ) |
---|
205 | a0(k,i) = ( 9.0 * sk_p(k,j,i+2) - 116.0 * sk_p(k,j,i+1) & |
---|
206 | + 2134.0 * sk_p(k,j,i) - 116.0 * sk_p(k,j,i-1) & |
---|
207 | + 9.0 * sk_p(k,j,i-2) ) * f1920 |
---|
208 | a1(k,i) = ( -5.0 * sk_p(k,j,i+2) + 34.0 * sk_p(k,j,i+1) & |
---|
209 | - 34.0 * sk_p(k,j,i-1) + 5.0 * sk_p(k,j,i-2) & |
---|
210 | ) * f48 |
---|
211 | a2(k,i) = ( -3.0 * sk_p(k,j,i+2) + 36.0 * sk_p(k,j,i+1) & |
---|
212 | - 66.0 * sk_p(k,j,i) + 36.0 * sk_p(k,j,i-1) & |
---|
213 | - 3.0 * sk_p(k,j,i-2) ) * f48 |
---|
214 | ENDDO |
---|
215 | ENDDO |
---|
216 | |
---|
217 | ! |
---|
218 | !-- Fluxes using the Bott scheme |
---|
219 | !-- *VOCL LOOP,UNROLL(2) |
---|
220 | DO i = nxl, nxr |
---|
221 | DO k = nzb+1, nzt |
---|
222 | cip = MAX( 0.0, ( u(k,j,i+1) - u_gtrans ) * dt_3d * ddx ) |
---|
223 | cim = -MIN( 0.0, ( u(k,j,i+1) - u_gtrans ) * dt_3d * ddx ) |
---|
224 | cipf = 1.0 - 2.0 * cip |
---|
225 | cimf = 1.0 - 2.0 * cim |
---|
226 | ip = a0(k,i) * f2 * ( 1.0 - cipf ) & |
---|
227 | + a1(k,i) * f8 * ( 1.0 - cipf*cipf ) & |
---|
228 | + a2(k,i) * f24 * ( 1.0 - cipf*cipf*cipf ) |
---|
229 | im = a0(k,i+1) * f2 * ( 1.0 - cimf ) & |
---|
230 | - a1(k,i+1) * f8 * ( 1.0 - cimf*cimf ) & |
---|
231 | + a2(k,i+1) * f24 * ( 1.0 - cimf*cimf*cimf ) |
---|
232 | ip = MAX( ip, 0.0 ) |
---|
233 | im = MAX( im, 0.0 ) |
---|
234 | ippb(k,i) = ip * MIN( 1.0, sk_p(k,j,i) / (ip+im+1E-15) ) |
---|
235 | impb(k,i) = im * MIN( 1.0, sk_p(k,j,i+1) / (ip+im+1E-15) ) |
---|
236 | |
---|
237 | cip = MAX( 0.0, ( u(k,j,i) - u_gtrans ) * dt_3d * ddx ) |
---|
238 | cim = -MIN( 0.0, ( u(k,j,i) - u_gtrans ) * dt_3d * ddx ) |
---|
239 | cipf = 1.0 - 2.0 * cip |
---|
240 | cimf = 1.0 - 2.0 * cim |
---|
241 | ip = a0(k,i-1) * f2 * ( 1.0 - cipf ) & |
---|
242 | + a1(k,i-1) * f8 * ( 1.0 - cipf*cipf ) & |
---|
243 | + a2(k,i-1) * f24 * ( 1.0 - cipf*cipf*cipf ) |
---|
244 | im = a0(k,i) * f2 * ( 1.0 - cimf ) & |
---|
245 | - a1(k,i) * f8 * ( 1.0 - cimf*cimf ) & |
---|
246 | + a2(k,i) * f24 * ( 1.0 - cimf*cimf*cimf ) |
---|
247 | ip = MAX( ip, 0.0 ) |
---|
248 | im = MAX( im, 0.0 ) |
---|
249 | ipmb(k,i) = ip * MIN( 1.0, sk_p(k,j,i-1) / (ip+im+1E-15) ) |
---|
250 | immb(k,i) = im * MIN( 1.0, sk_p(k,j,i) / (ip+im+1E-15) ) |
---|
251 | ENDDO |
---|
252 | ENDDO |
---|
253 | |
---|
254 | ! |
---|
255 | !-- Compute monitor function m1 |
---|
256 | DO i = nxl-2, nxr+2 |
---|
257 | DO k = nzb+1, nzt |
---|
258 | m1z = ABS( sk_p(k,j,i+1) - 2.0 * sk_p(k,j,i) + sk_p(k,j,i-1) ) |
---|
259 | m1n = ABS( sk_p(k,j,i+1) - sk_p(k,j,i-1) ) |
---|
260 | IF ( m1n /= 0.0 .AND. m1n >= m1z ) THEN |
---|
261 | m1(k,i) = m1z / m1n |
---|
262 | IF ( m1(k,i) /= 2.0 .AND. m1n < fmax(2) ) m1(k,i) = 0.0 |
---|
263 | ELSEIF ( m1n < m1z ) THEN |
---|
264 | m1(k,i) = -1.0 |
---|
265 | ELSE |
---|
266 | m1(k,i) = 0.0 |
---|
267 | ENDIF |
---|
268 | ENDDO |
---|
269 | ENDDO |
---|
270 | |
---|
271 | ! |
---|
272 | !-- Compute switch sw |
---|
273 | sw = 0.0 |
---|
274 | DO i = nxl-1, nxr+1 |
---|
275 | DO k = nzb+1, nzt |
---|
276 | m2 = 2.0 * ABS( a1(k,i) - a12(k,i) ) / & |
---|
277 | MAX( ABS( a1(k,i) + a12(k,i) ), 1E-35 ) |
---|
278 | IF ( ABS( a1(k,i) + a12(k,i) ) < fmax(2) ) m2 = 0.0 |
---|
279 | |
---|
280 | m3 = 2.0 * ABS( a2(k,i) - a22(k,i) ) / & |
---|
281 | MAX( ABS( a2(k,i) + a22(k,i) ), 1E-35 ) |
---|
282 | IF ( ABS( a2(k,i) + a22(k,i) ) < fmax(1) ) m3 = 0.0 |
---|
283 | |
---|
284 | t1 = 0.35 |
---|
285 | t2 = 0.35 |
---|
286 | IF ( m1(k,i) == -1.0 ) t2 = 0.12 |
---|
287 | |
---|
288 | !-- *VOCL STMT,IF(10) |
---|
289 | IF ( m1(k,i-1) == 1.0 .OR. m1(k,i) == 1.0 .OR. m1(k,i+1) == 1.0 & |
---|
290 | .OR. m2 > t2 .OR. m3 > T2 .OR. & |
---|
291 | ( m1(k,i) > t1 .AND. m1(k,i-1) /= -1.0 .AND. & |
---|
292 | m1(k,i) /= -1.0 .AND. m1(k,i+1) /= -1.0 ) & |
---|
293 | ) sw(k,i) = 1.0 |
---|
294 | ENDDO |
---|
295 | ENDDO |
---|
296 | |
---|
297 | ! |
---|
298 | !-- Fluxes using the exponential scheme |
---|
299 | CALL cpu_log( log_point_s(12), 'advec_s_bc:exp', 'continue' ) |
---|
300 | DO i = nxl, nxr |
---|
301 | DO k = nzb+1, nzt |
---|
302 | |
---|
303 | !-- *VOCL STMT,IF(10) |
---|
304 | IF ( sw(k,i) == 1.0 ) THEN |
---|
305 | snenn = sk_p(k,j,i+1) - sk_p(k,j,i-1) |
---|
306 | IF ( ABS( snenn ) < 1E-9 ) snenn = 1E-9 |
---|
307 | sterm = ( sk_p(k,j,i) - sk_p(k,j,i-1) ) / snenn |
---|
308 | sterm = MIN( sterm, 0.9999 ) |
---|
309 | sterm = MAX( sterm, 0.0001 ) |
---|
310 | |
---|
311 | ix = INT( sterm * 1000 ) + 1 |
---|
312 | |
---|
313 | cip = MAX( 0.0, ( u(k,j,i+1) - u_gtrans ) * dt_3d * ddx ) |
---|
314 | |
---|
315 | ippe(k,i) = sk_p(k,j,i-1) * cip + snenn * ( & |
---|
316 | aex(ix) * cip + bex(ix) / dex(ix) * ( & |
---|
317 | eex(ix) - EXP( dex(ix)*0.5 * ( 1.0 - 2.0 * cip ) ) & |
