1 |
afe |
1.1 |
subroutine greg_sqrt(xens,yo,H,ngp,mobs,RRs,nens) |
2 |
|
|
|
3 |
|
|
c implicit none |
4 |
|
|
|
5 |
|
|
include 'mpif.h' |
6 |
|
|
|
7 |
|
|
integer MASTER, FROM_MASTER, FROM_WORKER |
8 |
|
|
parameter (MASTER = 0) |
9 |
|
|
parameter (FROM_MASTER = 1) |
10 |
|
|
parameter (FROM_WORKER = 2) |
11 |
|
|
|
12 |
|
|
integer taskid, ierr, numtasks, numworkers, snumworkers |
13 |
|
|
integer source, dest, mtype, num, avenum, extra, offset |
14 |
|
|
integer status(MPI_STATUS_SIZE) |
15 |
|
|
|
16 |
|
|
! Arguments |
17 |
|
|
integer, intent(in) :: nens, mobs, ngp |
18 |
|
|
real, intent(inout) :: xens(ngp,nens) |
19 |
|
|
real, intent(in) :: yo(mobs), RRs(mobs), H(mobs,ngp) |
20 |
|
|
|
21 |
|
|
! Local Variables |
22 |
|
|
integer :: rad, xob(mobs), yob(mobs), ind, nx, ny |
23 |
|
|
integer :: k, j, i, r2, kk, jj, ko, jo, kj, g |
24 |
|
|
real :: PHT(ngp), HPHT, Ks(ngp), Khat(ngp), alpha |
25 |
|
|
real :: xp(ngp), xa(ngp), zp(ngp,nens), R, Rs, boost |
26 |
|
|
real, parameter :: eps = 1.e-6 ! corresponds to verr = 0.001 |
27 |
|
|
! Filter Stuff |
28 |
|
|
integer :: d, fexp |
29 |
|
|
real :: dr, fcoef, filt |
30 |
|
|
|
31 |
|
|
call MPI_COMM_RANK(MPI_COMM_WORLD,taskid,ierr) |
32 |
|
|
call MPI_COMM_SIZE(MPI_COMM_WORLD,numtasks,ierr) |
33 |
|
|
|
34 |
|
|
|
35 |
|
|
if (taskid.eq.MASTER) then |
36 |
|
|
|
37 |
|
|
|
38 |
|
|
write(6,*) 'a' |
39 |
|
|
|
40 |
|
|
nx = 33 |
41 |
|
|
ny = 65 |
42 |
|
|
|
43 |
|
|
Rs = RRs(1) |
44 |
|
|
R = sqrt(Rs) |
45 |
|
|
|
46 |
|
|
rad = 20 |
47 |
|
|
rad = 100 |
48 |
|
|
r2 = rad*rad |
49 |
|
|
|
50 |
|
|
fexp = 2 |
51 |
|
|
fcoef = 0.05 |
52 |
|
|
boost = 1.05 |
53 |
|
|
boost =1.0 |
54 |
|
|
|
55 |
|
|
zp = xens |
56 |
|
|
|
57 |
|
|
!*** Transform 2D state vector into 1D vector.... |
58 |
|
|
! do j = 1, nens |
59 |
|
|
! do k = 1, ny |
60 |
|
|
! zp(nx*(k-1)+1:k*nx,j) = xens(:,k,j) |
61 |
|
|
! end do |
62 |
|
|
! end do |
63 |
|
|
|
64 |
|
|
write(6,*) 'b' |
65 |
|
|
|
66 |
|
|
!*** Find the initial ensemble mean |
67 |
|
|
do j = 1, ngp |
68 |
|
|
xp(j) = sum(zp(j,:))/float(nens) |
69 |
|
|
end do |
70 |
|
|
|
71 |
|
|
!*** Apply inflation factor to initial ensemble |
72 |
|
|
do k = 1, nens |
73 |
|
|
zp(:,k) = boost*(zp(:,k) - xp) + xp |
74 |
|
|
end do |
75 |
|
|
|
76 |
|
|
write(6,*) 'c' |
77 |
|
|
|
78 |
|
|
!*** Find the xob and yob arrays from H |
79 |
|
|
do j = 1, ngp |
80 |
|
|
do k = 1, mobs |
81 |
|
|
if ( H(k,j) == 1. ) then |
82 |
|
|
xob(k) = mod(j-1,nx) + 1 |
83 |
|
|
yob(k) = (j-1)/nx + 1 |
84 |
|
|
end if |
85 |
|
|
end do |
86 |
|
|
end do |
87 |
|
|
|
88 |
|
|
|
89 |
|
|
write(6,*) 'd' |
90 |
|
|
|
91 |
|
|
!*** Now process each observation sequentially abiding by cut-off radius |
92 |
|
|
do j = 1, mobs |
93 |
|
|
ind = nx*( yob(j) - 1 ) + xob(j) |
