Sierra Club v. EPA ( 2020 )


Menu:
  •                                            PRECEDENTIAL
    UNITED STATES COURT OF APPEALS
    FOR THE THIRD CIRCUIT
    _____________
    No. 19-2562
    _____________
    SIERRA CLUB,
    Petitioner
    v.
    UNITED STATES ENVIRONMENTAL PROTECTION
    AGENCY
    *PENNSYLVANIA DEPARTMENT OF ENVIRONMENTAL
    PROTECTION,
    Intervenor Respondent
    *(Pursuant to the Court Order dated 8/5/19)
    ______________
    On Petition for Review of Final Agency Action
    of the United States Environmental Protection Agency
    (EPA-1: EPA-R03-OAR-2017-0290)
    ______________
    Argued May 21, 2020
    ______________
    Before: McKEE, BIBAS, and NYGAARD, Circuit Judges.
    (Opinion filed: August 27, 2020)
    Zachary M. Fabish, Esq.
    Sierra Club Environmental Law Program
    50 F Street, N.W.
    8th Floor
    Washington, DC 20001
    Charles McPhedran, Esq. [Argued]
    Earthjustice
    1617 John F. Kennedy Boulevard
    Suite 1130
    Philadelphia, PA 19103
    Mychal Ozaeta, Esq.
    Earthjustice
    707 Wilshire Boulevard
    Suite 4300
    Los Angeles, CA 90017
    Counsel for Petitioner
    Brandon N. Adkins, Esq. [Argued]
    United States Department of Justice
    Environment & Natural Resources Division
    P.O. Box 7611
    Ben Franklin Station
    Washington, DC 20044
    Counsel for Respondent
    Robert A. Reiley, Esq.
    Pennsylvania Department of Environmental Protection
    Office of Chief Counsel
    9th Floor
    400 Market Street
    Rachel Carson State Office Building
    Harrisburg, PA 17101
    Jesse C. Walker, Esq. [Argued]
    Office of Attorney General of Pennsylvania
    Bureau of Regulatory Counsel
    RCSOB 9th Floor
    P.O. Box 8464
    Department of Environmental Protection
    2
    Harrisburg, PA 17105
    Counsel for Intervenor
    _______________________
    OPINION
    ______________________
    McKEE, Circuit Judge.
    The Sierra Club petitions for review of the
    Environmental Protection Agency’s approval of new
    Pennsylvania National Ambient Air Quality Standards
    (NAAQS). The proposed standards govern pollution output at
    coal-burning power plants in the Commonwealth. The Sierra
    Club argues that the standards wrongly claim to reduce
    pollution output at Pennsylvania’s most advanced plants while
    simply rubberstamping an average of current pollution output
    as its supposed new gold standard. Further, the Sierra Club
    criticizes the proposal’s minimum temperature threshold—a
    measure that allows plants to nearly quintuple their pollution
    output when operating below 600 degrees Fahrenheit—as
    unsupported and unsupportable given the technical record
    before the agency. Finally, the Sierra Club claims that the
    approved standards lack enforceable reporting regulations.
    Because we agree that the regulatory regime which springs
    forth from these three defining characteristics is neither
    supported by adequate facts nor by reasoning found in the
    administrative record, we hold that the EPA’s approval was
    arbitrary and capricious. We will therefore grant the petition
    for review and remand to the agency with instructions that it
    develop enforceable pollution controls in accordance with its
    legal obligations.
    I.
    This dispute is governed by the Clean Air Act which
    gives Circuit Courts of Appeals original jurisdiction to review
    the EPA’s approval of a state’s proposed pollution standards.1
    CAA is a model of cooperative federalism. Indeed, that
    1
    
