United States Court of Appeals
         FOR THE DISTRICT OF COLUMBIA CIRCUIT



Argued October 3, 2018            Decided September 13, 2019

                         No. 16-1406

               STATE OF WISCONSIN, ET AL.,
                      PETITIONERS

                             v.

    ENVIRONMENTAL PROTECTION AGENCY AND ANDREW
       WHEELER, ADMINISTRATOR, UNITED STATES
         ENVIRONMENTAL PROTECTION AGENCY,
                   RESPONDENTS

           AMERICAN LUNG ASSOCIATION, ET AL.,
                     INTERVENORS



  Consolidated with 16-1428, 16-1429, 16-1432, 16-1436,
  16-1437, 16-1438, 16-1439, 16-1440, 16-1441, 16-1442,
      16-1443, 16-1444, 16-1445, 16-1448, 17-1066


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


    Neil Gormley argued the cause for petitioners
Conservation Groups. Valerie S. Edge, Deputy Attorney
General, Office of the Attorney General for the State of
Delaware, argued the cause for petitioner State of Delaware.
                              2
With them on the briefs were David Baron, Charles
McPhedran, Joshua R. Stebbins, and Zachary M. Fabish. Seth
L. Johnson entered an appearance.

     Misha Tseytlin, Solicitor General, Office of the Attorney
General for the State of Wisconsin, and Harvey M. Sheldon
argued the causes for State Petitioners, Cedar Falls Utilities,
and City of Ames, Iowa. With them on the briefs were Brad
D. Schimel, Attorney General, Luke N. Berg, Deputy Solicitor
General, Peter Michael, Attorney General, Office of the
Attorney General for the State of Wyoming, James Kaste,
Deputy Attorney General, Erik Petersen, Senior Assistant
Attorney General, Leslie Sue Ritts, Steve Marshall, Attorney
General, Office of the Attorney General for the State of
Alabama, Robert D. Tambling, Assistant Attorney General,
Leslie Rutledge, Attorney General, Office of the Attorney
General for the State of Arkansas, Nicholas J. Bronni, Deputy
Solicitor General, Michael DeWine, Attorney General, Office
of the Attorney General for the State of Ohio, Eric E. Murphy,
State Solicitor, Ken Paxton, Attorney General, Office of the
Attorney General for the State of Texas, Priscilla M. Hubenak,
and Craig J. Pritzlaff and Linda B. Secord, Assistant Attorneys
General. Andrew L. Brasher, Deputy Solicitor, Office of the
Attorney General for the State of Alabama, Michael J.
McGrady, Senior Assistant Attorney General, Office of the
Attorney General for the State of Wyoming, Lee P. Rudofsky,
Solicitor, Office of the Attorney General for the State of
Arkansas, and Ryan Walsh entered appearances.

    Norman W. Fichthorn, Aaron M. Streett, and C. Grady
Moore, III argued the causes for Industry Petitioners. With
them on the briefs were E. Carter Chandler Clements, Peter S.
Glaser, Margaret Claiborne Campbell, M. Buck Dixon, Scott
C. Oostdyk, E. Duncan Getchell, Jr., Michael H. Brady, Jane
E. Montgomery, J. Michael Showalter, Amy Antoniolli, P.
                              3
Stephen Gidiere, III, Julia B. Barber, David W. Mitchell,
Daniel J. Kelly, David M. Flannery, Kathy G. Beckett, Edward
L. Kropp, Megan H. Berge, Charles T. Wehland, Todd E.
Palmer, John A. Sheehan, Valerie L. Green, Ben H. Stone,
Terese T. Wyly, M. Brant Pettis, Louis E. Tosi, Cheri A.
Budzynski, and Michael A. Born. Alina Fortson and Jordan
Hemaidan entered appearances.

    Amy J. Dona and Chloe H. Kolman, Attorneys, U.S.
Department of Justice, argued the causes for respondents. With
them on the brief were Jonathan Brightbill, Deputy Assistant
Attorney General, and Stephanie L. Hogan, Attorney, U.S.
Environmental Protection Agency.

    E. Carter Chandler Clements argued the cause for Industry
Respondent-Intervenors. With her on the brief were Norman
W. Fichthorn, Peter S. Glaser, Margaret Claiborne Campbell,
M. Buck Dixon, Scott C. Oostdyk, E. Duncan Getchell, Jr.,
Michael H. Brady, Robert A. Manning, and Joseph A. Brown.

    Andrew G. Frank, Assistant Attorney General, argued the
cause for State Intervenors. With him on the brief were Eric T.
Schneiderman, Attorney General at the time the brief was filed,
Office of the Attorney General for the State of New York,
Barbara D. Underwood, Solicitor General, Steven C. Wu,
Deputy Solicitor General, Michael J. Myers, Senior Counsel,
Maura Healey, Attorney General, Office of the Attorney
General for the Commonwealth of Massachusetts, Jillian M.
Riley, Assistant Attorney General, Environmental Protection
Division, Peter F. Kilmartin, Attorney General, Rhode Island
Department of Attorney General, Gregory S. Schultz, Special
Assistant Attorney General, Brian E. Frosh, Attorney General,
Office of the Attorney General for the State of Maryland,
Michael F. Strande, Assistant Attorney General, Gordon J.
MacDonald, Attorney General, K. Allen Brooks, Assistant
                               4
Attorney General, New Hampshire Office of the Attorney
General, Thomas J. Donovan, Jr., Attorney General, Office of
the Attorney General for the State of Vermont, and Nicholas F.
Persampieri, Assistant Attorney General. Morgan A. Costello,
Assistant Attorney General, Office of the Attorney General for
the State of New York, entered an appearance.

    Charles McPhedran argued the cause for Public Health
and Environmental Intervenors. With him on the brief were
Sean H. Donahue, Susannah L. Weaver, Graham G. McCahan,
Vickie L. Patton, Ann Brewster Weeks, Neil Gormley, David
Baron, Howard Fox, Joshua R. Stebbins, and Zachary M.
Fabish.

     Hope M. Babcock was on the brief for amicus curiae
American Thoracic Society in support of respondent-
intervenors American Lung Association, Appalachian
Mountain Club, Environmental Defense Fund, and Sierra Club.

    Before: SRINIVASAN, MILLETT and WILKINS, Circuit
Judges.

     PER CURIAM: When upwind States pollute, downwind
States can suffer the consequences. Congress addressed that
problem in the Clean Air Act by enacting a “Good Neighbor
Provision.” The Provision requires upwind States to eliminate
their significant contributions to air quality problems in
downwind States.

      In 2016, the Environmental Protection Agency
implemented that requirement by promulgating a regulation
addressing the interstate transport of ozone, or smog. A
number of parties brought challenges to the Rule, some
contending that the Rule is too strict and others contending that
it is too lenient.
                              5

    We conclude that, in one respect, the Rule is inconsistent
with the Act: it allows upwind States to continue their
significant contributions to downwind air quality problems
beyond the statutory deadlines by which downwind States must
demonstrate their attainment of air quality standards. In all
other respects, though, we determine that EPA acted lawfully
and rationally.

                              I

     The Clean Air Act tasks EPA with setting national ambient
air quality standards, or NAAQS. See 42 U.S.C. § 7409(a).
Individual States must ensure that their ambient air quality
complies with the national standard. To that end, the Clean Air
Act requires States to adopt State implementation plans, or
SIPs, that provide for implementation, maintenance and
enforcement of the national standard. 

If a
State fails to submit a SIP, or if EPA disapproves it, EPA must
issue a federal implementation plan, or FIP, to correct any
deficiency. 

     State-level regulation of air quality faces a confounding
variable. Air pollution, once emitted, drifts with the wind.
Upwind pollutants affect air quality in downwind States via
various chemical processes. Ozone, for example, forms from
the interaction of nitrogen oxides (NOx) and volatile organic
compounds (VOCs) in the presence of sunlight. For downwind
States, upwind emissions of these ozone precursors can pose a
significant problem. According to a study referenced by EPA,
on average, over three-quarters of the ground-level ozone in
downwind States comes from upwind emissions. 81 Fed. Reg.
at 74,514.
                               6
     Congress included a Good Neighbor Provision in the
Clean Air Act to address the problem of upwind States’
pollution impairing downwind States’ air quality. The
Provision prohibits States from “emitting any air pollutant in
amounts” that will “contribute significantly to nonattainment”
or “interfere with maintenance” of air quality in other States.
42 U.S.C. § 7410(a)(2)(D)(i).

    EPA has addressed the Good Neighbor Provision’s
requirements in a series of rulemakings. In 2011, EPA
promulgated the Cross-State Air Pollution Rule (CSAPR),
which applied to States whose upwind pollution violated good
neighbor obligations under the 1997 ozone NAAQS and the
1997 and 2006 fine particulate matter NAAQS. See 76 Fed.
Reg. 48,208 (Aug. 8, 2011).

     In 2008, EPA reduced the ozone NAAQS from 80 parts
per billion (ppb) to 75 ppb. As a result, EPA promulgated the
rule at issue in this case: an update to the CSAPR for eastern
States that accounts for the stricter 2008 ozone NAAQS. See
Cross-State Air Pollution Rule Update for the 2008 Ozone
NAAQS, 81 Fed. Reg. 74,504 (Oct. 26, 2016) (Update Rule).
Under the Update Rule, EPA proceeded in four steps.

     At the first step, EPA identified downwind States expected
to have problems attaining or maintaining air quality in
compliance with the 2008 ozone NAAQS. To identify those
States, EPA had to estimate the future air quality in each State.

EPA devised a measure to turn 2011 ozone
measurements into 2017 projections.

    EPA started with 2011 modeled data from “receptors,”
devices in each State that measure air quality. EPA modeled
ozone concentrations in a three-by-three grid around each
receptor. EPA chose the ten days with the highest projected
                               7
ozone concentration, noted which of the nine 12-km2 grid cells
had the highest ozone concentration on that day, and averaged
the ten observations. See 

EPA then ran the
model for 2017, inputting 2011 environmental conditions (like
rainfall and fire emissions) but projected 2017 NOx emissions
rates. The percentage change from 2011 to 2017 was deemed
a receptor’s “relative response factor,” which measures the
sensitivity of an area to ozone formation. Multiplying a 2011
observation by the relevant response factor yielded a projection
for 2017 for the receptor.

     EPA projected 2017 ozone levels for each receptor by
applying the relative response factor to three “design values”
from a receptor across a five-year period. A “design value” is
a three-year historical average of monitored ozone data. The
selected design values represented 2009–2011, 2010–2012,
and 2011–2013. See 

Multiplying the three
design values by the applicable relative response factor resulted
in three different ozone projections for 2017.

     In light of the 2008 ozone NAAQS of 75 ppb, EPA
considered any projected value of up to 75.9 ppb to constitute
attainment.    

a receptor to be in
“nonattainment” status if the average of its three projected
design values and its most recent monitored design value
(2013-2015) exceeded the NAAQS. 

identified a
third category (beyond attainment and nonattainment): a
receptor would be deemed in “maintenance” status if the
highest of the three projected design values exceeded the
NAAQS but the other two values did not. 

second step, EPA identified those upwind States
whose pollution was linked to nonattainment or
nonmaintenance at downwind receptors. EPA quantified the
impact of each State’s pollution on downwind receptors using
                                8
a model that apportioned responsibility for ozone formation at
a given receptor to various categories of emitters. See 

EPA then multiplied a given State’s contribution
factor by the projected average ozone concentration at each
receptor (calculated in Step 1) to yield each State’s contribution
to ozone formation at each downwind receptor.

     Next, EPA screened out States whose contributions to
ozone formation in a downwind State comprised less than 1%
of the NAAQS (0.75 ppb) in that downwind State. See 

     In other words, EPA construed “contribute
significantly” and “interfere with” in the Good Neighbor
Provision to require an upwind State’s pollution to cause at
least 0.75 ppb of ozone formation in a downwind State. EPA’s
analysis concluded that twenty-three States and the District of
Columbia were linked to nonattainment or nonmaintenance in
at least one downwind State. Fourteen States’ contributions
fell below the 1% threshold and thus were screened out at this
step. See 

third step, EPA quantified the amount of emissions
reductions that the twenty-three upwind States and the District
of Columbia would be required to make under the Good
Neighbor Provision. To do so, EPA applied a multifactor test
that balanced costs against benefits.

