United States Court of Appeals
         FOR THE DISTRICT OF COLUMBIA CIRCUIT



Argued January 16, 2020                Decided May 19, 2020

                         No. 18-1285

                    STATE OF MARYLAND,
                        PETITIONER

                              v.

           ENVIRONMENTAL PROTECTION AGENCY,
                     RESPONDENT

                 CITY OF NEW YORK, ET AL.,
                       INTERVENORS


            Consolidated with 18-1287, 18-1301


        On Petitions for Review of an Action of the
       United States Environmental Protection Agency


    William J. Kassab, Deputy Attorney General, Office of the
Attorney General for the State of Delaware, argued the cause
and filed the briefs for petitioner State of Delaware.

    Joshua Berman argued the cause for Citizen Petitioners.
With him on the briefs were Ariel Solaski, Jon A. Mueller, Leah
Kelly, Ann Brewster Weeks, Graham McCahan, Sean H.
Donahue, and Susannah L. Weaver. Hayden W. Hashimoto
entered an appearance.
                              2
    Michael F. Strande, Assistant Attorney General, Office of
the Attorney General for the State of Maryland, argued the
cause for petitioner State of Maryland. With him on the briefs
were Brian E. Frosh, Attorney General, and Joshua M. Segal,
Special Assistant Attorney General.

     Letitia James, Attorney General, Office of the Attorney
General for the State of New York, Morgan A. Costello and
Claiborne E. Walthall, Assistant Attorneys General, Barbara
D. Underwood, Solicitor General, Steven Wu, Deputy Solicitor
General, David S. Frankel, Assistant Solicitor General, Gurbir
S. Grewal, Attorney General, Office of the Attorney General
for the State of New Jersey, Aaron A. Love, Deputy Attorney
General, and Christopher G. King were on the brief for
petitioner-intervenors State of New York, et al.

    Richard L. Revesz and Jack Lienke were on the brief for
amicus curiae Institute for Policy Integrity at New York
University School of Law in support of petitioners.

     Samara M. Spence, Attorney, U.S. Department of Justice,
argued the cause for respondent. With her on the brief were
Jeffrey Bossert Clark Sr., Assistant Attorney General,
Jonathan Brightbill, Principal Deputy Assistant Attorney
General, and Abirami Vijayan and Stephanie L. Hogan,
Counsel, U.S. Environmental Protection Agency.

     Norman W. Fichthorn, E. Carter Chandler Clements,
Garry S. Rice, Rae E. Cronmiller, and Janet J. Henry were on
the brief for respondent-intervenors.

    Before: HENDERSON, GARLAND, and KATSAS, Circuit
Judges.
                                3
    Opinion for the Court filed PER CURIAM.

     PER CURIAM: The Clean Air Act, 42 U.S.C. §§ 7401 et
seq., charges the United States Environmental Protection
Agency (EPA) with regulating air pollution, including ozone.
Clean Air Act section 110 requires individual states to adopt
plans for the implementation and enforcement of EPA-
mandated national air quality standards.

But
because pollutants are readily transported across large areas,
without regard to state boundaries, upwind emissions can
impede downwind states’ attainment of the national standards.
To address this unequal burden, section 110 includes a “Good
Neighbor Provision,” which requires state plans to prohibit
emissions that will “contribute significantly” to nonattainment
in any other state.

The EPA has
developed a four-step framework to address Good Neighbor
obligations in this context. At Step One, it identifies downwind
areas projected to have trouble attaining the relevant air quality
standard. At Step Two, the EPA determines which upwind
states are “linked” to the downwind nonattainment sites. At
Step Three, it calculates the optimal level of pollution control,
considering the marginal cost of emission reductions and
anticipated downwind air quality improvements. The EPA
then formulates an emissions budget for each state, accounting
for achievable reductions. Finally, at Step Four, the EPA
typically promulgates federal implementation plans that
require upwind states’ participation in a regional cap-and-trade
program to bring about compliance with their Good Neighbor
obligations.

     Separately, Clean Air Act section 126(b) authorizes “[a]ny
State” to petition the EPA for a finding that an upwind source
“emits or would emit” in violation of the Good Neighbor
Provision’s prohibition.

If the EPA makes the
requested finding, the offending source must cease operations
                               4
unless it complies with federally enforceable emission
limitations. In 2016 Maryland and Delaware filed section
126(b) petitions requesting that the EPA impose additional
limitations on certain upwind sources that were purportedly
contributing to the two States’ nonattainment of the national
ozone standards. Both States sought to require the optimization
of existing selective catalytic reduction controls; Maryland also
addressed the operation of selective non-catalytic reduction
controls at two facilities and Delaware requested that one
facility burn only natural gas.

     The EPA denied the petitions on October 5, 2018.
Because a section 126(b) petition seeks a finding that the
upwind source has violated the Good Neighbor provision, the
EPA applies the same four-step framework it developed in the
implementation of section 110. The EPA denied Delaware’s
petitions at Step One, finding that Delaware had not
demonstrated a current or future in-state air quality problem
and that, under the EPA’s own modeling, no such problem
would exist under either the 2008 or 2015 ozone standards.
Alternatively, the EPA concluded that denial was warranted
under Step Three because Delaware failed to identify any
available cost-effective controls at the named sources.
Although Maryland survived Steps One and Two, the EPA
denied its petition at Step Three. Like Delaware, Maryland
failed to identify further cost-effective emission reductions at
sources operating with catalytic controls. For the remaining
sources named in Maryland’s petition, the EPA explained that
non-catalytic controls were not cost-effective in this context.
Maryland, Delaware and a coalition of environmental groups
(Citizen Petitioners) petition for review of the EPA’s denials.
Although we reject some of the EPA’s Step One
determinations, we find, with one exception, that it reasonably
denied the petitions at Step Three. We conclude, however, that
the EPA’s explanation was inadequate with respect to non-
                                   5
catalytic controls. We therefore grant Maryland’s petition for
review in part and remand this issue to the EPA. We deny all
other petitions for review.

                           I. Background

A. Statutory Framework

     The Clean Air Act instructs the EPA to establish a primary
and secondary National Ambient Air Quality Standard
(NAAQS), see 42 U.S.C. § 7409, for each air pollutant “which
may reasonably be anticipated to endanger public health or
welfare,”

1 Once established by the EPA,
these standards “become the centerpiece of a complex statutory
regime aimed at reducing the pollutant’s atmospheric
concentration.” Am. Trucking Ass’ns, Inc. v. EPA, 

, 358–59 (D.C. Cir. 2002). The EPA first promulgated the
NAAQS for ground-level ozone, i.e., smog, in 1979. See
Revisions to the National Ambient Air Quality Standards for
Photochemical Oxidants, 44 Fed. Reg. 8202 (Feb. 8, 1979). In
1997 it set the ozone NAAQS at a level of 80 parts per billion
(ppb), measured over an eight-hour period. See National
Ambient Air Quality Standards for Ozone, 62 Fed. Reg. 38,856
(July 18, 1997). The EPA subsequently reduced the ozone
NAAQS to 75 ppb in 2008, see National Ambient Air Quality
Standards for Ozone, 73 Fed. Reg. 16,436 (Mar. 27, 2008), and,
in 2015, to 70 ppb, see National Ambient Air Quality Standards
for Ozone, 80 Fed. Reg. 65,292 (Oct. 26, 2015).



     1
        A “primary” NAAQS must specify the level of air quality
“requisite to protect the public health,” while “allowing an adequate
margin of safety.” 42 U.S.C. § 7409(b)(1). A “secondary” NAAQS,
on the other hand, “specif[ies] a level of air quality . . . requisite to
protect the public welfare.”

                               6
     To promote attainment and maintenance of the NAAQS,
the “EPA, in coordination with state governments, divides the
country geographically into ‘[a]ir quality control region[s].’”
NRDC v. EPA, 

, 458 (D.C. Cir. 2014) (alterations
in original) (quoting 42 U.S.C. § 7407). “Some areas lie within
a single state while others encompass portions of two or more
states.” Del. Dep’t of Nat. Res. & Envtl. Control v. EPA, 

, 94 (D.C. Cir. 2018). Once the EPA issues a new or
revised NAAQS, it “designates each area as ‘attainment,’
‘nonattainment,’ or ‘unclassifiable’ with respect to the
NAAQS.”

U.S.C. § 7407(d)(1)(A), (B)). An
“attainment” area meets the NAAQS, that is, the atmospheric
concentration of the regulated pollutant is less than the
allowable level; an “unclassifiable” area, as the name suggests,
cannot be classified due to the absence of available
information; and a “nonattainment” area exceeds the NAAQS
or contributes to a violation in a nearby area. 42 U.S.C.
§ 7407(d)(1)(A)(i)–(iii).    Ozone nonattainment areas are
further classified by operation of law, according to the severity
of their air quality problems, as marginal, moderate, serious,
severe, or extreme.

These classifications
determine how long the area has in order to attain the primary
NAAQS.

nonattainment area that misses its
attainment deadline is generally bumped up to the next highest
classification,

which “impose[s]
additional regulatory responsibilities on the states composing
that area,” Del. Dep’t of Nat. 

.

    Following the promulgation of a NAAQS, each state must
submit a state implementation plan (SIP) that “provides for
implementation, maintenance, and enforcement” of the
NAAQS within that state. 42 U.S.C. § 7410(a)(1). For states
in nonattainment areas, “SIPs must show how the areas will
achieve and maintain the relevant NAAQS.” S. Coast Air
Quality Mgmt. Dist. v. EPA, 

, 1143 (D.C. Cir.
                                 7
2018). In particular, SIPs for ozone nonattainment areas must
adopt certain measures, see, e.g.,

intended to
bring about attainment “as expeditiously as practicable” and
not later than specific statutory deadlines, 42 U.S.C.
§ 7511(a)(1). If a state fails to submit a SIP, or if its submission
is incomplete or disapproved, the EPA must issue a federal
implementation plan (FIP) that requires the state to correct the
identified deficiency.