---|
318 | ) & |
---|
319 | ) |
---|
320 | IF ( sterm == 0.0001 ) ippe(k,i) = sk_p(k,j,i) * cip |
---|
321 | IF ( sterm == 0.9999 ) ippe(k,i) = sk_p(k,j,i) * cip |
---|
322 | |
---|
323 | snenn = sk_p(k,j,i-1) - sk_p(k,j,i+1) |
---|
324 | IF ( ABS( snenn ) < 1E-9 ) snenn = 1E-9 |
---|
325 | sterm = ( sk_p(k,j,i) - sk_p(k,j,i+1) ) / snenn |
---|
326 | sterm = MIN( sterm, 0.9999 ) |
---|
327 | sterm = MAX( sterm, 0.0001 ) |
---|
328 | |
---|
329 | ix = INT( sterm * 1000 ) + 1 |
---|
330 | |
---|
331 | cim = -MIN( 0.0, ( u(k,j,i) - u_gtrans ) * dt_3d * ddx ) |
---|
332 | |
---|
333 | imme(k,i) = sk_p(k,j,i+1) * cim + snenn * ( & |
---|
334 | aex(ix) * cim + bex(ix) / dex(ix) * ( & |
---|
335 | eex(ix) - EXP( dex(ix)*0.5 * ( 1.0 - 2.0 * cim ) ) & |
---|
336 | ) & |
---|
337 | ) |
---|
338 | IF ( sterm == 0.0001 ) imme(k,i) = sk_p(k,j,i) * cim |
---|
339 | IF ( sterm == 0.9999 ) imme(k,i) = sk_p(k,j,i) * cim |
---|
340 | ENDIF |
---|
341 | |
---|
342 | !-- *VOCL STMT,IF(10) |
---|
343 | IF ( sw(k,i+1) == 1.0 ) THEN |
---|
344 | snenn = sk_p(k,j,i) - sk_p(k,j,i+2) |
---|
345 | IF ( ABS( snenn ) .LT. 1E-9 ) snenn = 1E-9 |
---|
346 | sterm = ( sk_p(k,j,i+1) - sk_p(k,j,i+2) ) / snenn |
---|
347 | sterm = MIN( sterm, 0.9999 ) |
---|
348 | sterm = MAX( sterm, 0.0001 ) |
---|
349 | |
---|
350 | ix = INT( sterm * 1000 ) + 1 |
---|
351 | |
---|
352 | cim = -MIN( 0.0, ( u(k,j,i+1) - u_gtrans ) * dt_3d * ddx ) |
---|
353 | |
---|
354 | impe(k,i) = sk_p(k,j,i+2) * cim + snenn * ( & |
---|
355 | aex(ix) * cim + bex(ix) / dex(ix) * ( & |
---|
356 | eex(ix) - EXP( dex(ix)*0.5 * ( 1.0 - 2.0 * cim ) ) & |
---|
357 | ) & |
---|
358 | ) |
---|
359 | IF ( sterm == 0.0001 ) impe(k,i) = sk_p(k,j,i+1) * cim |
---|
360 | IF ( sterm == 0.9999 ) impe(k,i) = sk_p(k,j,i+1) * cim |
---|
361 | ENDIF |
---|
362 | |
---|
363 | !-- *VOCL STMT,IF(10) |
---|
364 | IF ( sw(k,i-1) == 1.0 ) THEN |
---|
365 | snenn = sk_p(k,j,i) - sk_p(k,j,i-2) |
---|
366 | IF ( ABS( snenn ) < 1E-9 ) snenn = 1E-9 |
---|
367 | sterm = ( sk_p(k,j,i-1) - sk_p(k,j,i-2) ) / snenn |
---|
368 | sterm = MIN( sterm, 0.9999 ) |
---|
369 | sterm = MAX( sterm, 0.0001 ) |
---|
370 | |
---|
371 | ix = INT( sterm * 1000 ) + 1 |
---|
372 | |
---|
373 | cip = MAX( 0.0, ( u(k,j,i) - u_gtrans ) * dt_3d * ddx ) |
---|
374 | |
---|
375 | ipme(k,i) = sk_p(k,j,i-2) * cip + snenn * ( & |
---|
376 | aex(ix) * cip + bex(ix) / dex(ix) * ( & |
---|
377 | eex(ix) - EXP( dex(ix)*0.5 * ( 1.0 - 2.0 * cip ) ) & |
---|
378 | ) & |
---|
379 | ) |
---|
380 | IF ( sterm == 0.0001 ) ipme(k,i) = sk_p(k,j,i-1) * cip |
---|
381 | IF ( sterm == 0.9999 ) ipme(k,i) = sk_p(k,j,i-1) * cip |
---|
382 | ENDIF |
---|
383 | |
---|
384 | ENDDO |
---|
385 | ENDDO |
---|
386 | CALL cpu_log( log_point_s(12), 'advec_s_bc:exp', 'pause' ) |
---|
387 | |
---|
388 | ! |
---|
389 | !-- Prognostic equation |
---|
390 | DO i = nxl, nxr |
---|
391 | DO k = nzb+1, nzt |
---|
392 | fplus = ( 1.0 - sw(k,i) ) * ippb(k,i) + sw(k,i) * ippe(k,i) & |
---|
393 | - ( 1.0 - sw(k,i+1) ) * impb(k,i) - sw(k,i+1) * impe(k,i) |
---|
394 | fminus = ( 1.0 - sw(k,i-1) ) * ipmb(k,i) + sw(k,i-1) * ipme(k,i) & |
---|
395 | - ( 1.0 - sw(k,i) ) * immb(k,i) - sw(k,i) * imme(k,i) |
---|
396 | tendenz = fplus - fminus |
---|
397 | ! |
---|
398 | !-- Removed in order to optimize speed |
---|
399 | ! ffmax = MAX( ABS( fplus ), ABS( fminus ), 1E-35 ) |
---|
400 | ! IF ( ( ABS( tendenz ) / ffmax ) < 1E-7 ) tendenz = 0.0 |
---|
401 | ! |
---|
402 | !-- Density correction because of possible remaining divergences |
---|
403 | d_new = d(k,j,i) - ( u(k,j,i+1) - u(k,j,i) ) * dt_3d * ddx |
---|
404 | sk_p(k,j,i) = ( ( 1.0 + d(k,j,i) ) * sk_p(k,j,i) - tendenz ) / & |
---|
405 | ( 1.0 + d_new ) |
---|
406 | d(k,j,i) = d_new |
---|
407 | ENDDO |
---|
408 | ENDDO |
---|
409 | |
---|
410 | ENDDO ! End of the advection in x-direction |
---|
411 | |
---|
412 | ! |
---|
413 | !-- Deallocate temporary arrays |
---|
414 | DEALLOCATE( a0, a1, a2, a12, a22, immb, imme, impb, impe, ipmb, ipme, & |
---|
415 | ippb, ippe, m1, sw ) |
---|
416 | |
---|
417 | |
---|
418 | ! |
---|
419 | !-- Enlarge boundary of local array cyclically in y-direction |
---|
420 | #if defined( __parallel ) |
---|
421 | ngp = ( nzt - nzb + 6 ) * ( nyn - nys + 7 ) |
---|
422 | CALL MPI_TYPE_VECTOR( nxr-nxl+7, 3*(nzt-nzb+6), ngp, MPI_REAL, & |
---|
423 | type_xz_2, ierr ) |
---|
424 | CALL MPI_TYPE_COMMIT( type_xz_2, ierr ) |
---|
425 | ! |
---|
426 | !-- Send front boundary, receive rear boundary |
---|
427 | CALL cpu_log( log_point_s(11), 'advec_s_bc:sendrecv', 'continue' ) |
---|
428 | CALL MPI_SENDRECV( sk_p(nzb-2,nys,nxl-3), 1, type_xz_2, psouth, 0, & |
---|