     EPA first measured the emissions reductions that would
occur at various levels of cost controls. It began with a baseline
case (i.e., a cost-control level of $0). It then ran the model for
a cost-control level of $800 per ton—in other words, measuring
the emissions reductions that would occur if EPA required all
States to apply all possible emissions controls at its EGUs
(electric generating units, or power plants) up to the marginal
price of $800 per ton of NOx. See 

EPA then
ran the model for higher cost-control levels of $1,400/ton,
                                9
$3,400/ton, $5,000/ton, and $6,400/ton. For each of these cost-
control levels, EPA also estimated the air quality
improvements that would occur at each receptor in downwind
States. Comparing the cost-control levels against the resulting
emissions reductions and air quality improvements, EPA
concluded that the cost-control level of $1,400 per ton
represented the point at which upwind “NOx reduction
potential and corresponding downwind ozone air quality
improvements are maximized with respect to marginal cost” —
that is, the point at which EPA would get the biggest bang for
its buck. 

     As part of that analysis, EPA also performed an
“overcontrol” analysis to ensure that no upwind State would be
required to reduce its emissions more than called for by the
Good Neighbor Provision. Overcontrol would occur if either
(1) a State’s downwind receptors all reduced their ozone
concentrations below the NAAQS, or (2) a State’s
contributions to all downwind receptors was reduced below the
1% contribution threshold of 0.75 ppb. See EME Homer City
Gen., L.P. v. EPA, 

, 521 (2014) (“EME Homer
II”). EPA concluded that neither of those indicia of overcontrol
existed at the $800/ton and $1,400/ton cost-control levels. See
81 Fed. Reg. at 74,551–52.

     At the fourth step, EPA quantified State emissions
“budgets” by calculating the emissions amount that would
occur under $1,400/ton cost controls. Under the CSAPR,
States may emit more NOx than permitted by their budgets by
acquiring allowances from other States, and States may sell
allowances to other States if they reduce their emissions more
than required by their budgets. 

But to ensure
that a State does not entirely skirt its good neighbor obligations
by buying a large number of allowances, each State is limited
                               10
to emitting no more than 121% of the emissions budget (the
“assurance level”), irrespective of allowances. See 

promulgated the original CSAPR, various
parties brought a challenge to that rule. We initially vacated
the rule, see EME Homer City Generation, L.P. v. EPA, 

(D.C. Cir. 2012) (“EME Homer I”), but the Supreme
Court reversed our decision and upheld the rule in its entirety,
although it left open the possibility of as-applied challenges,
see EME Homer 

. On remand, we found that
emissions budgets for thirteen States were invalid due to
overcontrol. EME Homer City Generation, L.P. v. EPA, 

, 124 (D.C. Cir. 2015) (“EME Homer III”).

     In 2016, EPA promulgated the Update Rule at issue here.
A number of parties have petitioned for review of the Rule in
this court. See 42 U.S.C. § 7607(d)(9). One group of
petitioners, including environmental groups and the State of
Delaware (collectively, Environmental Petitioners), contends
that EPA was required to adopt a more stringent rule. Another
group of petitioners, including various States and industry
groups (collectively, State and Industry Petitioners), argues
that EPA was required to issue a more lenient rule.

                               II

     We begin by addressing Environmental Petitioners’
claims. Environmental Petitioners challenge the Update Rule
on the grounds that it: (i) fails to square with the statutory
attainment deadlines, (ii) makes impermissible modeling and
implementation choices, and (iii) incorrectly classifies
Delaware as an attaining downwind State.

    The Rule must be set aside if it is “arbitrary, capricious, an
abuse of discretion, or otherwise not in accordance with law”
                                 11
or “in excess of statutory jurisdiction, authority, or limitations.”
42 U.S.C. § 7607(d)(9). That standard is “essentially the same”
as the familiar standard of review contained in § 706(2) of the
Administrative Procedure Act. Ethyl Corp. v. EPA, 

, 1064 (D.C. Cir. 1995).

     Applying that standard, we agree with Environmental
Petitioners that the Rule is inconsistent with the Act’s
attainment deadlines. We reject Environmental Petitioners’
remaining claims.

                                 A

      We first consider Environmental Petitioners’ argument
that the Update Rule infringes the Good Neighbor Provision by
permitting upwind States to continue their significant
contributions to downwind air quality problems for too long—
i.e., past the statutory deadlines for nonattaining downwind
areas to meet the NAAQS for ozone. In fact, EPA set no
concrete deadline at all for upwind States to eliminate their
contributions to downwind States’ nonattainment.

     The Good Neighbor Provision requires States to submit
SIPs that “prohibit[], consistent with the provisions of this
subchapter [i.e., Title I of the Act], any source . . . from emitting
any air pollutant in amounts which will . . . contribute
significantly to nonattainment in . . . any other State with
respect to any” NAAQS. 42 U.S.C. § 7410(a)(2)(D)(i). If a
State fails to submit a SIP or submits one deemed inadequate,
such that EPA must then prepare a FIP, EPA must likewise
satisfy the Good Neighbor Provision in the FIP. See EME
Homer 

–14 & n.15.

    The question we face is one of timing: the Good Neighbor
Provision calls for upwind States to eliminate their significant
                               12
contributions to air pollution in downwind States, but by when
must upwind States do so? In particular, does the Provision
call for upwind States to eliminate their significant
contributions to downwind pollution by the deadlines for
downwind areas to comply with the relevant NAAQS—here,
the 2008 NAAQS for ozone?

     Those deadlines are prescribed by the Act. The Act first
tasks EPA with designating as “nonattainment” any area that
does not meet a NAAQS. See 42 U.S.C. § 7407(d)(1)(A). For
areas designated as nonattainment for ozone, the Act specifies
that each State must secure compliance “as expeditiously as
practicable but not later than” a date certain. 

(emphasis added). That date is July 20, 2018, for areas in
“moderate” nonattainment with respect to the 2008 NAAQS
(and was July 20, 2015 for areas in “marginal” nonattainment).
See Nat. Res. Def. Council v. EPA, 

, 465–66
(D.C. Cir. 2014); Implementation of the 2008 National
Ambient Air Quality Standards for Ozone: State
Implementation Plan Requirements, 80 Fed. Reg. 12,264,
12,268 (Mar. 6, 2015).

     The Update Rule does not require upwind States to
eliminate their significant contributions to downwind ozone
pollution by that date—or by any date, for that matter. EPA
acknowledges that, except for one State (Tennessee), it “is only
quantifying a subset of each State’s emission reduction
obligation pursuant to the good neighbor provision.” 81 Fed.
Reg. at 74,520 (emphasis added); see 

n.19. And
the Rule states that it represents only a “first, partial step to
addressing a given upwind State’s significant contribution to
downwind air quality impacts for the 2008 ozone NAAQS.”

That is in large part because the Update Rule
confines itself to addressing upwind contributions from EGUs
due to an ostensible lack of information about non-EGUs. EPA
                              13
“expects that a full resolution of upwind transport obligations
would require emission reductions from sectors besides
EGUs,” along with “further EGU reductions that are
achievable after 2017.” 

is that, while the Rule
calls for a certain level of reductions in upwind contributions
by the 2017 ozone year—“in time to assist downwind states to
meet the July 2018 attainment deadlines”—the Rule does not
purport to require upwind States to fully meet their good
neighbor obligations by that time. 

Update Rule, then, downwind States face a
dilemma. On one hand, they operate under a statutory
obligation to secure compliance with the ozone NAAQS by
July 20, 2018. But on the other hand, the Rule does not require
upwind States to eliminate their significant contributions to
downwind pollution by that deadline.            Environmental
Petitioners argue that the Rule is inconsistent with the Act in
failing to require upwind States to eliminate their significant
contributions in accordance with the deadline by which
downwind States must come into compliance with the
NAAQS. We agree.

                              1

     That conclusion follows from our decision in North
Carolina v. EPA, 

(D.C. Cir. 2008). There, we
considered essentially the same question we now face here:
whether EPA can allow upwind States to continue their
significant contributions to downwind pollution beyond the
statutory deadlines for downwind States to meet the NAAQS.
North Carolina involved the Clean Air Interstate Rule (CAIR),
a prior EPA effort to implement the Good Neighbor Provision
with regard to the then-applicable NAAQS for two pollutants,
fine particulate matter (PM2.5) and ozone. See 

Although the statutory deadline for nonattaining areas to
                               14
comply with those NAAQS was 2010, CAIR gave upwind
States until 2015 to eliminate their significant contributions to
downwind nonattainment. See 

     We held that CAIR’s “deadline of 2015 [was] unlawful.”

We explained that, under the terms of the Good
Neighbor Provision, upwind States must eliminate their
significant contributions “consistent with the provisions” of
Title I of the Act. See 

42 U.S.C. §
7410(a)(2)(D)(i). And the incorporated provisions of Title I,
we further explained, include ones setting the attainment
deadlines for downwind areas. See 42 U.S.C. § 7502(a)(2)(A)
(PM2.5); 

(ozone). But under CAIR, “downwind
nonattainment areas [were required to] attain NAAQS for
ozone and PM2.5” by 2010, “without the elimination” by then
“of upwind states’ significant contribution to downwind
nonattainment, forcing downwind areas to make greater
reductions than [the Good Neighbor Provision] requires.”
North 

. As a result, we concluded,
“EPA ignored its statutory mandate to promulgate CAIR
consistent with the provisions in Title I mandating compliance
deadlines for downwind states in 2010.” 

that is equally true here. Just as with CAIR, the
CSAPR Update Rule we consider in this case fails to eliminate
upwind States’ significant contributions to downwind pollution
by the statutory deadline for downwind States to meet the
NAAQS for ozone. That in turn “forc[es] downwind areas to
make greater reductions than [the Good Neighbor Provision]
requires.” 

at least imposed some deadline
for upwind States to fully satisfy their good neighbor
obligations, albeit a deadline we held was too late. Here, by
contrast, EPA established no deadline at all for upwind States
to eliminate their significant contributions. And while EPA
concluded that requiring upwind States to meet their good
                               15
neighbor obligations by the 2018 attainment deadline “simply
[was] not feasible in the existing timeframe,” 81 Fed. Reg.
74,523, in North Carolina, EPA had likewise sought
(unsuccessfully) to rely “on reasons of 

.

     EPA contends that North Carolina required it only to
“consider” the attainment deadline in some fashion when
establishing upwind States’ good neighbor responsibilities, not
to align the attainment deadline with the deadline for satisfying
good neighbor obligations. And EPA argues that the Update
Rule gives the requisite consideration to the July 2018
attainment deadline by requiring at least some level of good
neighbor reductions by that date.

     North Carolina, though, requires more than merely
“considering” attainment deadlines in that manner. In fact,
CAIR provided for a first phase of reductions in upwind
contributions to take place before the attainment deadlines. See

But that was not enough to satisfy the statute. The
problem was that the eventual elimination of significant
upwind contributions in the second phase of reductions would
occur only long after the attainment deadlines had passed.

     We explained that EPA needed to “harmonize” the “Phase
Two deadline for upwind contributors to eliminate their
significant contribution with the attainment deadlines for
downwind areas.” 

(emphasis added). Otherwise,
downwind areas would need to attain the NAAQS “without the
elimination of upwind states’ significant contribution.” 

here creates the same situation. (And we note it does
so with respect to both the 2018 and 2015 deadlines. Although
EPA contends that the claim as to the 2015 deadlines was
forfeited, we disagree. See Comment of Sierra Club et al., at
                               16
8–9, EPA Docket No. 2015-0500-0287 (Feb. 1, 2016), J.A.
1015–16.)