     State-level air quality regulation is an inherently
complicated endeavor because “[a]ir pollution is transient,
heedless of state boundaries. Pollutants generated by upwind
sources are often transported by air currents, sometimes over
hundreds of miles, to downwind States.” EPA v. EME Homer
City Generation, L.P., 

, 496 (2014). For example,
ground-level ozone is not emitted directly into air. Rather, it is
the product of chemical reactions between nitrogen oxides
(NOx) and non-methane volatile organic compounds in the
presence of sunlight. See New York v. EPA, 

, 989
(7th Cir. 1998). Thus, the upwind emission of ozone
precursors can seriously threaten downwind attainment of the
ozone NAAQS. “As the pollution travels out of state, upwind
States are relieved of the associated costs,” which “are borne
instead by the downwind States, whose ability to achieve and
maintain satisfactory air quality is hampered by the steady
stream of infiltrating pollution.” EME Homer 

.

    To alleviate this potential inequity, Congress included a
Good Neighbor Provision in the Clean Air Act. Under the
Good Neighbor Provision, SIPs must prohibit in-state sources
“from emitting any air pollutant in amounts which will . . .
contribute significantly to nonattainment in, or interfere with
maintenance by, any other State with respect to any
[NAAQS].” 42 U.S.C. § 7410(a)(2)(D)(i). If the SIP is
                                 8
incomplete or inadequate, a FIP may be necessary to address
the state’s Good Neighbor obligations. See

     Separate from the SIP and FIP process, Clean Air Act
section 126(b) authorizes “[a]ny State or political subdivision”
to petition the EPA “for a finding that any major source or
group of stationary sources emits or would emit any air
pollutant in violation of” the Good Neighbor Provision. 42
U.S.C. § 7426(b). 2 The EPA must “make such a finding or
deny the petition” within sixty days.

words, it
“must act quickly . . . and not wait the potential several years
that it would take for states to fully adopt SIPs implementing
new NAAQS.” GenOn REMA, LLC v. EPA, 

, 520
(3d Cir. 2013). But the EPA still must determine whether an
upwind source has violated the Good Neighbor Provision and,
accordingly, its evaluation of each section 126(b) petition is
tied to its interpretation and implementation of the Good
Neighbor Provision. Once a petition is granted, the offending
source must cease operations within three months unless it
complies with EPA-mandated “emission limitations.” 42
U.S.C. § 7426(c).

B. Regulatory History

     In 2011 the EPA issued the Cross-State Air Pollution Rule
(CSAPR), which placed emission limitations on upwind states
that violated their Good Neighbor obligations with respect to
the 1997 and 2006 fine particulate matter NAAQS and the 1997
ozone NAAQS. See Federal Implementation Plans: Interstate

    2
       Although § 7426(b) refers to any emission “in violation of the
prohibition of section 7410(a)(2)(D)(ii),” we have held that this
cross-reference is a scrivener’s error and should be read to refer to
the Good Neighbor Provision in § 7410(a)(2)(D)(i).               See
Appalachian Power Co. v. EPA, 

, 1040–44 (D.C. Cir.
2001) (per curiam).
                                9
Transport of Fine Particulate Matter and Ozone and Correction
of SIP Approvals, 76 Fed. Reg. 48,208 (Aug. 8, 2011). To
account for the revised ozone NAAQS, the EPA promulgated
an update to the CSAPR in 2016. See Cross-State Pollution
Rule Update for the 2008 Ozone NAAQS, 81 Fed. Reg. 74,504
(Oct. 26, 2016) [hereinafter Update Rule]. The Update Rule
finalized FIPs to address twenty-two states’ Good Neighbor
obligations with respect to the more-stringent 2008 ozone
NAAQS. Substantively, the Update Rule established further
limits on ozone season 3 NOx emissions from electric
generating units (EGUs) in those states.

To
quantify the necessary reductions, the EPA applied a four-step
framework.

     At Step One, the EPA considered current and modeled
future air quality data at downwind monitors (i.e., receptors) to
identify areas expected to have trouble attaining or maintaining
the 2008 ozone NAAQS.

The EPA utilized a
monitoring site’s earlier “design values”—a three-year
historical average of a receptor’s air quality data—to generate
multiple ozone-level projections for 2017.

With
the 2008 ozone NAAQS of 75 ppb, any receptor with a
projected design value of less than 76 ppb was determined to
be in attainment. See

was designated
nonattainment if its average projected design value and its most
recent monitored design value (2013–2015) equaled or
exceeded 76 ppb.

the EPA defined a
“maintenance” receptor as any site that is currently in

    3
       The ozone season runs May 1 through September 30. See
Update Rule, 81 Fed. Reg. at 74,507. “Ozone levels are generally
higher during the summer months” because “[t]he potential for
ground-level ozone formation increases during periods with warmer
temperatures and stagnant air masses.”

Reducing
emissions during this timeframe is thus a critical component of the
EPA’s regulatory approach.
                               10
attainment but has a projected average design value that
exceeds the NAAQS, or that has an average design value below
the NAAQS but a maximum projected design value of 76 ppb
or greater.

Two, the EPA identified the upwind states
“linked” to nonattainment or maintenance at downwind
monitors.
 Id. at 74,518. 
First, the EPA calculated each state’s
contribution to downwind ozone formation. Next, because the
Good Neighbor Provision prohibits only those emissions that
“contribute significantly to nonattainment” or “interfere with
maintenance,” the EPA screened out upwind states that
contributed less than one per cent of the 2008 ozone NAAQS
(i.e., 0.75 ppb) to ozone formation in a downwind state.
 Id. at
74,537. 
Simply put, only those “States . . . whose contributions
to a specific receptor meet or exceed the screening threshold
are considered linked to that receptor.”
 

Three, the EPA applied a multifactor test—
considering cost, NOx reduction potential, and downwind air
quality impacts—to quantify the magnitude of the emission
reductions required by the Good Neighbor Provision.
 Id. at
74,519. 
The EPA measured the expected reductions at
different cost-control levels: $800/ton, $1,400/ton, $3,400/ton,
$5,000/ton, and $6,400/ton.
 Id. at 74,540–42. 
“Each level . . .
represents an estimated marginal cost per ton of NOx reduced
and is characterized by a set of pollution control measures.”
 Id.
at 74,540. 
For each cost-control level, the EPA also estimated
corresponding air quality improvements at downwind
receptors. It ultimately concluded that a control cost of $1,400
per ton—which represents turning on and fully operating
existing, idled selective catalytic reduction controls—
constituted the point “at which incremental EGU NOx
reduction potential and corresponding downwind ozone air
quality improvements are maximized with respect to marginal
                              11
cost.”
 Id. at 74,550. 
The EPA then quantified each state’s
emissions “budget” by projecting the emissions that would
occur under $1,400 per ton cost controls.
 Id. at 74,553.
Emissions that can be reduced at or below the selected control
level are considered “significant” for purposes of Good
Neighbor compliance. See EPA Br. 10.

     Finally, at Step Four, the EPA implemented an allowance
trading program to achieve the required emission reductions.
See Update Rule, 81 Fed. Reg. at 74,521. Each state receives
an allocation of individual allowances authorizing the emission
of a designated quantity of ozone season NOx.
 Id. at 74,554.
The total allowances equal that state’s emissions budget and
are allocated among sources in that state. Because allowances
can be bought and sold through market transactions, sources
can emit more NOx than otherwise permitted by purchasing
additional allowances.
 Id. Each state, 
however, can emit no
more than 121 per cent of its emissions budget.
 Id.

     Importantly, the 
Update Rule was promulgated as a partial
remedy.      Because downwind states with a moderate
nonattainment classification faced a July 2018 attainment
deadline, the EPA focused solely on near-term emission
reductions. See
 id. at 74,540 
(“[T]he EPA limited its analysis
of potential NOx reductions in each upwind state to those that
could be feasibly implemented for the 2017 ozone season,
which is the last full ozone season prior to the July 20, 2018
attainment date.”). Accordingly, analysis of further controls
was anticipated “in any future action that may be necessary to
address upwind states’ full emission reduction obligations with
respect to the 2008 ozone standard.”
 Id. A number 
of parties
challenged the Update Rule and we invalidated it in part. See
Wisconsin v. EPA, 
938 F.3d 303
, 309 (D.C. Cir. 2019) (per
curiam). Because upwind states could continue to significantly
contribute to downwind air quality beyond the downwind
                               12
attainment deadline, we concluded the Update Rule was
inconsistent with the Clean Air Act, which “require[s] upwind
States to eliminate their significant contributions in accordance
with the deadline by which downwind States must come into
compliance with the NAAQS.”
 

     In December 2018, the EPA promulgated the Close-Out
Rule.     See Determination Regarding Good Neighbor
Obligations for the 2008 Ozone National Ambient Air Quality
Standard, 83 Fed. Reg. 65,878 (Dec. 21, 2018) [hereinafter
Close-Out Rule]. The Close-Out Rule found that, for the 2008
ozone NAAQS, it was not feasible to implement cost-effective
emissions controls before 2023—two years after the 2021
attainment deadline for serious areas—and, moreover, that all
downwind states would attain the NAAQS by 2023 even
without further upwind emission reductions.
 Id. at 65,904–05,
65,917. Due in part to its finding that regionwide NOx
emissions had declined twenty-one per cent in the Update
Rule’s first year,
 id. at 65,899, 
the EPA concluded that the
Update Rule fully resolved the Good Neighbor obligations for
twenty upwind states,
 id. at 65,879. 
We vacated the Close-Out
Rule because it relied on the same statutory interpretation of
the Good Neighbor Provision that Wisconsin rejected. New
York v. EPA, 781 F. App’x 4 (D.C. Cir. 2019) (per curiam).

C. Procedural History

     In 2016, Maryland and Delaware both filed section 126(b)
petitions with the EPA. Maryland’s sole petition alleged that
thirty-six EGUs, in “five upwind states that EPA ha[d] already
determined are significantly contributing to Maryland’s ozone
problem,” were violating the Good Neighbor Provision with
respect to the 2008 ozone NAAQS. Md. Cover Ltr at 1 (J.A.
48). According to Maryland, 2015 ozone season monitoring
data demonstrated that these units either were not optimizing
                               13
their existing controls or had ceased running these controls
regularly during the ozone season. Md. Pet. at 4–5 (J.A. 53–
54). Contending that the EPA’s regional cap-and-trade
approach did not prevent sources from emitting above
achievable limits on particularly bad ozone days,
 id. at 3 
(J.A.
52), Maryland requested source-specific limitations that would
require the “targeted EGUs to run their existing NOx control
technology effectively on each day of the ozone season,”
 id. at
4 
(J.A. 53).