429 | sk_p(nzb-2,nyn+1,nxl-3), 1, type_xz_2, pnorth, 0, & |
---|
430 | comm2d, status, ierr ) |
---|
431 | ! |
---|
432 | !-- Send rear boundary, receive front boundary |
---|
433 | CALL MPI_SENDRECV( sk_p(nzb-2,nyn-2,nxl-3), 1, type_xz_2, pnorth, 1, & |
---|
434 | sk_p(nzb-2,nys-3,nxl-3), 1, type_xz_2, psouth, 1, & |
---|
435 | comm2d, status, ierr ) |
---|
436 | CALL MPI_TYPE_FREE( type_xz_2, ierr ) |
---|
437 | CALL cpu_log( log_point_s(11), 'advec_s_bc:sendrecv', 'pause' ) |
---|
438 | #else |
---|
439 | DO i = nxl, nxr |
---|
440 | DO k = nzb+1, nzt |
---|
441 | sk_p(k,nys-1,i) = sk_p(k,nyn,i) |
---|
442 | sk_p(k,nys-2,i) = sk_p(k,nyn-1,i) |
---|
443 | sk_p(k,nys-3,i) = sk_p(k,nyn-2,i) |
---|
444 | sk_p(k,nyn+1,i) = sk_p(k,nys,i) |
---|
445 | sk_p(k,nyn+2,i) = sk_p(k,nys+1,i) |
---|
446 | sk_p(k,nyn+3,i) = sk_p(k,nys+2,i) |
---|
447 | ENDDO |
---|
448 | ENDDO |
---|
449 | #endif |
---|
450 | |
---|
451 | ! |
---|
452 | !-- Determine the maxima of the first and second derivative in y-direction |
---|
453 | fmax_l = 0.0 |
---|
454 | DO i = nxl, nxr |
---|
455 | DO j = nys, nyn |
---|
456 | DO k = nzb+1, nzt |
---|
457 | zaehler = ABS( sk_p(k,j+1,i) - 2.0 * sk_p(k,j,i) + sk_p(k,j-1,i) ) |
---|
458 | nenner = ABS( sk_p(k,j+1,i) - sk_p(k,j-1,i) ) |
---|
459 | fmax_l(1) = MAX( fmax_l(1) , zaehler ) |
---|
460 | fmax_l(2) = MAX( fmax_l(2) , nenner ) |
---|
461 | ENDDO |
---|
462 | ENDDO |
---|
463 | ENDDO |
---|
464 | #if defined( __parallel ) |
---|
465 | CALL MPI_ALLREDUCE( fmax_l, fmax, 2, MPI_REAL, MPI_MAX, comm2d, ierr ) |
---|
466 | #else |
---|
467 | fmax = fmax_l |
---|
468 | #endif |
---|
469 | |
---|
470 | fmax = 0.04 * fmax |
---|
471 | |
---|
472 | ! |
---|
473 | !-- Allocate temporary arrays |
---|
474 | ALLOCATE( a0(nzb+1:nzt,nys-1:nyn+1), a1(nzb+1:nzt,nys-1:nyn+1), & |
---|
475 | a2(nzb+1:nzt,nys-1:nyn+1), a12(nzb+1:nzt,nys-1:nyn+1), & |
---|
476 | a22(nzb+1:nzt,nys-1:nyn+1), immb(nzb+1:nzt,nys-1:nyn+1), & |
---|
477 | imme(nzb+1:nzt,nys-1:nyn+1), impb(nzb+1:nzt,nys-1:nyn+1), & |
---|
478 | impe(nzb+1:nzt,nys-1:nyn+1), ipmb(nzb+1:nzt,nys-1:nyn+1), & |
---|
479 | ipme(nzb+1:nzt,nys-1:nyn+1), ippb(nzb+1:nzt,nys-1:nyn+1), & |
---|
480 | ippe(nzb+1:nzt,nys-1:nyn+1), m1(nzb+1:nzt,nys-2:nyn+2), & |
---|
481 | sw(nzb+1:nzt,nys-1:nyn+1) & |
---|
482 | ) |
---|
483 | imme = 0.0; impe = 0.0; ipme = 0.0; ippe = 0.0 |
---|
484 | |
---|
485 | ! |
---|
486 | !-- Outer loop of all i |
---|
487 | DO i = nxl, nxr |
---|
488 | |
---|
489 | ! |
---|
490 | !-- Compute polynomial coefficients |
---|
491 | DO j = nys-1, nyn+1 |
---|
492 | DO k = nzb+1, nzt |
---|
493 | a12(k,j) = 0.5 * ( sk_p(k,j+1,i) - sk_p(k,j-1,i) ) |
---|
494 | a22(k,j) = 0.5 * ( sk_p(k,j+1,i) - 2.0 * sk_p(k,j,i) & |
---|
495 | + sk_p(k,j-1,i) ) |
---|
496 | a0(k,j) = ( 9.0 * sk_p(k,j+2,i) - 116.0 * sk_p(k,j+1,i) & |
---|
497 | + 2134.0 * sk_p(k,j,i) - 116.0 * sk_p(k,j-1,i) & |
---|
498 | + 9.0 * sk_p(k,j-2,i) ) * f1920 |
---|
499 | a1(k,j) = ( -5.0 * sk_p(k,j+2,i) + 34.0 * sk_p(k,j+1,i) & |
---|
500 | - 34.0 * sk_p(k,j-1,i) + 5.0 * sk_p(k,j-2,i) & |
---|
501 | ) * f48 |
---|
502 | a2(k,j) = ( -3.0 * sk_p(k,j+2,i) + 36.0 * sk_p(k,j+1,i) & |
---|
503 | - 66.0 * sk_p(k,j,i) + 36.0 * sk_p(k,j-1,i) & |
---|
504 | - 3.0 * sk_p(k,j-2,i) ) * f48 |
---|
505 | ENDDO |
---|
506 | ENDDO |
---|
507 | |
---|
508 | ! |
---|
509 | !-- Fluxes using the Bott scheme |
---|
510 | !-- *VOCL LOOP,UNROLL(2) |
---|
511 | DO j = nys, nyn |
---|
512 | DO k = nzb+1, nzt |
---|
513 | cip = MAX( 0.0, ( v(k,j+1,i) - v_gtrans ) * dt_3d * ddy ) |
---|
514 | cim = -MIN( 0.0, ( v(k,j+1,i) - v_gtrans ) * dt_3d * ddy ) |
---|
515 | cipf = 1.0 - 2.0 * cip |
---|
516 | cimf = 1.0 - 2.0 * cim |
---|
517 | ip = a0(k,j) * f2 * ( 1.0 - cipf ) & |
---|
518 | + a1(k,j) * f8 * ( 1.0 - cipf*cipf ) & |
---|
519 | + a2(k,j) * f24 * ( 1.0 - cipf*cipf*cipf ) |
---|
520 | im = a0(k,j+1) * f2 * ( 1.0 - cimf ) & |
---|
521 | - a1(k,j+1) * f8 * ( 1.0 - cimf*cimf ) & |
---|
522 | + a2(k,j+1) * f24 * ( 1.0 - cimf*cimf*cimf ) |
---|
523 | ip = MAX( ip, 0.0 ) |
---|
524 | im = MAX( im, 0.0 ) |
---|
525 | ippb(k,j) = ip * MIN( 1.0, sk_p(k,j,i) / (ip+im+1E-15) ) |
---|
526 | impb(k,j) = im * MIN( 1.0, sk_p(k,j+1,i) / (ip+im+1E-15) ) |
---|
527 | |
---|
528 | cip = MAX( 0.0, ( v(k,j,i) - v_gtrans ) * dt_3d * ddy ) |
---|
529 | cim = -MIN( 0.0, ( v(k,j,i) - v_gtrans ) * dt_3d * ddy ) |
---|
530 | cipf = 1.0 - 2.0 * cip |
---|
531 | cimf = 1.0 - 2.0 * cim |
---|
532 | ip = a0(k,j-1) * f2 * ( 1.0 - cipf ) & |
---|
533 | + a1(k,j-1) * f8 * ( 1.0 - cipf*cipf ) & |
---|
534 | + a2(k,j-1) * f24 * ( 1.0 - cipf*cipf*cipf ) |
---|
535 | im = a0(k,j) * f2 * ( 1.0 - cimf ) & |
---|
536 | - a1(k,j) * f8 * ( 1.0 - cimf*cimf ) & |
---|
