Version 2 Summary

[dt-woods]

The following summarizes the updates and improvements made to the ElectricityLCI Python package as completed towards the release of version 2, which included over 20 new issues on GitHub over the past six months¹. The updates since March 2024 resolve over 16 tracked issues with the goals of preserving the 2016 electricity baseline model run and refining the 2022 baseline. Several quality improvements, fixes, documentation updates, and miscellaneous changes to the code were made over this period. The key changes include:

1. Improve module and method documentation throughout the package (address Complete missing documentation #204).
2. Adopt a new temporal balancing authority list with FERC mapping using EIA's 930 reference table (resolve How to map balancing authorities within EIA and FERC regions? #215 and NCR ID column name mismatch #194).
3. Remove industrial co-generating facilities from inventory list (resolve Addressing the Industrial Cogeneration Problem and Implementing the filter in model_config #232).
4. Correct output exchange flows labeled as resources (resolve Fix output exchange flows mislabeled as resources #233).
5. Correct the electricity flow for at-grid consumption mixes from 120 V to 2300--7659 V (resolve Incorrect output flow for "at grid; consumption mix" #238).
6. Incorporate missing metadata from the Federal Elementary Flow List (e.g., CAS number, formula, synonyms, and FEDEFL version) (resolve FLCAC Metadata Requirements #239 and error accessing FLCAC api in _read_fedefl() #257).
7. Remove problematic electricity from NETL's 2016 plant water inventory (resolve Issues with the electricity column in generate_plant_water_use()  #242).
8. Implement 2023 coal model transportation and extraction and processing LCIs (address Update inventories | Coal #243).
9. Create facility-level coal transportation LCI from 2023 coal model for coal facilities since 2016, gap-filled with U.S. average transportation modes and distances (address Update inventories | Coal #243).
10. Correct the temporal representativeness of inventory processes to match the reporting agencies' inventory vintages and not the configured LCA modeling year (resolve Set temporal representativeness attribute for processes to inventory vintage, not target year #244).
11. Correct inventory sourcing, emissions aggregation, and facility matching across variable inventory years (resolve Who's putting NaNs in my exchange tables? #246).
12. Correct the unit conversion, aggregation method, and omit nuclear, wind, solar, and solar thermal from plant efficiency filter (resolve Error in calculate_plant_efficiency #247).
13. Introduce the Hawkins-Young uncertainty calculation for emissions inventories and make its calculation configurable (resolve Add configurable uncertainty calc #248).
14. Remove legacy code and outdated datasets (resolve What is the intended use of technosphereflows.py? #249).
15. Improve EIA bulk data filtering to support backwards compatibility---see EIA bulk data problem in Appendix (resolve Empty dataframe in eia_io_trading.py #251).
16. Implement EIA API call for U.S. Electric System Operating Data (resolve Introduce EIA API for managing U.S. Electric System Operating Data #253) and fix EPA API calls for CEMS data, including the reduction of data handling (i.e., daily data in place of hourly) and improvement of temporal representation (i.e., use U.S. time zone, not GMT).
17. Fix EIA923 fuel receipt CSV creation (resolve Error in reduced schedule CSV file #259).
18. Incorporate 2023 solar PV construction LCI fixes (resolve Update inventories | Solar PV construction #265).
19. Fix product flows labeled as elementary flows (resolve Product flows labeled as elementary flows in coal and natural gas extraction and processing #267).
20. Remove geothermal power plants from the efficiency filter, returning geothermal to the U.S. energy mix (resolve Add missing sources from renewables #278).
21. Remove electricity input flows from renewable generation processes (resolve Remove Electricity input from renewable O&M processes #281).
22. Purify power-plant construction processes found in fuel-specific generation processes, allowing multiple construction processes in mixed-fuel generation processes (resolve Clean up power plant construction at fuel-generation process level #282).
23. Scale renewable upstream operations and maintenace emissions on their data source year (resolve Renewable electricity O&M emissions scaling based on EIA year rather than source year #283).
24. Remove unmapped flows from the upstream coal inventories (address Remove non-FEDEFL elementary flows from coal mining and processing LCI #285).
25. Implement and test new regionalized power plant construction processes, with new renewable power plant construction metadata (address Split renewable upstream into separate system processes #150); see memory management challenge.
26. Add API sections and coal and renewable model vintages to model configuration YAML files and improve YAML file documentation.
27. Provide facility-level inventory return option in get_generation_process_df method (i.e., skips aggregation for facility-level analyses).
28. Correct the RCRAInfo inventory vintage from 2019 to 2021 for the 2021 and 2022 baselines in the YAML model configuration files.
29. Fix coal basin process proxy process gap-filling.
30. Correct zero-flow filtering to skip elementary flows.
31. Add EPA process and flow pedigrees to JSON-LD outputs.
32. Improve file and folder handling during startup (e.g., for saving new log files) and processing (e.g., missing FRS bridge file).
33. Add post-processing as a model configuration parameter.
34. Make main in main.py a runnable method.