     Delaware submitted four petitions, each addressing a
different upwind facility. All four petitions requested a finding
that EGUs at the named facilities violated the Good Neighbor
Provision with respect to both the 2008 and 2015 NAAQS.
Delaware alleged that three of the facilities were not optimizing
their existing controls. The fourth facility—the Brunner Island
power plant in Pennsylvania (Brunner Island)—did not have
catalytic controls installed. It was, however, then in the process
of adding natural gas capacity. Delaware maintained that
Brunner Island’s continued ability to burn coal warranted the
imposition of short-term NOx emission limits and asked the
EPA to impose an enforceable requirement that Brunner Island
burn only natural gas. Del. Brunner Island Pet. at 20, 22 (J.A.
238, 240).

     Although section 126(b) requires the EPA to act within
sixty days after receipt of a petition, the EPA sought to extend
its deadline by six months pursuant to 42 U.S.C. § 7607(d)(10).
But the EPA failed to hold a public hearing or otherwise act on
Maryland’s petition by the new deadline. Maryland filed suit
and the United States District Court for the District of
Maryland ordered the EPA to “sign a notice taking final agency
action on Maryland’s petition on or before September 15,
2018.” Maryland v. Pruitt, 
320 F. Supp. 3d 722
, 732 (D. Md.
2018). The EPA proposed to deny the Maryland and Delaware
                                   14
petitions on June 8, 2018, see Response to Clean Air Act
Section 126(b) Petitions from Delaware and Maryland, 83 Fed.
Reg. 26,666 (June 8, 2018) and finalized the denials on October
5, 2018, see Response to Clear Air Act Section 126(b) Petitions
from Delaware and Maryland, 83 Fed. Reg. 50,444 (Oct. 5,
2018) [hereinafter Response to Delaware and Maryland].

     The EPA, using the same four-step Good Neighbor
framework it applied in the Update Rule, concluded that neither
Maryland nor Delaware could establish a Good Neighbor
violation. First, the EPA determined that Maryland satisfied
Steps One and Two because the EPA modeled a 2017
maintenance problem at Maryland’s Harford County receptor
for the 2008 ozone NAAQS and the upwind states named in
Maryland’s petition are linked to that receptor according to the
EPA’s contribution modeling.
 Id. at 50,464. 
Maryland’s
petition failed, however, at Step Three. Maryland requested
that EGUs be required to operate and optimize existing
catalytic controls.4 But, because this same control strategy was
already reflected in the Update Rule’s emissions budgets, the
EPA determined that “all identified cost-effective emission
reductions have already been implemented for the 2008 ozone
NAAQS with respect to the” identified sources.
 Id. This was
so, the EPA said, based on both “a conceptual case as to why
those reductions will be achieved through the [Update Rule’s]
existing allowance trading program, and an evidence-based
case that reductions based on control optimization [were]
already achieved in 2017.”
 Id. at 50,462.

     4
       Catalytic and non-catalytic controls both involve injecting a
reagent into an exhaust flue, where it reacts with NOx to produce
molecular nitrogen and water. As their names suggest, catalytic
controls facilitate this reaction with a catalyst. Non-catalytic controls
do not. See Appalachian Power Co. v. EPA, 
135 F.3d 791
, 798 nn.7–
8 (D.C. Cir. 1998) (per curiam).
                               15
     As for EGUs operating selective non-catalytic reduction
controls, the EPA concluded “that fully operating with [non-
catalytic controls] is not a cost-effective NOx emissions
reduction strategy for these sources.”
 Id. at 50,469. 
Although
Maryland submitted a comment asking the EPA to assess its
petition under the 2015 ozone NAAQS, the EPA declined to do
so, finding that Maryland’s petition had requested a decision
with respect to the 2008 ozone NAAQS only.
 Id. at 50,463.
     The EPA denied Delaware’s petitions at Steps One and
Two and, in the alternative, at Step Three. First, Delaware
failed to satisfy its purported burden under section 126(b) to
demonstrate “that there is a current or future nonattainment or
maintenance problem in Delaware based on violations of the
[2008 or 2015 ozone] NAAQS, []or that the named sources are
improperly impacting downwind air quality on days when such
violations would be expected.”
 Id. at 50,457. 
Delaware’s
identification of individual exceedances (i.e., an eight-hour
measurement above the NAAQS) at its own monitors was
insufficient because the EPA considers “[v]iolations, rather
than exceedances” as “the relevant metric for identifying
nonattainment and maintenance problems.”
 Id. at 50,456. 
Nor
did the fact that areas of Delaware were designated
nonattainment for the 2008 NAAQS indicate, by itself, that the
State would have future attainment problems.
 Id. Under the
EPA’s interpretation of the Good Neighbor Provision, if a
downwind state will attain the NAAQS without further upwind
reductions, there is no air quality problem necessitating
additional emission limitations. Otherwise, the EPA would
over-control upwind states by imposing limitations on
emissions that do not significantly contribute to downwind
nonattainment.
 Id. For this 
reason, although commenters
identified Delaware monitors that were currently exceeding the
2015 ozone NAAQS, the commenters failed to “identify any
projected air quality violations in a future year associated with
                                16
the relevant attainment dates.”
 Id. Commenters also 
identified
monitors in the Philadelphia-Wilmington-Atlantic City
nonattainment area that were violating the 2008 and 2015
ozone NAAQS. Because Delaware’s New Castle County is
included in this multistate area, the commenters argued that
“Delaware’s attainment of the ozone NAAQS is tied to the
attainment of the other monitors in the nonattainment area.”
 

But the EPA interpreted section 126(b) to authorize
findings only with respect to downwind receptors within the
petitioning state and, accordingly, declined to consider these
data.
 Id.

     Although the 
EPA found Delaware’s conclusions
unsupported and technically deficient,
 id. at 50,456, 
it
nevertheless proceeded to analyze independently Delaware’s
petitions at Step One. Relying on the modeling it conducted in
connection with the Update Rule, the EPA determined that
Delaware was not projected to have any nonattainment or
maintenance receptors with respect to the 2008 ozone NAAQS
and its most recent design values were not to the contrary.
 Id.
at 50,458. 
And, although the modeling evinced air quality
problems in Delaware for the 2015 ozone NAAQS, the EPA
deemed upwind control unnecessary because it projected
Delaware to attain the 2015 ozone NAAQS by 2023.
 Id. at
50,459. 
Even though Delaware is first subject to the 2021
marginal nonattainment deadline,5 the EPA determined that
2023 is in fact the relevant future analytic year because it is the

    5
       New Castle County was designated as marginal nonattainment
for the 2015 ozone NAAQS given its inclusion in the Philadelphia-
Wilmington-Atlantic City multistate nonattainment area.          See
Additional Air Quality Designations for the 2015 Ozone National
Ambient Air Quality Standards, 83 Fed. Reg. 25,776, 25,794 (June
4, 2018). The three-year attainment deadline for marginal areas runs
from 2018, when the area was designated under the 2015 NAAQS.
See 42 U.S.C. § 7511(a)(1); 
NRDC, 777 F.3d at 464
–66.
                               17
last full year that will inform analysis of the 2024 moderate
attainment deadline—the first deadline requiring downwind
states to implement controls on existing sources.
 Id. at 50,461.
     Notwithstanding the EPA’s denial of Delaware’s petitions
at Step One, it found, in the alternative, that Step Three
constituted an independent basis for denial. Three of
Delaware’s petitions pertained to sources with catalytic
controls. As with Maryland’s petition, the EPA determined
that emission reductions associated with the operation of
catalytic controls were already implemented through the
Update Rule.
 Id. at 50,465. 
Delaware’s Brunner Island
petition also failed to show that the facility “emits or would
emit in violation” of the Good Neighbor Provision.
 

Brunner Island primarily burned natural gas during the
2017 ozone season, achieving emission reductions consistent
with Delaware’s proposed control strategy.
 Id. at 50,470–71.
Accordingly, the EPA concluded that “no additional feasible
and cost-effective NOx emissions reductions . . . have been
identified.”
 

And, because the EPA predicted that
Brunner Island would continue to burn natural gas for
economic reasons, it found that the facility would not emit in
violation of its Good Neighbor obligations.
 

.
    Maryland, Delaware, and Citizen Petitioners (collectively,
the Petitioners) petition for review, challenging the denial of
Delaware’s petitions at Step One of the Good Neighbor
framework and the denial of both States’ petitions at Step
Three. New York, New Jersey, and New York City6 intervened

    6
        The EPA recently denied a Section 126 petition from New
York on substantially similar grounds. See Response to Clean Air
Act Section 126(b) Petition from New York, 84 Fed. Reg. 56,058
(Oct. 18, 2019). New York, New Jersey, and New York City have
petitioned for review of that denial. New York v. EPA, No. 19-1231
(D.C. Cir. filed Oct. 29, 2019).
                                18
on behalf of the Petitioners and several power companies
intervened in support of the EPA.

      We have jurisdiction under the Clean Air Act. See 42
U.S.C. § 7607(b)(1). We may reverse any EPA action found
to be “arbitrary, capricious, an abuse of discretion, or otherwise
not in accordance with law.”
 Id. § 7607(d)(9)(A). 
To
determine whether an action is arbitrary and capricious, “we
apply the same standard of review under the Clean Air Act as
we do under the Administrative Procedure Act.” Allied Local
& Reg’l Mfrs. Caucus v. EPA, 
215 F.3d 61
, 68 (D.C. Cir. 2000).
In doing so, we must “give an ‘extreme degree of deference’ to
the EPA’s evaluation of ‘scientific data within its technical
expertise,’ especially where, as here, we review the ‘EPA’s
administration of the complicated provisions of the Clean Air
Act.’” Miss. Comm’n on Envtl. Quality v. EPA, 
790 F.3d 138
,
150 (D.C. Cir. 2015) (per curiam) (citation omitted) (first and
second quoting City of Waukesha v. EPA, 
320 F.3d 228
, 247
(D.C. Cir. 2003) (per curiam); then quoting Catawba Cty. v.
EPA, 
571 F.3d 20
, 41 (D.C. Cir. 2009) (per curiam)). Further,
we “review[] the EPA’s interpretation of the Clean Air Act
under the familiar two-step framework formulated in Chevron,
. . . defer[ring] to the EPA’s interpretation if the statutory text
is ambiguous and the EPA’s interpretation is reasonable.” Am.
Fuel & Petrochemical Mfrs. v. EPA, 
937 F.3d 559
, 574 (D.C.
Cir. 2019) (per curiam).