537 | + a2(k,j) * f24 * ( 1.0 - cimf*cimf*cimf ) |
---|
538 | ip = MAX( ip, 0.0 ) |
---|
539 | im = MAX( im, 0.0 ) |
---|
540 | ipmb(k,j) = ip * MIN( 1.0, sk_p(k,j-1,i) / (ip+im+1E-15) ) |
---|
541 | immb(k,j) = im * MIN( 1.0, sk_p(k,j,i) / (ip+im+1E-15) ) |
---|
542 | ENDDO |
---|
543 | ENDDO |
---|
544 | |
---|
545 | ! |
---|
546 | !-- Compute monitor function m1 |
---|
547 | DO j = nys-2, nyn+2 |
---|
548 | DO k = nzb+1, nzt |
---|
549 | m1z = ABS( sk_p(k,j+1,i) - 2.0 * sk_p(k,j,i) + sk_p(k,j-1,i) ) |
---|
550 | m1n = ABS( sk_p(k,j+1,i) - sk_p(k,j-1,i) ) |
---|
551 | IF ( m1n /= 0.0 .AND. m1n >= m1z ) THEN |
---|
552 | m1(k,j) = m1z / m1n |
---|
553 | IF ( m1(k,j) /= 2.0 .AND. m1n < fmax(2) ) m1(k,j) = 0.0 |
---|
554 | ELSEIF ( m1n < m1z ) THEN |
---|
555 | m1(k,j) = -1.0 |
---|
556 | ELSE |
---|
557 | m1(k,j) = 0.0 |
---|
558 | ENDIF |
---|
559 | ENDDO |
---|
560 | ENDDO |
---|
561 | |
---|
562 | ! |
---|
563 | !-- Compute switch sw |
---|
564 | sw = 0.0 |
---|
565 | DO j = nys-1, nyn+1 |
---|
566 | DO k = nzb+1, nzt |
---|
567 | m2 = 2.0 * ABS( a1(k,j) - a12(k,j) ) / & |
---|
568 | MAX( ABS( a1(k,j) + a12(k,j) ), 1E-35 ) |
---|
569 | IF ( ABS( a1(k,j) + a12(k,j) ) < fmax(2) ) m2 = 0.0 |
---|
570 | |
---|
571 | m3 = 2.0 * ABS( a2(k,j) - a22(k,j) ) / & |
---|
572 | MAX( ABS( a2(k,j) + a22(k,j) ), 1E-35 ) |
---|
573 | IF ( ABS( a2(k,j) + a22(k,j) ) < fmax(1) ) m3 = 0.0 |
---|
574 | |
---|
575 | t1 = 0.35 |
---|
576 | t2 = 0.35 |
---|
577 | IF ( m1(k,j) == -1.0 ) t2 = 0.12 |
---|
578 | |
---|
579 | !-- *VOCL STMT,IF(10) |
---|
580 | IF ( m1(k,j-1) == 1.0 .OR. m1(k,j) == 1.0 .OR. m1(k,j+1) == 1.0 & |
---|
581 | .OR. m2 > t2 .OR. m3 > T2 .OR. & |
---|
582 | ( m1(k,j) > t1 .AND. m1(k,j-1) /= -1.0 .AND. & |
---|
583 | m1(k,j) /= -1.0 .AND. m1(k,j+1) /= -1.0 ) & |
---|
584 | ) sw(k,j) = 1.0 |
---|
585 | ENDDO |
---|
586 | ENDDO |
---|
587 | |
---|
588 | ! |
---|
589 | !-- Fluxes using exponential scheme |
---|
590 | CALL cpu_log( log_point_s(12), 'advec_s_bc:exp', 'continue' ) |
---|
591 | DO j = nys, nyn |
---|
592 | DO k = nzb+1, nzt |
---|
593 | |
---|
594 | !-- *VOCL STMT,IF(10) |
---|
595 | IF ( sw(k,j) == 1.0 ) THEN |
---|
596 | snenn = sk_p(k,j+1,i) - sk_p(k,j-1,i) |
---|
597 | IF ( ABS( snenn ) < 1E-9 ) snenn = 1E-9 |
---|
598 | sterm = ( sk_p(k,j,i) - sk_p(k,j-1,i) ) / snenn |
---|
599 | sterm = MIN( sterm, 0.9999 ) |
---|
600 | sterm = MAX( sterm, 0.0001 ) |
---|
601 | |
---|
602 | ix = INT( sterm * 1000 ) + 1 |
---|
603 | |
---|
604 | cip = MAX( 0.0, ( v(k,j+1,i) - v_gtrans ) * dt_3d * ddy ) |
---|
605 | |
---|
606 | ippe(k,j) = sk_p(k,j-1,i) * cip + snenn * ( & |
---|
607 | aex(ix) * cip + bex(ix) / dex(ix) * ( & |
---|
608 | eex(ix) - EXP( dex(ix)*0.5 * ( 1.0 - 2.0 * cip ) ) & |
---|
609 | ) & |
---|
610 | ) |
---|
611 | IF ( sterm == 0.0001 ) ippe(k,j) = sk_p(k,j,i) * cip |
---|
612 | IF ( sterm == 0.9999 ) ippe(k,j) = sk_p(k,j,i) * cip |
---|
613 | |
---|
614 | snenn = sk_p(k,j-1,i) - sk_p(k,j+1,i) |
---|
615 | IF ( ABS( snenn ) < 1E-9 ) snenn = 1E-9 |
---|
616 | sterm = ( sk_p(k,j,i) - sk_p(k,j+1,i) ) / snenn |
---|
617 | sterm = MIN( sterm, 0.9999 ) |
---|
618 | sterm = MAX( sterm, 0.0001 ) |
---|
619 | |
---|
620 | ix = INT( sterm * 1000 ) + 1 |
---|
621 | |
---|
622 | cim = -MIN( 0.0, ( v(k,j,i) - v_gtrans ) * dt_3d * ddy ) |
---|
623 | |
---|
624 | imme(k,j) = sk_p(k,j+1,i) * cim + snenn * ( & |
---|
625 | aex(ix) * cim + bex(ix) / dex(ix) * ( & |
---|
626 | eex(ix) - EXP( dex(ix)*0.5 * ( 1.0 - 2.0 * cim ) ) & |
---|
627 | ) & |
---|
628 | ) |
---|
629 | IF ( sterm == 0.0001 ) imme(k,j) = sk_p(k,j,i) * cim |
---|
630 | IF ( sterm == 0.9999 ) imme(k,j) = sk_p(k,j,i) * cim |
---|
631 | ENDIF |
---|
632 | |
---|
633 | !-- *VOCL STMT,IF(10) |
---|
634 | IF ( sw(k,j+1) == 1.0 ) THEN |
---|
635 | snenn = sk_p(k,j,i) - sk_p(k,j+2,i) |
---|
636 | IF ( ABS( snenn ) .LT. 1E-9 ) snenn = 1E-9 |
---|
637 | sterm = ( sk_p(k,j+1,i) - sk_p(k,j+2,i) ) / snenn |
---|
638 | sterm = MIN( sterm, 0.9999 ) |
---|
639 | sterm = MAX( sterm, 0.0001 ) |
---|
640 | |
---|
641 | ix = INT( sterm * 1000 ) + 1 |
---|
642 | |
---|
643 | cim = -MIN( 0.0, ( v(k,j+1,i) - v_gtrans ) * dt_3d * ddy ) |
---|
644 | |
---|
645 | impe(k,j) = sk_p(k,j+2,i) * cim + snenn * ( & |
---|
646 | aex(ix) * cim + bex(ix) / dex(ix) * ( & |
---|
647 | eex(ix) - EXP( dex(ix)*0.5 * ( 1.0 - 2.0 * cim ) ) & |
---|
648 | ) & |
---|
649 | ) |
---|
650 | IF ( sterm == 0.0001 ) impe(k,j) = sk_p(k,j+1,i) * cim |
---|
651 | IF ( sterm == 0.9999 ) impe(k,j) = sk_p(k,j+1,i) * cim |
---|
652 | ENDIF |
---|
653 | |
---|
654 | !-- *VOCL STMT,IF(10) |
---|
655 | IF ( sw(k,j-1) == 1.0 ) THEN |
---|
656 | snenn = sk_p(k,j,i) - sk_p(k,j-2,i) |
---|