Unresolved new issues include failing to read the XML format of EIA's public coal workbooks (#230), adding energy storage to generation or storage mix process (#245), identifying unlinked coal and natural gas upstream processes (#252), addressing high tritium emissions in coal and renewable inventories (#255), and aggregating U.S. average inventory by fuel type (#256). In addition to these unresolved issues, a current list of challenges is discussed in the Appendix.

Additional key changes and improvements to the code base include the following, which were completed in the development branch in the months prior to March 2024:

1. Significant code formatting, cleaning and organizing (e.g., removal of unused code, commented code, duplicate methods, unused imports, and print statements and addition of log statements).
2. Abstract methods were introduced to avoid spaghetti code and improve modularity. Smaller, reusable functions were created, and the utils.py and globals.py modules were expanded for better usability.
3. Syntax updates were added for latest packages, including the new olca package with all new syntax and the latest pandas version, which required rewrites of defunct methods.
4. Error handling was improved (e.g., during math operations).
5. Documentation was added or updated across all methods, adding documentation to methods where it was lacking. Module documentation was copied from the wiki to the module documentation for accessibility and readability.
6. The StEWI package dependency was incorporated to support additional parts of the code, such as eGRID subregion generation by fuel category.
7. Legacy data transparency was improved by updating outdated data files, tracking down their sources, and reproducing methods for digesting them into the model. URLs that have changed since 2018 were also updated.
8. Missing data was addressed by finding new avenues to gap fill, such as using the EPA's API for CEMS data.
9. Bad data was documented, such as the latest versions of EIA's public coal data being published to 1995 Excel workbooks but appearing to be mistakenly saved as XML spreadsheets, which crash the model.
10. Logging was updated to allow both streaming and a rotating file handler at two different levels (i.e., info and debug, respectively).
11. New postprocess fixes were applied, including removing zero-value quantitative reference flows, fixing flows that are untracked, and correcting exchange internal IDs so they are consecutively numbered.
12. A product system generator was added to the code to improve transparency and match the JSON-LD output with the Fed Commons website.
13. Hotfixes were applied throughout the code to improve efficiency, make corrections, and amend mis-matched data types.
14. Impact assessment methods were removed from the JSON-LD as they were determined to be superfluous to the baseline dataset.
15. The Python setup script was updated for the latest list of dependencies and a contributor's guide was developed for the repository.

Limitations and challenges remaining with the proposed code base are outlined in Appendix A and in the unresolved issues on GitHub (i.e., not marked as "in v2"). The 2021–2022 baseline inventories were created using the proposed version 2 of the code. Select results from these runs are shown in Appendix B, including the reproducibility of the same model config across systems.