     EPA notes that, when we addressed the issue of the
appropriate remedy at the conclusion of our opinion in North
Carolina, we observed that EPA would need to “decide what
date, whether 2015 or earlier, is as expeditious as practicable
for states to eliminate their significant contributions to
downwind 

(emphasis added).
But our reference to 2015 did not suggest that EPA could delay
the deadline for upwind States to eliminate their significant
contributions until 5 years after the 2010 attainment deadline.
The entire object of our analysis was to reject the notion that
the Phase Two deadline of 2015 could be squared with the
Good Neighbor Provision. See 

(EPA operated “under
the assumption that 2015 was an appropriate deadline for CAIR
compliance. It is not.”). Rather, we presumably referred to
2015 because, as we had earlier specifically noted, EPA has
separate statutory authority to extend the deadline for attaining
the NAAQS for PM2.5—and thereby correspondingly also
extend the good neighbor deadline—for up to five years, or
until 2015. See 

(citing 42 U.S.C. § 7502(a)(2)(A)).

     In sum, under our decision in North Carolina, the Good
Neighbor Provision calls for elimination of upwind States’
significant contributions on par with the relevant downwind
attainment deadlines. The Update Rule fails to do so.

                               2

     North Carolina’s understanding of the Good Neighbor
Provision is confirmed by examining the Update Rule under
the framework set out in Chevron U.S.A. Inc. v. Nat. Res. Def.
Council, Inc., 

(1984). See EME Homer 

–14 (evaluating previous EPA implementation of
                                17
the Good Neighbor Provision under Chevron); see also Util.
Air Regulatory Grp. v. EPA, 

, 315 (2014) (“We
review EPA’s interpretations of the Clean Air Act using the
standard set forth in Chevron.”).

     “Under Chevron, 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.” Util. Air. Regulatory 

. The
question then “is whether in doing so the agency has acted
reasonably and thus has ‘stayed within the bounds of its
statutory authority.’” 

v. FCC, 

, 296 (2013)). Here, the Update Rule’s open-ended
compliance timeframe exceeds the bounds of EPA’s statutory
authority by allowing upwind States to continue their
significant contributions to downwind nonattainment well past
the deadline for downwind areas to comply with the NAAQS.

     The threshold question under Chevron ordinarily would be
whether the statute is ambiguous on that issue, such that the
agency then would have discretion to choose among reasonable
interpretations. E.g., 

is no need to resolve that
threshold issue in this case, because, regardless of ambiguity,
the Update Rule amounts to an unreasonable—and hence
impermissible—interpretation of the statute in any event. See
Michigan v. EPA, 

, 2707 (2015); Massachusetts
v. U.S. Dep’t of Transp., 

, 892 (D.C. Cir. 1996)
(declining to resolve how Chevron might apply because “the
agency’s determination here cannot be upheld with or without
deference”).

    The Good Neighbor Provision, as North Carolina
emphasized, requires upwind States to eliminate their
significant contributions to downwind pollution “consistent
with the provisions of this subchapter,” i.e., Title I of the Clean
                              18
Air Act. 42 U.S.C. § 7410(a)(2). One of the “provisions of
this subchapter” is § 7511(a)(1), which in turn requires
downwind areas in moderate nonattainment to attain the
NAAQS by July 20, 2018. See 

(a)(1); 81 Fed. Reg.
at 74,507. The statute cannot reasonably be understood to
enable upwind States to continue their significant contributions
outside of the statutory timeframe by which downwind areas
must achieve attainment, much less continue those
contributions with no deadline at all.

     We note that we do not conclude that the phrase
“consistent with” in the Good Neighbor Provision necessarily
effects an incorporation of the full contours of every provision
of Title I in pure, lockstep fashion. As we have observed
elsewhere in construing the same words in the context of the
same statute, the phrase “consistent with” other statutory
sections “calls for congruence or compatibility with those
sections, not lock-step correspondence.” Envtl. Def. Fund Inc.
v. EPA, 

, 460 (D.C. Cir. 1996); see Nuclear Energy
Institute, Inc. v. EPA, 

, 1270 (D.C. Cir. 2004)
(The “‘based upon and consistent with’ standard does not
require EPA to walk in lockstep”).

     Rather, it is the statutorily designed relationship between
the Good Neighbor Provision’s obligations for upwind States
and the statutory attainment deadlines for downwind areas that
generally calls for parallel timeframes. The Good Neighbor
Provision was enacted “to enable downwind States to keep
their levels of [air pollution] in check.” EME Homer 

–97. A “reasonable statutory interpretation” of the
Provision “must account for . . . the broader context of the
statute as a whole.” Util. Air Regulatory 

.
And the attainment deadlines, the Supreme Court has said, are
“the heart” of the Act. Train v. Nat. Res. Def. Council, 

, 66 (1975); see Sierra Club v. EPA, 

, 161
                              19
(D.C. Cir. 2002) (“the attainment deadlines are central to the
regulatory scheme”) (alteration and internal quotation marks
omitted). The Act’s central object is the “attain[ment] [of] air
quality of specified standards [within] a specified period of
time.” 

–65.

     EPA’s interpretation of the Good Neighbor Provision
subverts that scheme. Under the Update Rule, downwind
States face a crucial statutory obligation to secure attainment
of the ozone NAAQS by July 20, 2018 (or July 20, 2015), even
though upwind States face no symmetrical obligation to
eliminate their significant contributions to downwind
nonattainment by that deadline. The Rule thus puts downwind
States to the choice of flouting the attainment deadlines or
making greater reductions than the Good Neighbor Provision
requires. That choice is “incompatible with the substance of
Congress’ regulatory scheme.” Util. Air Regulatory 

(internal quotation marks omitted).

    That becomes particularly evident when one considers the
extent to which pollution from upwind States adversely affects
the ability of downwind States to attain the NAAQS.
According to a study EPA cited in the Update Rule’s preamble,
“on average 77 percent of each state’s ground-level ozone is
produced by precursor emissions from upwind states.” 81 Fed.
Reg. at 74,514.

    Consider, in this regard, the Rule’s 2017 projections for
one nonattainment area: Fairfield County, Connecticut. EPA
projects that, after the good neighbor reductions called for by
the Rule, a monitor in that area would observe an average
ozone concentration of 76.5 ppb, or 1.5 ppb more than the
NAAQS. Office of Air Quality Planning and Standards, U.S.
Environmental Protection Agency, Air Quality Modeling
Technical Support Document for the Final Cross State Air
                              20
Pollution Rule 14 (Aug. 2016). According to EPA, 53.82 ppb
of that 76.5 ppb would be caused by pollution from U.S. States
(including Connecticut itself). Appendix C, Contributions to
2017 8-Hour Ozone Design Values at Projected 2017
Nonattainment and Maintenance-Only Sites at C-4 (Aug.
2016). Yet Connecticut’s own emissions, according to EPA’s
projections, would account for only 3.89 ppb of that 53.82. 

The rest would come from upwind contributions, with
a significant share from one State alone (New York, which is
projected to contribute 17.22 ppb). 

does not get enough help from
upwind States under the Update Rule, if Connecticut wanted to
bring Fairfield County into attainment by the 2018 deadline, it
would need to reduce its own ozone precursor emissions by
almost 40% (1.5 ppb over 3.89). And missing that attainment
deadline carries serious consequences, triggering a host of
strict mandatory emissions controls and a bump-up from
“moderate” to “serious” nonattainment status. See South Coast
Air Quality Mgmt. Dist. v. EPA, 

, 887 (D.C. Cir.
2006); 42 U.S.C. § 7511(a). That has recently occurred for
eight nonattainment areas which failed to meet the NAAQS by
the July 2018 deadline, including Fairfield County. 84 Fed.
Reg. 44,239 (Aug. 23, 2019).

     The structure of the provision establishing the statutory
attainment deadlines for ozone reinforces the Update Rule’s
impermissibility. That provision calls for downwind States to
secure attainment “as expeditiously as practicable but not later
than” the deadline. 42 U.S.C. § 7511(a). The deadline, that is,
functions as the ultimate failsafe. By imposing a first-order
obligation to attain the NAAQS “as expeditiously as
practicable,” Congress “made clear that the States could not
procrastinate until the deadline approached. Rather, the
primary standards had to be met in less [time] if possible.”
                               21
Union Elec. Co. v. EPA, 

, 259–60 (1976). In light
of that mandate, EPA “does not dispute that it has an obligation
to implement [upwind States’] Good Neighbor emission
reductions as expeditiously as practicable to benefit affected
downwind areas.” EPA’s Br. 26.

     But if EPA must provide for upwind States to satisfy their
good neighbor obligations as “expeditiously as practicable,”
per the provision establishing the attainment deadlines, why is
it not also generally necessary for upwind States to satisfy their
good neighbor obligations by the ultimate attainment deadline
as a last resort, per the same provision? By structuring the
Update Rule to require upwind States to meet their good
neighbor requirements as expeditiously as practicable, but then
permitting States to continue their significant contributions
well past the attainment deadlines, EPA gives effect to the
“expeditiously as practicable” part of § 7511(a) without giving
any independent effect to the deadlines part of that provision.

    EPA argues that the § 7511(a) attainment deadlines need
not carry over to the Good Neighbor Provision because those
deadlines are subject to modification by other statutory
provisions. But those modification provisions do not render
the deadlines discretionary or otherwise rob them of legal
force. See Sierra 

(holding that
Ҥ 7511(a)(1), as written[,] sets a deadline without an
exception”). Rather, if a modified attainment deadline applies
to downwind States, EPA may be able, if justified, to make a
corresponding extension for an upwind State’s good neighbor
obligations. EPA makes no contention that any such
modification is applicable here.

     As a last argument, EPA contends that holding upwind
States to the downwind attainment deadlines would be
inconsistent with the timeframe for promulgating a FIP. Under
                               22
the Act, the first downwind attainment deadline can occur
within five years of the establishment of an ozone NAAQS.
See 42 U.S.C. § 7407(d)(1)(B)(i); § 7511(a)(1). But, EPA
notes, the statute contemplates that it might take more than six
years to promulgate a FIP. See 

(c)(1),
(k)(1)(B), (k)(2). Thus, EPA submits, upwind States cannot
have been expected to cease their significant contributions by
the downwind attainment deadline.

     There is no inconsistency. Under the statute, EPA need
not wait six years to issue a FIP. By shortening the deadline
for a SIP submission, see 

and by issuing a FIP
soon thereafter, see 

EPA could promulgate a
FIP well before the first downwind attainment deadline. See
EME Homer 

(“After EPA has disapproved
a SIP, the Agency can wait up to two years to issue a FIP . . . .
But EPA is not obliged to wait two years or postpone its action
even a single day.”).

     EPA’s argument, in this regard, also proves too much. By
EPA’s logic, EPA can wait until after the nonattainment
deadline to promulgate not only a FIP addressing unfulfilled
good neighbor obligations, but also a FIP addressing in-State
pollution control deficiencies. After all, the same provisions
EPA cites here also appear to allow EPA to take six years to
promulgate a FIP after finding that a State’s SIP will not bring
that same State into attainment. See 42 U.S.C. § 7410(k)(1)(A)
(FIPs may be promulgated when Administrator decides that a
SIP does not “compl[y] with the provisions of this
chapter”). This would suggest that no pollution control
requirement, let alone the Good Neighbor Provision, need be
complied with by the attainment deadline. Yet EPA does not
suggest that the timeframes for a FIP somehow render the
deadlines totally nonbinding. When EPA determines that a
State’s SIP is inadequate, EPA presumably must issue a FIP
                                23
that will bring that State into compliance before upcoming
attainment deadlines, even if the outer limit of the statutory
timeframe gives EPA more time to formulate the FIP. See
Sierra Club v. EPA, 

, 161 (D.C. Cir. 2002) (“the
attainment deadlines remain intact” even if procedural
deadlines are missed or changed). The same is true when a
State’s SIP fails to provide for the full elimination of the State’s
significant contributions to downwind nonattainment.

     For these reasons, we conclude that, by issuing a Rule that
does not call for upwind States to eliminate their substantial
contributions to downwind nonattainment in concert with the
attainment deadlines, EPA has strayed outside the bounds of its
statutory authority under the Good Neighbor Provision.

