Keystone-Conemaugh Projects LLC v. EPA

PRECEDENTIAL

UNITED STATES COURT OF APPEALS
FOR THE THIRD CIRCUIT
___________

No. 22-3026
___________

KEYSTONE-CONEMAUGH PROJECTS LLC,
Petitioner

v.

UNITED STATES ENVIRONMENTAL PROTECTION
AGENCY; ADMINISTRATOR ENVIRONMENTAL
PROTECTION AGENCY
_______________________

On Petition for Review of Actions of the United
States Environmental Protection Agency
______________

Argued: November 6, 2023

Before: RESTREPO, BIBAS, and SCIRICA, Circuit Judges.

(Filed: May 2, 2024)

Gina F. Buchman
Babst Calland
505 9th Street NW
Suite 602
Washington, DC 20004

Joseph V. Schaeffer [ARGUED]
Michael Winek
Babst Calland
603 Stanwix Street
Two Gateway Center, 6th Floor
Pittsburgh, PA 15222

Counsel for Petitioner

Robert A. Reiley
Pennsylvania Department of Environmental Protection
Office of Chief Counsel
9th Floor
400 Market Street
Rachel Carson State Office Building
Harrisburg, PA 17101

Jesse C. Walker [ARGUED]
Office of Attorney General of Pennsylvania
Bureau of Regulatory Counsel
RCSOB 9th Floor
P.O. Box 8464
Department of Environmental Protection
Harrisburg, PA 17105

Counsel for Intervenor Pennsylvania Department of
Environmental Protection

Brandon N. Adkins [ARGUED]
United States Department of Justice
Environment & Natural Resources Division

2
P.O. Box 7611
Ben Franklin Station
Washington, DC 20044

Counsel for Respondents

Zachary M. Fabish
Sierra Club Environmental Law Program
50 F Street NW
8th Floor
Washington, DC 20001

Charles McPhedran [ARGUED]
Earthjustice
1617 John F. Kennedy Boulevard
Suite 2020
Philadelphia, PA 19103

Mychal Ozaeta
Earthjustice
707 Wilshire Boulevard
Suite 4300
Los Angeles, CA 90017

Counsel for Intervenor Respondent Sierra Club

Stephen D. Daly
Katherine L. Vaccaro
Manko Gold Katcher & Fox
Three Bala Plaza East
Suite 700
Bala Cynwyd, PA 19004

3
 Counsel for Intervenor Respondent Montour LLC

Michael F. Strande
Office of Attorney General of Maryland
1800 Washington Boulevard
Suite 6048
Baltimore, MD 21230

Counsel for Amicus Appellee State of Maryland

Christian D. Wright
Office of Attorney General of Delaware
820 N French Street
Carvel Office Building
Wilmington, DE 19801

Counsel for Amicus Appellee State of Delaware

Claiborne E. Walthall
Office of Attorney General of New York
Environmental Protection Bureau
The Capitol
Albany, NY 12224

Counsel for Amicus Appellee State of New York

_________________

OPINION OF THE COURT
_________________

4
SCIRICA, Circuit Judge

The Commonwealth of Pennsylvania submitted a plan
to regulate emissions to the EPA for its approval, as the Clean
Air Act requires. At first, the EPA approved of the plan. But
then, in August 2020, this Court vacated that approval and
directed the Agency to either approve a new state-made plan
or “formulate a new federal . . . plan” to implement emissions
standards in the Commonwealth within two years. Sierra Club
v. EPA, 972 F.3d 290, 309 (3d Cir. 2020). The EPA ultimately
decided to make its own plan and promulgated it towards the
end of 2022. Now the Commonwealth joins one of the three
coal power companies affected by the plan to seek our review.
They argue first that the EPA exceeded its statutory authority
when it promulgated the plan and second that the plan is
arbitrary and capricious because the EPA failed to show its
work. But since the EPA acted in accordance with the Clean
Air Act, we will deny the petition for review.

I.

A. The Clean Air Act

This dispute is governed by the Clean Air Act, which
gives Courts of Appeals jurisdiction to review the EPA’s
promulgation of an implementation plan. See 42 U.S.C.
§ 7607(b)(1). The Act, which has been described as a model
of “cooperative federalism,” EPA v. EME Homer City
Generation, L.P., 572 U.S. 489, 537 (2014) (Scalia, J.,
dissenting), tasks federal, state, and local authorities with
combatting the challenging problem of interstate air pollution.
See 42 U.S.C. § 7401(a)–(c) (noting that a “primary goal” of
the Clean Air Act “is to encourage or otherwise promote

5
reasonable Federal, State, and local governmental
actions . . . for pollution prevention”); see also Bell v.
Cheswick Generating Station, 734 F.3d 188, 190 (3d Cir. 2013)
(“The Clean Air Act states that air pollution prevention and
control is the primary responsibility of individual states and
local governments but that federal financial assistance and
leadership is essential to accomplish these goals.”).

To that end, the Clean Air Act requires the EPA to
promulgate and revise National Ambient Air Quality Standards
(“NAAQs”) for certain pollutants, including ozone, to protect
public health, with higher requirements for areas that do not
attain those standards. 42 U.S.C. § 7409(a)–(b), (d)(1). Once
the EPA establishes the relevant standards, states have the
initial responsibility to devise state implementation plans
(“SIPs”) to address the EPA’s air pollutant reduction goals.
See Sierra Club, 972 F.3d at 293–94; see also 42 U.S.C.
§§ 7410, 7502.

A SIP must satisfy Reasonably Available Control
Technology (“RACT”) requirements. See 42 U.S.C.
§ 7502(c)(1). RACT is a “technology-forcing standard
designed to induce improvements and reductions in pollution
for existing sources.” Sierra Club, 972 F.3d at 294. Although
undefined in the Clean Air Act, the EPA has repeatedly
interpreted this term to mean “the lowest emission limit that a
particular source is capable of meeting by the application of
control technology that is reasonably available considering
technological and economic feasibility.” 87 Fed. Reg. 53381,
53382 (Aug. 31, 2022) (internal quotation marks omitted); see
also Sierra Club, 972 F.3d at 294, 300 (explaining that RACT
is not “the best possible emissions limit,” but “the best limit
that is also economically and technically achievable for plant

6
operators”).

To be RACT-compliant, an implementation plan must
satisfy technological and economic feasibility. Sierra Club,
972 F.3d at 295. Technological feasibility concerns the
application of an emission reduction method to a particular
source and “consider[s] the source’s process and operating
procedures, raw materials, physical plant layout, and any other
environmental impacts such as water pollution, waste disposal,
and energy requirements.” Id. (internal quotation marks
omitted). Economic feasibility “is largely determined by
evidence that other sources in a source category have in fact
applied the control technology in question” and “rests very
little on the ability of a particular source to ‘afford’ to reduce
emissions to the level of similar sources.” Id. (internal
quotation marks omitted).

Once a state submits a SIP or SIP revision, the EPA
must review it for completeness and compliance with the Clean
Air Act’s requirements. See 42 U.S.C. §§ 4710(k)(1)(B),
7410(k)(2)–(4). Specifically, within twelve months of
determining that a submission is complete, the EPA must
approve, disapprove, or conditionally approve, in whole or in
part, each SIP or SIP revision. Id. § 7410(k)(2)–(4). If the
EPA determines that a state failed to make a complete required
submission or “disapproves a [SIP] submission in whole or in
part,” the EPA “shall promulgate a Federal implementation
plan at any time within [two] years” of such determination
“unless the State corrects the deficiency, and the [EPA]
approves the plan or plan revision, before the [EPA]
promulgates” a Federal implementation plan (“FIP”). Id.
§ 7410(c)(1).

7
 B. Factual Background and Procedural History

In May 2016, the Pennsylvania Department of
Environmental Protection (“PADEP”) submitted a SIP (“the
2016 SIP”) to the EPA to satisfy certain pollution control
requirements for the 2008 revision to the 1997 ozone NAAQs.
Sierra Club, 972 F.3d at 296. PADEP’s plan proposed
nitrogen oxide (“NOx”) emission limits for the coal-fired
electric generating units (“EGUs”) equipped with selective
catalytic reduction (“SCR”) controls operating in
Pennsylvania. See id. at 293, 296. SCR is the “preferred
method for limiting coal-fired power plant pollution” and
involves “injecting a substance such as ammonia or urea as a
catalyst into the post-combustion flue gas” to break down
NOx, an ozone precursor emission, “into its component
nitrogen and water molecules and to be dispersed as vapor.”
Sierra Club, 972 F.3d at 295–96. In March 2018, the EPA
provisionally approved the 2016 SIP, which was opposed by a
number of groups, including the Sierra Club and neighboring
states. Id. at 296–97. In 2019, the EPA formally approved the
2016 SIP, and Sierra Club petitioned this Court for review. Id.
at 297.

