SP-mode not working as it should in ELM-FATES

[ckoven]

Based on some simulations that @JessicaNeedham has run, we've realized that ELM-FATES isn't running SP-mode properly. What appears to be happening is that LAI is set in the first timestep but then doesn't update to follow the phenology dataset, it just stays fixed. I just did a run using CTSM-FATES, and SP-mode is working as expected there, i.e. LAI follows the satellite dataset. So something in the ELM driver data does not seem to be passing the SP-mode LAI data properly.

[ckoven]

Just for reference/evidence, attached are jupyter notebook cells showing timeseries of LAI at some random gridcell from the two runs. The first one, with the flat line, is using the ELM-FATES output; the second one, with a periodic curve is the CTSM-FATES output. Interestingly, the initial values aren't even identical between the two runs either, not sure if that is relevant or not.

[glemieux]

Nice plots showing the issue Charlie. I'll prioritize this.

[glemieux]

@ckoven what's the e3sm hash that your are working from?

[JessicaNeedham]

@glemieux the problem occurs in E3SM runs that are on c63cce2.

[glemieux]

I think I determined part of the issue. The call to SatellitePhenology in elm_drv is in the wrong place and unreachable:
https://github.com/glemieux/E3SM/blob/3c7d4fd8efb65ab7bf919b7fac92b7aad6e2e51d/components/elm/src/main/elm_driver.F90#L1120-L1124.

I have a branch based off of c63cce2 that has a quick fix to correct this: https://github.com/glemieux/E3SM/tree/fates-satphen-call-fix. I'm not sure if the doalb check is really necessary here, as it isn't used in the interpMonthly call when using fates sp mode in elm-fates and it was recently removed from clm-fates iirc. Here's the quick plot near the lat/lon you were showing above, which is changing now, but doesn't really look the same as the clm-fates plot:

[rgknox]

Nice find @glemieux , that fix makes sense to me

[ckoven]

Thanks @glemieux, that is great news!

[JessicaNeedham]

Thanks @glemieux! I ran a 50 year spmode simulation using your branch and it looks like the issue is fixed. Attached figure is the last ten years of the simulation showing the same point as in @ckoven 's CTSM run. Periodicity and LAI values look good although a bit higher than the CTSM values. However, the two runs are using different surface data files which might explain this.