657 | IF ( ABS( snenn ) < 1E-9 ) snenn = 1E-9 |
---|
658 | sterm = ( sk_p(k,j-1,i) - sk_p(k,j-2,i) ) / snenn |
---|
659 | sterm = MIN( sterm, 0.9999 ) |
---|
660 | sterm = MAX( sterm, 0.0001 ) |
---|
661 | |
---|
662 | ix = INT( sterm * 1000 ) + 1 |
---|
663 | |
---|
664 | cip = MAX( 0.0, ( v(k,j,i) - v_gtrans ) * dt_3d * ddy ) |
---|
665 | |
---|
666 | ipme(k,j) = sk_p(k,j-2,i) * cip + snenn * ( & |
---|
667 | aex(ix) * cip + bex(ix) / dex(ix) * ( & |
---|
668 | eex(ix) - EXP( dex(ix)*0.5 * ( 1.0 - 2.0 * cip ) ) & |
---|
669 | ) & |
---|
670 | ) |
---|
671 | IF ( sterm == 0.0001 ) ipme(k,j) = sk_p(k,j-1,i) * cip |
---|
672 | IF ( sterm == 0.9999 ) ipme(k,j) = sk_p(k,j-1,i) * cip |
---|
673 | ENDIF |
---|
674 | |
---|
675 | ENDDO |
---|
676 | ENDDO |
---|
677 | CALL cpu_log( log_point_s(12), 'advec_s_bc:exp', 'pause' ) |
---|
678 | |
---|
679 | ! |
---|
680 | !-- Prognostic equation |
---|
681 | DO j = nys, nyn |
---|
682 | DO k = nzb+1, nzt |
---|
683 | fplus = ( 1.0 - sw(k,j) ) * ippb(k,j) + sw(k,j) * ippe(k,j) & |
---|
684 | - ( 1.0 - sw(k,j+1) ) * impb(k,j) - sw(k,j+1) * impe(k,j) |
---|
685 | fminus = ( 1.0 - sw(k,j-1) ) * ipmb(k,j) + sw(k,j-1) * ipme(k,j) & |
---|
686 | - ( 1.0 - sw(k,j) ) * immb(k,j) - sw(k,j) * imme(k,j) |
---|
687 | tendenz = fplus - fminus |
---|
688 | ! |
---|
689 | !-- Removed in order to optimise speed |
---|
690 | ! ffmax = MAX( ABS( fplus ), ABS( fminus ), 1E-35 ) |
---|
691 | ! IF ( ( ABS( tendenz ) / ffmax ) < 1E-7 ) tendenz = 0.0 |
---|
692 | ! |
---|
693 | !-- Density correction because of possible remaining divergences |
---|
694 | d_new = d(k,j,i) - ( v(k,j+1,i) - v(k,j,i) ) * dt_3d * ddy |
---|
695 | sk_p(k,j,i) = ( ( 1.0 + d(k,j,i) ) * sk_p(k,j,i) - tendenz ) / & |
---|
696 | ( 1.0 + d_new ) |
---|
697 | d(k,j,i) = d_new |
---|
698 | ENDDO |
---|
699 | ENDDO |
---|
700 | |
---|
701 | ENDDO ! End of the advection in y-direction |
---|
702 | CALL cpu_log( log_point_s(11), 'advec_s_bc:sendrecv', 'continue' ) |
---|
703 | CALL cpu_log( log_point_s(11), 'advec_s_bc:sendrecv', 'stop' ) |
---|
704 | |
---|
705 | ! |
---|
706 | !-- Deallocate temporary arrays |
---|
707 | DEALLOCATE( a0, a1, a2, a12, a22, immb, imme, impb, impe, ipmb, ipme, & |
---|
708 | ippb, ippe, m1, sw ) |
---|
709 | |
---|
710 | |
---|
711 | ! |
---|
712 | !-- Initialise for the computation of heat fluxes (see below; required in |
---|
713 | !-- UP flow_statistics) |
---|
714 | IF ( sk_char == 'pt' ) sums_wsts_bc_l = 0.0 |
---|
715 | |
---|
716 | ! |
---|
717 | !-- Add top and bottom boundaries according to the relevant boundary conditions |
---|
718 | IF ( sk_char == 'pt' ) THEN |
---|
719 | |
---|
720 | ! |
---|
721 | !-- Temperature boundary condition at the bottom boundary |
---|
722 | IF ( ibc_pt_b == 0 ) THEN |
---|
723 | ! |
---|
724 | !-- Dirichlet (fixed surface temperature) |
---|
725 | DO i = nxl, nxr |
---|
726 | DO j = nys, nyn |
---|
727 | sk_p(nzb-1,j,i) = sk_p(nzb,j,i) |
---|
728 | sk_p(nzb-2,j,i) = sk_p(nzb,j,i) |
---|
729 | ENDDO |
---|
730 | ENDDO |
---|
731 | |
---|
732 | ELSE |
---|
733 | ! |
---|
734 | !-- Neumann (i.e. here zero gradient) |
---|
735 | DO i = nxl, nxr |
---|
736 | DO j = nys, nyn |
---|
737 | sk_p(nzb,j,i) = sk_p(nzb+1,j,i) |
---|
738 | sk_p(nzb-1,j,i) = sk_p(nzb,j,i) |
---|
739 | sk_p(nzb-2,j,i) = sk_p(nzb,j,i) |
---|
740 | ENDDO |
---|
741 | ENDDO |
---|
742 | |
---|
743 | ENDIF |
---|
744 | |
---|
745 | ! |
---|
746 | !-- Temperature boundary condition at the top boundary |
---|
747 | IF ( ibc_pt_t == 0 .OR. ibc_pt_t == 1 ) THEN |
---|
748 | ! |
---|
749 | !-- Dirichlet or Neumann (zero gradient) |
---|
750 | DO i = nxl, nxr |
---|
751 | DO j = nys, nyn |
---|
752 | sk_p(nzt+2,j,i) = sk_p(nzt+1,j,i) |
---|
753 | sk_p(nzt+3,j,i) = sk_p(nzt+1,j,i) |
---|
754 | ENDDO |
---|
755 | ENDDO |
---|
756 | |
---|
757 | ELSEIF ( ibc_pt_t == 2 ) THEN |
---|
758 | ! |
---|
759 | !-- Neumann: dzu(nzt+2:3) are not defined, dzu(nzt+1) is used instead |
---|
760 | DO i = nxl, nxr |
---|
761 | DO j = nys, nyn |
---|
762 | sk_p(nzt+2,j,i) = sk_p(nzt+1,j,i) + bc_pt_t_val * dzu(nzt+1) |
---|
763 | sk_p(nzt+3,j,i) = sk_p(nzt+2,j,i) + bc_pt_t_val * dzu(nzt+1) |
---|
764 | ENDDO |
---|
765 | ENDDO |
---|
766 | |
---|
767 | ENDIF |
---|
768 | |
---|
769 | ELSEIF ( sk_char == 'sa' ) THEN |
---|
770 | |
---|
771 | ! |
---|
772 | !-- Salinity boundary condition at the bottom boundary. |
---|
773 | !-- So far, always Neumann (i.e. here zero gradient) is used |
---|
774 | DO i = nxl, nxr |
---|
775 | DO j = nys, nyn |
---|
776 | sk_p(nzb,j,i) = sk_p(nzb+1,j,i) |
---|
777 | sk_p(nzb-1,j,i) = sk_p(nzb,j,i) |
---|
778 | sk_p(nzb-2,j,i) = sk_p(nzb,j,i) |
---|
779 | ENDDO |
---|
780 | ENDDO |
---|
781 | |