94 |
|
|
|
95 |
|
|
write(6,*) 'e' |
96 |
|
|
|
97 |
|
|
!*** Find PH' first |
98 |
|
|
PHT = 0.0 |
99 |
|
|
do jj = yob(j)-rad, yob(j)+rad |
100 |
|
|
do kk = xob(j)-rad, xob(j)+rad |
101 |
|
|
jo = jj |
102 |
|
|
ko = kk |
103 |
|
|
|
104 |
|
|
write(6,*) 'f', j, jo, ko |
105 |
|
|
|
106 |
|
|
!*** Point is within block of radius, but it may not be within the |
107 |
|
|
! basin boundaries |
108 |
|
|
if ( ko>0 .and. ko<=nx .and. jo>0 .and. jo<=ny ) then |
109 |
|
|
|
110 |
|
|
!*** Since we've sequestered a square of side 2*rad and the |
111 |
|
|
! cut-off radius assumes a circle, we need to check to |
112 |
|
|
! make sure the point we're considering is actually |
113 |
|
|
! within the circle. |
114 |
|
|
|
115 |
|
|
write(6,*) 'g' |
116 |
|
|
|
117 |
|
|
d = (ko - xob(j))**2 + (jo - yob(j))**2 |
118 |
|
|
|
119 |
|
|
write(6,*) 'h' |
120 |
|
|
|
121 |
|
|
if ( d <= r2 ) then |
122 |
|
|
|
123 |
|
|
dr = sqrt( float( d ) ) |
124 |
|
|
|
125 |
|
|
!*** The element of interest in the 1D vector according to addresses |
126 |
|
|
! kk and jj is: |
127 |
|
|
kj = nx*(jo-1) + ko |
128 |
|
|
|
129 |
|
|
write(6,*) 'i' |
130 |
|
|
|
131 |
|
|
!*** Evaluate the filter coefficient based on distance from center d |
132 |
|
|
filt = 1.0 - exp( -fcoef*( (float(rad) - dr)**fexp ) ) |
133 |
|
|
filt = 1.0 |
134 |
|
|
|
135 |
|
|
write(6,*) 'j' |
136 |
|
|
|
137 |
|
|
!*** Now contribute to PHT sum |
138 |
|
|
do g = 1, nens |
139 |
|
|
PHT(kj) = PHT(kj) + filt*(zp(kj,g) - xp(kj))* |
140 |
|
|
& (zp(ind,g) - xp(ind)) |
141 |
|
|
end do |
142 |
|
|
|
143 |
|
|
write(6,*) 'k' |
144 |
|
|
|
145 |
|
|
end if |
146 |
|
|
end if |
147 |
|
|
end do |
148 |
|
|
end do |
149 |
|
|
PHT = PHT/float(nens - 1) |
150 |
|
|
|
151 |
|
|
!*** Now find HPH' from PH'. Because of cut-off radius, there is a |
152 |
|
|
! (good) chance that HPH' will be zero. |
153 |
|
|
HPHT = PHT(ind) |
154 |
|
|
|
155 |
|
|
!*** Evaluate Ks |
156 |
|
|
Ks = PHT/( HPHT + Rs ) |
157 |
|
|
|
158 |
|
|
!*** Update all effected elements in the mean |
159 |
|
|
xa = xp + Ks*( yo(j) - xp(ind) ) |
160 |
|
|
|
161 |
|
|
!*** Now update all ensemble members as perturbations about mean |
162 |
|
|
alpha = 1.0/( 1.0 + sqrt( Rs/( HPHT + Rs ) ) ) |
163 |
|
|
Khat = alpha*Ks |
164 |
|
|
|
165 |
|
|
do g = 1, nens |
166 |
|
|
zp(:,g) = ((zp(:,g) - xp) - Khat*( zp(ind,g) - |
167 |
|
|
& xp(ind) )) + xa |
168 |
|
|
end do |
169 |
|
|
|
170 |
|
|
!*** Use analysis ensemble as the background for the next observation |
171 |
|
|
xp = xa |
172 |
|
|
|
173 |
|
|
end do |
174 |
|
|
|
175 |
|
|
print*, 'EnSRF:: Done Mobs Loop' |
176 |
|
|
|
177 |
|
|
xens = zp |
178 |
|
|
|
179 |
|
|
endif |
180 |
|
|
|
181 |
|
|
return |
182 |
|
|
|
183 |
|
|
end subroutine greg_sqrt |
184 |
|
|
|
185 |
|
|
|