Appendix A

The Single Citation Problem

The electricity baseline, as it is found on the Federal LCA Commons, points to a single source, "Ingwersen et al. electricitylci: A Python package for U.S. regionalized electricity life cycle inventory data creation," which is labeled as "in development." This work is abandoned. In the interim, the following citation was recommended: "EPA, 'Electricity Life Cycle Inventory,' March 2022 [Online]. Available: 43ab081"

In 2024, a meeting was held between NETL and EPA to discuss the ElectricityLCI manuscript. A historical manuscript draft was shared between parties. NETL (if resources are available) intends to complete the manuscript, which will focus on describing the tool.

The Data Vintage Problem

In eLCI v2, the coal baseline inventory was updated to reflect the latest literature (see here). Users can select between the previous and current coal model inventories (both are hosted in the Python package's data).

A select number of renewable models (e.g., solar PV, solar thermal, and wind) were updated and corrected. Users can select between renewable model vintages. Inventories for geothermal, hydropower, and nuclear are unchanged.

The 2020 natural gas baseline (published in 2025) is not currently incorporated in v2 (see here for model details).

As more life cycle inventories are added to the model space, it becomes a challenge to host, organize, and track. A stretch goal for v2 is to move to a new data store for inventory models on NETL's Energy Data eXchange (EDX).

The EIA Bulk Data Problem

EIA's U.S. Electric System Operating Data for 2016–2018 are no longer supported in either the bulk EBA (http://api.eia.gov/bulk/EBA.zip) or via their Opendata API portal (https://www.eia.gov/opendata/), which are needed for legacy versions of the electricity baseline. A version of the EBA bulk data file was archived (from March 12, 2024) that has historical data dating back to 2015, which is the only known surviving copy of EIA bulk data available to support the 2016 electricity baseline.

The API Requirement Problem

The ElectricityLCI models require internet access to work. Some of the data downloaded are accessed through APIs (e.g.,EPA, EIA, EDX, and the Federal LCA Commons). API key registration is required from three of the services. The first API key required is from the EPA², which provides data from the Continuous Emission Monitoring Systems (CEMS) through the Clean Air Markets Program Data (CAMPD)—one of the reporting agencies for developing power plant emissions inventories. The second API key required is from EIA³, which provides bulk U.S. Electric Systems Operating Data necessary for net trading calculations. The third is NETL's EDX⁴, which is required for accessing the latest coal model transportation inventory workbook. The Federal LCA Commons utilizes a public API, which is used to gather units, unit groups, and data quality indicators.

The Tritium Problem

In Issue #255, high levels of tritium were identified across renewable and coal inventories, with values ranging from 1×10^9^ to 1×10^22^ kBq, which are orders of magnitude higher than the tritium emissions from the Fukushima disaster of 2011 (i.e., 3×10^12^ kBq), and were tracked to the aluminum production (e.g., from primary and secondary ingot) from 2009--2014 Gabi databases. These high emission values impact renewable inventories (e.g., wind, solar, solar thermal, and geothermal), and coal and gas inventories (via coal mining and petroleum extraction and processing).

The Memory Management Problem

There are a few computations in ElectricityLCI that spike a user's computer's memory, namely: upstream construction regionalization (new), uncertainty calculation, and electricity trading. Memory requirements can be in excess of 20 GB and crash a user's model execution if their computer does not have enough memory.

The NETL Water Use Problem

A significant amount of energy is unaccounted for in the plant water use methods in ElectricityLCI. This error is rooted in the aggregation method, which uses the water type category for grouping. The presence of "not a number" values in this category cause a significant portion of water entries in the dataset to get filtered out of the calculation. A proposed method is presented on GitHub that replaces this data file and calculates power-plant water consumption from EIA's water data (https://www.eia.gov/electricity/data/water/).

The EPA Trading Method Problem

While version 2 of the ElectricityLCI was updated to support the reproduction of the 2016 baseline (i.e., ELCI_1_config), the EPA trading method used in other model configurations needs to be updated, as it is not known to be working.