On August 27, 2020, this Court vacated the EPA’s
approval of the 2016 SIP provisions that Sierra Club
challenged, and remanded the matter to the Agency “with
instructions that it develop enforceable pollution controls in
accordance with its legal obligations.” Id. at 293. EPA was
given two years to “approve a revised, compliant SIP” or
“formulate a new federal implementation plan.” Id. at 309.

Following our decision in Sierra Club, PADEP and the
EPA worked together for months on a corrective SIP revision.

8
EPA Br. 9. In September 2021, the EPA proposed to
disapprove the invalidated portions of the 2016 SIP “to ensure
that [it] ha[d] authority to promulgate a FIP if Pennsylvania
d[id] not submit a timely or approvable SIP revision addressing
the Third Circuit’s decision.” 86 Fed. Reg. 51315, 51316
(Sept. 15, 2021). Between September and December 2021,
PADEP published draft permits that contained its proposed
RACT determinations and engaged in the comments process
for its proposals. 87 Fed. Reg. 31798, 31801–02 (May 25,
2022). The EPA expressed concern about the draft permits,
which the Agency believed were inconsistent with our decision
in Sierra Club. See, e.g., JA 525–27.

In February 2022, the EPA informed PADEP that rather
than continue working with the Commonwealth on a corrective
SIP, it would issue a FIP that it had been concurrently working
on. JA 760. In May 2022, the EPA published a proposed FIP.
See 87 Fed. Reg. 31798 (May 25, 2022). In May and June
2022, PADEP submitted case-by-case RACT SIP revisions
(“the 2022 SIP”). 87 Fed. Reg. 53381, 53382 (Aug. 31, 2022).
The same month, the EPA held a public hearing on its proposed
FIP, and, in July 2022, the comment period for the proposed
FIP closed. 87 Fed. Reg. 31798 (May 25, 2022). PADEP and
the affected facilities submitted comments, which included
concerns that the EPA’s move to promulgate a FIP “usurp[ed]
Pennsylvania’s regulatory primacy” and “violated the
cooperative-federalism structure underpinning the [Clean Air
Act].” Keystone-Conemaugh Br. 19. Moreover, the
commentators expressed concern about the assumptions,
methodology, and conclusions the EPA made in its calculation
of RACT. Id. at 19–20.

In August 2022, the EPA finalized the partial

9
disapproval of the 2016 SIP, and then issued the FIP. See 87
Fed. Reg. 50257 (Aug. 16, 2022); 87 Fed. Reg. 53381 (Aug.
31, 2022); see also Notice of Compliance, Sierra Club v. EPA,
No. 19-2562 (3d Cir. Aug. 26, 2022).

In October 2022, Petitioner Keystone-Conemaugh
Projects LLC (“Keystone-Conemaugh”), the operator of the
Keystone and Conemaugh Generating Stations, and Homer
City Generation LP (“Homer City”), the operator of the Homer
City Generating Station, filed petitions for review, which this
Court consolidated. 1 Docket No. 22-3026, Doc. Nos. 1-1, 8;
Docket No. 22-3039, Doc. No. 1-1. In December 2022,
PADEP intervened in support of Petitioners, and Sierra Club
and Montour, LLC, the operator of the Montour Generating
Station, intervened on behalf of Respondent, the EPA. Docket
No. 22-3026, Doc. Nos. 23, 24.

On February 15, 2024, after briefing and argument in
the above-captioned matter, the EPA issued a notice of
proposed rulemaking to disapprove the 2022 SIP. Docket. No.
22-3026, Doc. No. 92.

1
This case was consolidated for purposes of scheduling and
joint appendix with Appeal No. 22-3039, Homer City
Generation LP. v. EPA, wherein Homer City filed a petition
for review challenging the FIP. However, in accordance with
the agreement of the Parties, that appeal was dismissed under
Fed. R. App. P. 42(b), with prejudice and without cost to either
party. Because Keystone-Conemaugh and PADEP adopted
arguments advanced by Homer City, Homer City’s Opening
Brief and Reply Brief were filed in this appeal as well.

10
 C. The FIP

The FIP establishes NOx emission limits for SCR-
equipped coal-fired EGUs in Pennsylvania to address the
Clean Air Act’s RACT requirements for the 1997 and 2008
ozone NAAQs. See 87 Fed. Reg. 53381, 53402–05 (Aug. 31,
2022). There are nine remaining EGUs in Pennsylvania
operating at four facilities: three at Homer City Generating
Station, two each at Keystone and Conemaugh Generating
Stations, and two at Montour Generating Station. Id. at 53381.
In formulating the FIP, the EPA determined that SCR systems
“represent[ed] appropriate RACT level control technology”
because all the subject facilities are equipped with SCR
systems, which are “demonstrated and highly efficient control
technolog[ies] for the removal of NOx from emissions
associated with coal-fired boilers.” JA 482–83.

The FIP establishes two NOx emission limits: a
weighted, facility-wide rolling thirty-day average emission rate
expressed in pounds per million British Thermal Units
(“MMBtu”) (a unit of heat), and an EGU-specific daily mass
emission limit expressed in pounds. See 87 Fed. Reg. 53381,
53384–85, 53401–04 (Aug. 31, 2022). In addition, the FIP
establishes recordkeeping and reporting requirements such that
each facility must biannually submit reports documenting,
inter alia, daily operating time, daily NOx mass emissions,
heat input, and facility-wide thirty-day rolling emission rates.
Id. at 53404.

In its proposed FIP, the EPA explained that it was
difficult to formulate an enforceable RACT limit because each
EGU’s operations varied in a way that affected SCR
performance. See EPA Br. 12–13; JA 482–84; 87 Fed. Reg.

11
31798, 31803–04 (May 25, 2022). The EPA attributed this
variation in EGU performance to the units’ shift from
“baseload” to “cycling” operation. 87 Fed. Reg. 53381,
53382–83 (Aug. 31, 2022). When the EGUs were built some
sixty years ago, they were intended to operate as baseload
units, running at near-full capacity almost constantly with high
heat inputs and flue-gas temperatures. 87 Fed. Reg. 31798,
31804 (May 25, 2022). The SCR controls installed on the
Keystone, Homer City, and Montour EGUs in the early 2000s
and on the Conemaugh EGUs in 2014 were designed to operate
under these conditions and permitted the units to achieve very
low NOx emissions when they operated at baseload. 2 Id.;
JA 484–85. The EPA found that for the past decade, instead
of operating at baseload, the EGUs tended to “cycle” between
high heat inputs when demand for electricity is high and low
heat or full shutdowns when demand is low. JA 486–88. This
cycling behavior affected the units’ ability to operate their SCR
controls. Id. Indeed, the Agency observed that the EGUs
frequently cycled down to heat inputs below the threshold at
which they could run their SCR controls, leading to excess
emissions. Id. Accordingly, the EPA sought to develop RACT

2
The EPA observed that regulatory pressure also influenced
the units’ emission levels. During its review of historical
operating data, the EPA found that “[s]ome of the lowest NOx
emissions [it] observed coincided with high NOx allowance
prices associated with the NOx SIP Call which went into effect
in 2003.” 87 Fed. Reg. 53381, 53390 (May 25, 2022). The
EPA found that its observation of historical data and
corresponding regulations supported its view that the units had
no trouble meeting lower emission limits when it was
economically advantageous to do so. Id.

12
limits that addressed this cycling behavior and its effect on
SCR performance but also accounted for the lower emissions
rates the EGUs were able to achieve with their SCR controls
when operating at baseload. 87 Fed. Reg. 53381, 53383 (Aug.
31, 2022); 87 Fed. Reg. 31798, 31805–07 (May 25, 2022).

The EPA developed a weighted rates methodology that
considered each EGU’s historical baseload operation data and
more recent cycling operation data. JA 490–491. The EPA
analyzed data from each EGU’s ozone seasons and ranked each
unit’s ozone season performance based on its overall NOx
emission rate. An ozone season is the period from May 1 to
September 30 when meteorologic conditions are most
conducive to ozone formation. Sierra Club, 972 F.3d at 293–
94, 300. The EPA explained that it used ozone-season data as
opposed to full-year data for a number of reasons. First, the
EPA found that from 2017 onwards, the data showed that SCR
systems were often not operated out of ozone season. 87 Fed.
Reg. 53381, 53392 (Aug. 31, 2022). Second, the EPA
explained that the prevailing regulatory scheme 3 permitted the

3
The EPA explained the effect of other regulatory
requirements on non-ozone season data. 87 Fed. Reg. 53381,
53392 (Aug. 31, 2022). The Agency acknowledged that during
this period, the facilities were subject to the Clean Air
Interstate Rule’s (“CAIR”) annual NOx requirement beginning
in 2009, and the Cross-State Air Pollution Rule’s (“CSAPR”)
annual NOx requirements beginning in 2015. Id. CAIR set an
annual NOx emission rate of 0.15 lb/MMBtu, and CSPAR,
which replaced CAIR, utilized a cost-effectiveness level. Id.
These were cap and trade programs, however, and they
permitted individual sources to exceed their allocated
allowances by a certain percentage by purchasing additional

13
facilities to operate their EGUs without having to meet a
specific NOx emission rate, meaning that non-ozone season
data did not reflect the SCR systems’ true non-ozone season
capabilities. Id. The EPA’s decision to analyze ozone-season
data was also consistent with PADEP’s sole reliance on ozone-
season emissions in developing the 2016 SIP. Id.