       II. Denial of Delaware’s Petitions at Step One

     The Petitioners first argue that the EPA arbitrarily denied
Delaware’s section 126(b) petitions at Step One. They assert
that the EPA impermissibly refused to consider data from an
out-of-state receptor and data regarding nonattainment before
2023. The EPA contests both points. In addition, it contends
                                19
that, regardless of whether its own analysis was flawed,
Delaware bore the burden of proof and failed to meet it.

A. Burden of Proof

     We first consider whether the EPA permissibly assigned
the burden of proof to Delaware. On this question of statutory
construction, we ask only whether the EPA’s position
reasonably interprets the governing provisions of the Clean Air
Act. See Chevron U.S.A. Inc. v. NRDC, 
467 U.S. 837
, 842–44
(1984).

    We begin with statutory text. Section 126(b) provides that
“[a]ny State or political subdivision may petition the
Administrator for a finding” of a violation of the Good
Neighbor Provision. 42 U.S.C. § 7426(b). Section 126(b)
makes clear that the petitioning state must initiate the process.
As a general “default rule,” the burden of proof falls “upon the
party seeking relief.” Schaffer ex rel. Schaffer v. Weast, 
546
U.S. 49
, 57–58 (2005) (civil litigation). The same rule governs
formal proceedings under the Administrative Procedure Act. 5
U.S.C. § 556(d) (“[T]he proponent of a rule or order has the
burden of proof.”). And the provision of the Clean Air Act
governing section 126(b) proceedings does nothing to displace
the default rule. See 42 U.S.C. § 7607(d). This strongly
suggests that the default rule should apply. See Astoria Fed.
Sav. & Loan Ass’n v. Solimino, 
501 U.S. 104
, 108 (1991).

    Statutory context reinforces this conclusion. The Clean
Air Act requires the EPA to resolve a section 126(b) petition
“[w]ithin 60 days after receipt . . . and after public hearing,” 42
U.S.C. § 7426(b), though the EPA may extend the deadline to
six months,
 id. § 7607(d)(10). 
In New York v. EPA, 
852 F.2d
574 
(D.C. Cir. 1988), we held that this compressed timeline
supports requiring the petitioning state to bear the burden of
                               20
proof. There, three states argued that section 126(b) required
the EPA “to take the investigatory steps necessary to determine
whether” there was any violation of the Good Neighbor
Provision.
 Id. at 578. 
We rejected this contention that the EPA
must itself “conduct whatever data-gathering and research is
necessary to either prove . . . or affirmatively disprove” a
state’s allegations.
 Id. Given the 
sixty-day deadline, we
thought it “reasonable to conclude that Congress did not intend
that the Administrator be required to perform all these duties in
such a short period of time.”
 Id.

     We recognize 
that the petitioning states in New York
sought to compel the EPA to evaluate entire SIPs, whereas
Delaware seeks only a finding that individual upwind sources
emit excessively. This distinction makes little difference, for
any evaluation under the Good Neighbor Provision requires
time-intensive research and analysis assessing air quality
problems in the petitioning downwind state, the cause of those
problems in upwind states, and the cost-effectiveness of
possible solutions. As we explained in New York, these tasks
are at odds with a sixty-day deadline. We therefore hold that
the EPA reasonably interpreted section 126(b) to require
Delaware to bear the burden of proof.

B. Consideration of Out-of-State Monitors

     We now address the EPA’s decision not to consider air
quality data from non-attaining receptors outside Delaware,
“even if such monitors are located in a multistate
nonattainment area that includes [Delaware].” Response to
Delaware and Maryland, 83 Fed. Reg. at 50,460. The EPA
construes the “petition authority” set out in section 126(b) as
“limited to states and political subdivisions seeking to address
interstate transport of pollution impacting downwind receptors
within their geographical borders.”
 Id. (emphasis added).
                                 21
Delaware and Petitioner-Intervenors assert that section 126(b)
plainly authorizes Delaware to petition based on out-of-state
monitoring data or, alternatively, that the EPA’s statutory
interpretation is unreasonable.

     Again, we start with the text. See NRDC v. Browner, 
57
F.3d 1122
, 1125 (D.C. Cir. 1995). In relevant part, section
126(b) provides that “[a]ny State or political subdivision may
petition the Administrator for a finding that any major source
or group of stationary sources emits or would emit any air
pollutant in violation of [the Good Neighbor Provision].” 42
U.S.C. § 7426(b). Starting with the premise that “[a]ny State”
can file a section 126(b) petition, Delaware contends the EPA’s
interpretation conflicts with the statute’s plain meaning
because the text does not “explicitly bar[] a state from
petitioning EPA for a finding that a source is affecting
downwind receptors in another state.” Del. Br. 19. Delaware’s
emphasis is misplaced—a liberal construction of “[a]ny State”
does not mandate a similarly expansive scope for the petition
itself. 7 In other words, that any state can petition the EPA says

     7
        The parties’ discussion of Delaware Department of Natural
Resources is largely beside the point. There, we considered whether
an attainment-date extension, which may be granted “[u]pon
application by any State,” 42 U.S.C. § 7511(a)(5), required every
state in a multistate area to request the extension. Resolving the
question at Chevron step one, we held that “‘any State’
unambiguously permits EPA to consider an application filed by
fewer than all states in a multistate nonattainment area.” Del. Dep’t
of Nat. 
Res., 895 F.3d at 99
. Here, there is no dispute that a single
state within a multistate nonattainment area can file a section 126(b)
petition based on air pollution within its own borders. The fact that
“any State” was given unambiguous meaning in a different context
does not resolve whether “any State” can file a section 126(b)
petition to determine whether an upwind source is contributing to air
pollution in a different state.
                              22
nothing about whether the petitioning state can rely on air
quality data from a different state to support its requested
finding that an upwind source violates the Good Neighbor
Provision. Although Delaware construes this silence in its
favor, the fact remains that Congress “has [not] directly spoken
to the precise question at issue.” 
Chevron, 467 U.S. at 842
.

     We turn, therefore, to Chevron step two, where “we
presume that when an agency-administered statute is
ambiguous with respect to what it prescribes, Congress has
empowered the agency to resolve the ambiguity. The question
for a reviewing court is whether in doing so the agency has
acted reasonably and thus has ‘stayed within the bounds of its
statutory authority.’” Util. Air Regulatory 
Grp., 573 U.S. at
315 
(quoting City of Arlington v. FCC, 
569 U.S. 290
, 297
(2013)). Although we defer to a permissible construction of
the Clean Air Act, the EPA’s interpretation must be
“reasonable in light of the Act’s text, legislative history, and
purpose.” Allied Local & Reg’l Mfrs. 
Caucus, 215 F.3d at 68
(citation omitted).

     The EPA primarily argues that statutory context dictates a
narrow construction for section 126(b) petitions. First, other
Clean Air Act provisions “that contain petition authority . . .
expressly allow for any person to petition the EPA” whereas
section 126(b) is limited to states and political subdivisions.
Response to Delaware and Maryland, 83 Fed. Reg. at 50,460
(emphasis added); see, e.g., 42 U.S.C. § 7661d(b)(2). But this
observation has little bearing on the question before us. “Any
State” plainly includes Delaware and the fact that the section
126(b) petition process is comparatively circumscribed does
not mean an otherwise qualified petitioner is thereafter subject
to additional, implicit limitations. Next, the EPA contends that
“the context of . . . section 126 as a whole suggests these
provisions are meant to moderate interstate transport concerns
                               23
between affected states and upwind sources, not between any
third party (even if such party is another state) and upwind
sources.” Response to Delaware and Maryland, 83 Fed. Reg.
at 50,460 (emphasis added). For example, section 126(a)
requires certain upwind sources “to provide written notice to
all nearby States the air pollution levels of which may be
affected by such source.” 42 U.S.C. § 7426(a)(1). The EPA
imports this language into section 126(b), reading in a
requirement that the petitioning state be “directly affected by
upwind pollution.” Response to Delaware and Maryland, 83
Fed. Reg. at 50,460 (emphasis added). And, according to the
EPA, Delaware is not “affected” because it has provided no
evidence of an in-state air quality problem.

     The EPA’s position is unavailing. Although Congress, in
section 126(a), referred specifically to “States . . . which may
be affected by such [upwind] source,” 42 U.S.C. § 7426(a)(1),
it included no analogous limitation in section 126(b).
Moreover, it is not at all obvious when a state is “affected.”
Indeed, section 126(a) requires a new or modified source to
provide written notice if it “may significantly contribute to
levels of air pollution in excess of the [NAAQS] in any air
quality control region outside” the source’s own state.
 Id.
§ 7426(a)(1)(B). 
Because “air quality control regions” include
multistate areas,
 id. § 7407(b)(1), 
(c), it is unclear whether all
states in a shared nonattainment area are “affected”—and are
therefore owed written notice—regardless of where in the
multistate area the offending air pollution is measured. Thus,
even accepting the EPA’s proffered interpretation—that only
affected states may file section 126(b) petitions to protect
against violating upwind sources—the same ambiguity arises.

     Although statutory context, on its own, does not resolve
matters, the EPA also found that “the legislative history for
[section 126(b)] suggests the provision was meant to address
                                 24
adverse air impacts only in the petitioning state.” Response to
Delaware and Maryland, 83 Fed. Reg. at 50,460. Its brief foray
into the legislative history is, at best, unilluminating. It relies
solely on the Conference Report accompanying the Clean Air
Act Amendments of 1977, which added section 126. The
Senate’s proposed amendment is described in the Report as
cabining section 126(b)’s petition authority to findings that an
upwind source “adversely affect[s] the air quality in the
petitioning State.” H.R. Rep. No. 95-564, at 146 (1977) (Conf.
Rep.). This summary language, which the EPA erroneously
ascribes to the amendment itself, see Response to Delaware and
Maryland, 83 Fed. Reg. at 50,460 n.47, does not appear in the
enacted text, see Clean Air Act Amendments of 1977, Pub. L.
No. 95-95, § 123, 91 Stat. 685, 724. Nor does the EPA suggest
this limitation was considered and debated, beyond its brief
assertion that “[t]he House concurred with the Senate’s
amendment” and, despite making revisions elsewhere, “did not
indicate changes to this sentence.” Response to Delaware and
Maryland, 83 Fed. Reg. at 50,460 n.47.