---|
782 | ! |
---|
783 | !-- Salinity boundary condition at the top boundary. |
---|
784 | !-- Dirichlet or Neumann (zero gradient) |
---|
785 | DO i = nxl, nxr |
---|
786 | DO j = nys, nyn |
---|
787 | sk_p(nzt+2,j,i) = sk_p(nzt+1,j,i) |
---|
788 | sk_p(nzt+3,j,i) = sk_p(nzt+1,j,i) |
---|
789 | ENDDO |
---|
790 | ENDDO |
---|
791 | |
---|
792 | ELSEIF ( sk_char == 'q' ) THEN |
---|
793 | |
---|
794 | ! |
---|
795 | !-- Specific humidity boundary condition at the bottom boundary. |
---|
796 | !-- Dirichlet (fixed surface humidity) or Neumann (i.e. zero gradient) |
---|
797 | DO i = nxl, nxr |
---|
798 | DO j = nys, nyn |
---|
799 | sk_p(nzb,j,i) = sk_p(nzb+1,j,i) |
---|
800 | sk_p(nzb-1,j,i) = sk_p(nzb,j,i) |
---|
801 | sk_p(nzb-2,j,i) = sk_p(nzb,j,i) |
---|
802 | ENDDO |
---|
803 | ENDDO |
---|
804 | |
---|
805 | ! |
---|
806 | !-- Specific humidity boundary condition at the top boundary |
---|
807 | IF ( ibc_q_t == 0 ) THEN |
---|
808 | ! |
---|
809 | !-- Dirichlet |
---|
810 | DO i = nxl, nxr |
---|
811 | DO j = nys, nyn |
---|
812 | sk_p(nzt+2,j,i) = sk_p(nzt+1,j,i) |
---|
813 | sk_p(nzt+3,j,i) = sk_p(nzt+1,j,i) |
---|
814 | ENDDO |
---|
815 | ENDDO |
---|
816 | |
---|
817 | ELSE |
---|
818 | ! |
---|
819 | !-- Neumann: dzu(nzt+2:3) are not defined, dzu(nzt+1) is used instead |
---|
820 | DO i = nxl, nxr |
---|
821 | DO j = nys, nyn |
---|
822 | sk_p(nzt+2,j,i) = sk_p(nzt+1,j,i) + bc_q_t_val * dzu(nzt+1) |
---|
823 | sk_p(nzt+3,j,i) = sk_p(nzt+2,j,i) + bc_q_t_val * dzu(nzt+1) |
---|
824 | ENDDO |
---|
825 | ENDDO |
---|
826 | |
---|
827 | ENDIF |
---|
828 | |
---|
829 | ELSEIF ( sk_char == 'e' ) THEN |
---|
830 | |
---|
831 | ! |
---|
832 | !-- TKE boundary condition at bottom and top boundary (generally Neumann) |
---|
833 | DO i = nxl, nxr |
---|
834 | DO j = nys, nyn |
---|
835 | sk_p(nzb,j,i) = sk_p(nzb+1,j,i) |
---|
836 | sk_p(nzb-1,j,i) = sk_p(nzb,j,i) |
---|
837 | sk_p(nzb-2,j,i) = sk_p(nzb,j,i) |
---|
838 | sk_p(nzt+2,j,i) = sk_p(nzt+1,j,i) |
---|
839 | sk_p(nzt+3,j,i) = sk_p(nzt+1,j,i) |
---|
840 | ENDDO |
---|
841 | ENDDO |
---|
842 | |
---|
843 | ELSE |
---|
844 | |
---|
845 | WRITE( message_string, * ) 'no vertical boundary condi', & |
---|
846 | 'tion for variable "', sk_char, '"' |
---|
847 | CALL message( 'advec_s_bc', 'PA0158', 1, 2, 0, 6, 0 ) |
---|
848 | |
---|
849 | ENDIF |
---|
850 | |
---|
851 | ! |
---|
852 | !-- Determine the maxima of the first and second derivative in z-direction |
---|
853 | fmax_l = 0.0 |
---|
854 | DO i = nxl, nxr |
---|
855 | DO j = nys, nyn |
---|
856 | DO k = nzb, nzt+1 |
---|
857 | zaehler = ABS( sk_p(k+1,j,i) - 2.0 * sk_p(k,j,i) + sk_p(k-1,j,i) ) |
---|
858 | nenner = ABS( sk_p(k+1,j,i+1) - sk_p(k-1,j,i) ) |
---|
859 | fmax_l(1) = MAX( fmax_l(1) , zaehler ) |
---|
860 | fmax_l(2) = MAX( fmax_l(2) , nenner ) |
---|
861 | ENDDO |
---|
862 | ENDDO |
---|
863 | ENDDO |
---|
864 | #if defined( __parallel ) |
---|
865 | CALL MPI_ALLREDUCE( fmax_l, fmax, 2, MPI_REAL, MPI_MAX, comm2d, ierr ) |
---|
866 | #else |
---|
867 | fmax = fmax_l |
---|
868 | #endif |
---|
869 | |
---|
870 | fmax = 0.04 * fmax |
---|
871 | |
---|
872 | ! |
---|
873 | !-- Allocate temporary arrays |
---|
874 | ALLOCATE( a0(nzb:nzt+1,nys:nyn), a1(nzb:nzt+1,nys:nyn), & |
---|
875 | a2(nzb:nzt+1,nys:nyn), a12(nzb:nzt+1,nys:nyn), & |
---|
876 | a22(nzb:nzt+1,nys:nyn), immb(nzb+1:nzt,nys:nyn), & |
---|
877 | imme(nzb+1:nzt,nys:nyn), impb(nzb+1:nzt,nys:nyn), & |
---|
878 | impe(nzb+1:nzt,nys:nyn), ipmb(nzb+1:nzt,nys:nyn), & |
---|
879 | ipme(nzb+1:nzt,nys:nyn), ippb(nzb+1:nzt,nys:nyn), & |
---|
880 | ippe(nzb+1:nzt,nys:nyn), m1(nzb-1:nzt+2,nys:nyn), & |
---|
881 | sw(nzb:nzt+1,nys:nyn) & |
---|
882 | ) |
---|
883 | imme = 0.0; impe = 0.0; ipme = 0.0; ippe = 0.0 |
---|
884 | |
---|
885 | ! |
---|
886 | !-- Outer loop of all i |
---|
887 | DO i = nxl, nxr |
---|
888 | |
---|
889 | ! |
---|
890 | !-- Compute polynomial coefficients |
---|
891 | DO j = nys, nyn |
---|
892 | DO k = nzb, nzt+1 |
---|
893 | a12(k,j) = 0.5 * ( sk_p(k+1,j,i) - sk_p(k-1,j,i) ) |
---|
894 | a22(k,j) = 0.5 * ( sk_p(k+1,j,i) - 2.0 * sk_p(k,j,i) & |
---|
895 | + sk_p(k-1,j,i) ) |
---|
896 | a0(k,j) = ( 9.0 * sk_p(k+2,j,i) - 116.0 * sk_p(k+1,j,i) & |
---|
897 | + 2134.0 * sk_p(k,j,i) - 116.0 * sk_p(k-1,j,i) & |
---|
898 | + 9.0 * sk_p(k-2,j,i) ) * f1920 |
---|
899 | a1(k,j) = ( -5.0 * sk_p(k+2,j,i) + 34.0 * sk_p(k+1,j,i) & |
---|
900 | - 34.0 * sk_p(k-1,j,i) + 5.0 * sk_p(k-2,j,i) & |
---|
901 | ) * f48 |
---|
902 | a2(k,j) = ( -3.0 * sk_p(k+2,j,i) + 36.0 * sk_p(k+1,j,i) & |
---|
903 | - 66.0 * sk_p(k,j,i) + 36.0 * sk_p(k-1,j,i) & |
---|
904 | - 3.0 * sk_p(k-2,j,i) ) * f48 |
---|
905 | ENDDO |
---|
906 | ENDDO |
---|
