Minimizing emissions from grid-based hydrogen production in the United States

We use GenX to model system outcomes in the western US with a planning year of 2030, taking into account existing state policies as well as new federal subsidies established by IRA for carbon-free electricity production. A six-zone electricity system topology is used to represent key inter-regional transmission constraints in the US portion of the Western Interconnection (see figure 1), and electricity system operations are modeled at hourly resolution across 18 representative weeks, which are down-sampled from a full year of hourly data using a k-medoids clustering method. Model inputs, including regional demand profiles and cost, performance, and availability data for generators and storage, are compiled using PowerGenome [21] and are described in greater detail in Xu et al [22]. Only currently mature, commercially-available grid-scale generation and storage technologies are assumed to be available for deployment by 2030. Geothermal inputs have been adjusted from Xu et al [22] to reflect updated state policies and resource estimates [23, 24]. All carbon-free generator costs have been updated to reflect IRA tax credits for certain resources: onshore wind and solar power are assumed to receive a PTC of $26 MWh⁻¹ (2022 USD) for 10 years, which is represented in the optimization as the equivalent net-present value subsidy if provided over the full financial lifetime of the project; geothermal, offshore wind, and battery resources are assumed to receive an investment tax credit of 30%. All GenX input and results data relevant to this work are available at Ricks et al [27], and a high-level overview of input technology parameters is provided in supplementary tables 1 and 2.

Zoom In Zoom Out Reset image size

To explore the system impacts of grid-based hydrogen production we exogenously add a single large electrolysis load to a target model zone. The total electrolyzer capacity is fixed, as is the revenue that can be earned per unit of hydrogen produced. Electrolysis operations are co-optimized with the electricity system, and the hydrogen producer is able to curtail production at a given model timestep if the cost of consuming electricity exceeds the revenue that can be earned through hydrogen sales. In addition to purchasing grid electricity to run the electrolyzers, the hydrogen producer can also build on-site energy storage and contract directly with new grid-connected clean energy resources in the local model zone (via power purchase agreements (PPAs) or energy attribute certificates (EACs)) to meet any imposed policy requirements. Because grid zones in GenX are modeled as 'copperplates' with no internal transmission constraints, all locally-procured generation is implicitly assumed to be physically deliverable to the electrolysis facility at all times. We evaluate the emissions intensity of grid-connected hydrogen production in the model via two approaches [28]:

• (a)  
  Attributional Emissions: The share of total grid emissions that would be attributed to hydrogen producers based on their net consumption in a given hour, following a convention similar to the current Greenhouse Gas Protocol Scope 2 location-based emissions accounting guidance [29]. The attributional emissions intensity of produced hydrogen is calculated as:

  $$\begin{align} I\,^{\mathrm{Attr}} = \sum_{t}(\max(0,(L^{\mathrm{H_2}}_t - \mathrm{CFE_t}))\times E\,^{\mathrm{Avg}}_t)/\mathrm{H_{2}^{\mathrm{Tot}}} \end{align} \tag{ 1 }$$

  where $L^{\mathrm{H_2}}_t$ and ${\mathrm{CFE}_t}$ are the electrolysis load (including alterations from on-site energy storage) and procured carbon-free electricity at timestep t, $E^{\mathrm{Avg}}_t$ is the average grid supply emission rate at timestep t (as described in Xu et al [22]), and $\mathrm{H}_2^{\mathrm{Tot}}$ is the total hydrogen production in the system. Under this accounting framework hydrogen producers incur an emissions penalty whenever they use more electricity than is being concurrently supplied by procured clean resources. They are not able to achieve a negative hourly emission rate by procuring more clean electricity than they consume (aka 'offsets'). This accounting methodology is practically implementable and useful for allocating emissions resulting from direct consumption of fossil-sourced electricity, but does not capture the true marginal impact of new loads or the indirect effects of clean electricity procurement and sales on system-level emissions.
• (b)  
  Consequential Emissions: The true long-run electricity system-level emissions impact of hydrogen production, relative to a counterfactual scenario in which the hydrogen production does not occur. The consequential emissions intensity of hydrogen produced is calculated as:

  $$\begin{equation} I\,^{\mathrm{Cons}} = (E^{\mathrm{Tot}}_{\mathrm{H_2}} - E^{\mathrm{Tot}}_{\mathrm{Base}})/\mathrm{H_{2}^{\mathrm{Tot}}} \end{equation} \tag{ 2 }$$

  where $E^{\mathrm{Tot}}_{\mathrm{H_2}}$ is the total system-wide emissions in the case being investigated, and $E^{\mathrm{Tot}}_{\mathrm{Base}}$ is the total system-wide emissions in a counterfactual scenario where the hydrogen producer is not present in the system. Consequential emissions are impossible to measure in the real world due to a lack of observable counterfactuals, and therefore cannot be used for practical policy implementation or emissions accounting. However, the modeled consequential emissions results presented herein are a useful means of comparing the expected long-run emissions impacts of hydrogen production under alternative 45V PTC implementations.

In addition to emissions rates we also assess the impacts of possible 45V PTC implementations on the cost of hydrogen production. We use GenX results to calculate the levelized cost of hydrogen (LCOH), the total revenue per unit of hydrogen sold needed to make up all associated costs, for each model case under various electrolyzer cost assumptions. As electrolyzer capacity is an exogenously fixed quantity in the present study, these variations in assumed electrolyzer cost have no effect on other model outcomes presented in this study. The methodology used to calculate LCOH in this paper is described in detail in supplementary note 2. Cost and performance assumptions for hydrogen electrolyzers are detailed in table 1.

Table 1. Electrolysis financial and operational parameters used in this study. Financial assumptions are generally conservative: the large fixed grid connection fee is based on a cost analysis by LADWP for transmission-level customers in Los Angeles [30], and the CRF value assumes a weighted average cost of capital of 8% and a payback period of 10 years, equal to the length of the 45V PTC. The default sales revenue assumes that hydrogen is sold to the end consumer at $1 kg⁻¹. Default GenX inputs are shown in bold. Parameters without bolded entries are used only in results analysis and do not affect model outcomes.

Parameter	Units	Value
Installed capacity	GW	Varied: 1; 5
Efficiency	kWh kgH2 $^{-1}$; %LHV	50; 67
CAPEX	$ kW−1	Varied: 1200; 600; 300
Annual fixed O&M	% of CAPEX	5
Capital recovery factor	% of CAPEX	14.9
Grid connection fee	$ kWyr−1	85
Hydrogen sales revenue (including PTC)	$ kgH2 $^{-1}$	Varied: 3; 4; 5

In this study we examine five possible clean energy procurement regimes under which grid-connected electrolysis could be allowed to claim the 45V PTC:

• (a)  
  No Requirements: Electrolysis demand is added to the grid without requiring any new clean electricity additions beyond existing state policies, and is met by the least-cost mix of resources.
• (b)  
  100% Hourly Matching: Hydrogen producers are required to match their consumption with procured carbon-free generation at every hour of the year. This policy and its formulation in GenX are described in detail in Xu et al [22], where it is referred to as '24/7 Carbon-Free Electricity' (or '24/7 CFE'). The 100% hourly matching requirement effectively mimics the physical constraints on behind-the-meter hydrogen production by requiring producers to never consume more electricity than is being locally generated by a specific portfolio of clean resources. We assume by default that any excess procured carbon-free generation can be sold into the bulk electricity market, reducing overall costs, but we do not credit such sales with reducing the attributional emissions of the electrolysis load (i.e. no 'offsetting' of emissions during other periods is permitted). We also assess a subvariant of this regime in which no sales of excess procured generation are permitted, leading to effectively zero operational interaction with the larger electricity system.
• (c)  
  100% Weekly Matching: Hydrogen producers are required to procure enough carbon-free generation to match their total consumption in every week of the year. Net consumption of grid electricity in some hours may be offset by procurement of excess carbon-free generation in others.
• (d)  
  100% Annual Matching: Hydrogen producers are required to procure enough carbon-free electricity production to match their total consumption on an annual basis. This procurement strategy is commonly employed today in voluntary markets and for compliance with state policies (e.g. renewable portfolio standards).
• (e)  
  Net-Zero Short-Run Marginal Emissions (net-zero SRMEs): An alternative approach to time matching based on SRMEs accounting. Hydrogen producers' embodied emissions are evaluated in each hour based on the local SRMEs rate, the amount by which system-wide emissions would change due to a unit increase in local electricity demand. While this short-run rate is measurable and reflects the impact of changes in consumption or production on the operation of a static electricity system, it does not capture the (unobservable) potential impact of these changes on long-run capacity investment and retirement decisions. Under a Net-Zero SRME regime, hydrogen producers are assumed to incur SRME penalties or achieve SRME offsets in a given hour based on the net of total electricity consumed and clean electricity procured, and must have a non-positive total SRME impact over the course of a year, as described by the following equation:

  $$\begin{equation} \sum_{t}((L^{\mathrm{H}_2}_t - \mathrm{CFE}_t)\times \mathrm{SRME}_t) \leqslant 0. \end{equation} \tag{ 3 }$$

  The methodology used to calculate SRMEs in this paper is described in supplementary note 3.

We model system-level outcomes under each of these regimes for electrolysis facilities located in each of the six model zones shown in figure 1. In all cases, all contracted clean resources must be new (aka 'additional') and must be sited in the local model zone, thereby ensuring deliverability to the electrolysis facility. We explore the impact of relaxing these constraints via cases where additionality requirements are explicitly removed, as well as cases where hydrogen producers are allowed to procure non-local generation (e.g. located in a different model zone than the electrolysis load) that may be physically non-deliverable due to transmission constraints. We also include cases varying the revenue from hydrogen sales (and by extension the electricity cost at which producers are willing to curtail electrolysis) and the total installed electrolyzer capacity to assess the sensitivity of outcomes to variability along these dimensions. We explore variations in the cost and embodied emissions from hydrogen production across each of these modeled scenarios.

Model results indicate that both the attributional and consequential emissions intensities of hydrogen produced via electrolysis in a 2030 western US grid under a No Requirements policy are universally too large to meet statutory requirements for the full 45V PTC (figure 2, left column). This outcome occurs despite a large expansion of clean generation across the Western Interconnection driven by IRA subsidies (supplementary figure 1). Attributional emissions intensities can be very large, up to 20 kgCO₂e kgH₂ $^{-1}$ in model zones with high shares of coal-fired generation, but are notably small in the Pacific Northwest zone where hydropower is a majority of the energy mix. However, consequential emissions intensities are greater than 10 kgCO₂e kgH₂ $^{-1}$ in all zones, and nearly 40 kgCO₂e kgH$^{-1}$ in the Wyoming & Colorado zone, suggesting that it would always be more environmentally friendly to produce the same amount of hydrogen using SMR rather than grid-based electrolysis. This consequential outcome occurs because the marginal generation used to serve any new hydrogen load always comes in large part from fossil resources. This is the case in every model zone, as demonstrated in figure 3, which compares the generation portfolio procured to meet electrolysis load and actual system-level change in generation by technology resulting from the addition of the electrolysis load to the system for the same scenarios as figure 2. Replacing the average emissions rate used in equation (1) with the SRME rate (as shown in supplementary figure 2) may therefore be a more accurate means of evaluating hydrogen's true emissions impact from a 45V PTC compliance standpoint. Whereas hydrogen production with no accompanying clean energy procurements in the Pacific Northwest zone would qualify for the base PTC under an averages-based approach, it would not do so under a SRME-based approach.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We observe that the additional fossil electricity production used to supply new electrolysis load comes entirely from existing units, which are kept in service longer or operated at higher capacity factors to help meet the new electrolysis load than in the counterfactual without electrolysis. With sales revenue of at least $3 kgH₂ $^{-1}$ due to the 45V PTC alone, hydrogen producers are incentivized to consume electricity at prices up to and exceeding $60 MWh⁻¹. The PTC thereby motivates hydrogen producers to operate their electrolyzers at very high utilization rates year-round, and to continue consuming electricity even when high-price resources like coal and gas are on the margin. Sensitivity cases indicate that embodied emissions from grid-connected hydrogen production are fairly insensitive to changes in the final sales revenue above the PTC threshold, with major outcomes nearly unchanged for hydrogen sales revenues in the $3–5 kg⁻¹ range (supplementary figures 3–5). However, as shown in the same figures, outcomes are very different in cases without a PTC where hydrogen sales revenue is only $1 kg⁻¹. In these cases electrolyzers are only economical to operate when low-marginal-cost renewables or nuclear are on the margin, leading to very low consequential emissions from grid-based hydrogen production even under a No Requirements policy. The 45V PTC itself therefore appears to be the primary driver of unfavorable emissions outcomes from grid-based hydrogen production in the US.