     Delaware and Petitioner-Intervenors contend that, even if
a state must be “affected” in order to file a section 126(b)
petition, “upwind emissions that impact non-attaining
receptors within a multistate nonattainment area affect all of the
included states’ air quality . . . , regardless of where the
particular non-attaining receptor happens to be located.” Del.
Reply Br. 4. First, because ozone and its precursor pollutants
are easily transported over large areas, without regard to
political boundaries, selectively placed ozone monitors are
intended to “serve as proxies for identifying broader air quality
problems.” Pet’r-Intervenors Br. 29.8 Consequently, a
    8
        The EPA objects that Petitioner-Intervenors’ argument was
not raised in the comment period and is therefore forfeited. See 42
U.S.C. § 7607(d)(7)(B) (“Only an objection . . . raised with
reasonable specificity during the period for public comment . . . may
                                  25
violating monitor anywhere in the shared nonattainment area
signals that other locations may face similar problems.

     Second, for states in a shared nonattainment area, out-of-
state receptors can cause very real regulatory consequences.
Because a nonattainment area includes “any area . . . that
contributes to ambient air quality in a nearby area that does not
meet” the ozone NAAQS, 42 U.S.C. § 7407(d)(1)(A)(i), a
state’s attainment status is necessarily linked to all other states
comprising the multistate area. A non-attaining receptor
anywhere in the multistate area causes the entire area to be
designated nonattainment and, in turn, the states must
coordinate a collective response irrespective of the offending
monitor’s location. See
 id. § 7511a(j)(1). 
Thus, Delaware and
Petitioner-Intervenors attack the EPA’s notion that considering
monitoring data from outside the petitioning state would
impermissibly “allow states to act in the role of citizen
attorneys general on behalf of other states.” EPA Br. 39. They
maintain that, to the contrary, a state acts on its own behalf
when it raises a shared nonattainment problem because its own




be raised during judicial review.”). “But the word ‘reasonable’
cannot be read out of the statute in favor of a hair-splitting approach.
In other words, the Act does not require that precisely the same
argument that was made before the agency be rehearsed again, word
for word, on judicial review.” Appalachian Power 
Co., 135 F.3d at
817
–18. In its comments, Delaware argued that its attainment status
was tied to the status of the Philadelphia-Wilmington-Atlantic City
nonattainment area. See Del. Cmts. at 4 (J.A. 345). This is largely
the same argument now offered by Petitioner-Intervenors. And
because the EPA has “considered the particular challenge raised on
judicial review, it is of no import whether that challenge is phrased
in exactly the same way in each forum.” Appalachian Power 
Co.,
135 F.3d at 818
.
                              26
attainment status has been affected by the upwind source. We
agree.

     The untenability of the EPA’s interpretation is clear when
we consider its practical effect on the section 126(b) petition
process. Section 126(b) permits downwind states to petition
for a finding that an upwind source is violating the Good
Neighbor Provision, which prohibits emissions that will
“contribute significantly to nonattainment in . . . any other
State.” 42 U.S.C. § 7410(a)(2)(D)(i)(I). There is no dispute
that Delaware could file a section 126(b) petition to address an
upwind source’s contribution to an in-state non-attaining
receptor. But, according to the EPA, Delaware has no recourse
if its regulatory burden is attributable to a monitor in
Pennsylvania, even though Delaware remains bound by the
corresponding nonattainment designation. And, because
Delaware cannot compel another state to file a section 126(b)
petition, it is stuck in regulatory limbo, affected by an upwind
source yet unable to avail itself of the intended remedy for
addressing upwind contributions to nonattainment.

     The EPA’s efforts to explain away this incongruity fall
flat. First, the EPA contends that “concerns about the impacts
of upwind pollution on out-of-state monitors in a shared
multistate nonattainment area . . . can be addressed under other
statutory processes.” Response to Delaware and Maryland, 83
Fed. Reg. at 50,460. But the Clean Air Act creates the section
126(b) petition process as an alternative to the use of
implementation plans. See supra at 8. Indeed, with respect to
the requirements of the Good Neighbor Provision and section
126, “the EPA has consistently acknowledged that Congress
created these provisions as two independent statutory processes
to address the problem of interstate pollution transport.”
Response to Delaware and Maryland, 83 Fed. Reg. at 50,452.
The EPA cannot ignore section 126(b)’s standalone remedy
                                27
simply because it has other regulatory options at its disposal.
Next, the EPA asserts that Delaware’s obligations as a member
of a multistate nonattainment area should have no bearing on
the EPA’s interpretation of section 126(b).             Granted,
“[p]ortions of Delaware were included in the Philadelphia
nonattainment area because . . . those portions were themselves
contributing to the air quality problems in Pennsylvania.”
 
Id.
at 50,460. 
But, contrary to the EPA’s characterization,
Delaware is not trying to “relieve [itself] . . . of the specific
planning obligations associated with its inclusion in an area
designated nonattainment.”
 Id. Rather, it 
asks merely that
upwind sources contributing to air quality problems in the
multistate nonattainment area shoulder a comparable
regulatory burden, as the section 126(b) petition process
contemplates.

     In sum, states in a multistate nonattainment area share not
only a nonattainment designation but also the concomitant
responsibility to limit their own emissions. To equalize the
burdens between upwind and downwind states, the Clean Air
Act authorizes a state to petition the EPA for a finding that
upwind emissions significantly contribute to that state’s
nonattainment of the ozone NAAQS. But, under the EPA’s
interpretation, a state cannot file a section 126(b) petition if its
nonattainment status is caused by a receptor outside its political
boundaries, even as the state remains burdened by the
corresponding regulatory obligations. It is arbitrary for the
EPA to subject states like Delaware to this burden while
denying access to the intended remedy. Cf. Catawba 
Cty., 571
F.3d at 39 
(“[S]tatutory interpretation that is arbitrary and
capricious is unreasonable under Chevron step two.” (citing
Northpoint Tech., Ltd. v. FCC, 
412 F.3d 145
, 151 (D.C. Cir.
2005))). We therefore conclude that the EPA’s interpretation
of section 126(b) is unreasonable, at least if the petition
involves “monitors . . . located in a multistate nonattainment
                              28
area that includes the petitioning state.” Response to Delaware
and Maryland, 83 Fed. Reg. 50,460. Accordingly, the EPA
could not ignore Delaware’s evidence of non-attaining
receptors in the Philadelphia-Wilmington-Atlantic City
nonattainment area.

C. Selection of Year to Measure Air Quality

     We next consider a question of timing: In evaluating a
section 126(b) petition at Step One, for what year must the EPA
assess nonattainment in the downwind state? The Petitioners
argue that the EPA must focus on current nonattainment or, at
a minimum, nonattainment at the next future attainment
deadline applicable to the petitioning downwind state. We
disagree with the Petitioners on the first point, but we agree
with them on the second.

    1. Current Nonattainment

     The Petitioners contend that current nonattainment is
enough to satisfy Step One. The EPA responds that section
126(b), like the Good Neighbor Provision, concerns only
nonattainment that will last into the future. Again, the EPA’s
interpretation of the statute is reasonable.

     Section 126(b) requires a finding on whether an upwind
source “emits or would emit” a pollutant in violation of the
Good Neighbor Provision. 42 U.S.C. § 7426(b). By cross-
referencing the Good Neighbor Provision, “Congress clearly
hinged the meaning of § 126 on that of” the Good Neighbor
Provision. Appalachian Power Co. v. EPA, 
249 F.3d 1032
,
1049–50 (D.C. Cir. 2001) (per curiam). As a result, “the
substantive inquiry for decision is the same” under both
provisions.
 Id. at 1047 
(quotation marks omitted).
                               29
     The Good Neighbor Provision requires SIPs to prohibit
any in-state source “from emitting any air pollutant in amounts
which will . . . contribute significantly to nonattainment in, or
interfere with maintenance by, any other State with respect to
any [NAAQS].” 42 U.S.C. § 7410(a)(2)(D)(i). In determining
the temporal scope of this rule, the key word is “will.” In North
Carolina v. EPA, 
531 F.3d 896 
(D.C. Cir.) (per curiam),
modified on reh’g in part, 
550 F.3d 1176 
(D.C. Cir. 2008), we
held that the EPA reasonably construed “will”—which denotes
the future tense—to limit the Good Neighbor Provision to
downwind air quality problems (of nonattainment or
maintenance) that are currently present and will continue into
the future. See id at 913–14.

     North Carolina resolves the question presented here.
Section 126(b) requires a finding on whether an emission
causes a violation of the Good Neighbor Provision. And an
upwind source that currently contributes to downwind air
quality problems, but that will not contribute to these problems
in the future, does not cause such a violation. Thus, in its Step
One analysis, the EPA permissibly excluded downwind areas
that are not currently attaining the NAAQS but that will reach
attainment by a relevant future date.

    The Petitioners press four counterarguments. First, they
contend that the EPA’s approach of requiring future
nonattainment is inconsistent with inclusion of the present
tense in section 126(b), which covers any major source that
“emits or would emit” a pollutant in violation of the Good
Neighbor Provision. But to violate the Good Neighbor
Provision, as the EPA has permissibly construed it, the source
must contribute to both current and future downwind
nonattainment. See
 id. at 914. 
And if no such violation will
materialize, then the fact of current nonattainment is irrelevant.
                               30
     The Petitioners suggest that the word “will” must take on
a different meaning as incorporated into section 126(b) than it
does in the Good Neighbor Provision itself, lest the present-
tense “emits” be entirely collapsed into the future conditional
“would emit.” But statutes are not chameleons, acquiring
different meanings when presented in different contexts. See
Clark v. Martinez, 
543 U.S. 371
, 382 (2005). If present-only
nonattainment does not trigger a violation of the Good
Neighbor provision, then it cannot trigger a violation of that
same provision as incorporated into section 126(b).