The EPA sorted the hourly ozone season operating data
based on whether the EGU’s SCR system was operating to
calculate the rates and weights for the formula, which the
agency expressed as follows:

(SCR-on rate * SCR-on weight) + (SCR-off rate

* SCR-off weight) = emissions limit in

lb/MMBtu

JA 493; JA 491–92. The SCR-on and SCR-off rates represent
the EGU’s performance when its SCR system is or is not
running, respectively. JA 493. For each unit at the Keystone,
Homer City, and Montour facilities, the SCR-on rate is the
average hourly NOx rate for hours when the unit’s SCR system
was running during its third-best ozone season (i.e., the third
lowest ozone season average) from 2003 to 2021. Id. The EPA
used this time period because it encompassed all years of SCR

NOx allowances from other sources. Id. Based on the
foregoing, the EPA concluded that non-ozone season
emissions data beginning in 2009 did not necessarily reflect the
SCR control systems’ true non-ozone season capabilities
because the units were not required to meet a specific NOx
emission rate. Id.

14
system operation for the three facilities. 87 Fed. Reg. 53381,
53384 (Aug. 31, 2022). The EPA selected the third-best ozone
season to account for degradation of the SCR systems over
time and to avoid biasing the limit with “uncharacteristically
low emitting days, or under uncharacteristically optimal
operating conditions.” 4 Id. Because SCR controls were not
installed on Conemaugh’s EGUs until 2014, the EPA derived
the SCR-on rate for its EGUs from the units’ second-best
ozone season from 2015 to 2021. Id. at 53382; JA 493. The
EPA used the second-best ozone season for Conemaugh’s
EGUs after the units’ third-best ozone season yielded higher
NOx limits than those for the other facilities. 5 87 Fed. Reg.
53381, 53393 (Aug. 31, 2022).

For each unit, the SCR-off rate represents the average
of all hours the EGU’s SCR system was likely not running.
JA 493. As with the SCR-on rate, the EPA used the period
from 2003 to 2021 for the Keystone, Homer City, and Montour
EGUs, and the period from 2015 to 2021 for Conemaugh’s
EGUs. JA 493; 87 Fed. Reg. 53381, 53384 (Aug. 31, 2022).

4
The EPA also explained that it had used a similar approach in
two other recent rulemakings. See JA 493.
5
The EPA attributed the higher NOx limits for Conemaugh
resulting from the use of the third-best ozone season data to the
more limited data set for the facility’s EGUs. Id. The EPA
observed that Conemaugh’s average ozone season NOx rates
varied significantly between 2015 and 2021, and that the
shorter historical timeframe for the facility did not contain
periods “with high NOx allowance prices” that would have
encouraged Conemaugh to try to achieve lower emissions. Id.;
JA 493 n.10.

15
 The EPA then calculated SCR-on and SCR-off
“weights,” which represent the amount of heat input spent
above or below the threshold at which each unit could operate
its SCR system. 87 Fed. Reg. 53381, 53384 (Aug. 31, 2022).
The EPA formulated the weights by using data from the 2011
to 2021 ozone seasons, which the Agency determined were
likely representative of the time period when the EGUs began
cycling, and because it was likely reasonable to expect that the
EGUs would continue cycling operations in the future. Id.
Again, because Conemaugh’s SCR systems were not installed
until 2014, the EPA used data from 2015 to 2021 to calculate
its weights. JA 501–02. The EPA calculated the weights for
the Keystone, Homer City, and Montour EGUs from the ozone
seasons when each unit had its third-highest proportion of heat
input spent above the threshold for SCR operations. JA 493.
As with the rates, the EPA used Conemaugh’s EGUs’ second-
best ozone season to calculate the applicable weights. 6 87 Fed.
Reg. 53381, 53400 (Aug. 31, 2022).

Applying the formula, the EPA crafted limits using
historical data from each EGU that recognized the units’
present and future cycling behavior, but also weighted the rates
to discourage the EGUs from cycling down to a heat threshold
below which their SCR controls could not operate. EPA Br.
16. In response to comments, the EPA’s final emission rates

6
The EPA decided against using the Conemaugh’s third-best
ozone season to formulate the weights for its EGUs because
the third-best weight would be more analogous to the mean rate
over the shorter data set. 87 Fed. Reg. 53381, 53400 (Aug. 31,
2022). In contrast, the Agency found that using the second-
best ozone season would produce a rate that better reflected the
lowest rates the Conemaugh EGUs could achieve. Id.

16
were averaged over thirty days and across EGUs at each
facility to “smooth” variability in operations. 87 Fed. Reg.
53381, 53396, 53399 (Aug. 31, 2022).

In addition to the rolling thirty-day limit for each
facility, the FIP includes a limit on the pounds of NOx each
unit could produce each day. JA 494. The EPA explained that
this limit complemented the rolling thirty-day limit to ensure
that RACT was applied continuously. Id. The EPA derived
the daily limit by using the thirty-day rolling average limit for
an EGU, multiplying it by the maximum rated heat input for
that EGU (MMBtu/hr), and multiplying that figure by twenty-
four hours. 87 Fed. Reg. 53381, 53401 (Aug. 31, 2022).

II.

We have jurisdiction to review the FIP as it is a final
agency action applicable to the four facilities located within
this Circuit. See 42 U.S.C. § 7607(b)(1).

We review the contents of a final EPA rule to determine
whether it is “arbitrary, capricious, an abuse of discretion, or
otherwise not in accordance with the law.” 42 U.S.C.
§ 7607(d)(9)(A); see also Sierra Club, 972 F.3d at 298.
Although our review is deferential under this “narrow
standard,” GenOn REMA, LLC v. EPA, 722 F.3d 513, 525 (3d
Cir. 2013), “the agency cannot reach whatever conclusion it
likes and then defend it with vague allusions to its own
expertise,” Sierra Club, 972 F.3d at 298. Instead, it “must
examine the relevant data and articulate a satisfactory
explanation for its action including a ‘rational connection
between the facts found and the choice made.’” Motor Vehicle
Mfrs. Ass’n of U.S., Inc. v. State Farm Mut. Auto. Ins. Co., 463

17
U.S. 29, 43 (1983) (citation omitted). When an agency action
fails to consider an important aspect of the problem, cites no
data to support its conclusion, or fails to articulate a rational
basis for its conclusion, such an action is arbitrary and
capricious. See Sierra Club, 972 F.3d at 298 (collecting cases
finding agency action arbitrary and capricious).

“That said, arbitrary and capricious review is not meant
to be an exacting standard,” Logic Tech. Dev. LLC v. FDA, 84
F.4th 537, 549 (3d Cir. 2023), and we must be mindful not to
“substitute [our] judgment for that of the agency,” Motor
Vehicle Mfrs. Ass’n of U.S., 463 U.S. at 43. Accordingly, we
must defer to the agency’s expertise and “uphold agency action
even if its reasoning is ‘of less than ideal clarity’ as long as ‘the
agency’s path may reasonably be discerned.’” Logic Tech.
Dev. LLC, 84 F.4th at 549 (quoting Garland v. Ming Dai, 593
U.S. 357, 369 (2021)). This is “especially [true] in the context
of reviewing a federal agency’s scientific determinations.”
GenOn REMA, 722 F.3d at 526 (citing New Jersey Envt’l
Fed’n v. U.S. Nuclear Regul. Comm’n, 645 F.3d 220, 228 (3d
Cir. 2011)).

III.

Petitioners make two arguments that we must address
sequentially. First, Petitioners assert that the EPA exceeded its
authority under Clean Air Act by promulgating the FIP.
Second, Petitioners argue that even if the EPA’s promulgation
of the FIP were lawful, the Agency’s action was nonetheless
arbitrary and capricious because its calculations for NOx limits
relied on unsupported assumptions and failed to meet RACT
standards.

18
 A. Whether EPA was Statutorily Authorized to
Promulgate a FIP

The Parties’ arguments respecting the EPA’s statutory
authority under the Clean Air Act to promulgate a FIP turn on
the effect of our vacatur of the 2016 SIP in Sierra Club.