                                 3

     EPA invokes various justifications for allowing substantial
upwind contributions to continue beyond the downwind
attainment deadlines.       None of the agency’s asserted
justifications establishes cause to disregard the requirement
under the statute to align the deadline for satisfying good
neighbor obligations with the deadline for attaining the
NAAQS.

     First, EPA opted to require partial (rather than full)
satisfaction of upwind States’ good neighbor obligations due in
significant part to its decision to consider only upwind
emissions from EGUs. EPA decided against considering
emissions reductions from non-EGUs because, “[a]s compared
to EGUs, there is greater uncertainty in EPA’s current
assessment of non-EGU point-source NOx mitigation
potential.” 81 Fed. Reg. at 74,542; see 

                               24
     Scientific uncertainty, however, does not excuse EPA’s
failure to align the deadline for eliminating upwind States’
significant contributions with the deadline for downwind
attainment of the NAAQS.               “Questions involving the
environment are particularly prone to uncertainty,” but “the
statutes and common sense demand regulatory action to
prevent harm, even if the regulator is less than certain.” Ethyl
Corp. v. EPA, 

, 24–25 (D.C. Cir. 1976) (en banc).
As a result, “EPA [cannot] avoid its statutory obligation by
noting [scientific] uncertainty . . . and concluding that it would
therefore be better not to regulate at this time.” Massachusetts
v. EPA, 

, 534 (2007). It is only when “the
scientific uncertainty is so profound that it precludes EPA from
making a reasoned judgment” that it can excuse compliance
with a statutory mandate. 

invoke that exception,
EPA “must say so,” and it has not said so here. 

also concluded that “developing a rule that
would have covered additional sectors and emissions
reductions on longer compliance schedules would have
required more of the EPA’s resources over a longer rulemaking
schedule.” 81 Fed. Reg. at 74,522. But administrative
infeasibility, like scientific uncertainty, cannot justify the
Update Rule’s noncompliance with the statute.

     An agency cannot “shirk[] its duties by reason of mere
difficulty or inconvenience.” Am. Hosp. Ass’n v. Price, 

, 168 (D.C. Cir. 2017). When an agency faces a
statutory mandate, a decision to disregard it cannot be
grounded in mere infeasibility. Rather, the agency would need
to meet the “heavy burden to demonstrate the existence of an
impossibility.” Sierra Club v. EPA, 

, 462 (D.C.
Cir. 1983) (quoting Ala. Power Co. v. Costle, 

,
359 (D.C. Cir. 1979)).
                              25
    EPA has not attempted to meet that burden here. True,
EPA would need to devote “more of the EPA’s resources” in
order to quantify good neighbor obligations from non-EGU
sources. 81 Fed. Reg. at 74,522. And “greater uncertainty”
about reductions from non-EGUs might render EPA’s
calculations more inaccurate than it would prefer. 

But that does not amount to impossibility.

     EPA next contends that it should be permitted to address a
problem incrementally, one step at a time. EPA relies on two
of our decisions for support. First, in Grand Canyon Air Tour
Coal. v. FAA, 

(D.C. Cir. 1998), we affirmed an
FAA rule that only partly fulfilled the agency’s statutory
obligation to “restor[e] the natural quiet” to the Grand Canyon,

We were careful to note, however, that “Congress
had no specific timetable in mind.” 

The opposite is
true here—in fact, Congress has provided a literal timetable.
See 42 U.S.C. § 7511(a)(1). Second, in Las Vegas v. Lujan,

(D.C. Cir. 1989), we sustained the Interior
Secretary’s decision to list only one of two similar species of
tortoise as endangered, observing that “agencies have great
discretion to treat a problem partially,” 

But the
Endangered Species Act does not require the Secretary to list
all endangered species by a date certain. The Clean Air Act
requires upwind States to eliminate their significant
contributions to downwind ozone nonattainment by prescribed
deadlines.

     Finally, EPA cites delays occasioned by litigation. EPA
observes that its legal obligations under the Good Neighbor
Provision remained uncertain until the Supreme Court issued
its April 2014 decision in EME Homer II. And this court’s
ensuing decision in EME Homer III in July 2015 imposed
further obligations on EPA by invalidating budgets for certain
States. EPA released its Proposed Rule six months later, in
                                26
December 2015, and it released the Final Rule several months
thereafter, in October 2016. In this context, litigation delays
cannot justify EPA’s failure to bring the deadline for satisfying
good neighbor obligations into alignment with the 2018
attainment deadline. The timeframes do not amount to “the
existence of an impossibility.” Sierra 

.

     While EPA has not justified its failure to align the deadline
for upwind States to eliminate significant contributions with
the deadline for downwind areas to attain the NAAQS, the
agency retains some flexibility in administering the Good
Neighbor Provision. We acknowledge that the “realities of
interstate air pollution . . . are not so simple,” and EPA faces its
share of “thorny . . . problem[s]” in regulating it. EME Homer

–16. EPA, though, possesses a measure of
latitude in defining which upwind contribution “amounts”
count as “significant[]” and thus must be abated. See 

520 n.21. And the Supreme Court has indicated that EPA can
take into account, among other things, “the magnitude of
upwind States’ contributions and the cost associated with
eliminating them.” 

Additionally, in certain
circumstances, EPA can grant one-year extensions of the
nonattainment deadlines to downwind States. 42 U.S.C. §
7511(a)(5).         EPA grants those extensions fairly
commonly. E.g., 84 Fed. Reg. 44238, 44238 (Aug. 23, 2019);
81 Fed. Reg. 26,697, 26,697 (May 4, 2016). And finally, EPA
can always attempt to show “impossibility.” Sierra 

.

    It also bears reemphasizing that the Update Rule set no
deadline at all for upwind States to eliminate their significant
contributions (a result even more infirm than the five-year
extension struck down in North Carolina). We do not
foreclose the possibility that the statutory command we
construe here—that compliance with the Good Neighbor
                             27
Provision must be achieved in a manner “consistent with” Title
I—might reasonably be read, under particular circumstances
and upon a sufficient showing of necessity, to allow some
deviation between the upwind and downwind deadlines. Any
such deviation would need to be rooted in Title I’s framework,
cf. 42 U.S.C. § 7511(a) (allowing one-year extension of
attainment deadlines in particular circumstances), and of
course would still need to “provide a sufficient level of
protection to downwind States,” North 

. What EPA cannot do, in our view, is determine that
upwind States contribute to downwind nonattainment in a
manner the agency deems “significant,” but then still allow
those upwind contributions to persist out of step with the
deadline for downwind areas to come into attainment.

                              B

    Environmental Petitioners also challenge various of the
Update Rule’s modeling and implementation choices. We are
“at [our] most deferential” when reviewing an agency’s
predictions and scientific determinations. Balt. Gas & Elec.
Co. v. Nat. Res. Def. Council, Inc., 

, 103 (1983).
We find no basis to set aside the challenged determinations
here.

     First, Environmental Petitioners challenge EPA’s
assumption that turning on idled “Selective Catalytic
Reduction” (SCR) controls would reduce an EGU’s emissions
to 0.10 lbs/mmBtu. They contend that a lower rate would be
more accurate, and they argue that EPA failed to provide a
reasoned explanation for its choice. We conclude that EPA
adequately explained its choice. EPA acknowledged that
certain units could achieve a lower emissions rate with SCR,
but determined that the higher rate was “generally achievable”
and therefore more “appropriate” for EPA’s model, especially
                                28
because it calculated a rate on a fleet-wide basis. 81 Fed. Reg.
at 74,544. Additionally, EPA applied a unit’s historical rate
whenever it was lower. That explanation is “rational.” Motor
Vehicle Mfrs. Ass’n of U.S., Inc. v. State Farm Mut. Auto. Ins.
Co., 

, 56 (1983).

     Second, Environmental Petitioners challenge EPA’s
choice to limit its modeling to include only intrastate
generation shifting (i.e., the shifting of electricity generation to
cleaner power plants). Broader consideration of generation
shifting, they argue, would have yielded greater potential
emissions reductions. But EPA limited its modeling to
intrastate generation shifting because it thought “broader”
generation shifting would ignore near-term technological
feasibility. See 81 Fed. Reg. at 74,544–45. And EPA rejected
Environmental Petitioners’ preferred modeling approach
because it ignored the relevance of cost thresholds, rendering it
incompatible with an approach to modeling premised on
uniform cost-control thresholds. See Response to Comments
at 528, J.A. 572. Again, EPA’s choices were rational.

     Third, Environmental Petitioners challenge EPA’s
approach to converting allowances from previous allowance
trading programs. The Update Rule employs an allowance
trading program, which permits underpolluting EGUs to sell
unused allowances to overpolluting EGUs. Because this Rule
is more stringent than prior rules, it imposes a conversion
formula to mitigate the impact of plants losing out on
allowances banked under prior rules. The formula results in a
conversion ratio of approximately 3.5 old allowances per 1 new
allowance. See 81 Fed. Reg. at 74,557.

     Environmental Petitioners say that EPA’s approach will
create an “allowance glut” that will hinder the Rule’s salutary
effect on upwind emissions. In their view, EPA should have
                               29
declined to allow any conversion of old allowances. EPA’s
considered judgment, however, was that some conversion of
allowances was necessary to respect EGUs’ legitimate
“expectation that . . . banked allowances will have some value
in the future of th[e] program.” 

And if no
conversion were permitted, current EGUs would have the
incentive to use up their banked allowances all at once,
exacerbating downwind nonattainment problems. EPA has
demonstrated that its use of a 3.5-to-1 conversion ratio was
reasonable.

     Environmental Petitioners last raise a statutory challenge.
The Rule’s banked allowance program, they submit,
contradicts EPA’s statutory obligation to implement Good
Neighbor emissions reductions “as expeditiously as
practicable.” 42 U.S.C. § 7411(a)(1). That challenge fails
under Chevron. The Good Neighbor Provision does not speak
directly to these choices. It is unclear, for example, whether
the Provision requires a conversion ratio of 3.5 to 1, a different
ratio, or no conversion at all. EPA’s choices are reasonable and
merit deference. Our precedents read the Good Neighbor
Provision to grant EPA the authority to make precisely those
kinds of policy determinations. See, e.g., EME Homer 

.

                                C

     Delaware claims that it should have been designated a
nonattaining downwind State, triggering good neighbor
obligations from upwind States. Under EPA’s projections for
2017, no Delaware receptors were deemed problem receptors
because both average and maximum projected ozone
concentrations fell below 76.0 ppb. But under the Act, upwind
States’ SIPs—with corresponding Good Neighbor emissions
reductions—were initially due in 2011. See 42 U.S.C.
                              30
§ 7410(a)(1). Thus, Delaware argues, the Update Rule must
use 2011 data, not 2017 data, to designate receptors as problem
receptors. Otherwise, States upwind of nonattainment areas in
2011 can pollute without consequence, so long as those
downwind areas come into attainment by 2017.

     Delaware’s argument leans too heavily on the SIP
submission deadline. SIP submission deadlines, unlike
attainment deadlines, are “procedural” and therefore not
“central to the regulatory scheme.” Sierra 

. Nor can Delaware’s argument be reconciled with the text
of the Good Neighbor Provision, which prohibits upwind
States from emitting in amounts “which will” contribute to
downwind nonattainment.        42 U.S.C. § 7410(a)(2)(D)(i)
(emphasis added). Given the use of the future tense, it would
be anomalous for EPA to subject upwind States to good
neighbor obligations in 2017 by considering which downwind
States were once in nonattainment in 2011.

    Delaware also claims that EPA impermissibly relied on
only one year of modeling data to designate downwind
problem receptors. That argument mischaracterizes EPA’s
methodology. The Update Rule relies on a weighted average
of three design values (from 2009–2011, 2010–2012, and
2011–2013) in order to compute projected concentrations at
each downwind receptor for 2017. See 81 Fed. Reg. at 74,532.
Delaware’s challenge thus fails.

