# Stanford industrial heat study: integration in Radar

Integrated September 29, 2026, using the supplied `erlae65acsupp1.xlsx` and `erlae65acsupp2.pdf`.

**Result:** published 2022 useful process-heat estimates now supply the default assessment for **70 of Radar's 74 facilities**, matched by exact EPA GHGRP facility ID. Four facilities retain the 2023 EPA stationary-combustion gas proxy. Source, year, temperature profile, end uses, and fuel context are retained in the dossier, report, and CSV.

Saad, D. M., Brandt, A. R., and Azevedo, I. M. L. (2026), *Facility-level estimates of demand for industrial process heat in California*, Environmental Research Letters 21, 094023. [Publication and DOI](https://doi.org/10.1088/1748-9326/ae65ac). Supplementary [workbook](https://doi.org/10.1088/1748-9326/ae65ac/data1) and [methodology](https://doi.org/10.1088/1748-9326/ae65ac/data2) were supplied locally after the publisher downloads could not be accessed in the initial review.

## Source audit and coverage

| Item | Verified result |
| --- | --- |
| Workbook sheets | `Definitions`; `Unit-level` (2,345 allocated rows); `Facility-level` (120 facilities) |
| Reporting year | 2022 for every source record; Radar's EPA inventory is 2023 |
| Heat and fuel units | PJ/year; 1 PJ = 277,777.7778 MWh |
| Heat Emissions | Metric tonnes of CO2, distinct from the EPA inventory's CO2e |
| Published useful heat, all 120 facilities | 349.672835939 PJ/year, including Space Heat |
| Exact Radar matches | 70 of 74; all 70 pass identity and internal-reconciliation checks |
| Matched useful process heat | 23,966,724.7 MWh/year after excluding Space Heat |
| Matched Space Heat excluded | 5,121.5 MWh/year, retained separately |
| Study facilities outside Radar | 50 records retained in the research layer; no portfolio expansion |
| Numerical reconciliation | All 120 facility heat totals, heat-related CO2 totals, and published temperature bins reconcile to unit rows |
| Source identity | SHA-256 for both supplements and the Radar inventory stored in generated metadata |
| Reuse terms | No explicit license statement located in the supplied files; attribution retained and no license inferred |

The four unmatched Radar facilities are:

| GHGRP ID | Radar facility |
| --- | --- |
| 1002152 | PROCTER & GAMBLE PAPER PRODUCTS |
| 1004549 | Vitro Flat Glass LLC |
| 1014832 | Darling Ingredients Inc |
| 1012548 | Arcosa Lightweight - Frazier Park |

Matching uses the published GHGRP identifier, with identity and quantity checks recorded for review. Name, year, and NAICS differences remain auditable; no fuzzy name match silently changes the facility population. A missing, inconsistent, or review-required research record falls back to the EPA proxy. “Validated” means the source is internally reconciled and matched; it does not independently confirm the study's physical assumptions.

## How it changes the assessment

The old model inferred every facility's useful heat from stationary-combustion emissions using a natural-gas factor, then applied six broad sector profiles. This missed heat reported under other emissions subparts and did not reflect published unit/end-use allocations.

For a validated research match, Radar now calculates:

```text
useful process heat, MWh/year
  = sum(Unit-level Annual Heat Demand excluding Space Heat) * 1,000,000 / 3.6

addressable heat = useful process heat * scenario electrifiable share
electric energy = addressable heat / scenario COP
```

The study's Annual Heat Demand is already useful heat after its fuel-to-heat efficiencies. Its allocated rows already include Energy Share. Neither factor is applied again. Annual Fuel Demand is retained as context and is not substituted for useful heat. Published 2022 values are not scaled by the 2023 emissions inventory or added to the EPA estimate.