Petitioners argue that the EPA could not promulgate a
FIP without first considering the 2022 SIP. According to
Petitioners, our “vacatur wipe[d] the slate clean” and rendered
the 2016 SIP a “legal nullity.” Arg. Tr. 7:3–5; 7:17. As a
result, in Petitioners’ view, the EPA’s partial disapproval of the
2016 SIP following this Court’s vacatur in Sierra Club “had no
legal effect because the [2016 SIP] did not exist at that time as
a matter of law.” Keystone-Conemaugh Reply Br. 8; see also
Keystone-Conemaugh Br. 29. Assuming the 2022 SIP was the
only implementation plan before the EPA, Petitioners argue
that the Agency was obligated under the Clean Air Act to either
approve or disapprove it before promulgating a FIP.

The EPA counters, explaining that its partial
disapproval of the 2016 SIP permitted it to promulgate a FIP.
Implicit in the EPA’s argument is that our vacatur in Sierra
Club merely revoked the EPA’s approval of the 2016 SIP, and
that the Agency still needed to render a decision on the first
implementation plan. Sierra Club, 972 F.3d at 293.

As a threshold matter, a vacated agency action is a
nullity that has no force and effect. Alabama Power Co v. EPA,
40 F.3d 450, 456 (D.C. Cir. 1994). When a court vacates an
agency’s rule, it restores the status quo before the invalid rule
took effect and the agency must “initiate another rulemaking
proceeding if it would seek to confront the problem anew.”

19
Indep. U.S. Tanker Owners Comm. v. Dole, 809 F.2d 847, 854
(D.C. Cir. 1987). In other words, when a court vacates the
EPA’s approval of a SIP, it restores the status quo to that before
the Agency issued a final decision approving or disapproving
the SIP. See Sierra Club v. Johnson, 374 F. Supp. 2d 30, 33
(D.D.C. 2005).

Accordingly, our decision in Sierra Club is properly
understood as vacating the EPA’s approval of the 2016 SIP,
not the SIP itself. See Sierra Club, 972 F.3d at 293 (“[W]e
hold that the EPA’s approval was arbitrary and capricious.”).
As a result, the EPA needed to make a new final ruling
respecting the 2016 SIP. The EPA did so in August 2022 when
the Agency finalized the partial disapproval of the provisions
of the 2016 SIP at issue in Sierra Club. See 87 Fed. Reg. 50257
(Aug. 16, 2022). Once the EPA partially disapproved the 2016
SIP, it was statutorily authorized to promulgate a FIP. See 42
U.S.C. § 7410(c)(1)(B).

PADEP’s submission of the 2022 SIP revisions before
the EPA’s final, partial disapproval of the 2016 SIP did not
divest the EPA of its authority to promulgate a FIP for several
reasons. First, PADEP’s submission of the 2022 SIP did not
absolve the EPA of its responsibility to act on the 2016 SIP.
See Johnson, 374 F. Supp. 2d at 33 (explaining that the EPA’s
duty to act on an initial SIP submission whose approval was
vacated was “not [] mooted or overtaken by the fact that the
states made [additional] submissions”). Second, the plain
language of the Clean Air Act does not require the EPA to act
on a revised SIP before it can issue a FIP. See 42 U.S.C.
§ 7410(c). The Act provides that the EPA “shall promulgate”
a FIP “at any time within [two] years after” the Agency
“disapproves a State implementation plan submission in whole

20
or in part, unless the state corrects the deficiency, and the
[EPA] approves the plan or plan revision, before the [EPA]
promulgates such [a FIP].” Id. Although the Clean Air Act
indicates that submission of a revised SIP and approval of a
SIP revision by the EPA will vitiate the EPA’s obligation to
promulgate a FIP, the provision contains no language requiring
the EPA to act on the SIP revision before promulgating the
FIP. 7 See, e.g., SIH Partners LLLP v. Comm’r, 923 F.3d 296,
304, n.4 (3d Cir. 2019) (“We do not read absent words into a
statute ‘so that what is omitted . . . may be included within its
scope.’” (quoting Lamie v. U.S. Trustee, 540 U.S. 526, 538
(2004))). The EPA thus did not exceed its statutory authority
by promulgating a FIP without first addressing the 2022 SIP. 8

7
Indeed, if the EPA were required to act on each and every SIP
revision submitted before it could issue a FIP, an untenable
scenario could ensue. For instance, if a state were to submit
multiple inadequate SIP revisions, it could effectively nullify
the EPA’s ability to issue a FIP and thus delay the
implementation of any emission limits. See Amici Br. 10.
8
Petitioner PADEP also argues that the EPA violated the Clean
Air Act and abused its discretion by promulgating a FIP that
exceeded the scope of its partial disapproval of the 2016 SIP.
See PADEP Reply Br. 5–6; see also Arg. Tr. 20:12-21:10. In
particular, PADEP contends that the EPA’s FIP contained
source-specific RACT determinations even though it never
disapproved PADEP’s source-specific RACT submissions in
the 2016 SIP. PADEP Reply Br. 5. PADEP’s argument fails
for two reasons. First, PADEP waived this argument by first
raising it in its reply brief. See Garza v. Citigroup Inc., 881
F.3d 277, 284 (3d Cir. 2018) (“Raising an issue in a reply brief
is too late, for ‘[a]s a general matter, an appellant waives an
argument in support of reversal if it is not raised in the opening

21
 Relatedly, Petitioners argue that even if the EPA was
statutorily authorized to promulgate a FIP without first
considering the 2022 SIP, the Agency’s decisions to
discontinue its cooperation with PADEP and develop a FIP
without “a reasoned explanation for its actions,” were
arbitrary, capricious, and an abuse of discretion. Keystone-
Conemaugh Br. 31; PADEP Br. 20–24. Petitioners explain
that the EPA abused its discretion by contravening the
principles of “cooperative federalism” that animate the Clean
Air Act.

Petitioners’ abuse of discretion argument is predicated
on the “core principle” of cooperative federalism that
permeates the Clean Air Act, which gives states the “primary
responsibility” for assuring air quality by submitting SIPs that
specify the manner in which air quality standards will be
maintained. See Homer City, 572 U.S. at 542 (Scalia, J.,
dissenting) (quoting 42 U.S.C. § 7401(a)(3)); see also 42
U.S.C. §§ 7407(a). Petitioners assert that the EPA is
“relegated . . . to a secondary role in the process of determining
and enforcing [] specific, source-by-source emission

brief.’” (quoting In re Asbestos Prod. Liab. Litig. (No. VI), 873
F.3d 232, 237 (3d Cir. 2017))). Second, the plain language of
Section 7410(c) makes clear that even partial disapproval of a
SIP permits the EPA to authorize a FIP. See 42 U.S.C.
§ 7410(c)(1)(B) (allowing the Administrator to promulgate a
FIP at any time within two years after the Administrator
“disapproves a State implementation plan submission in whole
or in part” (emphasis added)). Section 7410(c) contains no
qualifying language that the EPA’s power to promulgate a FIP
is somehow constrained by the extent or scope of its
disapproval.

22
limitations,” Train v. Nat. Res. Def. Council, Inc., 421 U.S. 60,
79 (1975), and “only if a SIP fails to meet [the statute’s] goals
may the Agency commandeer a State’s authority by
promulgating a FIP,” Homer City, 572 U.S. at 538 (Scalia, J.,
dissenting). Consistent with the dissent in Homer City,
Petitioners urge that the EPA “has discretion to arrange things
so as to preserve the Clean Air Act’s core principle of state
primacy—and that it is an abuse of discretion to refuse to do
so.” Id. at 542 (citing 42 U.S.C. § 7607(d)(9)(A)). Indeed, the
dissent in Homer City viewed Section 7410(c)(1) as affirming
this principle of state primacy because the last clause of the
section terminates the EPA’s authority to promulgate a FIP if
the state submits a corrective SIP and EPA approves it. Id. at
542.

Petitioners thus argue that the EPA “could have
arranged things here to preserve Pennsylvania’s primacy over
RACT determinations for coal-fired stations,” but instead
chose to “federalize air emissions regulation in a way that
Congress never intended.” Keystone-Conemaugh Reply Br.
12–13. In their view, the EPA acted in bad faith by first
working with PADEP to promulgate a revised SIP and then
reversing course and issuing a FIP. Keystone-Conemaugh
Reply Br. 14. Next, Petitioners explain that the EPA abused
its discretion by failing to review and address any problems
with PADEP’s 2022 SIP in the time that the Agency spent
promulgating the FIP. Keystone-Conemaugh Reply Br. 14.
We disagree.