     In any event, the EPA’s interpretation does not turn
“emits” into surplusage. Sometimes, current emissions do
contribute to future nonattainment. For example, the EPA
evaluates attainment of ozone standards based on certain
concentrations averaged over three consecutive years. See
Response to Delaware and Maryland, 83 Fed. Reg. at 50,456.
Thus, to predict air quality for 2023, the EPA looks to actual or
predicted air quality between 2020 and 2022. So a source that
pollutes right now, in 2020, “emits” pollutants that “will
contribute” to pollution levels—and possibly nonattainment—
for 2023. That gives “emits” independent meaning.

     Second, the Petitioners argue that North Carolina itself
rejected the EPA’s current reading of section 126(b). After
holding that the EPA could permissibly construe the Good
Neighbor Provision as focused on future nonattainment, we
stated that this “does not mean that EPA may ignore present-
day violations for which there may be another remedy, such as
relief pursuant to section 
126.” 531 F.3d at 914
. But North
Carolina presented no question involving section 126, so our
statement about it was dictum. As a consequence, we had no
occasion to substantively address this interpretive question.
                               31
     Third, the Petitioner-Intervenors contend that the EPA’s
interpretation disregards the accelerated enforcement deadlines
in section 126. Under the Good Neighbor Provision,
implementation plans must ensure that upwind states do not
contribute significantly to downwind nonattainment at the next
future downwind deadline. See 
Wisconsin, 938 F.3d at 313
–
20. Those deadlines can be three to twenty years in the future.
See 42 U.S.C. § 7511(a)(1). In contrast, section 126 gives
upwind sources only three months, extendable to three years,
to eliminate significant contributions after the EPA finds a
violation of the Good Neighbor Provision.
 Id. § 7426(c). 
The
petitioner-intervenors argue that this shortened enforcement
horizon reflects heightened concern with current
nonattainment. They are right that section 126(b) provides for
fast remediation, but what is remediated must nonetheless be a
violation of the Good Neighbor Provision. And as we have
explained, its scope does not change depending on whether
enforcement is sought under section 126(b). See Appalachian
Power 
Co., 249 F.3d at 1047
.

     Fourth, the Citizen Petitioners invoke a past section 126(b)
finding that they say turns entirely on current nonattainment.
In 2011, the EPA granted New Jersey’s petition regarding
sulfur dioxide emissions from a nearby Pennsylvania power
plant. See Final Response to Petition from New Jersey
Regarding SO2 Emissions from the Portland Generating
Station, 76 Fed. Reg. 69,052, 69,053 (Nov. 7, 2011). Although
the EPA had promulgated a sulfur dioxide NAAQS in 2010, it
had not yet set future attainment deadlines. But the agency
nonetheless granted the petition because the power plant was
significantly contributing to current nonattainment. See
 id. at
69,053, 
69,058.

   The Citizen Petitioners are mistaken to suggest that the
EPA’s present position is inconsistent with agency precedent.
                             32
For several decades, the EPA has consistently interpreted the
Good Neighbor Provision to require future nonattainment. See
Close-Out Rule, 83 Fed. Reg. at 65,889 (discussing past agency
actions), vacated on other grounds, New York, 781 F. App’x at
4. The agency’s treatment of New Jersey’s petition fit that
pattern. Downwind sulfur dioxide pollution is usually caused
by one or a few nearby sources. See Data Requirements Rule
for the 2010 1-Hour Sulfur Dioxide (SO2) Primary National
Ambient Air Quality Standard (NAAQS), 80 Fed. Reg. 51,052,
51,057 (Aug. 21, 2015). In the case of the Pennsylvania plant,
actual emissions were causing pollution levels more than two
times greater than the applicable NAAQS, and otherwise
allowable emissions would have caused pollution levels more
than seven times greater. See Response to Petition from New
Jersey Regarding SO2 Emissions from the Portland Generating
Station, 76 Fed. Reg. 19,662, 19,668, 19,672 (Apr. 7, 2011).
Because a single source was causing New Jersey to exceed the
NAAQS by a wide margin, the EPA could reasonably have
concluded that, absent intervention, New Jersey’s current
nonattainment would persist into the future.

     No such extrapolation would be reliable for ozone. Ozone
precursors are transported over long distances, so downwind
ozone problems often are caused by numerous upwind sources.
See EME Homer 
City., 572 U.S. at 496
–97. And ozone
concentrations can be significantly affected by meteorological
variables. See Findings of Significant Contribution and
Rulemaking on Section 126 Petitions for Purposes of Reducing
Interstate Ozone Transportation, 64 Fed. Reg. 28,250, 28,292
(May 25, 1999). All of this makes predicting future levels for
ozone far more complex than doing so for sulfur dioxide. See
Michigan v. EPA, 
213 F.3d 663
, 674 (D.C. Cir. 2000) (per
curiam). The New Jersey finding thus does nothing to
undermine the EPA’s position in this case.
                               33
    2. Future Nonattainment

    The Petitioners next argue that, looking to the future, the
EPA was required to measure air quality in the year that
corresponds with the next applicable downwind attainment
deadline. We agree.

      The Good Neighbor Provision requires upwind states to
eliminate excess emissions “consistent with” Title I of the
Clean Air Act, which includes the deadlines for downwind
states to attain the ozone standards.           See 42 U.S.C.
§ 7410(a)(2)(D)(i). As we recently explained in Wisconsin, an
implementation plan violates the Good Neighbor Provision if
it fails to “eliminate upwind States’ significant contributions to
downwind pollution by the statutory deadline for downwind
States to meet the NAAQS for 
ozone.” 938 F.3d at 314
.
Because section 126(b) incorporates the Good Neighbor
Provision, the EPA must find a violation if an upwind source
will significantly contribute to downwind nonattainment at the
next downwind attainment deadline. Therefore, the agency
must evaluate downwind air quality at that deadline, not at
some later date.

     In this case, Delaware’s next attainment deadline under the
2015 ozone NAAQS is the deadline for marginal
nonattainment areas, which falls in 2021. See supra at 16 &
n.5. Yet when conducting its analysis of Delaware’s petitions
at Step One, the EPA chose to examine the State’s air quality
in 2023. This decision cannot be reconciled with the Good
Neighbor Provision as we construed it in Wisconsin.

     The EPA’s responses are unpersuasive. It argues that
marginal nonattainment areas often achieve the NAAQS
without further downwind reductions, so it would be
unreasonable to impose reductions on upwind sources based on
the next marginal attainment deadline. Nonetheless, Delaware
                              34
must achieve attainment “as expeditiously as practicable but
not later than” 2021, 42 U.S.C. § 7511(a)(1), so upwind
sources violate the Good Neighbor Provision if they will
significantly contribute to Delaware’s failure to meet that
deadline. See 
Wisconsin, 938 F.3d at 314
. The EPA adds that
if Delaware’s marginal area fails to reach attainment by 2021,
it will be automatically bumped up to a moderate
nonattainment status and then subjected to a 2024 deadline.
See 42 U.S.C. § 7511(b)(2). But that does not make
Delaware’s obligation to attain the NAAQS by 2021 any less
binding. And an upgrade from a marginal to a moderate
nonattainment area carries significant consequences, such as a
requirement to provide for annual emissions reductions in SIPs.
See
 id. § 7511a(b). 
So long as upwind sources significantly
contribute to Delaware’s nonattainment at its 2021 attainment
deadline, they violate the Good Neighbor Provision.

D. Whether Delaware Carried Its Burden

     In sum, here is the basic legal framework for analysis of
Delaware’s section 126(b) petitions at Step One. First,
Delaware bore the burden to prove a violation of the Good
Neighbor Provision. Second, the EPA was required to consider
data from out-of-state receptors in the Philadelphia-
Wilmington-Atlantic City Attainment Area, which includes
parts of Delaware. Third, the EPA was not required to consider
downwind areas reflecting only current nonattainment, but it
was required to consider areas reflecting nonattainment both
currently and at the next attainment deadline. Applying these
rules, we conclude that the EPA’s decision to reject Delaware’s
petitions at Step One was arbitrary with respect to petitions
under the 2008 ozone standards, but not with respect to
petitions under the 2015 ozone standards.
                               35
    1. 2008 NAAQS

     In its petitions under the 2008 NAAQS, Delaware failed to
identify any receptor connected to the State showing current or
future nonattainment.        See Response to Delaware and
Maryland, 83 Fed. Reg. at 50,456. The petitions simply noted
the number of days that air pollution levels in the State had
exceeded the NAAQS. See Del. Conemaugh Pet. at 3 (J.A.
160); Del. Homer City Pet. at 3(J.A. 193); Del. Brunner Island
Pet. at 3 (J.A. 221); Del. Harrison Pet. at 3 (J.A. 248). But the
EPA measures nonattainment based on the fourth-highest daily
maximum ozone concentration, averaged across three
consecutive years. See Response to Delaware and Maryland,
83 Fed. Reg. at 50,456. Delaware’s petitions failed to provide
that information and thus failed to carry the State’s burden of
proof at Step One.

     Delaware provided additional information in its comments
responding to the EPA’s proposed denial of its petitions. 9
Specifically, it highlighted data from a receptor in Bristol,
Pennsylvania that was part of the Philadelphia-Wilmington-
Atlantic City attainment area. The EPA declined to consider
this data because it came from an out-of-state receptor. See
Response to Delaware and Maryland, 83 Fed. Reg. at 50,456.
As explained above, that decision was arbitrary and capricious.
The EPA makes no suggestion that, had it considered data from
the Bristol receptor, it still might have concluded that Delaware
failed to carry its burden at Step One. We therefore hold that
Delaware carried its burden.




    9
        The EPA has not argued that Delaware forfeited reliance on
this information by failing to offer it in its petitions.
                               36
    2. 2015 NAAQS

     In its petitions under the 2015 NAAQS, Delaware also
failed to provide data showing any current or future
nonattainment. In its comments on the EPA’s proposed denial,
Delaware showed current nonattainment at both the Bristol
monitor and an in-state monitor in New Castle. Del. Cmts. at
4, 10 (J.A. 345, 351). But Delaware offered no data showing
nonattainment at either monitor in 2021, the deadline for both
areas under the 2015 NAAQS.
 Id. at 4, 
10 (J.A. 345, 351).
Because Delaware failed to show that upwind sources will
significantly contribute to nonattainment at its next future
deadline, it failed to carry its Step One burden. See 
Wisconsin,
938 F.3d at 314
.