    42 U.S.C. § 7607
    (b)(1).
    3
    philosophy has been described as the Act’s “core principle.”2
    Thus, although the EPA is charged with setting the NAAQS,3
    the individual states are afforded discretion in the planning and
    implementation of plans to achieve the EPA’s goals for
    reduction in air pollutants.4
    One such pollutant is ground level ozone, which is not
    a naturally occurring compound. This ozone is the result of a
    chemical reaction that occurs when power plants and industrial
    boilers mix nitrogen oxides (NOx) with the organic
    compounds they are emitting as air pollution.5 It is similar to
    the process that creates most anthropogenic (i.e., human
    produced) ozone, which primarily arises from the “photolysis
    of nitrogen dioxide by sunlight, occurring in the presence of
    hydrocarbons.”6
    A. Relevant standards and history
    This dispute has its beginnings in the EPA’s 2008
    revision to its 1997 ozone National Standard. There, the EPA
    tightened the cap on ozone pollution to an average of 75 parts-
    per-billion over an 8-hour period (the previous standards had
    2
    Miss. Comm’n on Envtl. Quality v. E.P.A., 
    790 F.3d 138
    ,
    156 (D.C. Cir. 2015) (quoting E.P.A. v. EME Homer City
    Generation, L.P., 
    572 U.S. 489
    , 511 n.14 (2014)).
    3
    See 
    42 U.S.C. § 7408
    (a) (directing the EPA to list “air
    pollutants” whose emissions “cause or contribute to air
    pollution which may reasonably be anticipated to endanger
    public health or welfare”); 
    42 U.S.C. § 7409
    (a)-(b) (for such
    listed air pollutants, EPA must set “ambient air quality
    standards the attainment and maintenance of which . . . are
    requisite to protect the public health” with an “adequate
    margin of safety”).
    4
    
    42 U.S.C. §§ 7410
    , 7502.
    5
    E.P.A., Ground-level Ozone Basics,
    https://www.epa.gov/ground-level-ozone-pollution/ground-
    level-ozone-basics.
    6
    Peter J. A. Rombout, et al., Rationale for an Eight-Hour
    Ozone Standard, 36 J. OF THE AIR POLLUTION CONTROL
    ASSOC. 913, 913 (1986),
    https://doi.org/10.1080/00022470.1986.10466130.
    4
    permitted 80 parts-per-billion7).8 Seventeen counties around
    Philadelphia and Pittsburgh were unable to achieve the new
    2008 requirements for pollution control.9 Accordingly, the
    Commonwealth was required to design a state implementation
    plan (SIP) for its major sources of NOx and volatile organic
    compounds.10 Power plants are one of the most significant
    sources of these pollutants, and were therefore most directly
    impacted by these proposals.
    Pennsylvania’s proposal had to satisfy Reasonably
    Available Control Technology (RACT) requirements.11 RACT
    is a technology-forcing standard designed to induce
    improvements and reductions in pollution for existing sources.
    It is a term of art at the foundation of the EPA’s decision-
    making, but is not defined in the Clean Air Act.12 The EPA
    itself originally defined this standard in internal guidance as
    requiring “the toughest controls considering technological and
    economic feasibility that can be applied to a specific situation
    . . . [a]nything less than this is by definition less than RACT.”13
    7
    EPA Proposes New National Ambient Air Quality Standards
    for Ozone, 20 No. 3 AIR POLLUTION CONSULTANT 2.12, 2.13
    (2010).
    8
    
    73 Fed. Reg. 16,436
     (Mar. 27, 2008). This 8-hour averaging
    period allows for variation in emissions throughout the day.
    Since 1997, the EPA has accepted such 8-hour averages for
    ozone emissions, and has judged compliance based on the
    fourth-highest (i.e., lowest) daily maximum from each 8-hour
    window, averaged over three years. 20 No. 3 AIR POLLUTION
    CONSULTANT at 2.13.
    9
    Pennsylvania DEP, “Attainment Status by Principal
    Pollutants,”
    http://www.dep.pa.gov/Business/Air/BAQ/Regulations/Pages
    /Attainment-Status.aspx.
    10
    See 42 U.S.C. §§ 7511c(a), 7502(c)(1).
    11
    