As a threshold matter, we note that Petitioners focus
their cooperative federalism arguments on the 2022 SIP, which
was PADEP’s second attempt at formulating a compliant plan.
PADEP had already exercised its primary regulatory

23
responsibility under the Clean Air Act by formulating the
emission limits in the 2016 SIP, the EPA’s approval of which
we vacated in Sierra Club. PADEP had the first shot at
developing emission limits, but because its SIP “fail[ed] to
meet [the Clean Air Act’s] goals,” Homer City, 572 U.S. at 538
(Scalia, J., dissenting), the EPA had the authority and
responsibility to promulgate a FIP at any time within the
following two years, id. at 508 (majority opinion). Thus,
consistent with Petitioners’ interpretation of cooperative
federalism, the EPA was already exercising its “secondary”
authority under the Clean Air Act. 9

Next, Petitioners’ arguments respecting the discretion
that the EPA retains under Section 7410(c)(1) after
disapproving a SIP are foreclosed by controlling caselaw and
the plain language of the statute. First, although the statute puts
priority on the states’ role, § 7401(a)(3), the majority in Homer
City dismissed the dissent’s view that the EPA adhere to
principles of cooperative federalism when acting under Section
7410(c)(1). The majority explained that “nothing in the statute
so restricts [the] EPA,” which retains “plenary authority to
issue a FIP ‘at any time’ within the two-year period that begins
the moment [the] EPA determines a SIP to be inadequate.”
Homer City, 572 U.S. at 511 n.14 (majority opinion); see also
Oklahoma v. EPA, 723 F.3d 1201, 1223 (10th Cir. 2013)
(holding that the filing of a SIP did not relieve the EPA of its
duty to promulgate a FIP, nor did the Agency need to delay
promulgation of a FIP until it ruled on a proposed SIP). This

9
As Petitioners readily admit, the EPA “acknowledged state
primacy in both word and deed” by initially cooperating with
PADEP in developing a revised SIP. Keystone-Conemaugh
Reply Br. 13.

24
is what transpired here.

Although Petitioners may take issue with the EPA’s
decision to discontinue its cooperation with the
Commonwealth on the 2022 SIP, the EPA did not abuse its
discretion by doing so. The Agency was under no obligation
to cooperate with PADEP, but in doing so, the EPA afforded
the Commonwealth more deference than required under
Section 7410(c)(1). 10 And when the EPA decided to
discontinue its cooperation with PADEP and promulgate a FIP,
it provided a rationale for its decision: this Court’s fast
approaching deadline. Faced with substantive concerns about
PADEP’s proposal and its ability to meet Sierra Club’s two-
year deadline, see 972 F.3d at 309, the EPA reasonably decided
to promulgate a FIP to ensure compliance with our Order. Cf.
Midwest Ozone Grp. v. EPA, 61 F.4th 187, 193 (D.C. Cir.
2023) (crediting the “limited amount of time [the] EPA had to
complete the rulemaking” in assessing the reasonableness of

10
PADEP complains that it expended substantial resources
over the course of its year-long cooperative effort with the EPA
to develop a revised SIP to meet its Clean Air Act obligations
in response to Sierra Club only to have its efforts squandered
when the EPA decided to discontinue its cooperation with the
Commonwealth. PADEP Br. 17–18. But the statutory
language anticipates and permits this potential outcome. As
previously discussed, after the EPA disapproves an initial SIP
submission, it must promulgate a FIP unless the state submits
a revised SIP submission, which the Agency may then approve.
See 42 U.S.C. § 7410(c)(1). Put differently, PADEP was
entitled to develop a revised SIP with or without the EPA’s
encouragement for the Agency’s consideration, but there was
no guarantee that the EPA would approve any SIP revision.

25
the EPA’s actions).

Although we conclude that the EPA’s conduct was
reasonable and did not contravene principles of cooperative
federalism, we do not necessarily find the Parties’ conduct
efficient. Nevertheless, we believe the EPA and PADEP
should endeavor to communicate more clearly about how they
will work together to attain the Commonwealth’s required
emissions reductions, recognizing that doing so benefits all
Pennsylvanians, as well as residents of downwind states.

B. Whether the EPA Acted Arbitrarily, Capriciously,
or Abused its Discretion in Promulgating the FIP

Because the EPA’s promulgation of the FIP was proper,
we proceed to Petitioners’ arguments respecting the
reasonableness of the FIP’s emission limits. Petitioners make
numerous arguments, but generally complain that the EPA’s
emission limits are arbitrary and capricious because: (1) they
are predicated on baseless assumptions and on a methodology
lacking evidentiary support; (2) they fail to satisfy RACT
standards; and (3) they impair the Petitioners’ legal obligations
to PJM Interconnection (“PJM”), its regional electric
transmission organization. See Keystone-Conemaugh Br. 31–
52; PADEP Br. 25–32. Because the daily NOx emission limits
are based on the rolling thirty-day limits, Petitioners focus their
arguments on the EPA’s explanations and the Agency’s
evidentiary support for its methodology. 11 See Keystone-

11
Petitioners urge that beyond the purportedly faulty
methodology used to develop the emission limits in the final
rule, the EPA violated the Clean Air Act by failing to provide
a sufficiently detailed explanation of its “novel” weighed-rates

26
approach and the underlying data used to develop the values in
its proposed FIP. Homer City Br. 29. Petitioners claim this
deprived them of a meaningful opportunity to review the
EPA’s decision and forced them to guess how the Agency
calculated the limits for their EGUs. See Homer City Br. 27–
30; JA 627–29, 636 (commenting that the description of the
proposed FIP’s methodology was inadequate to allow for
independent review). Under the Clean Air Act, the EPA must
include in its notice of proposed rulemaking: “the factual data
on which the proposed rule is based;” “the methodology used
in obtaining the data and in analyzing the data;” “the major
legal interpretations and policy considerations underlying the
proposed rule;” and “[a]ll data, information, and
documents . . . on which the proposed rule relies[.]” 42 U.S.C.
§ 7607(d)(3)(A)–(C) (explaining the requirements for
publication of a proposed rulemaking). The EPA did so here.
The Agency explained the challenges with determining RACT
limits for the facilities and its weighted-rates solution that
considered the EGUs’ baseload and cycling operations. 87
Fed. Reg. 31798, 31803–07 (May 25, 2022); JA 482–95. The
EPA also provided the formula it used to calculate the limits,
its calculations, and “[a]ll of the data [the] EPA used to develop
the proposed emission limits . . . was either available in the
docket, or, because of file type and size
limitations . . . available on request.” 87 Fed. Reg. 53381,
53394 (Aug. 31, 2022). So, contrary to Petitioners’ assertions,
having been provided with this information, they did not need
to speculate how the EPA calculated the limits for their EGUs.
Indeed, Montour requested the data from the EPA and clearly
understood the proposed methodology well enough to
“replicate and/or modify [the] EPA’s methodology.” 87 Fed.
Reg. 53381, 53394 (Aug. 31, 2022); see also Montour Br. 6,

27
Conemaugh Br. 51–52.

i. The FIP’s Emission Limits are Reasonable

First, Petitioners stress that the EPA’s use of historical
ozone-season data was arbitrary and capricious because the
EPA assumed that the data were representative of the facilities’
current operations and equipment capabilities. Homer City Br.
32; Keystone-Conemaugh Br. 33–34. Petitioners urge that the
EPA’s justification for its choice—that it had previously
determined in two other rulemakings that historical emissions
data were representative of current performance—was an
inadequate substitute for conducting a statistical analysis of the
available data in this matter to determine whether there were
any significant changes from year to year that could have
rendered the data unrepresentative. Homer City Br. 32–33;
Keystone-Conemaugh Br. 33. Petitioners claim that by failing
to conduct such an analysis, the EPA did not consider
technical, regulatory, and economic changes that occurred
during the nineteen-year data set that could have affected its
representativeness of current unit operations, including: the
EGUs’ shift from baseload to cycling operations; the higher
cost of NOx allowances during 2003 to 2011; and changes to
control technology. Homer City Br. 33–34. We disagree.

Upon review of the record, we find that the EPA
articulated a “satisfactory explanation” for why it could use

10, 20. Montour engaged with the EPA during the comments
process and even helped the Agency by submitting more
accurate SCR threshold data for its facility, which the EPA
used when calculating Montour’s final emission limits. Id. at
53397.