                              III

    In contrast to Environmental Petitioners and Delaware,
State and Industry Petitioners argue that the Update Rule—far
from doing too little to curb interstate air pollution—
unlawfully overregulates upwind emissions sources. They
                               31
present a smörgåsbord of arguments, which we address below.

                                A

    State Petitioners contend that EPA failed to rationally
analyze whether the environmental benefits of the Rule’s FIPs
justified their costs, and that the agency thus contravened the
Good Neighbor Provision and principles of administrative law.
The Good Neighbor Provision, State Petitioners note,
authorizes EPA to regulate emissions that “contribute
significantly to nonattainment.” State Pet’rs’ Br. 14 (quoting
42 U.S.C. § 7410(a)(2)(D)(i)) (emphasis in original). And in
Michigan v. EPA, 

(D.C. Cir. 2000), State
Petitioners add, we held that EPA may consider costs in
determining what contributions are “significant,” 

(quoting 

), and endorsed the
principle that (in the absence of a clear legislative statement to
the contrary) a regulation’s benefits must be “at least roughly
commensurate with [its] costs,” 

–79). Accordingly, State Petitioners conclude, the
Rule is unlawful because it irrationally subjects all regulated
States to costly FIPs, which impose a uniform $1,400/ton
control level on emissions sources, even if a FIP for a given
State forecasts de minimis emissions reductions. For example,
State Petitioners point out, the FIP that the Rule imposes on
Wisconsin projects to reduce the state’s emissions impact on
the sole downwind receptor to which it is linked by just two
ten-thousandths of a part per billion.

    State Petitioners’ argument fails. As they tell it, in
promulgating the Rule, EPA threw cost consideration to the
wind and rashly required certain states, like Wisconsin, to
expend great costs to achieve insignificant environmental
benefits. But the record belies this narrative. Indeed, EPA
settled on the Rule’s $1,400/ton control level precisely
                                32
because, the agency found, it maximized air quality
improvement achieved per increment of additional cost. 81
Fed. Reg. at 74550. Moreover, for states like Wisconsin, for
which the Rule admittedly predicts relatively few emissions
reductions, the Rule imposes relatively few costs. This is
because—as State Petitioners themselves explain—such States
“have already incorporated [most of] the [control] technology
available at $1,400 or less.” States’ Reply Br. 6. In addition,
with respect to such States, EPA determined that the Rule’s
projected emissions reductions are significant (even if they
appear modest relative to other regulated States’ projected
reductions), because any State subject to the Rule is
contributing at least one percent of the 2008 ozone NAAQS to
at least one downwind problem receptor. For these reasons, the
Supreme Court held that the original CSAPR, which relied on
a virtually identical uniform control level methodology, was a
“cost-effective . . . permissible . . . and equitable interpretation
of the Good Neighbor Provision.” EME Homer 

. State Petitioners fail to persuade us that the Update Rule
is anything different.

                                 B

    State Petitioners argue next that the Rule is unlawful
because, in quantifying upwind emissions, EPA’s source
apportionment model included ozone from biogenic (i.e.,
naturally occurring) sources. This contravenes the Good
Neighbor Provision, State Petitioners contend, because the
statute authorizes the agency to regulate only “emissions
activity,” i.e., anthropogenic (or human-caused) emissions.
State Pet’rs’ Br. 38 (quoting 42 U.S.C. § 7410(a)(2)(D)(i))
(emphasis in original). State Petitioners add that EPA
implicitly recognizes this limit on its authority, because the
agency stated in the Rule that at step two it sought to “quantify
                                  33
the contributions from anthropogenic emissions from upwind
states.” 

Fed. Reg. at 74,526).

    Assuming without deciding that the Good Neighbor
Provision authorizes EPA to regulate only human-caused
emissions,1 State Petitioners’ argument nevertheless fails. As
EPA explains, ozone is formed when ozone precursors, such as
NOx and VOCs, react to one another in the presence of sunlight.
81 Fed. Reg. at 74,513. Ozone precursors are emitted from
both anthropogenic and biogenic sources. 

possible,
therefore, for ozone to form from purely biogenic precursors,
purely anthropogenic precursors, or a mix of both. See 

n.123. State Petitioners complain that the Rule’s source
apportionment model contravened the Good Neighbor
Provision in classifying ozone formed from a mix of biogenic
and anthropogenic precursors as anthropogenic ozone, which
the Rule requires upwind States to reduce. We, however, see
no problem with this, because ozone formed from a mix of
biogenic and anthropogenic precursors is a product of human-
caused emissions. True, such ozone is only partially
anthropogenic. But Industry Petitioners point to no authority
indicating that the Good Neighbor Provision authorizes EPA to
regulate only emissions that are entirely attributable to human
activity. Moreover, EPA rationally explained that it selected
the Rule’s particular source apportionment model (as opposed
to a model that provided a separate classification for ozone
formed from a mix of anthropogenic and biogenic precursors)
because, in the agency’s view, it was the more appropriate of
the available source apportionment tools. 

That
is because it assigned culpability for downwind ozone to
specific upwind sources of emissions in a manner that best
advances the Good Neighbor Provision’s essential purpose of

     1
       EPA does not contest this assertion, but State Petitioners point
to no authority that definitively establishes its truth.
                                 34
curbing interstate air pollution. 

therefore, to
EPA’s modeling choice. Nat’l Ass’n of Mfrs. v. U.S. Dep’t of
Interior, 

, 1103 (D.C. Cir. 1998) (“[An] agency’s
choice of model . . . must be respected when the record
discloses that the agency examined the relevant data and
articulated a reasoned basis for its decision.”).2

                                 C

     We turn next to Industry Petitioners’ arguments that
certain aspects of the Rule’s methodology contravene EME
Homer II’s prohibition on overcontrol, which proscribes EPA
from requiring a State to reduce emissions below one percent
of the relevant NAAQS or by more than is necessary to achieve
attainment at every downwind receptor to which a state is
linked. EME Homer 

. First, Industry
Petitioners argue that “many” downwind problem receptors
would have attained the NAAQS had the Rule excluded
emissions attributable to international sources. Indus. Pet’rs’
Br. 16. Second, Industry Petitioners contend, had the Rule
accounted for emissions reductions required of States subject
to the Rule but not linked to a given problem receptor, a
reviewing court would be “far likelier” to find that the Rule
overcontrols problem receptors in general. 

Third,
Industry Petitioners assert, because the Rule did not reflect
reasonably expected downwind controls, “some” of the upwind


    2
      In a related argument, State Petitioners theorize that the Rule
double counts any ozone produced from a combination of biogenic
and anthropogenic sources because, even if the anthropogenic half
were to be reduced, the lingering biogenic partner would still react
with other biogenic sources to form ozone. State Pet’rs’ Br. 38–39.
But as EPA points out, this argument was never raised before the
agency. EPA’s Br. 71. Because it has not been preserved, we need
not address it. See Nuclear Energy 

.
                                35
emissions reductions that it requires “may be unnecessary.” 

     None of Industry Petitioners’ arguments succeed. As for
emissions from international sources, Industry Petitioners are
simply wrong that the Rule “identif[ies] as ‘problem’ receptors
many whose problems were actually attributable not to
upwind-state but to non-U.S. emissions.” Indus. Pet’rs’ Br.
16–17. That logic incorrectly assumes that an upwind State
“contributes significantly” to downwind nonattainment only
when its emissions are the sole cause of downwind
nonattainment. But an upwind State can “contribute” to
downwind nonattainment even if its emissions are not the but-
for cause. After all, “[m]any (or perhaps all) receptors would
also attain the NAAQS if all in-state contributions were
eliminated, or if all upwind contributions were eliminated, or
if     all     non-anthropogenic       contributions          were
eliminated.” EPA’s Br. 65. Under Industry Petitioners’
position, EPA could not require emissions reductions from any
of those sources because each of them could point the finger at
the others. See also Catawba County v. EPA, 

, 39
(D.C. Cir. 2009) (rejecting the argument “that ‘significantly
contribute’ unambiguously means ‘strictly cause’” because
there is “no reason why the statute precludes EPA from
determining that [an] addition of [pollutant] into the
atmosphere is significant even though a nearby county’s
nonattainment problem would still persist in its absence”);
Miss. Comm’n on Envtl. Quality v. EPA, 

, 163
n.12 (D.C. Cir. 2015) (observing that the argument that “there
likely would have been no violation at all . . . if it were not for
the emissions resulting from [another source]” is “merely a
rephrasing of the but-for causation rule that we rejected in
Catawba County”).
                              36
     Industry Petitioners’ other arguments fail because they are
too particularized. As we emphasized in EME Homer III, for
challengers who raise the possibility of overcontrol in only a
few instances, “the Supreme Court has made clear . . . that the
way to contest instances of over-control is not through
generalized claims that EPA’s methodology would lead to
over-control, but rather through a ‘particularized, as-applied
challenge.’” Homer 

(quoting EME Homer
II, 

–24). Accordingly, as we did when presented
with similar arguments in EME Homer III, we reject Industry
Petitioners’ arguments because they do no more than speculate
that aspects of “EPA’s methodology could lead to over-control
of upwind States.” 

                               D

    As noted in Part 

, in order to implement upwind
States’ good neighbor obligations, EPA devised a detailed
process to determine whether downwind pollution receptors
were in attainment, maintenance, or nonattainment status.
Under this framework, EPA identified thirteen maintenance
receptors. 81 Fed. Reg. at 74,533. Nine of those thirteen
measured in attainment, in that their most recent monitored
design value complied with the NAAQS. 

States—Iowa, Kentucky, Tennessee, and Wisconsin—were
linked exclusively to one or more of those nine maintenance
receptors. 

     Industry Petitioners take two jabs at the Update Rule’s
definition of “maintenance” receptors and its treatment of
States linked exclusively to them. First, they say EPA deviated
unreasonably from past agency practice in designating
receptors as maintenance, even when they monitored in
attainment. Second, they claim that imposing a single uniform
$1,400/ton control level on all upwind States necessarily leads
                              37
to overcontrol of those States linked exclusively to
maintenance receptors. See Indus. Pet’r’s Br. 8–15, 25–26.
Neither argument succeeds.

                              1

    Industry Petitioners accept, as they must, that EPA was
permitted to base its designations, at least in part, on
predictions about the state of air quality in 2017. See North

–14 (affirming as reasonable EPA’s
interpretation of “will” in the Good Neighbor Provision as
“indicat[ing] the future tense”).

    Industry Petitioners’ first objection is that the agency’s
exclusive reliance on projections constitutes an unreasonable
deviation from its past practice of relying on a combination of
modeled and monitored data. 63 Fed. Reg. 57,356, 57,375
(Oct. 27, 1998) (NOx SIP Call) (relying on both monitored and
modeled data); accord 70 Fed. Reg. 25,162, 25,241 (May 12,
2005) (CAIR); cf. 76 Fed. Reg. 48,208, 48,230 (Aug. 8, 2011)
(explaining that EPA had to “drop[]the ‘monitored’ part of the
modeled + monitored test” only because “the most recent
monitoring data” reflected effects of the unlawful Clean Air
Interstate Rule).

    Our decision in North Carolina squarely forecloses that
argument, and its reasoning fully explains the agency’s
purported switch to reliance only on projected air quality. The
Good Neighbor Provision directs EPA to regulate emissions
that both “contribute significantly to nonattainment,” and also
“interfere with maintenance,” of air quality standards. See 42
U.S.C. § 7410(a)(2)(D)(i); North 

–
10. In North Carolina, EPA had interpreted the Good
Neighbor Provision’s maintenance prong narrowly, as ensuring
only against retrogression by previously nonattaining
                              38
receptors. 

This court overturned that interpretation
because it failed to give “independent effect” to the “interfere
with maintenance” prong, leaving those areas “barely meeting
attainment” without any “recourse” against upwind States’
contamination of their air quality. Id.; accord EME Homer 

n.18 (describing EPA’s duty to “reduce”
emissions from upwind States sufficient to ensure that “an
already-attaining State [maintains] satisfactory air quality”)
(emphasis added); EME Homer 

(explaining that EPA rule “afford[ed] independent effect to the
‘interfere with maintenance’ prong”).

