[Dart-dev] DART/branches Revision: 10961

[dart at ucar.edu]

thoar at ucar.edu
2017-01-27 09:58:19 -0700 (Fri, 27 Jan 2017)
557
There are variable attributes that help understand the context of the variables.
The dimension and variable containing the list of ensemble sizes supported is now
called 'ens_sizes'.

Minor cleanup on formatting for the rest of the source codes.

Changed the html document to reflect the intended usage of the 
sampling_error_correction_table.nc lookup table.

Note that the actual values in the lookup table are a little different
than the values in the final_full.N files, and that the assim_tools
code does not actually use the netCDF lookup table yet.




Modified: DART/branches/rma_trunk/system_simulation/correl_error.f90
===================================================================
--- DART/branches/rma_trunk/system_simulation/correl_error.f90	2017-01-27 15:18:32 UTC (rev 10960)
+++ DART/branches/rma_trunk/system_simulation/correl_error.f90	2017-01-27 16:58:19 UTC (rev 10961)
@@ -6,9 +6,8 @@
 
 program correl_error
 
-use types_mod
-use random_seq_mod, only : random_seq_type, init_random_seq, random_gaussian, &          
-   twod_gaussians, random_uniform
+use      types_mod, only : r8
+use random_seq_mod, only : random_seq_type, init_random_seq, twod_gaussians 
 
 implicit none
 
@@ -40,83 +39,78 @@
    !!!ens_size = 20
    !!!ens_size = 128
 
+   allocate(pairs(2, ens_size), obs_inc(ens_size), unobs_inc(ens_size), new_unobs(ens_size))
+   write(*, *) 'stats for ensemble size ', ens_size
+   ! Loop through real correlations every 0.05
+   !!!do j = 0, 50
+   do j = 0, 0
+      ! Loop through a large number of samples to get mean
+      correl_mean  = 0.0_r8
+      correl_mean2 = 0.0_r8
+      ratio_mean   = 0.0_r8
+      ratio_mean2  = 0.0_r8
+      do k = 1, sample_size
+         t_correl = j * 0.02_r8
 
-allocate(pairs(2, ens_size), obs_inc(ens_size), unobs_inc(ens_size), new_unobs(ens_size))
-write(*, *) 'stats for ensemble size ', ens_size
-! Loop through real correlations every 0.05
-!!!do j = 0, 50
-do j = 0, 0
-   ! Loop through a large number of samples to get mean
-   correl_mean  = 0.0_r8
-   correl_mean2 = 0.0_r8
-   ratio_mean   = 0.0_r8
-   ratio_mean2  = 0.0_r8
-   do k = 1, sample_size
-      t_correl = j * 0.02_r8
+         ! Generate the covariance matrix for this correlation
+         cov(1, 1) = 1.0;    cov(2, 2) = 1.0_r8
+         cov(1, 2) = t_correl; cov(2, 1) = t_correl
 
-      ! Generate the covariance matrix for this correlation
-      cov(1, 1) = 1.0;    cov(2, 2) = 1.0_r8
-      cov(1, 2) = t_correl; cov(2, 1) = t_correl
-   
-      ! Loop to generate an ensemble size sample from this correl
-      ! Generate a random sample of size ens_size from something with this correlation
-      do i = 1, ens_size
-         call twod_gaussians(ran_id, zero_2, cov, pairs(:, i))
-      end do
-   
-      ! Compute the sample correlation
-      call comp_correl(pairs, ens_size, sample_correl)
-   
-      ! Keep a sum of the sample_correl
-      correl_mean = correl_mean + abs(sample_correl)
-      correl_mean2 = correl_mean2 + sample_correl**2
+         ! Loop to generate an ensemble size sample from this correl
+         ! Generate a random sample of size ens_size from something with this correlation
+         do i = 1, ens_size
+            call twod_gaussians(ran_id, zero_2, cov, pairs(:, i))
+         enddo
 
+         ! Compute the sample correlation
+         call comp_correl(pairs, ens_size, sample_correl)
 
-      !-----------------
-      ! Also interested in finding out what the spurious variance reduction factor is
-      ! First need to compute sample s.d. for the obs and unobs variable
-      do i = 1, 2
-         call sample_mean_var(pairs(i, :), ens_size, s_mean(i), s_var(i))
-      end do
-      ! Next, compute the increments for obs variable given reduction in s.d. by a
-      a = 0.0_r8
-      do i = 1, ens_size
-         obs_inc(i) = (a - 1.0_r8) * (pairs(1, i) - s_mean(1))
-      end do
+         ! Keep a sum of the sample_correl
+         correl_mean = correl_mean + abs(sample_correl)
+         correl_mean2 = correl_mean2 + sample_correl**2
 
-      ! Regress these onto the unobserved variable
-      do i = 1, ens_size
-         unobs_inc(i) = sample_correl * sqrt(s_var(2) / s_var(1)) * obs_inc(i)
-         new_unobs(i) = pairs(2, i) + unobs_inc(i)
-      end do
+         !-----------------
+         ! Also interested in finding out what the spurious variance reduction factor is
+         ! First need to compute sample s.d. for the obs and unobs variable
+         do i = 1, 2
+            call sample_mean_var(pairs(i, :), ens_size, s_mean(i), s_var(i))
+         enddo
+         ! Next, compute the increments for obs variable given reduction in s.d. by a