| 1 |
afe |
1.2 |
program hadley3 |
| 2 |
afe |
1.1 |
|
| 3 |
afe |
1.2 |
c*** This subroutine calculates analysis errors using greg`s |
| 4 |
afe |
1.1 |
c*** ensemble square root filter, or the traditional ensemble |
| 5 |
|
|
c*** kalman filter. |
| 6 |
|
|
c*** |
| 7 |
|
|
c*** Written by: Sai Ravela |
| 8 |
|
|
c*** Last modified: Feb 2, 2003 |
| 9 |
|
|
|
| 10 |
|
|
implicit none |
| 11 |
|
|
|
| 12 |
afe |
1.2 |
integer i, j, k, ijuli, isteps |
| 13 |
|
|
integer nens ! ensemble size |
| 14 |
|
|
integer iters ! number of iterations |
| 15 |
|
|
integer n ! state size, calculated from |
| 16 |
|
|
integer nx, ny, nz ! physical dimension sizes, and |
| 17 |
|
|
integer nf ! number of observed variables |
| 18 |
|
|
integer mobs, isai |
| 19 |
afe |
1.1 |
integer ii, jj, zob, xob, yob |
| 20 |
|
|
#include "hadley3.h" |
| 21 |
afe |
1.2 |
integer iobsloc(mobs) ! obs locations, in terms of |
| 22 |
|
|
! dimensions in the state vector. |
| 23 |
|
|
! Indicated with the matrix H in KF |
| 24 |
|
|
! literature. |
| 25 |
|
|
! In practical terms, the numbers |
| 26 |
|
|
! the locations of the words in the |
| 27 |
|
|
! pickup file |
| 28 |
|
|
integer indxa |
| 29 |
|
|
real*8 ytrue(n) ! state vector |
| 30 |
|
|
real*8 yobs(mobs) ! current observations of state |
| 31 |
|
|
real*8 ytmp(n) ! tmp state vector |
| 32 |
afe |
1.1 |
real*8 pert(n), R(mobs), yobsfull(n), ave(n), var(n) |
| 33 |
|
|
real*8 xens(n,nens) |
| 34 |
|
|
real*8 var_obs, eps, dt, x |
| 35 |
|
|
real*8 obserr, analerr, distsub, distobs |
| 36 |
|
|
integer iret, system |
| 37 |
|
|
real*4 mask(ny,nx) |
| 38 |
|
|
|
| 39 |
|
|
c*** open some output files |
| 40 |
|
|
open(unit=2,file='test.dat',status='unknown') |
| 41 |
|
|
c open(unit=3,file='fields.dat',status='unknown') |
| 42 |
|
|
|
| 43 |
|
|
c*** initialisations |
| 44 |
afe |
1.2 |
eps=0.0005 ! obs error variance - 1/2 mm/s |
| 45 |
|
|
! only good for u and v |
| 46 |
|
|
! need to vary this for |
| 47 |
|
|
! different variables |
| 48 |
|
|
|
| 49 |
|
|
c*** Here is where you read iobsloc and set the corresponding |
| 50 |
|
|
c variances. |
| 51 |
|
|
call ReadObsLoc(3,mobs,iobsloc) |
| 52 |
afe |
1.1 |
write (*,*) 'iobsloc set',mobs, iobsloc(mobs), nens |
| 53 |
|
|
var_obs=eps**2 |
| 54 |
|
|
do i=1,mobs |
| 55 |
afe |
1.2 |
c why the if/then? |
| 56 |
|
|
if (iobsloc(i) < nx*ny*nz*3) then |
| 57 |
afe |
1.1 |
R(i)=var_obs |
| 58 |
|
|
else |
| 59 |
afe |
1.2 |
R(i) = 0.1 ! 0.9 deg other |
| 60 |
afe |
1.1 |
end if |
| 61 |
afe |
1.2 |
enddo |
| 62 |
afe |
1.1 |
|
| 63 |
afe |
1.2 |
c*** get initial observations -- assumes that model has |
| 64 |
|
|
c already spun up (how long?) |
| 65 |
afe |
1.1 |
|
| 66 |
afe |
1.2 |
write (*,*) 'getting Truth' |
| 67 |
|
|
call ReadPickup(0,n,nx,ny,nz,ytrue) |
| 68 |
|
|
write (*,*) 'got Truth' |
| 69 |
|
|
|
| 70 |
|
|