    EPA’s Rule does what the rule in North Carolina did not.
It gives effect to the upwind States’ independent duty not to
impede downwind States’ maintenance of air quality standards.
As EPA explained, “the possibility of failing to maintain the
NAAQS in the future, even in the face of current attainment of
the NAAQS, is exactly what the maintenance prong of the good
neighbor provision is designed to guard against.” 81 Fed. Reg.
at 74,531. So the Rule’s designation method for maintenance
receptors was reasonable, and its decision to change its
approach to protect receptors in maintenance status was a
sensible response to North Carolina’s requirement that EPA
give full effect to the statute’s distinct maintenance command.
See 42 U.S.C. § 7410(a)(2)(D)(i).

     As for State Petitioners’ insistence that EPA should have
relied on a combination of monitored and modeled data, that
argument overlooks that the agency’s projections were
predicated     directly    upon      monitored    data    from
2009–2013. See 81 Fed. Reg. at 74,532. Any standard more
demanding—a requirement, for example, that maintenance
receptors actually monitor in nonattainment between
2013–2015—would run headlong into North Carolina’s
directive that EPA give “independent effect” to the
                               39
maintenance prong.       North 

.

                                2

    As previously noted, EPA applied a uniform cost threshold
of $1,400/ton to identify necessary emission reductions in
contributing upwind States. In other words, the Rule requires
that States deploy all available technologies capable of
reducing emissions at a cost of $1,400 or less per ton of NOx
reduced. 81 Fed. Reg. at 74,541.

    Industry Petitioners contend that this uniform control
threshold led to overcontrol in the four States linked
exclusively to maintenance receptors (rather than to receptors
showing nonattainment).          Indus. Pet’rs’ Br. 25–26.
Specifically, after EME Homer II, the maintenance prong only
authorizes EPA to “limit emissions ‘by just enough to permit
an already-attaining State to maintain satisfactory air quality.’”
EME Homer 

(quoting EME Homer 

n.18).

    Industry Petitioners say the agency’s approach ran afoul of
that mandate in two interrelated ways. First, for States linked
exclusively to maintenance-only receptors, they argue that their
existing upwind emission levels are by definition compatible
with attainment in the downwind States, so that any additional
reductions beyond “existing . . . levels” constitute overcontrol.
Indus. Pet’rs’ Br. 26. Second, Industry Petitioners contend
that, as a conceptual matter, if the $1,400/ton control level were
sufficient to resolve issues at nonattainment receptors, then that
same standard would, by definition, lead to overcontrol of
those States linked exclusively to maintenance receptors. 

                                 40
    Both arguments fail because they ignore key aspects of the
agency’s reasoning. Industry’s insistence that current levels
suffice for maintenance wrongly assumes that maintenance
receptors will violate the NAAQS only if upwind emissions
increase beyond the existing baseline. But things are not that
simple. Variations in atmospheric conditions and weather
patterns can bring maintenance receptors into nonattainment
even without elevated emissions. See 81 Fed. Reg. at 74,513–
14, 74,532, 74,537–38.

    Likewise, the argument that the uniform control standard
necessarily overshoots for maintenance receptors presupposes
that the Update Rule fully satisfies upwind States’ Good
Neighbor responsibilities. Not so—as the Rule repeatedly self-
describes, it is only a partial remedy. See, e.g., 81 Fed. Reg. at
74,505, 74,508, 74,520–22.

     More fundamentally, as we discussed with respect to
Industry Petitioners’ over-particularized claims of overcontrol,
these arguments fail to identify a single “actual . . . instance[]
of over-control,” which is what EME Homer II calls for. EME
Homer 

(emphasis added) (interpreting
EME Homer 

–24). In the Update Rule, EPA
conducted a rigorous overcontrol analysis, and concluded that
even with the new $1,400/ton control level, only a small subset
of maintenance and nonattainment receptors were projected to
succeed in fully resolving their air quality problems. See 81
Fed. Reg. at 74,551–74,552. Of all the upwind States, only
Tennessee was linked exclusively to those fully resolved
receptors. 

then, the agency confirmed that
problems at Tennessee’s linked receptors could not be resolved
at a less stringent level of control. 

record illustrates,
where evidence of “actual” overcontrol is needed, Industry
Petitioners’ conceptual objections alone cannot suffice. See
EME Homer 

n.18.
                                41
                                E

     State Petitioners challenge EPA’s use of a grid-cell
approach for identifying maintenance and nonattainment
receptors. State Pet’rs’ Br. 24–29. EPA uses that method to
calculate a monitor’s relative response factor. It does so by
putting the monitor at the center of a twelve-square-kilometer
grid, which consists of nine four-square-kilometer cells, and
then analyzing the air quality in each of the individual cells. 81
Fed. Reg. at 74,526–27. For “coastal” monitors near the ocean,
the grids included “offshore” cells in the air quality
measurements. See 

     State Petitioners claim that these offshore cells artificially
inflated projected ozone concentrations at the coastal monitors,
and that Iowa and Wisconsin were linked exclusively to those
misidentified air quality strugglers. State Pet’rs’ Br. 24–29; 81
Fed. Reg. at 74,534, 74,538–39 (Tables V.E-2, V.E-3). In the
States’ view, EPA should have based its designation decisions
on data derived solely from the cell in which the monitor was
located or from all of the over-land cells within the grid. State
Pet’rs’ Br. 24–29.

    That objection fails for four reasons.

     First, the agency offered a reasonable explanation for why
its grid-cell approach was “most representative” of onshore
ozone concentration levels. For starters, these models can be
imprecise at the “grid cell level”—that is, small variations in
the model may influence whether ozone is shown to form in
one particular cell rather than its neighbor. 81 Fed. Reg. at
74,534. That precision problem is further compounded by the
fact that monitors are often located close to the border of
several cells. 

two problems into account, the
agency reasonably worried that it would miss data “most
                               42
representative” of ozone concentrations, if it were to disregard
high concentrations in neighboring cells and focus exclusively
on the individual monitor cell. See id.; EME Homer 

(describing the considerable “deferen[ce]” owed to
agency “modeling choices”).

     Second, EPA found that over-water ozone often blows
onto the land above coastal monitors, and so capturing that
input is critical to accurately gauging air quality in the monitor
area. 81 Fed. Reg. at 74,534; North 

(affording “substantial deference to EPA’s technical
expertise”).

      Third, the accuracy of EPA’s judgment is confirmed by the
fact that, even under State Petitioners’ single-cell approach, the
lone receptor to which Iowa and Wisconsin are linked would
still demonstrate maintenance, and thus a need to preserve that
status. Cf. 81 Fed. Reg. at 74,534–35. Having pressed the
single-cell approach on the agency, State Petitioners cannot
seriously complain about a methodology that produced a
functionally identical result.

     Fourth, trying a different tack, State Petitioners latch onto
Michigan v. EPA’s command that the agency regulate only
“onshore state 

. But
Petitioners omit critical context. In Michigan, we struck
Wisconsin’s inclusion in a downwind attainment plan because
EPA had shown only that Wisconsin contributed significantly
to nonattainment in Lake Michigan itself. The record did not
reveal a downwind contribution to “any other State[.]” 

(quoting 42 U.S.C. § 7410(a)(2)(D)(i)(I)). The fatal blow
for EPA’s approach was when the agency “conceded” at oral
argument that it had provided no record “explanation to
support” a linkage “between the Lake Michigan receptor area
and the onshore states.” 

marks omitted).
                                43

     EPA was careful not to make that same mistake a second
time. So here, EPA explained carefully and on the record how
data from those offshore cells could be reasonably probative of
air quality at any given coastal receptor. See 81 Fed. Reg. at
74,534. Michigan is not the categorical bar Petitioners
describe, and EPA’s approach was fully consistent with its
holding.3

                                 F

     Industry Petitioners next complain that EPA failed to fully
consider the emissions-reducing effects of a 2016 Pennsylvania
rule. Indus. Pet’rs’ Br. 17–18. That argument falls wide of the
mark.

    EPA took into account State-level efforts to reduce
emissions when forecasting 2017 air quality during the
rulemaking process. But for reasons of modeling-reliability,
EPA included only those State emissions rules in effect through
February 1, 2016. 81 Fed. Reg. at 74,528 n.108. That was
EPA’s “cutoff date” because, after February 1, 2016, “it
[would] no longer [be] possible to incorporate updates into the
input databases” before EPA had to run the model. 

a footnote, Industry Petitioners argue that Arkansas,
Mississippi, and Pennsylvania were also negatively affected by the
inclusion of offshore cells because those States were linked to
monitors located in close proximity to large bodies of water. Indus.
Pet’rs’ Br. 29 n.15. But that lone, underdeveloped footnote does not
even show whether these States were actually impacted by the
agency’s methodology. In any event, “cursory arguments made only
in footnotes” generally do not preserve an issue, and there is no
reason to make an exception here. Abdelfattah v. U.S. Dep’t of
Homeland Sec., 

, 532 (D.C. Cir. 2015) (formatting
modified).
                              44

    In April 2016—several months after that cutoff date—
Pennsylvania finalized a rulemaking to implement
“Reasonably Available Control Technology” (“Control
Technology”) that was designed to limit the emission of NOx
and VOCs. That rule was scheduled to be implemented on
January 1, 2017, almost a year after EPA’s modeling window
had closed.

    While unable to include the Pennsylvania rule in its
modeling, EPA acknowledged its potential magnitude, and so
chose to conduct a “robust separate analysis to evaluate [the
Control Technology’s] impacts.” EPA, Memorandum on
Pennsylvania Rulemaking, J.A. 463. That study concluded that
the Pennsylvania rule did “not affect EPA’s identification of
[any] nonattainment or maintenance receptors.” 

      Industry Petitioners now fault the agency for considering
only the rule’s effect on NOx emissions, while ignoring its (far
more modest and unquantified) effect on emissions of VOCs.
That argument is a nonstarter. First off, Industry Petitioners
fail to explain how the agency could have figured reductions in
VOCs emissions into its analysis since the Pennsylvania rule
makes no effort to quantify them.

     Anyhow, the regulated electric utilities, which account for
the majority of the Control Technology’s emission reductions,
were projected to emit almost sixty times more NOx than VOCs
in the absence of the Pennsylvania rulemaking. Given that
even the reductions in NOx had no appreciable effect on EPA’s
receptor designations, it was perfectly reasonable for the
agency to conclude that factoring in the even more nominal
effects of VOCs regulation would not be worth the candle. Cf.
Thompson v. Clark, 

, 408 (D.C. Cir. 1984) (The
Administrative Procedure Act “has never been interpreted to
                                 45
require the agency to . . . analyse [sic] every . . . alternative
raised by the comments, no matter how insubstantial.”). For
those reasons, EPA’s decision was well within legal bounds.

                                 G

    Industry Petitioners lodge several objections to EPA’s
methodology for calculating States’ emissions budgets. But a
comprehensive picture of the agency’s approach exposes
where those arguments fall short.

     EPA’s $1,400/ton control level reflects the costs
associated with turning on idled selective catalytic reduction
equipment, as well as with the installation of “state-of-the-art
combustion controls,” like “low-NOx burners” and “over-fire
air.” 81 Fed. Reg. at 74,541.4 EPA used an Integrated Planning
Model (“Integrated Model”) that simulated the electricity
market to project both (i) a “baseline case” of what 2017
emissions would be without any additional pollution controls,

74,532; and (ii) a “control case” that incorporated
the selective catalytic reduction and combustion control
equipment, see 

74,548–49 (Tables VI.C-1-2).

     In setting a given State’s emissions budget, the agency
took the difference between the baseline and control cases—
what it calls the “relative-rate delta”—and subtracted it from
the State’s actual 2015 emission rate. 81 Fed. Reg. at 74,547–
48. That emissions rate—expressed in terms of pounds of NOx
per one million British thermal units of emitted heat
(“lbs/mmBtu”)—was then multiplied by the State’s 2015 heat
    4
       Selective catalytic reduction takes place when a reagent is
injected into a pollutant gas flue, inducing a chemical reaction that
transforms the pollutant into a more palatable chemical or chemicals.
See J.A. 1462–66.
                              46
input to produce the individual State emissions budget. 