For significantly larger installed electrolyzer capacities (e.g. 5 GW), both attributional and consequential emissions intensities of grid-based hydrogen production decline somewhat (supplementary figure 6). In these cases, the large additional demand cannot be fully met by existing fossil generators, requiring significant amounts of new capacity to be deployed. Because clean resources are more competitive against new fossil resources than against existing ones, they make up a larger relative share of this new capacity (supplementary figure 7). Despite the lower emissions intensities, the total additional emissions in these cases are still larger than in cases with less hydrogen production.

We find that enforcing a 100% Hourly Matching requirement (as described in section 2.2) leads to zero attributional emissions and, in some cases, near-zero consequential emissions as well (figure 2, columns two and three). Attributional emissions are effectively zero by definition in these cases, as electrolyzers never consume more electricity than is being concurrently and locally generated by clean resources. Hydrogen producers procure the mix of local clean generation and storage that is able to meet their demand cost-effectively in as many hours as possible (figure 3 and supplementary figure 9). Operational profiles for hourly-matched electrolysis systems are shown in the left column of figure 4 and supplementary figures 11–15, and illustrate how electrolyzers occasionally reduce consumption during periods of clean electricity scarcity to avoid drawing power from the grid mix.

Zoom In Zoom Out Reset image size

While hydrogen production under a 100% Hourly Matching requirement therefore never directly consumes electricity from emitting resources, there are still scenarios in which it can have a high consequential emissions impact. This can occur through two mechanisms: first, via sales of excess clean electricity into the market, and second, via competition for limited high-quality clean resources. The comparison between 100% Hourly Matching outcomes in cases with and without excess sales permitted (figure 2 and supplementary figure 6) indicates that it is primarily the second mechanism that leads to large consequential emissions impacts. We observe that sales of excess clean electricity play an unpredictable but secondary role, likely determined by the specific competing resources that these excess sales displace in the market. In cases without excess sales, the only interaction between hydrogen producers and the rest of the electricity system is through competition for limited renewable resource development. When high-quality renewable resources are scarce, procurement of these resources by hydrogen producers can lead other system users to rely on fossil resources, rather than lower-quality clean resources, to make up the difference. This phenomenon is illustrated in supplementary figure 10, which shows the modeled wind supply curve in the Wyoming & Colorado zone as well as the observed wind buildout in both the base case without electrolysis load and cases with 100% Hourly Matching of a 1 GW electrolysis load. When a portion of the highlighted high-quality resource is procured for hydrogen production and cannot be used for grid supply, the system chooses to retire less coal capacity rather than developing significantly more expensive wind resources in the next step of the step-wise approximation of the wind supply curve used in this study. As illustrated in the third column of figure 3, these interactions occur primarily with wind resources, which generally have significantly more variance in quality and steeper supply curves than solar resources. Impacts of resource procurement in the real world may be less stark than those observed here, as the step-wise supply curves used in the GenX model do not reflect the more continuous spectrum of real resource quality. Additionally, while we model capacity additions and retirements as continuous variables in GenX to ensure computational tractability, these are in reality discrete decisions that pull an entire generating unit into or out of the capacity mix. This can lead to large discontinuities in real-world impacts if, for example, the marginal unit of electrolysis load makes the difference between a GW-scale coal plant being kept online and retired.