28
historical ozone-season data in developing its emission
limits. 12 Prometheus Radio Project v. FCC, 373 F.3d 372,

12
Petitioners’ claim that the EPA failed adequately to explain
why it used different time periods and averaging periods to
develop its weighted rate approach also fails. Homer City Br.
40–41. The EPA explained how it developed its weighted rate
methodology and why it relied on certain data sets when
calculating certain variables. Specifically, the EPA observed
SCR thresholds from the 2003 to 2021 ozone seasons to
determine the operating threshold at which the facilities could
run their SCR controls. 87 Fed. Reg. 31798, 31804 (May 25,
2022). The EPA considered the 2015 to 2021 ozone seasons
for Conemaugh’s EGUs because SCR controls were not
installed until 2014. JA 493. When the EPA developed the
SCR-on rate for the facilities (except Conemaugh), it explained
that it chose the third-best ozone season from the 2003 to 2021
ozone seasons because it accounted for degradation of control
equipment over time and avoided biasing the limit with
uncharacteristically low emitting days or uncharacteristically
optimal operating conditions. 87 Fed. Reg. 53381, 53384
(Aug. 31, 2022); JA493. The EPA used the second-best ozone
season from the period from 2015 to 2021 for Conemaugh
because its SCR controls on its facilities were not installed
until 2014, and because the average ozone season NOx
emission rates varied significantly over the time period.
JA 493. The EPA derived the SCR weights from a more
limited data set—the 2011 to 2021 ozone seasons—because
the Agency determined that consistent with the trends it
observed in the historical data set, this period was likely
representative of the period during which the EGUs began to
exhibit greater cycling operations. 87 Fed. Reg. 53381, 53384
(Aug. 31, 2022). In sum, the EPA articulated a basis for why

29
389–90 (3d Cir. 2004), as amended (June 3, 2016). The EPA
made clear that a major challenge in developing emission
limits for the facilities was their shift from baseload to cycling
operations. 87 Fed. Reg. 31798, 31802–03 (May 25, 2022);
JA 482–84. The EPA further explained that to identify the
appropriate RACT limits for these sources, it needed to
examine the units’ historical data to determine the emission
rates these sources were capable of meeting when operating
their SCR controls. 87 Fed. Reg. 53381, 53383, 43392 (Aug.
31, 2022). This involved analyzing when and how the units
ran their SCR controls, as well as the regulatory landscape that
influenced operators’ decisions to use the SCR systems.
Indeed, the record indicates that the EPA examined how the
changing regulatory framework influenced SCR performance
during the data set, concluding that the units were capable of
meeting lower emission limits when it was economically
advantageous to do so. Id. at 53390; see also id at. 53395
(acknowledging a “correlation between increased SCR
operation (and correspondingly lower NOx emissions), and
periods when new regulatory requirements . . . have created
meaningfully more stringent NOx emission
budgets[, which] . . . can compel EGUs to operate their SCRs
more often and at lower NOx emission rates[.]”). EPA also
considered whether the regulations in question affected the
representativeness of the data. Compare id. at 53397
(concluding that the facilities could comply with Mercury and
Air Toxin Standards (“MATS”) and reduce their NOx
emissions simultaneously, meaning that emissions data
preceding MATS was still representative), with id. at 53392
(concluding that cap and trade programs that did not require

the different historical periods were selected and how they
were used to develop its methodology.

30
the facilities to meet firm emissions limits rendered non-ozone
season data unrepresentative of SCR performance). This
included considering how NOx allowance prices either
encouraged or discouraged SCR system use. See, e.g., id. at
53390. The EPA also accounted for cycling behavior by
observing the periods in the data set that reflected these
operations and moderated the historical performance by using
more recent data to weight the emission limits. In these
circumstances, we find that the EPA’s use of historical data to
set emissions limits was not arbitrary or capricious.

Relatedly, Petitioners fault the EPA for failing to cite
evidence to support its use of historical ozone-season data to
set emission limits applicable to the entire year. Keystone-
Conemaugh Br. 35; Homer City Br. 35. Petitioners criticize
the Agency’s justifications for the choice: that historical non-
ozone-season did not reflect SCR systems’ true performance
because the SCR controls were not often operated and the
regulatory scheme at the time permitted facilities to purchase
NOx emissions allowances that discouraged emission
reduction; that EGUs in other states were subject to similar
regulations and could achieve similar limits; and that historical
data showed that certain facilities, like Keystone, were capable
of meeting the NOx limits in non-ozone-season months when
operating their SCR controls. 87 Fed. Reg. 53381, 53392,
53395 (Aug. 31, 2022). In response to the Agency’s
justifications, Petitioners urge that that the EPA’s distinctions
between unit operations as reflected in historical ozone- and
non-ozone season data demonstrate the logical fallacy of
applying rates based exclusively on ozone-season data to the
whole year. Keystone-Conemaugh Br. 34. Additionally,
Petitioners argue that the EPA did not articulate a rational basis
for why EGUs in other states and subject to different regulatory

31
regimes were appropriate or relevant points of comparison to
Petitioners’ EGUs. Keystone-Conemaugh Br. 38.

Here, the EPA offered a reasonable explanation for why
it used ozone-season data to calculate the FIP’s year-round
emission limits. See Prometheus Radio Project, 373 F.3d at
389–90. This ozone-season data reflected a period of increased
electrical operations from greater demand and demonstrated
the emission levels the EGUs could achieve when there are
additional regulatory constraints and economic incentives to
encourage operators to reduce emissions. 87 Fed. Reg. 53381,
53392, 53395 (Aug. 31, 2022); see also Sierra Club, 972 F.3d
at 301 (faulting the EPA for relying on emission rates that were
“achieved voluntarily” and “[a]bsent any regulatory pressure
whatsoever” in setting emission limits). Moreover, the EPA
moderated the ozone-season data’s impact on the final
emission limits by weighting those values to account for the
EGUs increased cycling operations, thus acknowledging the
present and likely performance of the units. The EPA also
noted that non-ozone season data was unrepresentative of the
SCR systems’ true performance capabilities because for
substantial periods of time during the data set, the units were
not required to meet a specific NOx emission rate, nor did they
operate their controls for significant periods of time outside of
the ozone season. See id. at 53392.

Contrary to Petitioners’ assertion that the EPA engaged
in the same speculative comparison to out-of-state facilities
that resulted in our vacatur of the 2016 SIP in Sierra Club, see
972 F.3d at 301–03, the EPA provided a reasoned explanation
for the comparison. See 87 Fed. Reg. 31798, 31808 (May 25,
2022). The EPA analyzed operating and emissions data from
out-of-state facilities, which are equipped with SCR controls

32
operating on a year-round basis and are also contractually
obligated to PJM. 87 Fed. Reg. 31798, 31808 (May 25, 2022).;
87 Fed. Reg. 53381, 53395 (Aug. 31, 2022). After reviewing
the out-of-state facilities’ data, the EPA determined that those
facilities experienced comparable changes in operations and
load levels but did not exhibit the same level of excess
emission levels as the Pennsylvania units. 13 87 Fed. Reg.
31798, 31808 (May 25, 2022); 87 Fed. Reg. 53381, 53395
(Aug. 31, 2022). After this data analysis, the EPA was able to
conclude there “was nothing unique about the operating
patterns of the units in Pennsylvania” which would explain this
difference. 87 Fed. Reg. 31798, 31808 (May 25, 2022). It was
not arbitrary or capricious for the EPA to use these out-of-state
facilities—which are subject to similar market forces and
operational trends as the Pennsylvania EGUs and achieve
emission levels similar to the final emission limits in the FIP—
as points of comparison. Their performance and emission
levels reinforced the reasonableness and technological and
economic feasibility of the FIP’s limits derived from the
Pennsylvania facilities’ own operating data.

Next, Petitioners urge that the emission limits are
arbitrary and capricious because the EPA failed to cite any
evidence supporting its selection of the affected units’ second-

13
In these circumstances, where the EPA compared the
Pennsylvania EGUs performance to similar, SCR-equipped,
out-of-state EGUs running their controls year-round, in an
effort to determine whether there were any state-specific
conditions that caused Pennsylvania’s EGUs to emit greater
NOx emission levels, we believe that the agency has
sufficiently “show[n] its work.” Nat’l Parks Conservation
Ass’n v. EPA, 803 F.3d 151, 167 (3d Cir. 2015).

33
or third-best ozone season “as a proxy for the SCR systems’
current performance condition.” Keystone-Conemaugh Br.
34; Homer City Br. 38. Petitioners claim that it was
unreasonable for the EPA to justify its choice by citing to its
use in two previous rulemakings and to assume that these
ozone years accurately reflected the current performance of the
facilities’ SCR systems, which degrade over time. Keystone-
Conemaugh Br. 33–34.

