first pass of the statewide synthesis/meta-analysis calibration runs

[divine7022]

Report status and follow-up — 20 October 2026

Historical 61-site cover-only forward comparison. Keep its 0.11 Mg C ha⁻¹ yr⁻¹ result separate from the later 65-site, 0.13 comparison. This is not an executed synthesis calibration.

Results and implementation: six practice/synthesis comparisons #288, calibration report/generator PR #8, pinned results manifest, and artifact retrieval. Target/likelihood choices belong to #245 and implementation to #247. The requested other-practice/N₂O/CH₄ comparisons are recorded in #288; remaining estimand, pool, duration and uncertainty corrections stay there. Plot-level SD must not silently become the uncertainty of a replicated synthesis mean.

---

using the model inputs generated by Chris on 2026-09-14 , I have completed the first statewide forward pass for cover-crop soil C using paired ON/OFF runs (2016–2023). there are 61 usable paired sites from the available dated cover-crop events. ( among 1000 design points, 517 annual, 94 had ever been COVER flagged , 61 had a dated, event visible cover planting, those 61 became the usable paired ON/OFF runs )

model gives +0.11 Mg C ha-1 yr-1 v/s the Poeplau & Don synthesis estimate of +0.32, so the gap is 0.21 Mg C ha-1 yr-1. The model gets the direction right, compared with ~91% positive in the observed synthesis data.
supplementary data give an observed spread of SD around 0.744 Mg C ha-1 yr-1. So the 0.21 gap is only ~0.28 SD, ie. well within the observed variability; that said we’re therefore not calling this definitive underprediction at this point.
given that the forward model gets the direction right, but the mean is 0.11 versus the 0.32 synthesis estimate, do you think the next step should be a small EKI calibration against the cover-crop synthesis . For the EKI observation uncertainty, I now have the observed spread of about 0.744 from the supplementary plot-level rates. my suspection is that to use that as sigma for the pseudo site observations is the next question ?

anyway In parallel, we’re looping back with other practice/outcome combinations in the synthesis table to identify the next candidates for calibration, including cover-crop N2O, N2O/CH4 effects for the other practices, and the other soil C practice effects

event files are here:
/projectnb/dietzelab/ccmmf/usr/akash/statewide_cover/cover_onoff_events/
on/ is chris's monitoring events for the 61 sites. off/ is the same file with the detected cover season deleted. detection is the LandIQ COVER flag joined to the planting table chris's events were generated from, on site/year/season; deletion is that season's plant line plus its termination harvest, with fert/irrig/till/leafon/leafoff left identical in both arms so the contrast is just the cover vegetation. all_diffs.txt shows exactly what got removed per site (97 plant and 89 harvest lines across 97 cover seasons), and deletion_accounting.csv has the dates and payloads. README in the dir has the full construction chain

[divine7022]

moving the slack David's replay here:

The 0.1 vs 0.32 is definitely sufficient.

The next step is to quantify what the other practice change effects are, if that can be done in a straightforward way.

kCN stays at the sipnet default 80

I wouldn't call that a 'default', it is just a number; I don't think it has any rationale behind it.

unless we bring N2O/CH4 into calibration

we should start by evaluating what these numbers look like w/ current calibration

[dlebauer]

Closing, follow up in #288