Although a 100% Hourly Matching requirement therefore cannot guarantee zero long-run emissions impact from hydrogen production, it does lead to consequential emissions outcomes that are universally superior to those under every alternative 45V PTC implementation investigated in this work, and often by wide margins. In several cases, a 100% Hourly Matching requirement reduces consequential emissions to near-zero when they would be worse than those of SMR under any alternative policy. These benefits are more pronounced in scenarios with greater total hydrogen production, as illustrated in supplementary figure 6. A 100% Hourly Matching requirement is therefore likely to be the best practical means of minimizing the real emissions impact of grid-based hydrogen production in the US. It should be noted that this requirement (in combination with deliverability and additionality requirements, as discussed in sections 3.4 and 3.5) also ensures consequential emissions outcomes no worse than those from hydrogen production supplied exclusively by behind-the-meter carbon-free resources. Like grid-supplied clean generation under an hourly matching requirement, behind-the-meter clean generation both competes for high-quality renewable resource sites that could instead supply power to the grid and can influence system-level emissions outcomes positively or negatively by exporting excess clean generation to the grid. The large consequential emissions impacts observed in some of our modeled cases are therefore a potential consequence of any electrolysis-based hydrogen production in the current US policy environment. Possible approaches to mitigating these negative outcomes through updated federal policy are discussed in section 4.

We find that the three potential alternatives to a 100% Hourly Matching standard investigated in this work, 100% Weekly Matching, 100% Annual Matching, and Net-Zero SRME, are universally ineffective at reducing consequential emissions from grid-based hydrogen production (figure 2 and supplementary figure 6, columns 4–6). In some cases they do achieve reduced attributional emissions rates, though only insofar as they lead hydrogen producers to directly reduce consumption of grid power in hours when carbon-free generation is plentiful. Importantly, these three alternative strategies differ from 100% Hourly Matching in that they allow hydrogen producers to offset net consumption of grid electricity in some hours with excess production of clean electricity in others. As shown in figure 4 and supplementary figures 11–17, producers typically meet these requirements by running their electrolyzers at high utilization rates year-round while procuring enough of the cheapest local renewable generation to fully offset their use in terms of the relevant metric—either megawatt-hours for Weekly and Annual Matching or marginal emissions for Net-Zero SRME. The compliance strategy of procuring the cheapest available renewable electricity in bulk, regardless of the timing of this generation, is remarkably consistent across the three policy cases. However, as illustrated in figure 3 and supplementary figure 7, this excess procurement almost never translates into real changes in the energy mix. The actual new generation used to meet new hydrogen load is instead almost identical to that observed in the No Requirements cases. This is true even for scenarios where procurement made under a 100% Hourly Matching requirement does successfully translate into actual changes in generation.

The general ineffectiveness of these 'offsets' based approaches as an emissions mitigation strategy is explained by the anticipated evolution of US electricity markets in a post-IRA world. The economic impetus provided by IRA subsidies leads clean electricity penetration in our modeled systems to far exceed levels mandated under current state policies by 2030. In this scenario the market for state-level policy compliance EACs is fully saturated, so simply adding demand for clean electricity attributes does not provide any economic incentive to increase supply. Instead, hydrogen producers are able to pay effectively zero to procure excess clean power from generators that would already have been built. These observations suggest that hydrogen producers should be considered fully responsible for any emissions induced by direct consumption of grid electricity, and that offsets should not be considered a credible means of eliminating the embodied emissions of hydrogen electrolysis.

We find that allowing resource procurement over large geographic areas can lead to significant consequential emissions from hydrogen production even when a 100% Hourly Matching requirement would otherwise ensure low consequential impact, as the introduction of transmission constraints prevents physical delivery of procured clean electricity. Transmission congestion can lead to different marginal generating units supplying power on each side of a constrained pathway, and persistent congestion can affect capacity retirements and additions in the long run. Consumption and production on different sides of frequent transmission constraints can thus lead to divergent emissions impacts. We demonstrate this through a set of controlled test cases (supplementary figure 18) where we show that allowing hydrogen production in each of two model zones using local solar power with a 100% Hourly Matching requirement leads to zero or negative consequential emissions impact, but allowing production in one zone using solar procured from the other zone leads to a very large consequential impact. In the case with non-local procurement, transmission constraints lead to the hydrogen producer consuming local fossil generation in some hours even while claiming full matching.