As an initial matter, Petitioners are mistaken that the
EPA selected the units’ second- or third-best ozone season
exclusively as proxy for the units’ SCR systems’ current
operational performance. 87 Fed. Reg. 53381, 53395 (Aug.
31, 2022); see also id. at 53395 (explaining that equipment
degradation was a “contributing factor” but “not the only
consideration [the EPA] evaluated when selecting the third-
best approach”). Rather, the EPA explained that it selected the
units’ third-best ozone season to determine RACT because
doing so would avoid “biasing the SCR-on limit with
uncharacteristically low emitting ozone seasons, or under
uncharacteristically optimal operating conditions.” Id. at
53391. The EPA relied on the RACT standard, which does not
require the sources to “achieve the absolute lowest level of
emissions that is technologically possible . . . to satisfy
RACT,” Sierra Club, 972 F.3d at 302, when it decided against
using ozone seasons that reflected better SCR system
performance, as it concluded that emission performances
reflected in those ozone seasons might not again be achievable
due to aging control technology. 87 Fed. Reg. 53381, 53392
(Aug. 31, 2022). Lastly, the EPA’s referral to previous
rulemakings, which it used to support the reasonableness of its
selection, reflects a reasonable reliance on its expertise where,
as here, its selection of the third-best ozone season was

34
supported by historical data. See Prometheus Radio Project,
373 F.3d at 417 (“[W]e must uphold an agency's line-drawing
decision when it is supported by the evidence in the record.”).

Petitioners also argue that the EPA’s use of the second-
best ozone season data for the Conemaugh facility instead of
the third-best ozone season data used for the other facilities
invalidates its entire emissions limit methodology. Keystone-
Conemaugh Reply Br. 21–24; Homer City Br. 39. Petitioners
point out that the EPA decided to use the second-best ozone
season data for Conemaugh’s EGUs after its calculations
relying on the EGUs’ third-best ozone season data returned
comparatively higher emission limits for Conemaugh’s units.
Keystone-Conemaugh Reply Br. 22; 87 Fed. Reg. 53381,
53393 (Aug. 31, 2022). Petitioners urge that the EPA’s
explanation for the Conemaugh calculation does not relate to
equipment degradation, which Petitioners contend was the
primary reason the Agency offered for using the third-best
ozone season. Homer City Br. 39. Petitioners explain that
these unexpected results for the Conemaugh facility obligated
the EPA to reevaluate its methodology or identify facts in the
record that supported its conclusion to use the second-best
ozone season data for the facility. Keystone-Conemaugh
Reply Br. 23.

Having reviewed the record, we are satisfied with the
EPA’s explanation for its use of the second-best ozone season
data for the Conemaugh facility. See Motor Vehicle Mfrs.
Ass’n, 463 U.S. at 43. The Agency explained that the
comparatively high proposed emission limits for Conemaugh
calculated using the weight derived from the third-best ozone
season stemmed from selecting the third-best ozone season
from a more limited data set. 87 Fed. Reg. 53381, 53400 (Aug.

35
31, 2022). Conemaugh’s shorter data set contained periods
when NOx allowance prices declined, which would have
disincentivized Conemaugh from achieving the lowest
possible emissions. Id. at 53393; see also JA 490 (observing
an increase in NOx emission rates at some SCR-controlled
EGUs as NOx allowance prices declined between 2017 and
2020). The Agency determined that using weights derived
from the third-best ozone season from Conemaugh’s more
limited data set produced emission limits that were analogous
to the facility’s mean rates. 87 Fed. Reg. 53381, 53400 (Aug.
31, 2022). Having been admonished for approving emission
limits in the 2016 SIP that effectively amounted to the average
pollution output of three of the affected facilities, it was
reasonable for the EPA to adjust the formula for Conemaugh
to ensure that its limits were RACT-compliant. See Sierra
Club, 972 F.3d at 300 (“an average of the current emissions
being generated by existing systems, will not usually be
sufficient to satisfy the RACT standard.”).

Petitioners also urge that the EPA made unsupported
assumptions about the facilities’ obligations to PJM.
Keystone-Conemaugh Br. 39–42. PJM is a grid operator that
provides wholesale electricity in thirteen states and the District
of Columbia, and “generally directs the day-to-day and hour-
to-hour dispatch” of Petitioners’ units. 87 Fed. Reg. 53381,
53397 (Aug. 31, 2022); see also id. at 53382 n.5. The EGUs
supply their energy by bidding into the PJM electricity market.
Id. at 53382. Petitioners contend the EPA failed to consider
the operational control that PJM exerts over the facilities when
it formulated the FIP’s emission limits. Keystone-Conemaugh
Br. 28. They argue that because the EPA does not have special
expertise in the realm of power delivery and grid reliability, its
assumptions that Petitioners would be able to “consider their

36
emission limits when developing the information they supply
to PJM” and that PJM would in turn consider those limits when
determining “subsequent dispatch instructions” were arbitrary
and capricious. 87 Fed. Reg. 53381, 53398 (Aug. 31, 2022).

Although the EPA understood that PJM was generally
responsible for dispatch of the affected EGUs, it explained that
the facilities had the ability to influence PJM’s dispatch
instructions through their offer prices and with operating
parameters they provide to PJM. 87 Fed. Reg. 53381, 53397
(Aug. 31, 2022). PJM considers these parameters—which
relate to operations such as unit start-ups, extended low-load
operations, and output changes—when making its dispatch
decisions. Id.; see also id. at 53398 & n.51. The EPA also
corroborated its assumption based on a comparative analysis
of 2021 and 2022 ozone season data from the Keystone and
Conemaugh units. Id. at 53398–99. The EPA found that the
Keystone and Conemaugh Generating Stations appeared to
receive an exception in 2022 based on operating parameters
provided to PJM, as the units’ so-called “Turn Down Ratio”
was below the ratio’s default floor value applicable in the
absence of such an exception. Id. at 53398 n.51; see also id. at
53399. The EPA also considered Keystone-Conemaugh’s
claim that it would be obligated to forfeit dispatch
opportunities an admission that PJM did not have undisputed
control over the facilities’ operations. Id. at 53398–99. As a
result, the EPA reasoned that the facilities could provide PJM
with operational parameters to help them avoid dispatch
instructions from PJM that would require low-load operations
(with potentially suboptimal SCR performance). Id. In these
circumstances, we find that the EPA articulated a rational basis
for its conclusion. See Sierra Club, 972 F.3d at 298.

37
 ii. The FIP’s Emission Limits Satisfy RACT Standards

In addition to their claims that the emission limits in the
FIP are unreasonable, Petitioners also urge that the limits do
not meet RACT standards because they are not technologically
or economically feasible. 14 Keystone-Conemaugh Br. 42.

First, Petitioners claim that the FIP does not comply
with RACT because the EPA did not conduct unit-specific
technological and economic feasibility analyses, and that its
statistical analysis is an inadequate substitute. Homer City Br.
30–31; PADEP Br. 26–29. Petitioners explain that without this

14
PADEP goes a step further, arguing not only that the
emission limits are technologically and economically
unfeasible, but that the EPA’s selection of SCR controls
without analysis or consideration of other control technologies
renders its RACT analysis invalid. PADEP Br. 31–32. Yet
PADEP itself determined in its 2016 SIP and 2022 SIP that
SCR controls represent appropriate RACT-level control
technology. 87 Fed. Reg. 53381, 53387 (Aug. 31, 2022);
JA 482–83. The EPA agreed with that assessment, which was
never challenged, when it approved the 2016 SIP, and
continued to agree with it when formulating the FIP. 87 Fed.
Reg. 53381, 53387 (Aug. 31, 2022); JA 483. PADEP also
conducted an analysis of other potential control technologies
and, apart from boiler tuning, rejected them. 87 Fed. Reg.
53381, 53387–88 (Aug. 31, 2022). Because SCR controls are
present and operating at each of the EGUs, and because the
EPA explained that its adoption of SCR controls as RACT-
level control technology was consistent with PADEP’s
previous analysis and findings, it was reasonable for the EPA
not to evaluate other technologies in its RACT analysis.

38
EGU-specific analysis, the EPA’s presumptions about each
source’s operational and emissions-reduction capabilities are
speculative and not RACT-compliant. Homer City Br. 31.
Here, it is difficult to envision how the EPA’s limits are not
source-specific when they are derived from the individual
EGUs’ historical operating data. 15 JA 491–95. Similarly, we
find the EPA’s use of statistical analysis of historical data
reasonable when it is not prohibited by statute or the EPA rules
or guidance, and where the EPA explained the adjustments it
made to account for factors in the units’ historical data that may
have affected the data set’s reflection of the units’ lowest
achievable rates. 87 Fed. Reg. 53381, 53387 (Aug. 31, 2022).

Next, Petitioners argue that the EPA’s weighted rates
approach fails to account for the EGUs’ shift from baseload to
cycling operations and results in emission limits that are not
technically and economically feasible. Keystone-Conemaugh
Br. 43–45. Petitioners explain that the EPA’s use of ozone-
season data biased the results of its analysis toward EGU
operations with SCR controls engaged and thus “arrived at
limits that are not technologically feasible.” Keystone-
Conemaugh Reply Br. 19. Additionally, Petitioners urge that
the EPA’s selection of the second- or third-best ozone season
data to set the SCR-on and -off weights further biased the limits
toward SCR operation because those data “are not typical
of . . . cycling operations” and produced technologically
infeasible limits. Keystone-Conemaugh Reply Br. 20–21;
Keystone-Conemaugh Br. 44–45.

