In every hour of 2023, California's marginal emission rate exceeded the annual inventory emission factor

Annual inventory accounting and EPA's hourly avoided-emissions method applied to synthetic load reductions in California in 2023, each at its own accounting boundary.

CheAI Research Inc. | Research article | November 12, 2024

Keywords: marginal emissions; avoided emissions; Scope 2 accounting; eGRID; AVERT; California electricity

Abstract

Industrial energy studies usually count electricity emissions with an annual average grid factor, as an emissions inventory requires. Decisions to add or remove electrical load, however, concern the margin: the plants that run more or less because of the change. We applied both methods to synthetic load reductions on California's grid in 2023. A flat 10 MW reduction for the year, applied at each method's own boundary (87,600 MWh of delivered consumption for the inventory, the same quantity at the fossil generation bus for AVERT), corresponds to 19,569 short tons of CO2 under inventory accounting with the eGRID2023 CAMX total-output rate and grid-loss gross-up, and to 41,337 short tons avoided under EPA's AVERT method, 2.1 times as much. An hour-by-hour reconstruction of the AVERT result shows the marginal rate ranging from 0.276 to 0.814 short tons per MWh across the 8,760 hours of the year; even the lowest hourly rate was 1.24 times the inventory factor. Concentrating the change in the top five percent of load hours raises the avoided rate to 0.592; a solar-shaped profile gives 0.469. The inventory emission factor per MWh is identical for all three profiles because it has no time dimension. The two methods answer different questions, cover different geographies and should not be netted against each other.

1. Introduction

An emissions inventory attributes a share of a grid's annual emissions to each megawatt-hour consumed, at the grid's average rate [1]. That is appropriate for reporting the emissions associated with a year of purchases. It is not an estimate of what changes when a site adds or removes load. A marginal megawatt-hour is supplied by the plants dispatched at the margin in that hour, whereas the annual average also includes generation, such as solar, hydro and nuclear, that rarely responds to a change in load [2].

The distinction matters for any comparison between onsite generation and grid power, or between fuel-fired and electric heat sources. A study that uses the average factor to decide whether to electrify, or whether to run onsite generation harder, applies a reporting factor to a decision question. This study quantifies the gap for one grid and one year.

2. Methods

2.1 Inventory accounting

Inventory emissions use the eGRID2023 Revision 2 CAMX annual total-output CO2 rate [3], 194.3 kg per MWh of net generation, with delivered electricity grossed up to generation by the Western Interconnection's 4.1 percent grid gross loss. This gives 0.223 short tons per delivered MWh, applied uniformly to every hour. The gross-up includes the emissions of grid losses, which the GHG Protocol assigns to an end user's Scope 3 rather than Scope 2 [1]; without it the flat-profile inventory total is 18,767 short tons.

2.2 Avoided-emissions accounting

Avoided emissions follow EPA's AVERT v4.3 method for its California region in 2023 [4], which covers 237 fossil units in California, Nevada and Utah. For each hour, the regional expected-value tables give CO2 at the hour's load and at the load reduced by the displacement; the difference is the avoided CO2. Ozone-season tables apply from May to September, 3,672 hours, and non-ozone tables otherwise, as the AVERT manual specifies for annual totals. The displacement is applied at the fossil generation bus, without a transmission-loss adjustment, so the two methods apply the nominal reduction at different boundaries. Because a delivered reduction corresponds to a larger reduction at the generation bus, applying both at the delivered boundary would increase the avoided figure, and the reported ratios are therefore lower bounds.

2.3 Load profiles

Three synthetic profiles were evaluated: a flat 10 MW reduction in every hour, the same reduction concentrated in the top five percent of load hours, and a daytime solar-shaped profile.

2.4 Verification

The annual avoided-emissions rate differs from EPA's published California 2023 uniform energy-efficiency rate [5] by 0.36 percent, within a 0.5 percent acceptance criterion. The residual has not been attributed to a single cause: it combines the difference between a 10 MW displacement and EPA's half-percent reference size with the interpolation used to reconstruct the method between load bins, to which the result is sensitive (step selection changes the annual total by 3.3 percent and bin-midpoint interpolation by 0.3 percent). EPA's own Excel engine and its Monte Carlo smoothing were not re-executed. A second, independently written implementation matches the first to a relative 2.5 × 10^-15. The hourly profile in Figures 2 and 3 was computed by a third implementation, which reproduces the annual total exactly before reporting any hour.

3. Results

3.1 The avoided-emissions method attributes about twice the inventory figure

Table 1. Synthetic load reductions under inventory accounting (delivered basis) and avoided-emissions accounting (generation-bus basis).

