NCAR
bug: Sampling Error Correction doesn't affect adaptive inflation updates
Sampling error correction does not appear to impact the adaptive inflation updates.
The value of the sampling error correction is read from the table in subroutine update_from_obs_inc in assim_tools_mod.f90. It is then used to modify the value of the reg_coef which is subsequently multiplied by the obs_inc to get the state_inc which is returned. The updated value of the reg_coef is also returned to the calling location in subroutine obs_updates_ens. However, the reg_coef is a local variable array in obs_update_ens and is not returned from this subroutine. Subroutine update_varying_state_space_inflation does not make any use of the sampling error correction table. It seems like it would be appropriate to apply sampling error correction to reduce the strength of the relation between the observation and state variable when updating the inflation, too.
The current behavior has been unchanged since the original implementation of the adaptive inflation algorithms. Should we update the behavior to include the use of sampling error correction, or should we document that it is not being done?
I can confirm that the regression coefficient used to update the inflation does not apply SEC. I've always thought that SEC is used to update the inflation but I was mistaken. Question: Do we actually need SEC in the inflation update? To answer this question, we'll need some testing. I'll conduct some experiments with low-order models to gain some insights. More to follow ..
It is my opinion that the most correct thing would be for SEC to apply to limit the inflation update, just like other adjustments to the regression coefficient. However, the impact of losing backward compatibility for all applications is a concern. We would need to do a very broad set of tests before deciding to implement this.
On Mon, Dec 9, 2024 at 2:46 PM Moha Gharamti @.***> wrote:
I can confirm that the regression coefficient used to update the inflation does not apply SEC. I've always thought that SEC is used to update the inflation but I was mistaken. Question: Do we actually need SEC in the inflation update? To answer this question, we'll need some testing. I'll conduct some experiments with low-order models to gain some insights. More to follow ..
— Reply to this email directly, view it on GitHub https://github.com/NCAR/DART/issues/778#issuecomment-2529589808, or unsubscribe https://github.com/notifications/unsubscribe-auth/ANDHUIWNVH55B6ZCGFR7RB32EYFRXAVCNFSM6AAAAABSVMM7D6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKMRZGU4DSOBQHA . You are receiving this because you were assigned.Message ID: @.***>