The Manufactured Uncertainty Problem

Flow uncertainties are computed for life cycle processes based on the Hawkins-Young method, which models uncertainty assuming an underlying log-normal distribution designed to reproduce the calculated emission factor as the expected value of the log-normal distribution and the 95th percentile as the 90% confidence interval (upper). These uncertainty values are designed to provide specific distributions in the results from Monte-Carlo runs and are not to be confused with other common definitions of uncertainty (e.g., random or systematic).

The DQI Problem

The use of the '5;5' process DQI (undocumented) is used throughout the electricity baseline.
Only electricity generation processes have DQI flow values; DQI values for upstream flows are not provided (e.g., extraction, processing, and construction).

The Reproducibility Problem

Running the same ElectricityLCI model configuration on two different computing environments can have different results. This is due to the dependence on background data, which is taken by a variety of providers (e.g., EIA, EPA, Canada Energy Regulator, and StEWI). The data pulled from these sources typically have versions (e.g., StEWI v1.1.4) or accessed date (e.g., when downloading data via an API).

In order to get the same model specifications to return the same results, the background and supporting data (including the Federal Elementary Flow List and facilitymatcher) must be the same in both environments. When this condition is met, the differences in model runs are insignificant (<0.000001%).

The 2022 electricity baseline was generated using the following vintages:

• FacilityMatcher: forStewi v1.1.4 (13 Feb 2025)
• FedElemFlowList v1.2.0, e57a542 (hash)
• Stewi
  • eGRID 2021 v1.1.3, 6710a0f
  • NEI 2020 v1.1.0, 084a311
  • RCRAInfo 2021 v1.1.3, 6710a0f
  • TRI 2021 v1.1.0, 084a311
• EIA Historical Coal Production Data: 2022 (accessed 6 Feb 2025)
• EIA's U.S. Electric System Operating Data (accessed via API 6 Feb 2025)
• EIA Form 860 and 923 (accessed 11 Feb 2025)
• EPA Continuous Emission Monitoring Systems data (accessed via API 6 Feb 2025)
• Canada Energy Regulator Electricity Generation 2023 (accessed 28 Oct 2024)

The NETL coal transportation life cycle inventory, available online here was identified having systematic errors in its results (i.e., functional unit mismatch in coal train construction). The correction was published on NETL's Energy Data eXchange, found here.

The Reverse Compatibility Problem

The ability to reproduce model runs of ELCI_1 (i.e., the 2016 baseline, as found on the Fed Commons) is a goal, but will likely result in a new inventory. Furthermore, the UUIDs of processes have changed in v2, due to a correction in how UUIDs are defined.

Appendix B

Reproducibility of Impact Results

The TRACI 2.1 impacts (from FLCAC) were calculated for ELCI_2022 in openLCA 2.4 on two model runs (win10 and macOS), both using Python 3.12 and the package dependency versions as mentioned in the repository README. The results from both model runs are given below.

Category	Units	win10	macOS
Acidification	kg SO2 eq	0.977314	0.977314
Eutrophication	kg N eq	0.0904227	0.0904227
Freshwater ecotoxicity	CTUeco	23730.5	23730.5
Global warming	kg CO2 eq	470.793	470.793
Human health - cancer	CTUcancer	5.18908e-5	5.18908e-5
Human health - non-cancer	CTUnoncancer	0.00947411	0.00947411
Human health - particulate matter	PM 2.5 eq	0.0618760	0.0618760
Ozone depletion	kg CFC-11 eq	7.36090e-6	7.36090e-6
Smog formation	kg O3 eq	22.4373	22.4373

Impacts Over Time Results

The following table shows the same TRACI 2.1 mid-point impact categories as described above, but for the 2020–2022 baseline runs.