    42 U.S.C. § 7502
    (c)(1) (imposing the RACT requirement).
    12
    
    42 U.S.C. § 7502
    (c)(1). See also Natural Res. Def. Council,
    Inc. v. E.P.A., 
    571 F.3d 1245
    , 1252 (D.C. Cir. 2009) (finding
    “reasonably available control technology” under the Clean
    Air Act to be ambiguous).
    13
    Memorandum from Roger Strelow, Assistant Admin. for
    Air and Waste Mgmt., U.S. E.P.A., to Regional Admins.,
    5
    Because the parties do not dispute this long-standing
    definition, we assume without deciding that the EPA’s
    definition is correct. Therefore, RACT is “the lowest emission
    limitation that a particular source is capable of meeting by the
    application of control technology that is reasonably available
    considering technological and economic feasibility.”14
    Pennsylvania calls the SIP’s implementation of that standard
    the RACT II Rule, as it is a second generation approach (since
    RACT must be reconsidered at each NAAQS revision).15
    RACT is not designed to rubber-stamp existing control
    methods. It is a technology-forcing mechanism. When
    originally introducing the standard, the EPA noted that “the
    control agency, using the available guidance, should select the
    best available controls, deviating from those controls only
    where local conditions are such that they cannot be applied
    Regions I - X, at 2-3 (Dec. 9, 1976),
    https://www3.epa.gov/ttn/naaqs/aqmguide/collection/cp2/197
    61209_strelow_ract.pdf.
    14
    E.P.A., State Implementation Plans; Nitrogen Oxides
    Supplement to the General Preamble for the Implementation
    of Title I of the Clean Air Act Amendments of 1990, 
    57 Fed. Reg. 55,620
    , 55,624/3 (proposed Nov. 25, 1992); Navistar
    Intern. Transp. Corp. v. E.P.A., 
    941 F.2d 1339
    , 1343 (6th Cir.
    1991) (“Since 1976, the EPA has interpreted reasonably
    available control technology to be the lowest emission
    limitation that a particular source is capable of meeting by the
    application of control technology that is reasonably available
    considering technological and economic feasibility.”)
    (internal quotation marks omitted); Washington Envtl.
    Council v. Bellon, 
    732 F.3d 1131
    , 1137 (9th Cir. 2013)
    (describing same description of the standard adopted into
    state law as part of an EPA-approved SIP).
    15
    JA38; see also JA247 (“The evaluation or reevaluation of
    what constitutes RACT-level control for affected sources is a
    requirement that must be fulfilled each time the EPA
    promulgates a new NAAQS . . .”). For clarity, given the
    potential for confusion between the general RACT approach
    and Pennsylvania’s specific implementation here, dubbed
    RACT II, we refer to the proposed changes as the SIP
    whenever possible.
    6
    there and imposing even tougher controls where conditions
    allow.”16
    Thus, in order to satisfy RACT, a proposed SIP must
    satisfy both technological and economic feasibility. The EPA
    explains, “[t]he technological feasibility of applying an
    emission reduction method to a particular source should
    consider the source’s process and operating procedures, raw
    materials, physical plant layout, and any other environmental
    impacts such as water pollution, waste disposal, and energy
    requirements.”17 Whereas the economic feasibility “considers
    the cost of reducing emissions and the difference in costs
    between the particular source and other similar sources that
    have implemented emission reduction.”18 When calculating the
    cost, the EPA has previously cautioned: “Economic feasibility
    rests very little on the ability of a particular source to ‘afford’
    to reduce emissions to the level of similar sources. Less
    efficient sources would be rewarded by having to bear lower
    emission reduction costs if affordability were given high
    consideration. Rather, economic feasibility . . . is largely
    determined by evidence that other sources in a source category
    have in fact applied the control technology in question.”19
    A key technological element of this discussion is the
    development of selective catalytic reduction (SCR). SCR
    originated in the 1970s and is currently a preferred method for
    limiting coal-fired power plant pollution. More than 300 coal
    power plants in the U.S. feature the technology.20 SCR can,
    potentially, remove up to 100% of the ozone pollution
    produced by a plant. However, most systems are only designed
    for 90% reduction. By comparison, another potential pollution
    control, SNCR, achieves 25-75% reductions.21 The SCR
    16
    Memorandum from Roger Strelow, supra n.10.
    17
    E.P.A., State Implementation Plans; General Preamble for
    the Implementation of Title I of the Clean Air Act
    Amendments of 1990; Supplemental, 
    57 Fed. Reg. 18,070
    ,
    18,073 (proposed Apr. 28, 1992) (first introducing RACT as a
    standard to regulate emissions from existing sources).
    18
    