The EPA fallback remains selectable for matched facilities and supplies unmatched sites:

```text
EPA useful heat, MWh/year
  = 2023 stationary-combustion tCO2e / 0.05306 * 0.293071 * boiler efficiency
```

The factor is 53.06 kg CO2/MMBtu for natural gas. Combining it with reported CO2e and mixed fuels remains an explicit screening proxy. The displayed EPA comparison uses the same boiler-efficiency assumption as the active scenario. Comparing the two baselines reveals differences in year, emissions boundary, fuel, and efficiency; it is not a measured year-over-year trend.

Source choice persists with the facility scenario. Presets retain the selected source, and Reset restores the default source and Base assumptions. Portfolio ranking stays fixed at the default baseline and Base assumptions when data loads; active scenario edits and source switching affect facility calculations and decision cards without re-ranking the portfolio.

## Temperatures and ranking

The importer preserves each supplied numeric temperature and sums its already-allocated useful process heat. It retains original 100-degree facility bins separately, including their published Space Heat boundary. Portfolio bands use process-heat shares at below 160°C, 160–250°C, and above 250°C. The largest share sets the heat-band filter.

Temperature-fit score points use the whole distribution:

```text
temperature fit = 15 * low share + 10 * medium share + 4 * high share
```

The remaining score is unchanged: heat scale * 45, electrifiable share * 20, shiftable share * 10, and sector confidence points, clamped to 1–99. The heat scale is `log10(addressable MW + 1) / 2`, clamped to 0–1. Unmatched facilities retain their sector temperature band and its 15/10/4 points.

Electrifiable share, COP, operating hours, peak factor, and shiftable share remain scenario assumptions. A temperature-compatible quantity alone does not establish available heat sources, equipment feasibility, retrofit suitability, flexibility, or economic value. Most published temperatures are literature and sector/end-use estimates, not facility measurements (supplementary S2.4, printed pp.18–19 / PDF pp.19–20).

## Cost, emissions, and CHP boundaries

The energy cost model compares electrification with a hypothetical gas boiler delivering the selected addressable heat:

```text
gas-reference cost = addressable heat / 0.293071 / boiler efficiency * gas price
electric cost = addressable heat / COP * electricity price
energy-only difference = gas-reference cost - electric cost
```

Changing the boiler-efficiency control changes the reference cost and EPA proxy, but never the study's published useful heat. This comparison is not actual mixed-fuel or CHP project savings. It excludes demand charges, capital, maintenance, incentives, grid work, and the costs or value of displaced CHP electricity.

The study applies a 40% fuel-to-heat conversion to CHP. Its useful heat therefore excludes electricity, but its associated fuel and combustion CO2 include both products. Radar retains the useful thermal load while withholding research CHP emissions displacement as **“Needs CHP study.”** It does not count the entire CHP fuel or CO2 as avoidable by replacing thermal output. Facility-specific thermal/electric allocation and replacement electricity are needed (S2.5 and Table S3, printed pp.19–21 / PDF pp.20–22).

For research sites without CHP, the gross on-site screening ceiling is published process-heat CO2 * electrifiable share, excluding Space Heat. The EPA fallback uses stationary-combustion CO2e * electrifiable share. These are explicitly different units; neither is a net electricity-system or life-cycle estimate. Proportional allocation remains an assumption until actual displaced duties and fuels are known.

Research-based implied equipment utilization is withheld: dividing the study's 2022 heat/fuel boundary by the separately reported 2023 equipment capacity would not establish a valid utilization rate. Reported capacity remains visible as evidence.

## Cement improvement and source discrepancy

The original proxy substantially understates cement heating where kiln combustion is combined with process emissions outside stationary combustion. The study separates a modeled 40% combustion share from combined cement emissions and applies a national cement-fuel factor of 76.842 kg CO2/MMBtu. It does not classify all calcination CO2 as fuel combustion (S1.2.2 and S4.25, printed pp.10 and S75–S77 / PDF pp.11 and 76–78).