(Tables VI.C-1-2). Industry Petitioners challenge
several steps in the analysis, but none of their arguments
succeed.

     First, they claim that EPA was unrealistic to expect that
the relevant emission controls could be fully installed during
the less-than-eight-month period between when the Update
Rule was finalized and when it was set to take effect. They say
at least eighteen months is needed, citing supporting anecdotes.

     But all those anecdotes show is that installation can drag
on when companies are unconstrained by the ticking clock of
the law. That does not establish how much time is technically
required to complete installation. EPA reasonably based its
determination on a real-world example identified during an
earlier rulemaking. Industry Petitioners fail to explain, by
reference to the actual mechanics of installation, why that EPA
judgment was so plainly wrong as to demonstrate arbitrary,
capricious, or unreasoned decisionmaking. In what is
effectively a war of competing anecdotes, EPA wins because
“we are forbidden from substituting our judgment for that of
the agency.” Ass’n of Am. R.R.s v. Interstate Commerce
Comm’n, 

, 740 (D.C. Cir. 1992) (formatting
modified).

     Second, Industry Petitioners take issue with EPA’s
“idling” assumption—that is, the proposition that certain less
efficient electric generating units would temporarily cease
operations once energy supply exceeds demand. Indus. Pet’rs’
Br. 23–24. By way of background, Industry Petitioners raised
concerns, during the comment period, that the agency’s model
assumed an unrealistic number of imminent unit retirements.
In response, EPA promised to “constrain[ ] the model to
prevent . . . retirement projections” before 2020. J.A. 361. The
                               47
Integrated Model nevertheless assumed that certain units
would be “idled” whenever supply outstrips demand.
According to Industry Petitioners, treating those units as idled
“amount[s] to the same thing” as deeming them permanently
retired from production. Indus. Pet’rs’ Br. 24.

     That argument mixes apples and oranges. Idling is a
natural component of modeling programs, like the Integrated
Model, that are designed to reflect electricity markets “as
accurately as possible.” 81 Fed. Reg. at 74,528. To capture
actual market mechanics, the model determines the least-cost
method of anticipating electricity demand over a given period,
and it assumes that less efficient units will be “idled” in the
short run when they are not needed to meet demand. That
temporary, on-again-off-again idling is quite distinct from
permanent retirement and closure of a facility. That a model
overestimates the rate of long-run retirements thus says nothing
about whether it accurately projects the ebb and flow of short-
run supply and demand. So EPA’s decision to limit near-term
retirement projections based on long-run unprofitability says
nothing about the use of temporary, market-driven idling in its
economic models.

     Industry Petitioners supplement with an argument that
EPA failed to fully explain its idling assumption on the record.
Indus. Reply Br. 12. But that is neither here nor there, because
Industry Petitioners make no showing that the idling
assumption actually altered State emissions budgets. EPA used
the Integrated Model only to determine the delta between a
State’s baseline case and the control case, which it then applied
to the State’s historical 2015 emission rates. Because any
projected idling was held constant between the baseline case
and the control case, it could not affect how much units were
expected to reduce their emissions relative to their historical
baseline. Cf. 81 Fed. Reg. at 74,547.
                              48

     Third, Industry Petitioners claim the agency promised to
treat .1 lbs/mmBtu as a ceiling on the emissions-reduction
potential for units equipped with selective catalytic reduction.
But instead the agency sometimes assumed rates as low as .075
lb/mmBtu in its actual emission-budget analysis. Indus. Pet’rs’
Br. 24–25. That argument does not stand up to scrutiny.

     EPA initially proposed calculations based on the
assumption that selective catalytic reduction-equipped units
could achieve a NOx emissions level of .075 lbs/mmBtu. See
81 Fed. Reg. at 74,544. In the Final Rule, EPA required a less-
demanding threshold for Industry of .1 lbs/mmBtu. 

At the same time, EPA had learned during the
comment period that certain, newer plants had a proven track
record of achieving superior emissions rates. So EPA decided
in the Final Rule that those units would be assigned their
historical rate if lower than .1 lbs/mmBtu. That decision not to
license plants with better emission controls to emit more NOx
than they already do was eminently reasonable.

     In a related argument, Industry Petitioners complain that
EPA applied its .075 lbs/mmBtu limit for selective catalytic
reduction-controlled units that share “common stacks” with
uncontrolled units. Indus. Pet’rs’ Br. 24. But these common
stack arrangements prevent the agency from gathering “reliable
data to determine the emission rates of the individual units.”
J.A. 461. EPA therefore adopted a .075 lbs/mmBtu estimate in
both the base and control cases, effectively concluding that
those units equipped with selective catalytic reductions were
incapable of “achiev[ing] any additional reductions.” J.A.
461–62.

    Because that emissions-reduction capacity was held
constant between the base and control measures, Industry
                               49
Petitioners have failed to show how the statutory or regulatory
scheme required EPA to take a different approach to dealing
with the lack of empirical data from combined stacks. Nor did
they show how the underlying emissions rate could have
affected the delta EPA relied on in calculating a State’s
emissions-reduction potential, given that the reduction was
held constant. EPA’s Br. 87–88; cf. 81 Fed. Reg. at 74,547.

                               IV

                                A

     Industry Petitioners challenge the emissions budgets for
Mississippi, Oklahoma, Indiana, and Illinois, as well as several
specific emissions-allowance allocations to units in those
States. The gist of the argument is that the complaining States
want larger emissions budgets, and the complaining units want
a larger share of those budgets in the form of increased
allowances. But a majority of these challenges arise in direct
response to EPA’s decision in the Final Rule to use the relative-
rate method in calculating emissions limits. And because that
methodological change was introduced for the first time in the
Final Rule and the decision to adopt it was made in response to
comments, 81 Fed. Reg. at 74,547–48; J.A. 419, Industry
Petitioners’ challenges to the relative-rate method are not yet
ripe for judicial review. See 42 U.S.C. § 7607(d)(7)(B).

     The Clean Air Act requires, as a predicate for judicial
review, that EPA first be afforded the opportunity to address
objections to its rules, and that those objections be raised with
“reasonable specificity during the period for public comment.”
42 U.S.C. § 7607(d)(7)(B). Where “it was ‘impracticable to
raise a particular objection’ or if ‘the grounds for the objection
arose after that [comment] period,’” the party challenging the
agency action “still must petition EPA for administrative
                               50
reconsideration before raising the issue before this Court.”
EME Homer 

(quoting 42 U.S.C.
§ 7607(d)(7)(B)).      Should EPA choose not to grant
reconsideration, that decision is independently reviewable. Id.;
42 U.S.C. § 7607(d)(7)(B).

     So no matter how EPA responds, a petition for
reconsideration is “what the statute requires and what [this
court] therefore must insist upon,” even if it might “seem a
roundabout” way of doing things. EME Homer 

. After all, we cannot fairly review how the agency
responded to an argument that was never presented to it. See
Smith v. Berryhill, 

, 1779 (2019) (“Fundamental
principles of administrative law . . . teach that a federal court
generally goes astray if it decides a question that has been
delegated to an agency if that agency has not first had a chance
to address the question.”); see also Util. Air Regulatory Grp. v.
EPA, 

, 747 (D.C. Cir. 2014) (“[T]he only
objections that may immediately be raised upon judicial review
are those that were raised during the public comment period.
Objections raised for the first time in a petition for
reconsideration must await EPA’s action on that petition.”).
That administrative exhaustion requirement is “strictly”
enforced. Nat. Res. Def. Council v. EPA, 

, 1259
(D.C. Cir. 2009) (per curiam).

    Industry Petitioners run headlong into this exhaustion
requirement when they complain that Oklahoma’s and
Mississippi’s emissions in the agency’s 2017 base cases far
exceeded those States’ actual 2015 emissions. Those inflated
bases, they say, translated into unrealistically low emissions
budgets for both States. Indus. Pet’rs’ Br. 36, 39–40. EPA
responds that its methodology neutralizes any errors that might
have caused those distortions by holding constant, between the
base and control cases, those erroneous inputs that caused the
                                  51
purported inflation in the base. EPA’s Br. 89, 96–97. Industry
Petitioners beg to differ, asserting that the artificial inflation is
not, in fact, neutralized or cancelled out because: (i) if the base
and control cases are multiplied, then the delta between the two
will increase; and (ii) the marginal cost of emissions reduction
goes up as overall emissions go down, see Indus. Reply Br. 17–
18.

     The questions of whether, and the extent to which, the
relative-rate method actually neutralizes distortions are not
properly before us. Because the challenged methodology first
appeared in the Final Rule, Industry Petitioners’ arguments
should have been raised in a petition for agency
reconsideration. Just like EME Homer III, this court is
“without authority” to decide a challenge that petitioners “did
not and could not have raised . . . during the period for public

. Instead, agency reconsideration
is “the only appropriate path” forward. 

Presumably that exhaustion requirement is why at least two of
the Industry Petitioners have already sought agency
reconsideration.5 Until EPA acts on those reconsideration
petitions, the challenges are not yet ripe for our review. Id.; cf.

     5
       See Oklahoma Gas and Electric Company, Petition for
Reconsideration, EPA–HQ–OAR–2015–0500–0589, at 5 (“Because
the 2017 budget-setting base case was unrealistically high, the ‘delta’
between that number and the 2017 cost threshold was also too
high.”), https://www.regulations.gov/document?D=EPA-HQ-OAR-
2015-0500-0589; 

“[i]t was impossible . . . to
comment on EPA’s revised emissions budget calculation
methodology” because it “was first presented . . . in the Final Rule”)
(emphasis added); Western Farmers Electric Cooperative, Petition
for Reconsideration, EPA–HQ–OAR–2015–0500–0588, at 4
(describing the relevant disparity as the “Perverse IPM Result”)
(emphasis in original), https://www.regulations.gov/document?
D=EPA-HQ-OAR-2015-0500-0588.
                                 52
Appalachian Power Co. v. EPA, 

, 818 (D.C. Cir.
1998) (per curiam) (“The purpose of the exhaustion
requirement is to ensure that the agency is given the first
opportunity to bring its expertise to bear on the resolution of a
challenge to a rule.”). And those Petitioners who have not yet
moved for reconsideration run into that same bar on judicial
review. EME Homer 

.

    The arguments by Energy Association and Indiana Utility
Group (“Indiana Petitioners”) meet the same fate. They
complain that (i) the relative-reduction method yielded a
budget far below the one proposed under EPA’s initial formula;
and (ii) EPA’s reliance on 2015 heat-input data, rather than the
2014 data cited in the proposed rule, harmed Indiana units
because the State’s heat input fell considerably between the
two years. Indus. Pet’rs’ Br. 31.

     Petitioners, of course, enjoy no special entitlement to
either the initial emissions figure or the less-current data
referenced in the proposed rule. And to the extent that the
Indiana Petitioners mean to challenge the reasonableness of the
relative-rate method or the representativeness of the 2015 data,
those empirically laden propositions must first be exhausted in
a motion for reconsideration. Cf. Weinberger v. Salfi, 

, 765 (1975) (“Exhaustion is generally required as a matter
of preventing premature interference with agency processes, so
that the agency may function efficiently and so that it may have
an opportunity to correct its own errors, to afford the parties
and the courts the benefit of its experience and expertise, and
to compile a record which is adequate for judicial review.”).6

    6
      Indiana Petitioners recognize this implicitly when, in a single
sentence in their reply brief, they suggest that EPA was required to
resubmit its relative-reduction methodology for additional comment.
                                  53

     The Indiana Petitioners’ remaining contentions do not
advance the ball. They criticize EPA’s decision to assign
emissions rates between .07 and .075 lb/mmBtu to units newly
equipped with selective catalytic reduction, even though those
technologies had not yet been put into operation in 2015 and
2016. Indus. Pet’rs’ Br. 32 & n.30. But EPA reasonably
assumed that these technologies would be in use when the Rule
took effect in 2017, and so EPA assigned these units an
emission rate of .075 lbs/mmBtu to “reflect” the full effects of
the new technology. J.A. 420. As for the .07 lbs/mmBtu rate,
the figure was used not to adjust the State’s 2015 emissions
data, but rather to calculate its relative-rate delta. EPA claims
that, because it was assigned in both the base and control cases,
the method neutralized any inflating effect. EPA’s Br. 94 n.24.
If Petitioners have any colorable quibble, it is with the
proposition that the relative reduction methodology neutralizes
the effects of mistaken inputs. And as we have said, that
argument must first be raised in a motion for reconsideration.