    Id.
    19
    
    Id.
    20
    JA108.
    21
    
    Id.
    7
    process involves injecting a substance such as ammonia or urea
    as a catalyst into the post-combustion flue gas. That causes the
    harmful NOx pollution to break down into its component
    nitrogen and water molecules and to be dispersed as vapor.22
    Plants that are unable to meet specified pollution limits
    are allowed to average their emissions with other plants of the
    same owner or operator to comply with the limits.23 A plant is
    also permitted under section 129.99 of the RACT II Rule to
    request an alternative RACT limit that applies only to its own
    facility. These are called “source-specific RACT limits.”24 It is
    important to stress that the overall RACT limit is therefore not
    a hard cap if certain older plants could not otherwise satisfy the
    requirements.
    B. The Pennsylvania proposal
    Pennsylvania’s SIP was submitted in May 2016.25 It
    proposed that NOx emission limits for SCR-equipped coal-
    fired power plants operating at 600 degrees or higher be
    required to emit less than 0.12 pounds of NOx per million
    British Thermal Units (MMBtu) (a unit of heat).26 Yet when
    SCR-equipped boilers operate below 600 degrees Fahrenheit,
    a higher limit between 0/16 and 0.4 lb/MMBtu limit would be
    permitted.27 Pennsylvania had originally not offered any
    reduced limit on SCR-equipped facilities but added these
    provisions (and an accompanying temperature exception,
    discussed below) in response to critical feedback from the
    EPA, which suggested a 0.12 lb/MMBtu limit for such plants.28
    In March 2018, the EPA provisionally approved
    Pennsylvania’s proposal.29 A number of groups opposed it.
    They expressed particular concern over the regulations for
    22
    E.P.A., Air Pollution Control Technology Fact Sheet at 3,
    https://www3.epa.gov/ttncatc1/dir1/fscr.pdf.
    23
    JA39. The other plants must be subject to the same NOx
    presumptive limits and be at the same facility or within the
    same nonattainment area. 
    Id.
    24
    