907 | |
---|
908 | ! |
---|
909 | !-- Fluxes using the Bott scheme |
---|
910 | !-- *VOCL LOOP,UNROLL(2) |
---|
911 | DO j = nys, nyn |
---|
912 | DO k = nzb+1, nzt |
---|
913 | cip = MAX( 0.0, w(k,j,i) * dt_3d * ddzw(k) ) |
---|
914 | cim = -MIN( 0.0, w(k,j,i) * dt_3d * ddzw(k) ) |
---|
915 | cipf = 1.0 - 2.0 * cip |
---|
916 | cimf = 1.0 - 2.0 * cim |
---|
917 | ip = a0(k,j) * f2 * ( 1.0 - cipf ) & |
---|
918 | + a1(k,j) * f8 * ( 1.0 - cipf*cipf ) & |
---|
919 | + a2(k,j) * f24 * ( 1.0 - cipf*cipf*cipf ) |
---|
920 | im = a0(k+1,j) * f2 * ( 1.0 - cimf ) & |
---|
921 | - a1(k+1,j) * f8 * ( 1.0 - cimf*cimf ) & |
---|
922 | + a2(k+1,j) * f24 * ( 1.0 - cimf*cimf*cimf ) |
---|
923 | ip = MAX( ip, 0.0 ) |
---|
924 | im = MAX( im, 0.0 ) |
---|
925 | ippb(k,j) = ip * MIN( 1.0, sk_p(k,j,i) / (ip+im+1E-15) ) |
---|
926 | impb(k,j) = im * MIN( 1.0, sk_p(k+1,j,i) / (ip+im+1E-15) ) |
---|
927 | |
---|
928 | cip = MAX( 0.0, w(k-1,j,i) * dt_3d * ddzw(k) ) |
---|
929 | cim = -MIN( 0.0, w(k-1,j,i) * dt_3d * ddzw(k) ) |
---|
930 | cipf = 1.0 - 2.0 * cip |
---|
931 | cimf = 1.0 - 2.0 * cim |
---|
932 | ip = a0(k-1,j) * f2 * ( 1.0 - cipf ) & |
---|
933 | + a1(k-1,j) * f8 * ( 1.0 - cipf*cipf ) & |
---|
934 | + a2(k-1,j) * f24 * ( 1.0 - cipf*cipf*cipf ) |
---|
935 | im = a0(k,j) * f2 * ( 1.0 - cimf ) & |
---|
936 | - a1(k,j) * f8 * ( 1.0 - cimf*cimf ) & |
---|
937 | + a2(k,j) * f24 * ( 1.0 - cimf*cimf*cimf ) |
---|
938 | ip = MAX( ip, 0.0 ) |
---|
939 | im = MAX( im, 0.0 ) |
---|
940 | ipmb(k,j) = ip * MIN( 1.0, sk_p(k-1,j,i) / (ip+im+1E-15) ) |
---|
941 | immb(k,j) = im * MIN( 1.0, sk_p(k,j,i) / (ip+im+1E-15) ) |
---|
942 | ENDDO |
---|
943 | ENDDO |
---|
944 | |
---|
945 | ! |
---|
946 | !-- Compute monitor function m1 |
---|
947 | DO j = nys, nyn |
---|
948 | DO k = nzb-1, nzt+2 |
---|
949 | m1z = ABS( sk_p(k+1,j,i) - 2.0 * sk_p(k,j,i) + sk_p(k-1,j,i) ) |
---|
950 | m1n = ABS( sk_p(k+1,j,i) - sk_p(k-1,j,i) ) |
---|
951 | IF ( m1n /= 0.0 .AND. m1n >= m1z ) THEN |
---|
952 | m1(k,j) = m1z / m1n |
---|
953 | IF ( m1(k,j) /= 2.0 .AND. m1n < fmax(2) ) m1(k,j) = 0.0 |
---|
954 | ELSEIF ( m1n < m1z ) THEN |
---|
955 | m1(k,j) = -1.0 |
---|
956 | ELSE |
---|
957 | m1(k,j) = 0.0 |
---|
958 | ENDIF |
---|
959 | ENDDO |
---|
960 | ENDDO |
---|
961 | |
---|
962 | ! |
---|
963 | !-- Compute switch sw |
---|
964 | sw = 0.0 |
---|
965 | DO j = nys, nyn |
---|
966 | DO k = nzb, nzt+1 |
---|
967 | m2 = 2.0 * ABS( a1(k,j) - a12(k,j) ) / & |
---|
968 | MAX( ABS( a1(k,j) + a12(k,j) ), 1E-35 ) |
---|
969 | IF ( ABS( a1(k,j) + a12(k,j) ) < fmax(2) ) m2 = 0.0 |
---|
970 | |
---|
971 | m3 = 2.0 * ABS( a2(k,j) - a22(k,j) ) / & |
---|
972 | MAX( ABS( a2(k,j) + a22(k,j) ), 1E-35 ) |
---|
973 | IF ( ABS( a2(k,j) + a22(k,j) ) < fmax(1) ) m3 = 0.0 |
---|
974 | |
---|
975 | t1 = 0.35 |
---|
976 | t2 = 0.35 |
---|
977 | IF ( m1(k,j) == -1.0 ) t2 = 0.12 |
---|
978 | |
---|
979 | !-- *VOCL STMT,IF(10) |
---|
980 | IF ( m1(k-1,j) == 1.0 .OR. m1(k,j) == 1.0 .OR. m1(k+1,j) == 1.0 & |
---|
981 | .OR. m2 > t2 .OR. m3 > T2 .OR. & |
---|
982 | ( m1(k,j) > t1 .AND. m1(k-1,j) /= -1.0 .AND. & |
---|
983 | m1(k,j) /= -1.0 .AND. m1(k+1,j) /= -1.0 ) & |
---|
984 | ) sw(k,j) = 1.0 |
---|
985 | ENDDO |
---|
986 | ENDDO |
---|
987 | |
---|
988 | ! |
---|
989 | !-- Fluxes using exponential scheme |
---|
990 | CALL cpu_log( log_point_s(12), 'advec_s_bc:exp', 'continue' ) |
---|
991 | DO j = nys, nyn |
---|
992 | DO k = nzb+1, nzt |
---|
993 | |
---|
994 | !-- *VOCL STMT,IF(10) |
---|
995 | IF ( sw(k,j) == 1.0 ) THEN |
---|
996 | snenn = sk_p(k+1,j,i) - sk_p(k-1,j,i) |
---|
997 | IF ( ABS( snenn ) < 1E-9 ) snenn = 1E-9 |
---|
998 | sterm = ( sk_p(k,j,i) - sk_p(k-1,j,i) ) / snenn |
---|
999 | sterm = MIN( sterm, 0.9999 ) |
---|
1000 | sterm = MAX( sterm, 0.0001 ) |
---|
1001 | |
---|
1002 | ix = INT( sterm * 1000 ) + 1 |
---|
1003 | |
---|
1004 | cip = MAX( 0.0, w(k,j,i) * dt_3d * ddzw(k) ) |
---|
1005 | |
---|
1006 | ippe(k,j) = sk_p(k-1,j,i) * cip + snenn * ( & |
---|
1007 | aex(ix) * cip + bex(ix) / dex(ix) * ( & |
---|
1008 | eex(ix) - EXP( dex(ix)*0.5 * ( 1.0 - 2.0 * cip ) ) & |
---|
1009 | ) & |
---|
1010 | ) |
---|
1011 | IF ( sterm == 0.0001 ) ippe(k,j) = sk_p(k,j,i) * cip |
---|
1012 | IF ( sterm == 0.9999 ) ippe(k,j) = sk_p(k,j,i) * cip |
---|
1013 | |
---|
1014 | snenn = sk_p(k-1,j,i) - sk_p(k+1,j,i) |
---|
1015 | IF ( ABS( snenn ) < 1E-9 ) snenn = 1E-9 |
---|
1016 | sterm = ( sk_p(k,j,i) - sk_p(k+1,j,i) ) / snenn |
---|
1017 | sterm = MIN( sterm, 0.9999 ) |
---|
1018 | sterm = MAX( sterm, 0.0001 ) |
---|
1019 | |
---|
1020 | ix = INT( sterm * 1000 ) + 1 |
---|
1021 | |
---|
1022 | cim = -MIN( 0.0, w(k-1,j,i) * dt_3d * ddzw(k) ) |
---|
1023 | |
---|
