Skip to content

fix: anchor MECS sector shares on 2022 with an annual path back to 2018 (#1210) - #1318

Open
catherinebirney wants to merge 9 commits into
mainfrom
cb_mecs
Open

catherinebirney wants to merge 9 commits into
mainfrom
cb_mecs

Conversation

@catherinebirney

@catherinebirney catherinebirney commented Nov 19, 2026 •

Copy link
Copy Markdown
Member

cc: @MoLi7
Closes: #1210

What changed? Why?

Problem

Sector shares in Energy_manufacturing_national_nowcast_{year} only changed when the method switched surveys: MECS 2018 through 2020, then MECS 2022 from 2021. So the whole 2018–2022 change landed in a single year (e.g. 32511 HGL share jumped at 2021). MECS 2022 also reports the chemical sector split far more fully than 2018 (natural gas and HGL), so the two surveys are not equally good end points.

Fix

  • All years use MECS 2022 as the anchor, and the Use table that splits MECS parents is fixed at census 2022 (split_year: 2022), so there is no 2017→2022 switch in the split either.
  • New clean_fba_w_sec step, apply_mecs_annual_share_path (EIA_MECS.py), runs on the MECS activity sets (table_3_1_coal, table_3_1_ng, table_2_1, table_2_1_ng) after parent-incomplete-child remainders and sector mapping:
    • From 2022 on, shares are MECS 2022 unchanged.
    • Before 2022, each piece's share within its fuel moves from 2022 toward 2018 by w = lambda x min(2022 - t, 4) / 4 (geometric when both shares are above 0, linear when one survey reports 0), then shares are renormalized so each fuel total stays at the MECS 2022 value.
    • lambda = max(0, 1 - (v_2022 + v_2018) / (s_2018 - s_2022)^2), using EIA's published RSEs: the part of the survey-to-survey change larger than its sampling noise. * cells count as RSE 100%; a remainder adds its published children's variances.
    • lambda = 0 (keep the 2022 share) where the 2022 piece is 0, the 2018 cell is D/Q or GHGRP-filled, the piece is missing in 2018, its published children differ between surveys, or the fuel is Petroleum (2018 "Other" includes hydrogen, which 2022 reports in its own column).
  • NAICS 2017→2022 renames (322121+322122 → 322120, 336111+336112 → 336110) come from NAICS_Year_Concordance, using the catalog's by_year vintages. A NAICS code that splits into several newer codes raises an error rather than being skipped.

Checked and ruled out

  • GHGRP subpart C as the annual path: it follows steel activity closely, but outside steel it does not predict the 2018→2022 feedstock change (e.g. 325211: MECS gas ratio 2.13 vs GHGRP C 1.02; 325199: 0.54 vs 1.03). We use a flat path, with RSE weighting deciding how much of the 2018 end point to trust.
  • RSE zeros: many cells have Spread = 0. The raw EIA RSE sheets publish those zeros, for cells where every plant is surveyed (e.g. all of 331110, 311 coal), so they are used as published, with no floor.
  • Sectors dropped in 2022 (e.g. 315 coal): 2018 * (under 0.5, RSE 99.4%) vs a reported 0 in 2022. Not real changes, so they are not added back. After the renames, no 2018-only piece with a real value remains.
  • Table 3.2 (TBtu) instead of 3.1 for coal: finer resolution, but kept Table 3.1.
  • Fuel totals: the 2020–2024 builds show each fuel total equal to MECS 2022 in every year (natural gas table_2_1_ng 705.75 bcf, table_3_1_ng 6,796 bcf; HGL 976.75 million bbl).

Testing

Fuel totals, every year 2017–2024

Activity set Flowable Unit Total
table_2_1_ng Natural gas billion cu ft 705.75
table_3_1_ng Natural gas billion cu ft 6,796.00
table_2_1 HGL million bbl 976.75
table_3_1_coal Coal kg 19,277,675,725
table_2_1 Petroleum MJ 1.375136e12

Petroleum energy shares stay on the 2022 values in every year (lambda_zero_flownames). The largest HGL, natural-gas, and coal shares:

HGL share of table_2_1, %

Sector 2017 2018 2019 2020 2021 2022 2023 2024
32511 44.95 44.95 47.32 49.41 51.19 52.62 52.62 52.62
325211 30.60 30.60 30.74 30.62 30.27 29.69 29.69 29.69
32519 16.87 16.87 15.76 14.93 14.45 14.36 14.36 14.36

Natural gas feedstock share of table_2_1_ng, %

Sector 2017 2018 2019 2020 2021 2022 2023 2024
325211 4.46 4.46 5.76 7.37 9.35 11.76 11.76 11.76
32512 17.25 17.25 15.92 14.58 13.23 11.90 11.90 11.90
32519 28.81 28.81 28.22 27.64 27.07 26.50 26.50 26.50
32531 46.30 46.30 47.01 47.37 47.35 46.90 46.90 46.90

Natural gas fuel share of table_3_1_ng, %

Sector 2017 2018 2019 2020 2021 2022 2023 2024
32411 15.30 15.30 15.15 14.96 14.73 14.48 14.48 14.48
32519 12.27 12.27 11.78 11.30 10.82 10.36 10.36 10.36
325211 6.45 6.45 6.63 6.81 6.97 7.12 7.12 7.12
32511 4.66 4.66 4.93 5.20 5.47 5.74 5.74 5.74
32212 3.53 3.53 3.30 3.08 2.87 2.66 2.66 2.66

Coal share of table_3_1_coal, %

Sector 2017 2018 2019 2020 2021 2022 2023 2024
32731 24.85 24.85 26.05 27.01 27.74 28.24 28.24 28.24
32519 13.36 13.36 12.30 11.27 10.31 9.41 9.41 9.41
311221 4.44 4.44 5.44 6.59 7.91 9.41 9.41 9.41
32212 8.88 8.88 7.69 6.59 5.59 4.71 4.71 4.71
32518 8.88 8.88 7.69 6.59 5.59 4.71 4.71 4.71
31121 0.00 0.00 0.41 0.82 1.24 1.64 1.64 1.64

The biggest single-year share step in this set is 2.41 points (325211 natural gas feedstock, 2021 to 2022). That step is one year of a 7.30-point move from 2018 to 2022, spread across 2019–2022. Coal sectors 31121, 311224, 311225, and 31123 are 0 in 2017 and 2018, so those four rows are absent from those two parquets, and they rise linearly after that.

@catherinebirney
catherinebirney marked this pull request as ready for review November 20, 2026 14:02
@catherinebirney

Copy link
Copy Markdown
Member Author

@WesIngwersen I've only run the Energy_manufacturing_national_nowcast_{2017–2023} FBS with this new approach to the MECS data. I havent had a chance to test on the GHG FBS.

I'm headed to the airport, but will try to update testing/run the ghg fbs when/if i can

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Non-energy use and manufacturing gas split: anchor sector shares on MECS 2022 and give them an annual path

1 participant