For Tehachapi Cement Plant (GHGRP 1002566):

| Quantity | Value and boundary |
| --- | --- |
| Published 2022 useful heat | 2.578070589 PJ/year = 716,130.7 MWh/year, including Space Heat |
| Radar 2022 process-heat baseline | 715,734.5 MWh/year |
| Excluded Space Heat | 396.2 MWh/year |
| 2023 stationary-combustion emissions | 79.70 tCO2e; a much narrower boundary |
| 2023 EPA proxy at 80% efficiency | Approximately 352.2 MWh/year |

There is a material **workbook/text discrepancy**: the supplementary cement description states a 9:1 kiln-to-dryer energy split at roughly 1,483°C and 163°C. The actual workbook gives approximately 5:1, or 83.33% at 1,482°C and 16.67% at 163°C, across its cement process-heat records. Radar preserves the supplied workbook numbers and records the discrepancy rather than silently replacing them with the narrative ratio. Site-level or author clarification would improve this temperature allocation.

## Other research assumptions to retain

- **Efficiencies:** natural-gas boilers use 94%, process heating 75%, CHP thermal output 40%, and Space Heat 90%. Boiler efficiencies vary by fuel. These are research assumptions, not site measurements (S2.5/Table S3).
- **Grouped units:** CP/GP/OCS entries can combine duties. The study distributes unresolved units among sector boiler and process-heating profiles. Workbook row counts are allocated observations, not equipment counts (S2.2–S2.4).
- **Multi-fuel treatment:** the study selects a representative unit fuel using the largest reported methane-emissions contribution. A facility's Main Fuel Type is determined by largest heat contribution. Neither describes a complete measured fuel mix (S1.2.1/S2.2; workbook Definitions).
- **Automotive temperatures:** S4.38.2 describes paint curing up to 180°C but assigns all heat 500°C under limited data. This can suppress an apparent lower-temperature opportunity; verify actual duties before excluding a technology (printed S103 / PDF p.104).
- **Semiconductor temperatures:** the study assumes 500°C without a measured steam temperature. Its supporting reference [65] is titled as a chipboard production plant, a further source concern. Preserve the published value while flagging it for site validation (S4.37.2, printed S101 / PDF p.102; reference on PDF p.110).
- **Coverage:** the study includes refining and oil/gas extraction; Radar's existing population excludes them. A statewide share or the 120-versus-74 count cannot be treated as a site default or a coverage-gap count.
- **Historical operations:** 2022 production levels may differ from current operations, and small sites outside GHGRP reporting are not represented.

## Files, reproducibility, and validation

[`scripts/build_stanford_heat_layer.py`](scripts/build_stanford_heat_layer.py) reads the two supplied files and [`data/facilities.json`](data/facilities.json), then generates [`data/stanford_heat.json`](data/stanford_heat.json). Run from this folder with Python and `openpyxl` installed:

```powershell
python scripts/build_stanford_heat_layer.py
python -m unittest discover -s tests -p "test_stanford_heat_layer.py"
```

Optional arguments are `--workbook`, `--pdf`, `--facilities`, and `--output`. The JSON contains:

- `metadata`: attribution, source hashes, years, scope, conversions, tolerances, counts, aggregate totals, and caveats.
- `records`: all 120 source-facility aggregates, source workbook row references, exact GHGRP match, validation status, process and published totals, excluded Space Heat, exact temperatures, published bins, end uses, fuels, contributors, warnings, and reconciliation differences.
- `unmatchedRadarFacilities`: four explicit fallback records and reasons.

The importer checks unique identities, reporting years, finite nonnegative quantities, facility/unit heat and CO2 reconciliation, temperature-bin reconciliation, and process-plus-Space-Heat accounting. It retains quantities without rounding during calculations. The importer test suite covers source coverage, reconciliation, unit conversions, Space Heat separation, cement, and invalid/review-required records. The browser model equations are documented in [`documentation.html`](documentation.html#heat-model).

The next quantitative improvement is replacing remaining sector technology and operating assumptions with verified site heat duties, fuel bills, current operating hours, and CHP allocation. Published useful heat substantially improves the starting point while preserving the need for that diligence.