                                   B

    Industry Petitioners from Oklahoma and Illinois challenge
other aspects of EPA’s budget and unit allocation decisions.
Their arguments are no more successful.

    As previously explained, EPA computed unit allocations
based on a plant’s projected share of its State’s overall ozone-
season heat input, capped at the unit’s actual emissions
between 2011 and 2015. 81 Fed. Reg. at 74,562. In setting that


But Petitioners have forfeited this cursory reframing of the argument
by failing to raise it until their reply brief. See World Wide Minerals,
Ltd. v. Republic of Kazakhstan, 

, 1160 (D.C. Cir.
2002).
                               54
cap, EPA relied principally upon measured data reported
directly by industry to the agency under 40 C.F.R. Part 75,
Subpart G. “Where EPA data [were] unavailable,” the agency
said it would also rely on data from the United States Energy
Information Administration. J.A. 260.

     Western Farmers Electric Cooperative, one of the
Oklahoma Industry Petitioners, complains that, for certain
units, EPA relied on just one year of available reported data,
and refused to fill in the gaps with data from the Energy
Information Administration. Indus. Pet’rs’ Br. 40–42. This,
they claim, was contrary to the purpose of EPA’s multi-year
averaging approach, which was designed to avoid aberrations
and to best approximate each unit’s true heat input. Indus.
Reply Br. 21.

     The Cooperative is mistaken. EPA reasonably prioritized
its own data, which “relies on unmodified historic data reported
directly by the vast majority of covered sources, whose
designated representatives have already attested to [its]
validity.” 76 Fed. Reg. at 48,288. In deciding whether to use
the Energy Information Administration’s estimates to fill in the
gaps, EPA faced a tradeoff between accuracy, on the one hand,
and long-run representativeness, on the other. We see no
reason to disturb the balance that EPA struck. Cf. Catawba

(describing the “extreme degree of
deference [given] to [EPA] when it is evaluating scientific data
within its technical expertise”) (formatting modified).

     In an effort to evade that deference, the Cooperative
frames its challenge as the agency “[d]eparting” from its own
internal standards. Indus. Pet’rs’ Br. 40–41. But absent
evidence that the agency ever committed to or even applied the
Cooperative’s preferred approach, the deviation claim falls flat.
                               55
     Last, Prairie State Generating Company (“Prairie State”),
an Illinois-based petitioner, claims it was unfairly
disadvantaged by EPA’s unit classification system. Indus.
Pet’rs’ Br. 29–30. That system divided units into two general
categories: “existing” and “new.” 81 Fed. Reg. at 74,564–65
(describing also a third category for new units in Indian
country, not at issue here). Over ninety percent of each State’s
budget went to “existing units”—that is, those units that started
operation prior to January 1, 2015, and for which EPA
possessed at least one year of measured emissions data. 81
Fed. Reg. at 74,564.

     By comparison, “new units” are ones for which EPA lacks
even this first year’s worth of data. Because the agency lacked
reliable emissions data on these new units, it established a new-
unit “set-aside” for each State. EPA calculated it based on (i)
a uniform two percent baseline, which “reflect[s] a reasonable
upper bound of state-level share of emissions from new units;”
and (ii) state-specific additions based on amounts that EPA
“projects to be emitted from ‘planned’ units in 2020.” J.A. 257.
Should new-unit allowances go unallocated, they are then
redistributed to existing units before the relevant compliance
deadline. 81 Fed. Reg. at 74,565. Relatedly, EPA shifts to the
set-aside all allowances from units that have ceased operations
for over five years. 

dormancy
requirement was necessary because a sudden loss of
allowances might “cause a unit, which would otherwise retire,
to continue operations in order to retain ongoing allowance
allocations.” 

    Prior rulemaking had defined Prairie State as a “new” unit.
But by January 2015, when it was well into “normal
operations,” it was reclassified as an “existing unit” under the
Update Rule, with allocation based upon its actual heat input.
Indus. Pet’rs’ Br. 28–29.
                                 56

     Prairie State’s principal complaint is that it would have
been able to enjoy the benefits of new-unit set asides if it were
classified as a new, rather than an existing, unit. See Indus.
Pet’rs’ Br. 29–31. But a preference for more youthful
treatment is not a legal argument. In 2011, during the prior
rulemaking, Prairie State was new; five years later, not so
much.

     Prairie State’s various ancillary contentions amount to no
more than explanations for why it would have been better off
if EPA had deviated from its even-handed approach and treated
it as a new unit, despite its several years of operation. To the
extent Prairie State is attacking, indirectly, the Rule’s definition
of “new units” or its allocation for retiring units, both of those
agency choices were reasonable and sufficiently explained. 81
Fed. Reg. at 74,565. Because “new units” were defined as
those for which EPA lacks a single year’s worth of reliable
emissions data, EPA necessarily could not rely on actual
emissions data to make an allocation. J.A. 256. And a set aside
for retiring units was necessary to ensure that the allowance
allocations did not have the perverse incentive of deterring
retirement. 81 Fed. Reg. at 74,566.7


    7
       Prairie State’s challenge to EPA’s budget-setting decision is
equally fruitless. According to Prairie State, EPA deflated Illinois’
budget allocation by averaging Prairie State’s highest heat inputs
between 2011 and 2015, which were “artificially low due to issues
with [Prairie State’s] advanced technology.” Indus. Pet’rs’ Br. 29.
But Prairie State’s first premise is wrong. EPA relied only on the
State’s most recent measured data from 2015, not the three-year
average, in setting State budgets. 81 Fed. Reg. at 74,547. And by
Prairie State’s own admission, it “began normal operations in 2014.”
Indus. Pet’rs’ Br. 28. There was neither error nor discernible
prejudice to Prairie State in that budget-setting decision.
                               57
     To make that long story short, all of the Industry
Petitioners’ State- and unit-specific arguments fail.

                               V

     State Petitioners offer up a pair of procedural challenges
to the Rule. But both claims suffer fatal jurisdictional defects.

     First, Texas, Ohio, and Wisconsin claim that EPA sat on
their timely SIP submissions beyond the twelve-month
statutory deadline, see 42 U.S.C. § 7410(k)(2), so that it could
develop the data and methodology necessary to justify rejecting
the SIPs and to impose FIPs in their place. State Pet’rs’ Br.
29–38; see 81 Fed. Reg. 53,309 (Aug. 12, 2016) (disapproval
of Wisconsin SIP); 81 Fed. Reg. 53,284 (Aug. 12, 2016)
(disapproval of Texas SIP); 81 Fed. Reg. 38,957 (June 15,
2016) (disapproval of Ohio SIP). They ask that we vacate the
Rule and instruct the agency to revisit these SIPs based
exclusively upon data acquired before the Section 7410(k)
deadline. State Pet’rs’ Br. 37–38.

     Those arguments are, in effect, collateral attacks on EPA’s
SIP denials—and they come too late in the game. The SIP
denials were finalized in June and August of 2016. The States
filed their petitions for review in this case in November and
December of 2016, far outside of the sixty-day jurisdictional
window for challenging SIP denials.             See 42 U.S.C.
§ 7607(b)(1); Dalton Trucking, Inc. v. EPA, 

,
879–80 (D.C. Cir. 2015) (sixty-day window is jurisdictional).
To the extent the States challenge these SIP denials, their
untimely arguments lie beyond our jurisdiction.

    The States insist “[i]t is the FIP that is defective, and the
FIP that is attacked here.” States’ Reply Br. 14. Even if
accurate, that would be entirely beside the point. In EME
                               58
Homer II, the States claimed that EPA, as a condition for
promulgating Good Neighbor FIPs, had to give them a second
bite at compliant SIPs once the agency had calculated their
significant contributions to downwind 

–07. The Supreme Court explained that this was
not a collateral attack because “[t]he gravamen of the . . .
challenge” was that EPA failed to timely comply with a
condition precedent for promulgating FIPs, “not that
[its] disapproval of any particular SIP was erroneous.” 

As evidence, the Court emphasized that the States’
argument “does not depend on the validity of the prior SIP
disapprovals. Even assuming the legitimacy of those
disapprovals, the question remains whether EPA was required
to do more . . . to trigger the Agency’s statutory authority to
issue a FIP.” 

     By contrast, here, even if the States’ argument were to lead
to the invalidation of the Update Rule, its success would
“depend on the [in]validity of the prior SIP disapprovals,” and
their argument expressly “assum[es]” the illegitimacy of
EPA’s decisions. 

That is the hallmark of an
improper collateral attack. The true gravamen of the claim lies
in the agency’s failure to timely act upon the States’ SIP
submissions and, relatedly, its reliance on data compiled after
the SIP action deadline. Both go directly to the legitimacy of
the SIP denials. And, critically, those problems exist whether
or not EPA follows up with a FIP of its own.

     Lastly, we cannot decide Wyoming’s claim that EPA
“misled western States into believing that [it] would not apply
the [the relevant] modeling to the West,” State Pet’rs’ Br. 42,
and yet “applied the CSAPR modeling directly to the West
without performing a regional or state-specific analysis and
disapproved parts of [Wyoming’s] SIP revision,” 

                               59
    Wyoming lacks standing to press that argument here
because its injury is traceable not to the Update Rule, but rather
to separate rulemakings in which EPA “solicit[ed] public
comment” on the appropriateness of applying the CSAPR
modeling to Wyoming, 81 Fed. Reg. 81,712, 81,716 (Nov. 18,
2016), and disapproved Wyoming’s SIP, 82 Fed. Reg. 9,142
(Feb. 3, 2017). Because the Update Rule has not caused the
complained-of injury, Wyoming cannot demonstrate a key
element of standing, see Friends of the Earth, Inc. v. Laidlaw
Environmental Servs. (TOC), Inc., 

, 180–81
(2000), and we are without authority to consider its argument.

                               VI

     Finally, in light of our invalidation of the Update Rule in
one respect, we take up the question of the proper remedy. As
a general rule, we do not vacate regulations when doing so
would risk significant harm to the public health or the
environment. See Allied-Signal, Inc. v. Nuclear Regulatory
Comm’n, 

, 150–51 (D.C. Cir. 1993). For that
reason, we have remanded without vacatur in previous Good
Neighbor Provision cases. See EME Homer 

; North Carolina v. EPA, 

, 1178 (D.C. Cir.
2008) (per curiam). And we have done the same in other cases
involving the Clean Air Act. See, e.g., Envtl. Def. Fund, Inc.
v. EPA, 

, 190 (D.C. Cir. 1990) (remanding without
vacatur because vacatur would undermine “the enhanced
protection of the environmental values covered by the [Clean
Air Act]”).

     We follow the same course here. Vacatur of the Update
Rule “could cause substantial disruption to the [allowance]
trading markets that have developed.” EME Homer 

. And “some good neighbor obligations [imposed
by the Rule] may be appropriate for some of the relevant
                               60
upwind States.”    

conclude that vacatur is
inappropriate.

     We decline Environmental Petitioners’ request, however,
to impose a six-month timeframe on EPA’s promulgation of a
revised rule. But of course, “we do not intend to grant an
indefinite stay of the effectiveness of this court’s decision.”
North 

. And Environmental
Petitioners could attempt to “bring a mandamus petition to this
court in the event that EPA fails to modify [the Rule] in a
manner consistent with our . . . opinion.” 

  *    *   *

    For the foregoing reasons, the petitions for review are
granted in part and denied in part.

                                                   So ordered.