Bug report

From: David Lawrence <dlawren@ucar.edu>
Date: Wednesday, March 14, 2018 at 3:55 PM
To: Staff Member <Ruby.Leung@pnnl.gov>
Subject: Critical bug in CLM that likely affects ELM

Hi Ruby,
I'm not really sure who is in charge of land model development in E3SM anymore so I am sending this to you. Please forward to whoever is responsible.
We found a significant bug (actual series of bugs) in CLM5 with respect to the energy imbalance over land (global energy imbalance of +0.08 W/m2 into land, up from 0.01 W/m2 in CLM4). This energy imbalance will cause the oceans to lose an equivalent amount of heat if the TOA is balanced. We traced the imbalance to features that were introduced in CLM4.5, so these bugs are likely to exist in ELM. See the following pull request for the code modifications required to fix the problems.

https://github.com/ESCOMP/ctsm/pull/307
Also see the following issue for further discussion.

https://github.com/ESCOMP/ctsm/issues/304
I can provide additional information if necessary.
Regards,
Dave

Analysis

16 Mar 2018 : William Riley (Unlicensed) and Gautam Bisht

Gautam and I just spoke with Dave Lawrence, who graciously explained what they found regarding the energy imbalance. Apparently there were several problems that were not captured by the column level energy check, because it wasn't properly accounting for all energy transitions. The problems were mostly associated with phase changes in snow on standing water and snow interactions with radiation absorption in the lake. Apparently this was causing a comparable imbalance in the ocean in coupled mode.

Here's the figure of the problem in CLM5 (of 0.087 W/m2) showing very high spatial variability, pointing to snow, and being pretty large at high latitudes.

20 Mar 2018 : Gautam Bisht

Attached below is the plot of ELM's radiation error for the water cyle experiment. The error is an average of 20-year simulation results from 390-409 years.

The values in the plot corresponds to the following equation, which was provided by Dave Lawrence:

NET:         FSA + FLDS –FIRE – FSH_TO_COUPLER – EFLX_LH_TOT – SNOW_FROM_ATM * latice
                 + QRUNOFF_ICE_TO_COUPLER * latice
Note that FSH_TO_COUPLER = FSH + FSH_PRECIP_CONVERSION
                         + FSH_RUNOFF_ICE_TO_LIQ – EFLX_DYNBAL

The highlighted variables are 0 for ELM.

I have implemented CLM's bug fixes in ELM (bishtgautam/lnd/radiation-fix branch) and am running two land-only simulations at 2-deg grid with and without the bugfix on Edison. Use of land-only and coarse resolution simulation at 2 deg is justified because Dave Lawrence mentioned the radiation error was reproducible in a land only compset with CLM. I will get back to you all once the two simulations complete.

21 Mar 2018 : Gautam Bisht

Attached below is plot showing 20-year average net radiation error for two recently completed ELM runs. The top panel shows radiation error for simulation with master and the bugfix branch. The bottom left panel shows the difference of top two panels.

20 Mar 2018 : Chris Golaz

Comparison of net flux of energy at TOA with the change in ocean heat content for the first 400 years of the control simulation.

Years	TOA	Ocean	Difference
0001-0100	0.0000	0.0285	0.0284
0101-0200	0.0065	0.0350	0.0285
0201-0300	-0.0307	-0.0017	0.0290
0301-0400	-0.0149	0.0098	0.0246
0001-0400	-0.0098	0.0179	0.0276

Over the entire 400 year period, there is a net energy loss at TOA of 0.0098 W/m2 while the (liquid) oceans gain the equivalent of 0.0179 W/m2 of heat. The imbalance between TOA and ocean heat is +0.0276 W/m2. This is a very small number for a climate model, much smaller than any external forcing we are interested in. At least on a globally averaged scale, the impact of the energy conservation bug in ELM is small.

Browser not supported