Address hardcoded 2018 MECS in GHG methods - #688

Energy FBS national totals (2022 MECS) — follow-up fix incoming

While building the NEU activity-set pivot for this review, I found a pipeline bug that inflates 2022 MECS energy weights (e.g. Table 2.1 HGL national FBS ≈ 2,176 mmbbl vs MECS 31-33 control ≈ 870 mmbbl, ~2.5×).

Root cause: EIA_MECS_Energy is assigned NAICS_2022_Code for the 2022 survey (source_catalog.yaml). Energy methods target Cornerstone (target_naics_year: 2017), so map_to_sectors calls convert_naics_year (2022 → 2017). That path is correct for NAICS vintage conversion, but the final aggregation step reassigns group_id = df.index on every fan-out row after the Cornerstone crosswalk maps one MECS activity to multiple target sectors. Proportional BEA attribution then treats each fan-out row as an independent full group instead of splitting one group_total.

2018 MECS does not hit this path (source and target are both NAICS 2017, so convert_naics_year is skipped and group_id survives fan-out). National HGL stays ≈ 888 mmbbl.

Separate note: Change in disclosure for some sectors causes significant jumps between MECS years.

There are just a few cases, but data suppression, especially D (which is to avoid revealing confidential information) is different between 2018 and 2022. For example, **32531 has a +10.7 Mt jump in which aligns with 2022 MECS disclosing Table 2.1 NG feedstock for 325311 that 2018 suppresses (D → 0). We may want to consider alternate approaches to handle these cases (which are different than the * which reflect small amounts).

Also turns out there are some extra blank columns in the 2022 Non-energy tables so the fuels were mis aligned

Ok after we pulled in those additional fixes, I am reviewing this with a few frames in mind:

1. When we use this new approach, but fix it to 2018 MECS, how different is it from our current pinned v0.3 GHG FBS? This should help tease out differences that result from the methodology of this revision, separate from the vintage update. (I've created a separate FBS method that does this).
2. I'm looking to confirm that there are no non-MECS related changes in the results (note this also includes understanding how the manufacturing and non-manufacturing shares may have changed).
3. I'm trying to tease out what, if any, mapping differences that are rolled in to this change might impact.
4. And finally, the update to use 2022 MECS data will likely significantly impact some of the splits of emissions - this is expected - we just need to understand how and why.

I'm referring to these methods as:
A - our pinned v0.3 GHG FBS
B - 2024 GHG FBS with the new approach, but forcing it to use 2018 MECS (to tease out methodology changes only)
C - the new designed 2024 GHG FBS, using 2022 MECS.

Verdict so far: Isolation is working as designed. National totals match (4,924.087 Mt). Only T_3_11 and T_3_14 move. Everything else is an explicit no-change. A→C decomposes cleanly into A→B + B→C (zero residual).

What the three arms are saying

Arms: A = v0.3 pinned (CEDA, 2018 MECS); B = GHG_national_Cornerstone_2024_mecs2018_isolation (flowsa, 2018 MECS, ratio_mecs_year: 2022); C = cb_mecs production (flowsa, 2022 MECS).

Findings comparing A (pinned v0.3 method using old CEDA allocators and 2018 MECS data) to B (new approach, still using 2018 MECS data). Results are B relative to A:

Stationary Combustion

Non-Manufacturing

1. Non-mfg petroleum +7 Mt at uniform +4.9% — Non-mfg petroleum_industrial rises 142.614 → 149.615 Mt (+7.001). Every non-mfg petroleum sector scales by the same +4.91% (factor 1.049); within-non-mfg shares match A exactly. Those non-mfg stationary-combustion increases are only this pool increase — not a remapping among non-mfg sectors.
2. No change in NG or Coal non-manufacturing.

Manufacturing

3. Coal manufacturing combustion emissions spreads to more sectors — Total emissions are flat, but are spread across more sectors in new approach where those sectors report * or < 0.5 TBtu (assigned as 0.25 TBtu), instead of dropped in v0.3. 44 new sectors report minimal coal combustion emissions (60 vs 16)
4. Natural gas manufacturing combustion emissions, adjustments to residual approach and mapping — Mfg ng_manufacturing pool is flat (378 Mt); same ~247 sectors on both arms; |sector churn| ≈ 7.4 Mt (share remapping only). Arm B weights come from Energy FBS table_3_1_ng: published MECS leaves keep their amounts; parent incomplete-child residuals are split with BEA_Detail_Use purchases of natural gas. Arm A uses curated CEDA industry→MECS lists (often equal among target leaves).

Non-energy use

1. F01000 lubricants fully explain the change in transportation lubricants— Routing transport lubricants to households adds F01000 +3.442 Mt on NEU (was 0 under v0.3). That gain is exactly offset by losses on the prior absorbers: transport 48/49 (−2.155 Mt), GSLGO (−0.837), and special S00* (−0.450).

in progress