     After rejecting Delaware’s petitions on this ground, the
EPA went on to conduct its own independent analysis of future
attainment, which erroneously considered pollution levels in
2023 rather than 2021. But because the EPA independently
rested its decision on Delaware’s failure to carry its burden of
proof, the agency’s error on this point was harmless.

         III. Denial of All Petitions at Step Three

     Having concluded our review at Step One of the EPA’s
Good Neighbor framework, we proceed to Step Three: the
identification of cost-effective reductions at the named sources.
On this front, the parties have separate disputes with respect to
the sources that have catalytic controls; the sources that have
non-catalytic controls; and the one facility, Brunner Island,
with neither. We take each in turn.

A. Catalytic Controls

    Delaware’s three petitions not related to Brunner Island
involve sources that, the State claims, are failing to optimize
                               37
their catalytic controls. Similarly, thirty-two out of the thirty-
six sources in Maryland’s petition have catalytic controls that
the State says they are failing to optimize. The EPA concluded
in the Update Rule that such optimization is a cost-effective
strategy for reducing NOx emissions. This, Petitioners say, is
more or less the ball game. If optimization is the measure of
Good Neighbor compliance, and if the named sources are
failing to optimize, then it necessarily follows that those
sources are not currently in compliance with the Good
Neighbor Provision.

     The EPA offered two answers in its denial. First, it said,
the latest data showed that “the control optimization and the
emission reductions anticipated from the [Update Rule] are
being realized from the 34 units with [catalytic controls].”
Response to Delaware and Maryland, 83 Fed. Reg. at 50,465
(emphasis added). Second, it explained, “even in the event of
any single-unit variation in performance, the overall reductions
[attributable to optimization] are occurring within the same
airshed due to the fact that state budgets and assurance levels
were set to ensure those reduction levels statewide and
regionwide” through the Update Rule’s trading program.
 
Id. at
50,466. 
In other words, the logic of a cap-and-trade program
is that not all sources will reduce their individual emissions to
the same extent. We uphold the EPA’s first answer as
reasonable and do not address the second.

     The parties disagree about how to tell if any given source
is optimizing its controls. (No one has knocked on the plants’
doors to see for themselves.) How high do a source’s emissions
need to be before we can conclude that it is failing to optimize?
For purposes of the Update Rule, the EPA concluded that
optimizing catalytic controls would result in an average NOx
emissions rate of 0.10 lb/mmBtu. Update Rule, 81 Fed. Reg.
at 74,543. That figure was based on the third-best average
                                 38
performance of the covered sources between 2009 and 2015, a
choice we upheld in Wisconsin over the objection that the EPA
should have chosen an even lower 
figure. 938 F.3d at 320
–21.

     Now suppose a source is found emitting above the EPA’s
estimated average—at 0.11 lb/mmBtu, for instance—after the
Update Rule. Is the source failing to optimize? Petitioners
seem to think so. See, e.g., Md Br. 29; Citizen Pet’rs Br. 16–
17; Pet’r-Intervenors Br. 46–47. But the EPA explains why
that may not be so:

     The optimized rate for any particular unit depends on
     the unit-specific characteristics, such as boiler
     configuration, burner type and configuration, fuel
     type, capacity factor, and control characteristics such
     as the age, type, and number of layers of catalyst and
     reagent concentration and type.

Response to Delaware and Maryland, 83 Fed. Reg. at 50,466
n.62. As a result, the bare fact that a source emits above 0.10
lb/mmBtu is equivocal. It could be evidence that the source is
not optimizing, or it could be evidence that that particular
source’s optimized rate is higher than average.

    In that light, the EPA approached the question from the
opposite direction: What sorts of rates are likely to result if a
source does not consistently operate its catalytic controls? The
agency pegged that figure at 0.20 lb/mmBtu, a rate that sources
can usually hit by operating only their combustion controls.
See id.10 And as the agency observed, virtually all of the

     10
         Of course, any threshold the agency chose would present a
trade-off. Short of determining each source’s true optimized rate, an
undertaking in some tension with the general principle that section
126 allocates the initial evidentiary burden to Petitioners, 
see supra
Part II.A, the EPA could not avoid either under- or over-inclusion.
                                39
sources equipped with catalytic controls nationwide beat that
mark in 2017, after the Update Rule went into effect,
“including every unit with [catalytic controls] named in
Delaware’s and Maryland’s petitions.”
 
Id. at 50,466. 
The
remaining few either achieved as much based on preliminary
2018 data, are not located in the states targeted by the petitions,
or have since retired. See
 id.

     Maryland and 
Citizen Petitioners do not respond directly
to the EPA’s explanation for applying a 0.20 lb/mmBtu rule of
thumb, as opposed to the 0.10 lb/mmBtu average it articulated
in the Update Rule. But they mount a further argument that
these particular sources are not achieving their individual
optimized rates because they have performed more efficiently
in the past. In particular, the Petitioners calculated the “highest
30-day rolling average rate of emissions experienced by [each]
source during its best ozone season,” Md. Br. 23, and argue that
the EPA should have required sources to match that
performance. The trouble with this argument is that it mirrors
one we rejected in Wisconsin. There, in choosing to look to
each source’s third-best ozone season for purposes of the
Update Rule, the EPA explained that sources’ very best rates
are not, in the main, consistently achievable. See 
Wisconsin,
938 F.3d at 320
–21; Update Rule, 81 Fed. Reg. at 74,544.
Having found EPA’s reasoning on that question reasonable
once, we do so again here.

     Delaware’s objection is somewhat different. The State
acknowledges EPA’s use of the 0.20 lb/mmBtu threshold, but
points out that some of the named sources emitted above that
mark on individual days of the ozone season. See Del. Br. 30–
31. Since the EPA bases attainment on daily concentrations,
not seasonal emissions, Delaware argues that the EPA should
likewise control daily emissions to avoid the possibility that
sources will idle their controls on days with high electricity
                                40
demand. We did not have the occasion to consider the Update
Rule’s reliance on a seasonal cap in Wisconsin, since no party
challenged that decision there. But as the EPA explained here,
there appears to be “very little difference” between “NOx rates
for EGUs for hours with high energy demand” and “seasonal
average NOx rates.” Response to Delaware and Maryland, 83
Fed. Reg. at 50,466. In other words, Delaware’s concern
makes sense but has not been observed in practice. The EPA
also noted that there may be valid operational reasons not to
operate catalytic controls on particular days, “e.g., to avoid
damaging or plugging of the [control] or taking a forced outage
where a breakdown leaves the unit unavailable to produce
power.”
 Id. at 50,466-67. 
As a result, that a source ends up
emitting above 0.20 lb/mmBtu on a particular day is not
necessarily evidence of a failure to optimize. The EPA’s
explanation was reasonable. 11

B. Non-Catalytic Controls

     We now turn to the next control strategy at issue.
Maryland contends that the four electric generating units in its
petition that do not have catalytic controls should be required
to operate their non-catalytic controls. Maryland argues that
the EPA cannot rely on the Update Rule’s conclusion that such
controls are not cost-effective in light of our decision in
Wisconsin. We agree, and because we do not read the denial

    11
         Maryland argues, relatedly, that the Clean Air Act itself
requires daily limits—that a seasonal cap is not a valid “emission
limitation” within the meaning of the statute because it does not
“limit[] the quantity, rate, or concentration of emissions of air
pollutants on a continuous basis.” 42 U.S.C. § 7602(k). But
Maryland did not present this interpretive claim in its comments; it
said only that daily limits would be a good idea. See Md. Reply Br.
6 (citing Md. Cmts. at 6, 11, 39, 45 (J.A. 290, 295, 323, 329)). The
argument is therefore forfeited. See 42 U.S.C. § 7607(d)(7)(B).
                               41
to have offered an adequate argument in the alternative, we
remand it to the agency with respect to this issue.

     The EPA explained concisely its conclusion that operating
non-catalytic controls at these sources would not be cost-
effective. The Update Rule, the agency said, had already
concluded as much on a regional level. See Response to
Delaware and Maryland, 83 Fed. Reg. at 50,469–70. And there
was no reason to think these sources were any different, it
continued, because the units identified “are relatively small in
size and have low emission levels, indicating that the units have
a relatively limited ability to substantially reduce NOx
emissions.”
 

     We cannot endorse this explanation after Wisconsin.
There, we concluded that the Update Rule was impermissibly
“partial.” 
Wisconsin, 938 F.3d at 318
. As relevant here, in
choosing a cost-effectiveness threshold, the Update Rule did
not consider control strategies that could not have been
implemented in time for the 2017 ozone season.
 
Id. at 313. 
As
the EPA concedes, the agency’s judgment about the cost-
effectiveness of non-catalytic controls may change when it
conducts a new comparative analysis in response to the
Wisconsin remand. See Recording of Oral Arg. at 35:06;
accord Close-Out Rule, 83 Fed. Reg. at 65,898
(acknowledging that the EPA could have revisited the Update
Rule’s conclusion about non-catalytic controls in the Close-
Out Rule if it had found a continuing air quality problem). In
that light, the EPA cannot rely mechanically on the Update
Rule for the proposition that non-catalytic controls are not cost-
effective. And the denial does not seriously suggest that its
brief discussion of these particular units amounted to a
standalone cost-effectiveness analysis. See EPA Br. 78
(suggesting only that the EPA “reviewed the specific emissions
                                42
levels of the named sources to determine if anything had
changed” since the Update Rule was promulgated).

     We recognize that Wisconsin does not imply that non-
catalytic controls are cost-effective—or even that the EPA, on
remand, will choose a different cost threshold than the one it
originally did.12 Indeed, Counsel suggested at argument that
the EPA’s judgment about non-catalytic controls “probably”
would not change. See Recording of Oral Arg. at 52:29. But
under the familiar rule of SEC v. Chenery Corp., 
318 U.S. 80
(1943), we must review an agency’s action on the basis of
reasons it actually gave, not ones it hypothetically could. And
while there are gestures in the denial at an alternative argument
that Petitioners bear the burden to establish that a control is
cost-effective at Step Three (and failed to do so here), the EPA
ultimately based its decision on its own conclusion that non-
catalytic controls are not cost-effective. In that light, we need
not decide what burden a petitioner may have to show cost-
effectiveness. 13

C. Brunner Island

     Finally, we address Delaware’s Brunner Island petition.
Brunner Island, which has installed neither catalytic nor non-
catalytic controls, added natural gas capacity to augment—and
eventually replace—its coal-firing generation units. Delaware
argues that voluntary conversion is not enough and that the
EPA must affirmatively prohibit the facility from burning coal.