This finding demonstrates that clean resources subject to transmission constraints that prevent delivery of the procured energy cannot be relied on to eliminate emissions from hydrogen production. In this study, the deliverability condition is operationalized by requiring procurement of clean electricity from within the local model zone. However, unlike the model system studied in this work, the real grid is not divided neatly into well-connected zones with perfect internal deliverability, and transmission bottlenecks of varying severity exist at all spatial scales. When implementing a 100% Hourly Matching requirement for grid-based hydrogen production, prior determination of qualifying grid regions within which transmission constraints are minimized could help to mitigate instances of non-deliverable procurement. If these regions are internally well-connected, then locality (i.e. procurement from within the same region) could stand in as a reasonably proxy for deliverability. A more robust deliverability enforcement mechanism could instead rely on real-time monitoring via existing metrics like locational marginal electricity prices (LMPs), which diverge when congestion exists between two points in the electricity grid. Under this system, grid-based hydrogen production would be allowed to claim use of a non-colocated clean resource only during periods when the LMPs at the point of generation and point of delivery show that the procured energy is physically deliverable.

In our baseline scenarios we assume that only new clean resources (i.e. not in operation at the beginning of the model planning period) may be procured to meet a 100% Hourly Matching requirement for grid-based hydrogen production. We additionally assume that resources used to meet state capacity installation mandates (e.g. California's recent 1 GW clean baseload procurement order [23]) cannot also be counted towards clean hydrogen production. In modeled scenarios cases where we remove each of these requirements individually (see supplementary figure 19), we find that a 100% Hourly Matching requirement loses all of its consequential impact. This is because contracts with existing or mandated clean energy resources have no causal impact on the continued operation of these resources in the electricity system as long as they are not under threat of economic retirement. Any credible implementation of the 45V PTC that allows grid-based hydrogen production to qualify for subsidies should therefore enforce strict additionality requirements, bounding the installation dates of resources that can be counted toward clean hydrogen production and considering overlap with state-level capacity procurement mandates. Minor exceptions to this rule may be acceptable, specifically in the case of existing plants that would be forced to retire or curtail their generation without offtake agreements from hydrogen producers.

While existing and mandated resources can therefore be considered non-additional due to their lack of causal linkage to procurement by hydrogen producers, this definition could be extended to encompass even new, non-mandated resources that would have been built with or without having being procured for hydrogen production specifically. For example, the procurement of high-quality wind resources discussed in section 3.2 could be considered non-additional under this definition, as those resources would have been deployed regardless due to their economic favorability. In fact, by specifically banning procurement of these high-quality wind resources for hydrogen production (supplementary figures 20 and 21), we can significantly improve the observed consequential emissions outcomes. However, this broader definition of additionality is likely difficult if not impossible to enforce, as it requires counterfactual knowledge of which resources would have been developed had the hydrogen producer not made certain procurement choices. However, zero or near-zero market-based prices for EACs are a likely indicator that procured resources are non-additional, as such sales deliver little-to-no additional revenue to clean generators, and thus cannot materially affect capacity entry/exit decisions.

Enforcing a 100% Hourly Matching requirement leads to moderately increased costs for grid-based hydrogen production in some cases. Figure 5 shows the LCOH of hydrogen produced in the system under the same scenarios shown in figures 2 and 3, for installed electrolyzer system costs of $1200 kW⁻¹ (reflecting current costs), $600 kW⁻¹ (a 'moderate' possible cost in 2030), and $300 kW⁻¹ (a 'low' possible cost in 2030) [12, 13]. LCOH outcomes are fairly consistent across modeled regions, and are nearly identical among non hourly-matched cases.

Zoom In Zoom Out Reset image size

The observed differences in cost between cases with No Requirements and 100% Hourly Matching indicate that enforcing a 100% Hourly Matching requirement generally adds $0–1 kgH₂ $^{-1}$ to the LCOH. The additional costs can be near-zero when clean firm resources like geothermal power are available for procurement. Even in regions where only wind, solar and batteries can be relied on, the additional cost of 100% Hourly Matching is not substantial. Costs are somewhat greater when excess sales are forbidden, or when the total hydrogen production is larger (leading to procurement of clean generation from 'higher' up the supply curve; supplementary figure 8). For sales prices of $1 kg⁻¹ or greater, which would slightly undercut conventional grey hydrogen, and assuming an additional $3 kg⁻¹ PTC, clean hydrogen producers in all regions would likely break even or make a profit on their investments as long as electrolyzer costs continue to decline [12, 13]. The US Clean Hydrogen Strategy and Roadmap suggests that there may be large markets for clean hydrogen in the US at sales prices well above $1 kg⁻¹ [7], which could serve as viable initial markets even at current electrolyzer costs.