Profile	Energy, MWh	Inventory, short tons	Avoided, short tons	Avoided per MWh	Ratio
Flat load reduction, every hour	87,600	19,569	41,337	0.472	2.11
Concentrated in the top five percent of load hours	21,900	4,892	12,975	0.592	2.65
Daytime solar shape	69,311	15,484	32,512	0.469	2.10

For the flat profile the avoided rate of 0.472 short tons per MWh, 428 kg CO2 per MWh, is about 2.1 times the inventory rate (Table 1). The ratio is 2.65 for the peak-hour profile and 2.10 for the solar profile. Inventory accounting assigns the same rate to every profile; avoided-emissions accounting does not (Figure 1).

Grouped bars for three load profiles: inventory rate constant near 0.22 short tons per MWh, avoided rate near 0.47, 0.59 and 0.47 — **Figure 1. Avoided and inventory emission rates for three load profiles.** Short tons of CO2 per MWh for each profile under each method. The inventory rate is identical for all three because an annual average factor has no time dimension.

3.2 No hour of the year was as clean as the annual average

The marginal rate shows one strong pattern, a summer peak: in July and August, from late afternoon into the evening, the month-hour means rise well above the rest of the year (Figure 2). Outside those months they lie in a narrow band with no pronounced daily shape. Across the year the hourly rate ranges from 0.276 to 0.814 short tons per MWh, with the tenth percentile at 0.336 and the ninetieth at 0.588. The inventory factor of 0.223 lies below the entire distribution (Figure 3). The comparison is conservative: avoided rates are per MWh at the fossil bus, whereas the inventory rate is per delivered MWh after the loss gross-up, so on a common delivered basis the gap would be larger.

Heat map of avoided CO2 per MWh by month and hour of day for California 2023, mostly uniform with a bright band on July and August afternoons and evenings — **Figure 2. Marginal emission rate by month and hour of day, California 2023.** Mean avoided CO2 per MWh for a 10 MW reduction, by month and hour of day, from the hour-by-hour reconstruction of the AVERT result. Hour of day is derived from the hour-of-year index in the AVERT data file; alignment with local clock time is assumed, as the source does not state a time zone or daylight-saving convention.

Histogram of month-hour mean avoided rates between about 0.3 and 0.75, with the inventory factor marked near 0.22 to the left of every bar — **Figure 3. Distribution of marginal rates and the annual inventory factor.** Distribution of the month-hour mean rates in Figure 2, with the annual inventory factor marked.

4. Discussion

Both figures are correct, and they answer different questions. Inventory accounting states what a year of purchases was responsible for; avoided-emissions accounting estimates what a change caused. The gap is large on a grid such as California's, where zero-carbon generation lowers the average but rarely sets the margin. For engineering decisions that add or remove load, the average understates the consequence, here by about half.

The direction of the bias matters for industrial decisions. A study comparing onsite gas-fired generation with grid import at the average factor makes grid power appear cleaner than the power a change would actually displace or draw on, so electrification appears more favorable and onsite generation less favorable than the margin indicates. On a grid where coal sets the margin the comparison shifts again. The two accountings cannot be netted against each other: they cover different geographies, eGRID's CAMX subregion against AVERT's three-state fossil fleet, and different questions.

5. Limitations

Results describe one historical year and do not predict future years.
The load profiles are synthetic and describe no real site, and the avoided figures are not a savings potential for California or elsewhere.
The hourly reconstruction does not resolve effects below the AVERT load-bin width, depends on the interpolation between bins, and was not compared with EPA's Excel engine.
The clock alignment of AVERT hours is assumed, because the source states neither a time zone nor a daylight-saving convention; the timing of the summer evening peak is therefore uncertain by an amount that remains unquantified.
The inventory figures include grid-loss emissions that a strict Scope 2 inventory would report in Scope 3.

6. Conclusions

In California in 2023, the hourly marginal emission rate exceeded the annual inventory factor in every hour, and a flat load reduction avoided 2.1 times the emissions attributed to it by inventory accounting. Studies of electrification or onsite generation should use marginal rates for decisions and reserve average factors for inventories.

References

World Resources Institute and World Business Council for Sustainable Development. GHG Protocol Scope 2 Guidance. Washington, DC, 2015.
Siler-Evans K, Azevedo IL, Morgan MG. Marginal emissions factors for the U.S. electricity system. Environmental Science and Technology, 46(9):4742-4748, 2012.
U.S. Environmental Protection Agency. Emissions and Generation Resource Integrated Database, eGRID2023 Revision 2. Washington, DC.
U.S. Environmental Protection Agency. AVoided Emissions and geneRation Tool (AVERT), version 4.3: User Manual and Regional Data Files. Washington, DC, 2024.
U.S. Environmental Protection Agency. Avoided Emission Rates Generated from AVERT. Washington, DC, April 2024.