Stan Benjamin and Ethan Gutmann asked about the possibility of running CTSM with the same processor decomposition as the atmosphere, in order to avoid the overhead of a global processor redistribution. Negin and Bill S answered: LILAC provides that flexibility, though some work would be needed in CTSM to allow it to reuse the atmosphere decomposition.

The below notes capture some of the discussion, but do not attempt to provide a summary of the presentations.

Negin points out: we should have a WRF configuration that uses SNICAR and VOCs.

Are there any significant concerns with the idea of evolving CLM and NWP configurations in tandem? Nothing raised.

Andy: Is there a second phase that is about evaluation / refinement of this configuration? For example, Christa's group runs a number of tests of a proposed configuration to ensure consistency with a NLDAS configuration.

• Mike & Dave: Yes, we'll be doing more along these lines.

Fei: what about plant breeding and its importance for crop yields? Suggestion that plant breeding is responsible for about 50% of historical yield increases (e.g., allowing increased planting density). Dave L: agrees that that is something we're missing.

Fei: What about increasing urban areas? Answer: this is included in the same global data set, but CTSM doesn't yet allow transient urban areas.

Ned: How many other modeling groups are pursuing this? Dave L: Most/all major ESMs are going down this path. CLM may have more emphasis than many others, but there are differences in terms of who is focusing on what: e.g., CLM doesn't have much in the way of pasture, but has more in the way of crops than many others.

Roy views this and the CLM efforts as very complementary: global vs. high-resolution regional. It will be good to work together moving forward.

(These notes just capture a subset of the presentation: mostly things that were discussed beyond what Danica's slides showed.)

CLM does well at capturing FAO crop yields from around 1960 - 1990, but then CLM levels off in 1990s and early 2000s, whereas FAO yields continue to increase. A likely cause is that CLM doesn't capture intensification: new varieties, greater planting density.

This includes irrigation; this is separate per crop type, because each crop is on its own column in CLM.

In the US: CLM underestimates corn yields (corn yield distribution is one of the bigger issues in CLM), overestimates wheat yields, but overall reasonable. (This run used GSWP3 atmosphere forcing.)

Global crop yield evaluation: Based on EarthStat (which blends multiple data products).

Note that we have NOT spent much effort on parameter adjustments: we were happy to get global yield distributions in the right ballpark, and there's certainly still a lot of room for improvement.

The model performs poorly in places like India, where water is a bigger control than temperature; this is a high-priority item to address.

A big question in carbon cycle science is why there has been an increase in the amplitude of the CO2 seasonal cycle; CLM supports the hypothesis that active crop management could be a partial cause.

Note that crop management here includes irrigation and fertilization, as well as active planting and harvesting, as well as having multiple crop types.

How expensive is it to turn on the crop model? Maybe something like 40% cost increase; some (much?) of this comes from the fact that there are multiple soil columns, so need to do the soil calculations that many more times.

• Question: will Sam Levis's current work to allow the collapse of vegetation types allow collapsing crop types to fewer types? No, because that would have been complex / confusing. There are other mechanisms for collapsing crop types, though.

Question: how are state variables initialized when a new crop enters a grid cell? Soil carbon and nitrogen are taken from columns that are shrinking to make room for this new crop. Currently, water and temperature states are just taken from the grid cell's natural vegetation column (with some flux adjustments for conservation); eventually, would like to do more rigorous state initializations for water and energy, too.

Question: how do you deal with changes in irrigation practices over time? Currently that isn't handled. However, there are now hooks in place that will start to allow that. But we still don't really represent flood irrigation (as for rice).

Question: Have you looked at this in the fully-coupled model? Danica looked at this a while ago, and yields were somewhat low. Note that the crop model is active in all of our cmip6 runs; we expect to see (possibly large) biases.

Question: Is there any thought about developing a simpler, less expensive crop model for NWP applications? Answer: the crop model isn't really fundamentally more expensive; it's mainly that there are multiple crop columns. For NWP applications, you'd typically have just one crop per grid cell, so this wouldn't be an issue.

(These notes just capture a subset of the presentation: mostly things that were discussed beyond what their slides showed.)

Lesson about irrigation: model doesn't need to be very sophisticated to capture the total annual irrigation, but the timing is harder to get right. One issue is that the timing of irrigation relative to annual growth period differs for different crops.

• Question: could you have a crop-dependent LAI threshold rather than crop-dependent GDD thresholds? Fei feels it's a lot harder to get that right.

The actual farmer decision for irrigation is much more complicated than just being based on soil moisture - e.g., based on crop health/state and weather forecast.

You can calibrate parameters to get a better match of regional irrigation, but you're then masking missing processes - such as the absence of rice in this model.

Watersmart system: Mike brings this up because he could imagine possibly tackling this in parallel with CTSM as with Noah-MP.

Farmers want maps of 3-day total ET, to help them determine whether to irrigate.

Question: does this incorporate current vegetation state? Not currently: it always goes back to planting; but plan is to try to assimilate current (or slightly-in-the-past) state.

What is the vision for merging the Noah-MP and CTSM efforts? Some short-term things are:

• Introduce ability for specified planting and harvest dates in CTSM
• Tile drainage (not present in either model yet)

Is there a barrier to merging these if there are aspects - like C & N cycling - that are important to the climate community, but not important to the NWP community, and it carries too much performance overhead? Dave L: This remains an open question; he suspects that computational time is less of an issue than scientific complexity, which makes it harder to massage the model into "correct" states. (Note that Noah-MP has a very simple C cycle model.)