Opened 11 years ago
Closed 10 years ago
#356 closed defect (fixed)
Is there an error in ionospheric correction?
| Reported by: | sti | Owned by: | sti,ssy,idculv |
|---|---|---|---|
| Priority: | normal | Milestone: | 8.0 |
| Component: | ROPP (all) | Version: | 7.1 |
| Keywords: | Cc: |
Description
ropp_pp_ionospheric_correction
Line 384
ba_12 = (/ ba_low(1,i) - ba_is(i), &
ba_low(2,i) - ba_is(i)*(f_L1/f_L2)2 /)
I think this is incorrect, and not consistent with user-guide (eq.240).
Please investigate.
Attachments (11)
Change history (23)
comment:1 by , 11 years ago
| Owner: | changed from to |
|---|---|
| Status: | new → assigned |
comment:2 by , 11 years ago
| Owner: | changed from to |
|---|
comment:3 by , 11 years ago
comment:4 by , 11 years ago
In other words, use the actual obs - MSIS bangle diff, as in Eqn 2.43, not the low-frequency smoothed version.
There seems to be general agreement about this, eg these T+3 refrac O-Bs (courtesy Stig) show much smaller std devs when using ba_diff rather than ba_low:
For obs within 1 hour of the analysis time the effect is even bigger:
by , 11 years ago
| Attachment: | refrac_3h4d_testxxa.png added |
|---|
by , 11 years ago
| Attachment: | refrac_1h4d_testxxa.png added |
|---|
comment:5 by , 10 years ago
Making this change to ropp_pp_ionospheric_correction.f90 at r4214 causes the following difference in refractivity when the 50 GRAS profiles in IT-PP-02.nc are passed through ropp_pp_invert_tool:
average ROPP8.0 - ROPP7.1
standard deviation ROPP8.0 - ROPP7.1
An sd of ~ 0.01 N-units at 15km (where N ~ 25 N-units) is about 0.04%, which is the same ballpark as the thin-thick differences shown in Stig's plots above.
by , 10 years ago
| Attachment: | refrac_av_diff_356.png added |
|---|
by , 10 years ago
| Attachment: | refrac_sd_diff_356.png added |
|---|
comment:6 by , 10 years ago
| Resolution: | → fixed |
|---|---|
| Status: | assigned → closed |
All looks good, so closing ticket.
comment:7 by , 10 years ago
| Resolution: | fixed |
|---|---|
| Status: | closed → reopened |
comment:8 by , 10 years ago
Note the large increase in noisiness of the refractivity, esp below 25km, when ropp_pp_invert_tool is run on the single COSMIC occ in the first test of the make test procedure:
ROPP7.1
ROPP8.0
Some of this noise must be because we start from (presumably) smoothed bending angles. When we process from phases, using ropp_pp_occ_tool, the differences narrow:
ROPP7.1
ROPP8.0
by , 10 years ago
| Attachment: | ropp_pp_comp1_ROPP71.jpg added |
|---|
by , 10 years ago
| Attachment: | ropp_pp_comp2_ROPP71.jpg added |
|---|
by , 10 years ago
| Attachment: | ropp_pp_comp1_ROPP80.jpg added |
|---|
by , 10 years ago
| Attachment: | ropp_pp_comp2_ROPP80.jpg added |
|---|
comment:9 by , 10 years ago
This increased noisiness is enough to fail the test folder, eg
IT-PP-02 ROPP7.1
IT-PP-02 ROPP8.0
by , 10 years ago
| Attachment: | IT-PP-02_ifort12_ROPP71.png added |
|---|
by , 10 years ago
| Attachment: | IT-PP-02_ifort12_ROPP80.png added |
|---|
comment:10 by , 10 years ago
Since everyone's agreed that this change should go ahead, we therefore need to update the reference data.
by , 10 years ago
| Attachment: | figure_1.png added |
|---|
comment:11 by , 10 years ago
The jump in behaviour at 20 km is consistent with what Sean found earlier ('old' = ROPP7.1; 'new' = ROPP8.0):
comment:12 by , 10 years ago
| Resolution: | → fixed |
|---|---|
| Status: | reopened → closed |
Updated ref figs at r4226.
Re-closing ticket.
The key thing appears to be a need to replace
ba_12 = (/ ba_low(1,i) - ba_is(i), & ba_low(2,i) - ba_is(i)*(f_L1/f_L2)**2 /)by
ba_12 = (/ ba_diff(1,i) - ba_is(i), & ba_diff(2,i) - ba_is(i)*(f_L1/f_L2)**2 /)