Category	Units	2020	2021	2022
Acidification	kg SO2 eq	0.956813	0.978505	0.977314
Eutrophication	kg N eq	0.0834738	0.0869581	0.0904227
Freshwater ecotoxicity	CTUeco	25784.3	24785.3	23730.5
Global warming	kg CO2 eq	470.029	477.628	470.793
Human health - cancer	CTUcancer	5.55542e-5	5.41828e-5	5.18908e-5
Human health - non-cancer	CTUnoncancer	0.0102233	0.00990965	0.00947411
Human health - particulate matter	PM 2.5 eq	0.0620631	0.0632773	0.0618760
Ozone depletion	kg CFC-11 eq	9.33308e-6	8.09468e-6	7.36090e-6
Smog formation	kg O3 eq	21.8015	21.2501	22.4373

The spike in GWP in 2021 is caused by a the spike in MIXED fuel category emissions (e.g., from PJM, ERCOT, and Louisville Gas & Electric).

Electricity Mix Over Time Results

The following table shows the mix fractions for "Electricity; at grid; consumption mix - US - US" process across the three baseline vintages. Note that the 'ALL' fuel category represents Canadian imports (not discretized by fuel type).

Fuel	2020	2021	2022
ALL	0.01503	0.01332	0.01374
BIOMASS	0.00510	0.00500	0.00468
COAL	0.17336	0.16644	0.16117
GAS	0.35326	0.33736	0.35006
GEOTHERMAL	0.00605	0.00520	0.00464
HYDRO	0.08246	0.07459	0.07387
MIXED	0.05231	0.07940	0.06682
NUCLEAR	0.19291	0.18601	0.17767
OFSL	0	0	0
OIL	0.00012	0.00013	0.00021
OTHF	0.00014	0.00016	0
SOLAR	0.02629	0.03245	0.03841
SOLARTHERMAL	0.00117	0.00095	0.00101
WIND	0.09179	0.09900	0.10771

The figure below shows a graphical representation of the table above.

Footnotes

1. https://github.com/USEPA/ElectricityLCI/issues ↩

2. https://www.epa.gov/power-sector/cam-api-portal#/api-key-signup ↩

3. https://www.eia.gov/opendata/ ↩

4. https://edx.netl.doe.gov/ ↩

[dt-woods]

@bl-young or @m-jamieson, let me know if you want to review the PR #310; otherwise, 🎊 on version 2.0

[bl-young]

I don't expect I'll be able to review again anytime soon. Plus I'm sure there's nothing that can't be resolved with a patch if anything gets discovered!

[m-jamieson]

Agreed on not holding it up. I wouldn't be able to review for a little while either. 🎊

[dt-woods]

[dt-woods]

The ElectricityLCI model run for 2022 is now public (note the DOI URLs may take 24–48 hr to publish).

• 2022 Electricity Baseline
  • https://edx.netl.doe.gov/dataset/electricity-baseline-2022
  • https://doi.org/10.18141/2569335
• 2022 Electricity Baseline Background Data and Log file
  • https://edx.netl.doe.gov/dataset/electricity-baseline-2022-background-data-and-log-file
  • https://doi.org/10.18141/2569193
• 2021 Electricity Baseline
  • https://edx.netl.doe.gov/dataset/electricity-baseline-2021
  • https://doi.org/10.18141/2569580
• 2021 Electricity Baseline Background Data and Log file
  • https://edx.netl.doe.gov/dataset/electricity-baseline-2021-background-data-and-log-file
  • https://doi.org/10.18141/2569576
• 2020 Electricity Baseline
  • https://edx.netl.doe.gov/dataset/electricity-baseline-2020
  • https://doi.org/10.18141/2569609
• 2020 Electricity Baseline Background Data and Log file
  • https://edx.netl.doe.gov/dataset/electricity-baseline-2020-background-data-and-log-file
  • https://doi.org/10.18141/2569605

[dt-woods]

@bl-young, please see the public URLs for the 2020–2022 baselines above. Let me know if there's anything more I can do to assist getting these into the Fed Commons.

[bl-young]

thanks - will let you know!