1024 | imme(k,j) = sk_p(k+1,j,i) * cim + snenn * ( & |
---|
1025 | aex(ix) * cim + bex(ix) / dex(ix) * ( & |
---|
1026 | eex(ix) - EXP( dex(ix)*0.5 * ( 1.0 - 2.0 * cim ) ) & |
---|
1027 | ) & |
---|
1028 | ) |
---|
1029 | IF ( sterm == 0.0001 ) imme(k,j) = sk_p(k,j,i) * cim |
---|
1030 | IF ( sterm == 0.9999 ) imme(k,j) = sk_p(k,j,i) * cim |
---|
1031 | ENDIF |
---|
1032 | |
---|
1033 | !-- *VOCL STMT,IF(10) |
---|
1034 | IF ( sw(k+1,j) == 1.0 ) THEN |
---|
1035 | snenn = sk_p(k,j,i) - sk_p(k+2,j,i) |
---|
1036 | IF ( ABS( snenn ) .LT. 1E-9 ) snenn = 1E-9 |
---|
1037 | sterm = ( sk_p(k+1,j,i) - sk_p(k+2,j,i) ) / snenn |
---|
1038 | sterm = MIN( sterm, 0.9999 ) |
---|
1039 | sterm = MAX( sterm, 0.0001 ) |
---|
1040 | |
---|
1041 | ix = INT( sterm * 1000 ) + 1 |
---|
1042 | |
---|
1043 | cim = -MIN( 0.0, w(k,j,i) * dt_3d * ddzw(k) ) |
---|
1044 | |
---|
1045 | impe(k,j) = sk_p(k+2,j,i) * cim + snenn * ( & |
---|
1046 | aex(ix) * cim + bex(ix) / dex(ix) * ( & |
---|
1047 | eex(ix) - EXP( dex(ix)*0.5 * ( 1.0 - 2.0 * cim ) ) & |
---|
1048 | ) & |
---|
1049 | ) |
---|
1050 | IF ( sterm == 0.0001 ) impe(k,j) = sk_p(k+1,j,i) * cim |
---|
1051 | IF ( sterm == 0.9999 ) impe(k,j) = sk_p(k+1,j,i) * cim |
---|
1052 | ENDIF |
---|
1053 | |
---|
1054 | !-- *VOCL STMT,IF(10) |
---|
1055 | IF ( sw(k-1,j) == 1.0 ) THEN |
---|
1056 | snenn = sk_p(k,j,i) - sk_p(k-2,j,i) |
---|
1057 | IF ( ABS( snenn ) < 1E-9 ) snenn = 1E-9 |
---|
1058 | sterm = ( sk_p(k-1,j,i) - sk_p(k-2,j,i) ) / snenn |
---|
1059 | sterm = MIN( sterm, 0.9999 ) |
---|
1060 | sterm = MAX( sterm, 0.0001 ) |
---|
1061 | |
---|
1062 | ix = INT( sterm * 1000 ) + 1 |
---|
1063 | |
---|
1064 | cip = MAX( 0.0, w(k-1,j,i) * dt_3d * ddzw(k) ) |
---|
1065 | |
---|
1066 | ipme(k,j) = sk_p(k-2,j,i) * cip + snenn * ( & |
---|
1067 | aex(ix) * cip + bex(ix) / dex(ix) * ( & |
---|
1068 | eex(ix) - EXP( dex(ix)*0.5 * ( 1.0 - 2.0 * cip ) ) & |
---|
1069 | ) & |
---|
1070 | ) |
---|
1071 | IF ( sterm == 0.0001 ) ipme(k,j) = sk_p(k-1,j,i) * cip |
---|
1072 | IF ( sterm == 0.9999 ) ipme(k,j) = sk_p(k-1,j,i) * cip |
---|
1073 | ENDIF |
---|
1074 | |
---|
1075 | ENDDO |
---|
1076 | ENDDO |
---|
1077 | CALL cpu_log( log_point_s(12), 'advec_s_bc:exp', 'pause' ) |
---|
1078 | |
---|
1079 | ! |
---|
1080 | !-- Prognostic equation |
---|
1081 | DO j = nys, nyn |
---|
1082 | DO k = nzb+1, nzt |
---|
1083 | fplus = ( 1.0 - sw(k,j) ) * ippb(k,j) + sw(k,j) * ippe(k,j) & |
---|
1084 | - ( 1.0 - sw(k+1,j) ) * impb(k,j) - sw(k+1,j) * impe(k,j) |
---|
1085 | fminus = ( 1.0 - sw(k-1,j) ) * ipmb(k,j) + sw(k-1,j) * ipme(k,j) & |
---|
1086 | - ( 1.0 - sw(k,j) ) * immb(k,j) - sw(k,j) * imme(k,j) |
---|
1087 | tendenz = fplus - fminus |
---|
1088 | ! |
---|
1089 | !-- Removed in order to optimise speed |
---|
1090 | ! ffmax = MAX( ABS( fplus ), ABS( fminus ), 1E-35 ) |
---|
1091 | ! IF ( ( ABS( tendenz ) / ffmax ) < 1E-7 ) tendenz = 0.0 |
---|
1092 | ! |
---|
1093 | !-- Density correction because of possible remaining divergences |
---|
1094 | d_new = d(k,j,i) - ( w(k,j,i) - w(k-1,j,i) ) * dt_3d * ddzw(k) |
---|
1095 | sk_p(k,j,i) = ( ( 1.0 + d(k,j,i) ) * sk_p(k,j,i) - tendenz ) / & |
---|
1096 | ( 1.0 + d_new ) |
---|
1097 | ! |
---|
1098 | !-- Store heat flux for subsequent statistics output. |
---|
1099 | !-- array m1 is here used as temporary storage |
---|
1100 | m1(k,j) = fplus / dt_3d * dzw(k) |
---|
1101 | ENDDO |
---|
1102 | ENDDO |
---|
1103 | |
---|
1104 | ! |
---|
1105 | !-- Sum up heat flux in order to order to obtain horizontal averages |
---|
1106 | IF ( sk_char == 'pt' ) THEN |
---|
1107 | DO sr = 0, statistic_regions |
---|
1108 | DO j = nys, nyn |
---|
1109 | DO k = nzb+1, nzt |
---|
1110 | sums_wsts_bc_l(k,sr) = sums_wsts_bc_l(k,sr) + & |
---|
1111 | m1(k,j) * rmask(j,i,sr) |
---|
1112 | ENDDO |
---|
1113 | ENDDO |
---|
1114 | ENDDO |
---|
1115 | ENDIF |
---|
1116 | |
---|
1117 | ENDDO ! End of the advection in z-direction |
---|
1118 | CALL cpu_log( log_point_s(12), 'advec_s_bc:exp', 'continue' ) |
---|
1119 | CALL cpu_log( log_point_s(12), 'advec_s_bc:exp', 'stop' ) |
---|
1120 | |
---|
1121 | ! |
---|
1122 | !-- Deallocate temporary arrays |
---|
1123 | DEALLOCATE( a0, a1, a2, a12, a22, immb, imme, impb, impe, ipmb, ipme, & |
---|
1124 | ippb, ippe, m1, sw ) |
---|
1125 | |
---|
1126 | ! |
---|
1127 | !-- Store results as tendency and deallocate local array |
---|
1128 | DO i = nxl, nxr |
---|
1129 | DO j = nys, nyn |
---|
1130 | DO k = nzb+1, nzt |
---|
1131 | tend(k,j,i) = tend(k,j,i) + ( sk_p(k,j,i) - sk(k,j,i) ) / dt_3d |
---|
1132 | ENDDO |
---|
1133 | ENDDO |
---|
1134 | ENDDO |
---|
1135 | |
---|
1136 | DEALLOCATE( sk_p ) |
---|
1137 | |
---|
1138 | END SUBROUTINE advec_s_bc |
---|