    12
        Cf. 
Michigan, 213 F.3d at 680 
(noting that EPA’s “selection
of the cut-off point [is] essentially unbounded,” given the sense in
which we have permitted the agency to consider cost-effectiveness).
    13
        Given the possible alternative rationales identified by the
EPA, we do not vacate the denial. See Allied-Signal, Inc., v. U.S.
Nuclear Regulatory Comm’n, 
988 F.2d 146
, 150-51 (D.C Cir. 1993).
                               43
The EPA declined to impose the requested limitation,
concluding that Brunner Island had not violated its Good
Neighbor obligations. The EPA’s determination is reasonable.

     Brunner Island significantly reduced its pollution footprint
by operating primarily on natural gas during the 2017 ozone
season. Compared to the 2016 season, NOx emissions fell from
3,765 tons to 877 tons, as the emissions rate declined from
0.370 lb/mmBtu to 0.090 lb/mmBtu. See Response to
Delaware and Maryland, 83 Fed. Reg. at 50,470–71. In other
words, consistent with Delaware’s proposed control strategy,
Brunner Island utilized its natural gas capabilities to achieve a
low NOx emissions rate during the 2017 ozone season.
Because Delaware did not identify any “additional feasible and
cost-effective NOx emissions reductions,” it could not
“demonstrate that, at this current level of emissions, Brunner
Island emits in violation of” the Good Neighbor Provision.
 Id.
The EPA 
next explained that favorable natural gas prices,
coupled with the incentive for Brunner Island to sell unused
emissions allowances, supported its “belie[f that] Brunner
Island will continue to primarily use natural gas as fuel during
future ozone seasons for economic reasons.”
 

.
Thus, Delaware’s petition also failed to show that Brunner
Island “would emit” in violation of the Good Neighbor
Provision.

    According to Delaware, the installation of a NOx control
technology does not, by itself, prevent a future violation of the
Good Neighbor Provision. Rather, a corresponding regulatory
requirement is necessary to ensure optimal operation and, on
this front, “passive market forces are an unacceptable
emissions limitation under the [Clean Air] Act.” Del. Br. 34.
Delaware asserts that the EPA, by crediting Brunner Island’s
voluntary choice to burn natural gas, contravened section
126(c)’s instruction that a violating source may continue
                                44
operating only if it “complies with such emission limitations
. . . as may be provided by the Administrator.” 42 U.S.C.
§ 7426(c) (emphasis added). As Delaware sees it, the EPA
must affirmatively ensure that Brunner Island maintains its
NOx reductions in the future. But section 126(c) contemplates
EPA-mandated limitations only for “a source referred to in
paragraph (2),”
 id., that is, 
“any major existing source . . . after
such finding has been made with respect to it,”
 id. § 7426(c)(2).
And because the requisite finding is one “made under
subsection (b),”
 id. § 7426(c)(1), 
the imposition of limitations
under section 126(c) is therefore predicated on finding, under
section 126(b), that a source “emits or would emit” in violation
of the Good Neighbor Provision,
 id. § 7426(b). 
Put differently,
the remedies to cure a Good Neighbor violation are distinct
from whether a violation has occurred in the first place. See
Response to Delaware and Maryland, 83 Fed. Reg. at 50,472
(“[T]he EPA only implements federally enforceable limits
under step four of the four-step framework for sources that the
EPA determines have emissions that significantly contribute to
nonattainment or interfere with maintenance of the ozone
NAAQS downwind under steps one, two, and three.”). The
EPA determined that Brunner Island does not, and would not,
emit in violation of the Good Neighbor Provision.
Accordingly, it was not required to issue federally enforceable
limitations under the Clean Air Act.

     Petitioner-Intervenors dispute the EPA’s Good Neighbor
determination, alleging that Brunner Island could, at any time,
revert to burning coal which, considering its 2016 NOx
emissions rate of 0.370 lb/mmBtu, proves it “would emit” in
violation of the Good Neighbor Provision. See Pet’r-
Intervenors Br. 48. The EPA concluded such a reversion was
unlikely, considering Brunner Island’s strategic decision to
invest in natural gas, the economic incentive to burn natural gas
and sell unused emission allowances through the NOx trading
                              45
program, and price projections indicating that natural gas will
remain a less expensive fuel source. See Response to Delaware
and Maryland, 83 Fed. Reg. at 50,471. The EPA also noted
that, pursuant to a settlement agreement, Brunner Island’s
owner “agree[d] to operate only on natural gas during the ozone
season . . . starting on January 1, 2023, . . . and cease coal
operations after December 31, 2028.”
 Id. at 50,471 
n.79.
Petitioner-Intervenors offer no evidence contradicting the
EPA’s record determination beyond unsupported conjecture
that Brunner Island could decide to burn coal.

     Moreover, Delaware and Petitioner-Intervenors provide
no reason that the EPA could not consider all known conditions
affecting Brunner Island’s anticipated emissions. “Our review
under the ‘arbitrary and capricious’ standard is narrow and
does not permit us to substitute our policy judgment for that of
the Agency.” Bluewater Network v. EPA, 
370 F.3d 1
, 11 (D.C.
Cir. 2004). “Thus, when an agency’s decision is primarily
predictive, our role is limited; we require only that the agency
acknowledge factual uncertainties and identify the
considerations it found persuasive.” Rural Cellular Ass’n v.
FCC, 
588 F.3d 1095
, 1105 (D.C. Cir. 2009). Here, the EPA
discussed the economic incentives built into the cap-and-trade
program, outlined Brunner Island’s business strategy and
highlighted natural gas price projections prepared by
independent analysts.        See Response to Delaware and
Maryland, 83 Fed. Reg. at 50,471. And it acknowledged the
fallibility of its predictions, conceding that “Brunner Island’s
operations [could] change such that the facility is operating
primarily on coal during future ozone seasons.”
 Id. at 50,472.
In that case, if “future emission levels increase,” Delaware can
“submit[] another petition regarding Brunner Island’s
impacts.”
 Id. Accordingly, we 
find that the EPA reasonably
explained its predictive judgment.
                               46
       IV. Nonconsideration of Maryland’s Petition
                 Under 2015 NAAQS

     Finally, Maryland argues that the EPA’s refusal to
evaluate its petition under the 2015 ozone NAAQS was
arbitrary and capricious. We disagree.

     To preserve an issue for our review, a party generally must
raise the issue before the agency. The Clean Air Act
specifically provides that issues must be raised during the
period for public comment in order to be reviewable here. See
42 U.S.C. § 7607(d)(7)(B). Likewise, general administrative-
law principles require timely preservation of issues before the
agency. See United States v. L.A. Tucker Truck Lines, Inc., 
344
U.S. 33
, 37 (1952); Nuclear Energy Inst., Inc. v. EPA, 
373 F.3d
1251
, 1297 (D.C. Cir. 2004).

     In this case, Maryland’s section 126(b) petition did not ask
the EPA to make a finding that upwind sources were
significantly contributing to its nonattainment of the 2015
NAAQS. On the contrary, the petition sought a finding only
“with respect to the 2008 ozone [NAAQS].” Md. Cover Ltr. at
1 (J.A. 48). Moreover, Maryland repeatedly described its
emissions problem, its requested finding, and its proposed
remedy by reference to the 2008 NAAQS. See Md. Pt. at 1, 3,
4, 17 (J.A. 50, 52, 53, 66). Because Maryland did not ask for
a finding under the 2015 standards, the EPA permissibly
declined to make one. See Appalachian Power Co. v. EPA, 
251
F.3d 1026
, 1036 (D.C. Cir. 2001) (“An agency cannot be
faulted for failing to address such issues that were not raised by
petitioners.”).

    Maryland contends that its petition did request a finding
under the 2015 NAAQS, if not in so many words. Specifically,
Maryland noted that a timely remedy might allow some in-state
areas also to reach attainment under the 2015 standards. See
                               47
Md. Pet. at 9, 13, 14 (J.A. 58, 62, 63). But that does not amount
to requesting a finding under those standards. Instead, it
suggests only that a finding under the 2008 standards might
have produced further in-state benefits under other standards.

    Alternatively, Maryland responds that its petition did not
need to request a finding under the 2015 NAAQS. Maryland
reasons that, when it filed the petition in November 2016, the
EPA had not yet issued attainment designations for it under the
2015 NAAQS. But whether it would have been premature for
Maryland to request a finding under the 2015 standards at that
time is beside the point. In its November 2016 petition—the
only one at issue here—Maryland did not request a finding
under the 2015 standards. Now that the EPA has designated
areas in Maryland as failing to attain those standards, the State
remains free to file a separate petition requesting a finding
under them.

     Finally, Maryland notes that it asked for a finding under
the 2015 standards in its comments to the EPA’s proposed
denial of its petition. But that request came too late. Consistent
with Maryland’s petition, the EPA’s proposed action did not
address the 2015 standards and failed to give public notice that
it might do so. See Response to Delaware and Maryland, 83
Fed. Reg. at 50,463. With the proceeding so far along,
Maryland could not properly request an entirely new finding.
We recognize that Maryland was required to preserve
arguments in its response to the proposed denial, 42 U.S.C.
§ 7607(d)(7)(B), but that hardly entitled it to raise points
wholly outside the scope of the notice generated by its own
prior petition. Because Maryland’s request for a finding under
the 2015 NAAQS fell outside the scope of the pending
rulemaking—as triggered by Maryland’s own petition—the
EPA had no obligation to address it. See Am. Fuel &
Petrochemical 
Mfrs., 937 F.3d at 585
–86.
                             48
                       V. Conclusion

     For the foregoing reasons, we grant Maryland’s petition
for review in part and remand the non-catalytic controls issue
to the EPA. We otherwise deny the petitions for review.