c*** fantasy observations |
| 71 |
|
|
write (*,*) 'generating observations' |
| 72 |
|
|
call random2(pert,n) |
| 73 |
|
|
yobsfull = ytrue+eps*pert |
| 74 |
|
|
c*** apply observation mask |
| 75 |
|
|
do i = 1,mobs |
| 76 |
|
|
indxa = iobsloc(i) |
| 77 |
|
|
yobs(i) = yobsfull(indxa) |
| 78 |
|
|
end do |
| 79 |
|
|
|
| 80 |
|
|
write (*,*) 'generating ensemble' |
| 81 |
|
|
do i=1,nens |
| 82 |
|
|
call random2(pert,n) |
| 83 |
|
|
do j=1,n |
| 84 |
|
|
xens(j,i)=yobsfull(j)+eps*pert(j) |
| 85 |
|
|
ytmp(j)=xens(j,i) |
| 86 |
|
|
enddo |
| 87 |
|
|
call WritePickUp(i,n,nx,ny,nz,ytmp) |
| 88 |
|
|
enddo |
| 89 |
|
|
|
| 90 |
|
|
c*** main iteration loop |
| 91 |
|
|
do i=1,iters |
| 92 |
afe |
1.1 |
|
| 93 |
afe |
1.2 |
c*** step truth forward |
| 94 |
|
|
write (*,*) 'Stepping Truth' |
| 95 |
afe |
1.1 |
call Model(0) |
| 96 |
afe |
1.2 |
c call ReadPickup(0,n,nx,ny,nz,ytmp) |
| 97 |
|
|
c call WritePickUp(0,n,nx,ny,nz,ytmp) |
| 98 |
afe |
1.1 |
write (*,*) 'Stepped Truth' |
| 99 |
afe |
1.2 |
|
| 100 |
|
|
write (*,*) 'getting Truth' |
| 101 |
|
|
call ReadPickup(0,n,nx,ny,nz,ytrue) |
| 102 |
|
|
write (*,*) 'got Truth' |
| 103 |
|
|
|
| 104 |
|
|
c*** fantasy observations |
| 105 |
|
|
write (*,*) 'generating observations' |
| 106 |
|
|
call random2(pert,n) |
| 107 |
|
|
yobsfull = ytrue+eps*pert |
| 108 |
|
|
c*** apply observation mask |
| 109 |
|
|
do j = 1,mobs |
| 110 |
|
|
indxa = iobsloc(j) |
| 111 |
|
|
yobs(j) = yobsfull(indxa) |
| 112 |
|
|
end do |
| 113 |
|
|
|
| 114 |
|
|
c*** step ensemble forward |
| 115 |
|
|
write (*,*) 'Stepping Ensemble' |
| 116 |
|
|
do j=1,nens |
| 117 |
|
|
call Model(j) |
| 118 |
|
|
call ReadPickUp(j,n,nx,ny,nz,ytmp) |
| 119 |
afe |
1.1 |
do k=1,n |
| 120 |
afe |
1.2 |
xens(k,j)=ytmp(k) |
| 121 |
afe |
1.1 |
enddo |
| 122 |
|
|
enddo |
| 123 |
|
|
write (*,*) 'Stepped Ensemble' |
| 124 |
afe |
1.2 |
|
| 125 |
|
|
|
| 126 |
|
|
call EnSRF3d(xens,yobs,iobsloc,n,mobs,R,nens,nx,ny,nz,mask) |
| 127 |
|
|
|
| 128 |
|
|
do j=0,nens |
| 129 |
|
|
do k=1,n |
| 130 |
|
|
ytmp(k)=xens(k,j) |
| 131 |
|
|
enddo |
| 132 |
|
|
call WritePickUp(j,n,nx,ny,nz,ytmp) |
| 133 |
|
|
enddo |
| 134 |
|
|
|
| 135 |
|
|
|
| 136 |
|
|
c*** get ensemble mean and variance |
| 137 |
|
|
call ranmean2(xens,ave,n,nens) |
| 138 |
|
|
call ranvar(xens,ave,n,nens,var) |
| 139 |
|
|
|
| 140 |
|
|
c*** calculate obs and anal err |
| 141 |
|
|
obserr=distsub(yobs,yobsfull,n) |
| 142 |
|
|
analerr=distsub(yobs,ave,n) |
| 143 |
|
|
write(6,*) "\tOBSERR\t", "\tANALERR" |
| 144 |
|
|
write(6,*) obserr, analerr |
| 145 |
|
|
write(2,*) iters, obserr, analerr |
| 146 |
|
|
write(3,'(250f18.10)') (ave(i), i=1,n) |
| 147 |
|
|
|
| 148 |
|
|
call flush() |
| 149 |
|
|
|
| 150 |
afe |
1.1 |
enddo |
| 151 |
|
|
|
| 152 |
|
|
return |
| 153 |
|
|
end |
| 154 |
|
|
|
Parent Directory
| 
Revision Log
| 
Revision Graph