15
We also note that Petitioner PADEP appears to concede that
emission limits in the FIP are in fact source-specific. See
PADEP Reply Br. 8–9.

39
 Here, EPA took steps to account for the EGUs’
increased cycling operations when developing its emissions
methodology. EPA’s weighted rate limits are designed to
encourage improvements in SCR performance by
acknowledging the historical effectiveness of the control
technology’s emission reduction capabilities and the more
recent cycling behavior that can hinder SCR performance.
This goal is consistent with RACT, which, as we have already
explained, is a technology-forcing standard designed to reduce
pollution from existing sources. Sierra Club, 972 F.3d at 294.
EPA recognized that the units have exhibited more frequent
cycling behavior—and thus more periods of operation without
SCR controls engaged—over the past decade and used data
from this timeframe 16 to calculate the SCR-on and -off weights
in its formula to account for this operational reality. See 87
Fed. Reg. 53381, 53384, 53396 (Aug. 31, 2022); see also
JA 493 (explaining that the 2011 to 2021 ozone seasons were
“likely representative of the time period when the units began
to exhibit a greater cycling pattern”).

In addition, the EPA’s decision to average the NOx
emission rate over thirty days and across units within a facility
gives Petitioners an easier path to compliance. 87 Fed. Reg.
53381, 53396 (Aug. 31, 2022). This accounts for cycling
behavior in two ways. First, the thirty-day average permits
operators to offset periods where their EGUs operate above the
limit (and presumably without their SCR controls running)
with periods of optimal operation where the units operate

16
Petitioner Keystone-Conemaugh incorrectly asserts that the
EPA calculated the SCR-on and -off weights by considering
the entire period since the EGUs’ SCR systems were installed.
Keystone-Conemaugh Br. 43.

40
below the limit (when their SCR controls presumably are
running). Id. Second, the facility-wide aspect of the average
permits operators to average the suboptimal, above-limit
performance of one EGU, which may be exhibiting more
frequent cycling behavior, with the optimal, below limit
performance of another. Id. The FIP’s limits thus accounted
for the EGUs cycling behavior. See Sierra Club, 792 F.3d at
295 (explaining that technological feasibility considers, inter
alia, the “source’s process and operating procedures”).

In addition, data in the record shows that these sources
are capable of meeting the limits. Indeed, past historical data
reflects that each affected unit has met the FIP limits. 17
JA 495. The EPA also explained that very recent performance
data from the affected units indicates that they can comply with
the FIP limits, even while engaging in more frequent cycling
behavior. See 87 Fed. Reg. 53381, 53390–91 (Aug. 31, 2022)
(“[D]ata for some of these units from May through June of the
2022 ozone season generally indicate SCR operating patterns
(and, as a result NOx emissions) that match or are among their
best in the recent data record.”). In these circumstances, we
find that the EPA did not bias the limits toward SCR-on
operations in such a way that rendered them technologically
infeasible.

17
Moreover, in response Keystone-Conemaugh’s comments
that some of its EGUs would not be able to meet the thirty-day
facility-wide average emission rate if there was one cold start-
up in a thirty-day period, the EPA reviewed unit startup data
during non-ozone season months and found that one of the
EGUs was able to achieve the thirty-day emission rate in the
final rule after a cold startup, even without the regulatory
pressure to do so. 87 Fed. Reg. 53381, 53396 (Aug. 31, 2022).

41
 Petitioners also argue that the limits are not RACT-
compliant because the EPA did not conduct an appropriate
economic feasibility analysis. Homer City Br. 41. In
particular, Petitioners claim that the EPA’s failure to consider
source-specific capital 18 and maintenance costs by assuming
that the EGUs could meet the emission limits simply by
optimizing their existing operations was arbitrary and
capricious. Homer City Br. 41. In addition, Petitioners urge
that the limits are not economically feasible because the cost
of new or upgraded equipment is disproportionate to the
amount of pollutant reduction achieved. Homer City Br. 43.
Moreover, Petitioners urge that the EPA failed to consider that
the units’ cycling operations would require them to increase
ammonia injections in order to more aggressively control NOx
emissions, and that this action would lead to accelerated
maintenance costs. Keystone-Conemaugh Br. 48.

18
After the issuance of the FIP, an engineering firm retained
by Homer City to evaluate its facility’s ability to comply with
the FIP’s emission limits concluded that two of the facility’s
three units would require costly upgrades to their ammonia
vaporizers to comply. Homer City Br. 42–43 n.7. This study
was issued after the notice and comment period, however, so
we may not consider it when evaluating the EPA’s decision.
See Am. Petroleum Inst. v. Costle, 665 F.2d 1176, 1186 n.3
(D.C. Cir. 1981); 42 U.S.C. § 7607(d)(7)(A). And, if we were
to consider the report, it is noteworthy that it concluded that
one of Homer City’s EGUs can comply with the FIP without
any additional upgrades, see Homer City Br. 42–43 n.7,
demonstrating that at least one of the “sources in [the] source
category ha[s] in fact applied the control technology in
question.” Sierra Club, 972 F.3d at 295.

42
 Here, the EPA considered record evidence and did not
find that capital expenditures were necessary for compliance.
87 Fed. Reg. 53381, 53388–89 (Aug. 31, 2022). The Agency
also found that historical operating data, as well as data from a
portion of the 2022 ozone season, indicated that the EGUs
could meet the limits with existing controls and without
injecting excessive amounts of ammonia during unfavorable
SCR operating conditions. Id. at 53390. Although Petitioners
may wish to upgrade their equipment or inject additional
ammonia to maintain their current levels of profitability, these
market considerations have no bearing on the RACT analysis.
See Sierra Club, 972 F.3d at 295.

Petitioners also argue that the limits are economically
infeasible because they would have to forfeit potential dispatch
opportunities with PJM to comply with the limits. Keystone-
Conemaugh Br. 44. But Petitioners’ purported costs are
actually profit opportunities that Petitioners may have to
forego, and not costs associated with reducing emissions to
comply with the FIP. See Sierra Club, 972 F.3d at 295
(explaining that economic feasibility concerns the cost of
reducing emissions and not the ability of a particular source to
afford to reduce emissions). Although Petitioners will have to
reevaluate their operations to comply with the FIP, lost
potential revenues do not factor into the economic-feasibility
analysis. See 57 Fed. Reg. 18070, 18074 (Apr. 28, 1992).

Lastly, Petitioners argue that the EPA’s daily mass limit
for each EGU fails to satisfy RACT because it is based on the
facility-wide thirty-day limit, which Petitioners contend is not
source-specific and does not consider source-specific design
parameters. Keystone-Conemaugh Br. 51–52; Homer City Br.

43
46–50. We disagree. The daily mass limit for each unit is
based on two source-specific metrics: each unit’s maximum
capacity and the facility-wide limit. 87 Fed. Reg. 31798,
31806 (May 25, 2022); JA 494–95. Again, as we have already
discussed, the facility-wide thirty-day limit is source-specific
and considers source-specific design parameters because it is
based on actual operating data from each affected unit. 87 Fed.
Reg. 31798, 31806 (May 25, 2022); JA 494–95. The EPA also
observed from emission data following its FIP proposal that
when the EGUs operated within the thirty-day limits, they
generally achieved the daily mass limits. 87 Fed. Reg. 53381,
53396 (Aug. 31, 2022).

In sum, we reject Petitioners’ critiques of the FIP’s
emission limits, as they satisfy RACT and are not arbitrary and
capricious. 19 Cf. Midwest Ozone Grp., 61 F.4th at 193
(denying petition for review when the record demonstrated that

19
We find it compelling that Montour does not challenge the
emission limits for the two EGUs at its generating station,
which it believes may be reasonably achieved and satisfy
RACT as technologically and economically feasible. Montour
Br. 10. Petitioners do not challenge the limits for the Montour
EGUs, but those limits were determined using the same
underlying weighted rates methodology. Montour Br. 12; see
87 Fed. Reg. 53381, 53402–03 (Aug. 31, 2022) (applying the
same weighed-rate methodology to each affected facility).
Although Montour’s acquiescence to the emission limits does
not make them reasonable or RACT-compliant per se, its
engagement with the EPA in the same rulemaking process that
resulted in the emission limits for the other affected EGUs
further supports our finding that the EPA acted reasonably in
promulgating comprehensible emission limits in the FIP.

44
the EPA “chose[] analytical techniques rationally connected”
to the rule and “appropriately explained” methodology).

IV

For the foregoing reasons, we will uphold the FIP and
deny the petition for review. We hold that the EPA properly
exercised its authority under the Clean Air Act by partially
disapproving the 2016 SIP and promulgating the FIP. We also
hold that the contents of the FIP are not arbitrary, capricious,
or abusive of the EPA’s discretion.

45