    Id.
    25
    JA26.
    26
    
    25 Pa. Code § 129.97
    (g)(1)(viii).
    27
    
    Id.
     § 129.97(g)(1)(vi)(C).
    28
    JA330.
    29
    JA26.
    8
    coal-fired combustion plants, which generate a large
    percentage of Pennsylvania’s emissions.30 Surrounding states
    opposed Pennsylvania’s proposal and submitted negative
    public comments during the ensuing comment period.
    For instance, New York took issue with the EPA’s
    claim that the Pennsylvania limits were “comparable to NOx
    emission limitations in other states.”31 New York compared its
    own limits to those proposed by Pennsylvania and argued that
    its limits were “more stringent nearly across the board.”32
    According to New York, Pennsylvania was proposing limits
    that were “comparable to limits that were effective in New
    York from 1995 to 2014.”33
    New York argued that the EPA’s finding that
    Pennsylvania’s proposal was comparable to New York and
    Connecticut ignored the fact that both states are partway
    through a multi-year emission limit reduction plan. Additional
    substantial reductions are already completed in Connecticut
    and expected in New York by 2023.34 Finally, New York
    criticized the relatively lenient 30-day averaging of pollution
    proposed by Pennsylvania. New York, by contrast, requires a
    24-hour average, which the state claimed prevents emitters
    from relying on the month-long average to balance out days
    when they do not use pollution control mechanisms.35
    Maryland also opposed Pennsylvania’s proposal. It
    argued: “all of Pennsylvania’s coal-fired boilers equipped with
    SCR have demonstrated that they can achieve NOx emission
    rates far below 0.12 lb/MMBtu, including hours where the
    SCR inlet temperature may be lower than 600 F.”36 Maryland
    noted that the proposed limits were “on average, nearly 60%
    higher than what they have achieved in the past.”37 The
    submission included data showing that all of Pennsylvania’s
    30
    JA47-52.
    31
    JA351.
    32
    JA352.
    33
    Id.
    34
    Id.
    35
    Id.
    36
    JA354.
    37
    JA355.
    9
    powerplants were capable of emissions lower than the
    proposed limit, down to as low as 0.043 lb/MMBtu.38
    C. Petitioner’s critiques
    The Sierra Club offered even more pointed criticism. It
    submitted lengthy comments against the proposal. In
    particular, the Sierra Club used data from the EPA’s own
    sources to demonstrate that at least some of Pennsylvania’s
    plants have historically been able to emit NOx at 50% of the
    proposed limit.39 The Sierra Club also argued that other
    neighboring states have not received a temperature threshold
    for their plants. It is therefore not at all clear why it is
    appropriate or reasonable for Pennsylvania to be given one.
    This is particularly important because the Sierra Club provided
    evidence regarding at least one Pennsylvania plant showing
    that it has operated below the proposed limit (600 degrees) in
    the past.40 Lastly, the Sierra Club complained that the
    Pennsylvania proposal failed to require plants to report their
    temperature records. There would be no reliable method for
    monitoring whether plants were even complying with these
    emission limits. Consequently, a plant could explain away
    higher emissions by claiming it was operating below 600
    degrees at the time, without having to substantiate that claim.
    The combination of this lack of mandatory reporting and the
    temperature waiver created a potent loophole for polluters to
    walk through.41
    Despite these concerns, the EPA eventually gave the
    Pennsylvania plan formal approval in 2019.42 Within the 60-
    day statutory window to contest that decision, the Sierra Club
    filed the present petition for review.43
    38
    Id.
    39
    JA336-41.
    40
    JA341-45 (modeling temperatures and pollution at the
    Cheswick plant). This data suggests that Cheswick would be
    able to evade the 0.12 lb/MMBtu limit by operating at lower
    temperatures, as it has in the past.
    41
    JA345-46
    42
    JA7.
    43
    See 
    84 Fed. Reg. 20,274
     (May 9, 2019) (JA007); 
    42 U.S.C. § 7607
    (b)(1).
    10
    II.
    The EPA had jurisdiction to review and approve
    Pennsylvania’s SIP.44 The SIP is “locally or regionally
    applicable” within the Third Circuit.45 Accordingly, we have
    jurisdiction to review the EPA’s final approval of the
    Commonwealth’s proposal.
    A. Standard of Review
    We accord due deference to a final EPA action, in light
    of the agency’s expertise. We will remand only when the
    agency’s decision “[i]s ‘arbitrary, capricious, an abuse of
    discretion, or otherwise not in accordance with the law.’”46
    But to receive such deference, the agency cannot reach
    whatever conclusion it likes and then defend it with vague
    allusions to its own expertise; instead, the agency must support
    its conclusion with demonstrable reasoning based on the facts
    in the record. When it fails to do so, an agency action is
    arbitrary and capricious—examples include the agency
    “entirely fail[ing] to consider an important aspect of the
    problem.”47 Or offering only a “conclusory statement” which
    “failed to articulate a rational basis for its conclusion.”
    48 Or 44
    JA028; see 
    42 U.S.C. §§ 7502
    , 7410(k).
    45
    
    42 U.S.C. § 7607
    (b)(1).
    46
    Nat’l Parks Conservation Ass’n v. E.P.A., 
    803 F.3d 151
    ,
    158 (3d Cir. 2015) (internal citation omitted). This is
    illustrated, for example, in Berks County v. E.P.A., an
    unpublished 2015 decision, where we denied a petition for
    review because we found the EPA had considered all the data
    and made a reasonable decision. 
    619 Fed. Appx. 179
    , 184 (3d
    Cir. 2015).
    47
    Motor Vehicle Mfrs. Ass’n v. State Farm Mut. Auto. Ins.
    Co., 
    463 U.S. 29
    , 43 (1983); see, e.g., E. Bay Sanctuary
    Covenant v. Barr, 
    964 F.3d 832
    , 857 (9th Cir. 2020)
    (remanding an agency action because it “entirely failed to
    consider an important aspect of the problem”) (citing 
    463 U.S. at 43
    ); Sierra Club, Inc. v. U.S. Forest Serv., 
    897 F.3d 582
    , 605 (4th Cir. 2018) (same); Genuine Parts Co. v. E.P.A.,
    
    890 F.3d 304
    , 341 (D.C. Cir. 2018) (same).
    48
    W.R. Grace & Co. v. E.P.A., 
    261 F.3d 330
    , 342 (3d Cir.
    2001).
    11
    indeed if “cit[ing] no data whatsoever in support of its
    decision.”49 We have held that “courts are ‘not obliged to stand
    aside and rubber-stamp their affirmance of administrative
    decisions that they deem inconsistent with a statutory mandate
    or that frustrate the congressional policy underlying a
    statute.’”50
    However, we cannot “substitute [our] judgment for that
    of the agency.”51 We will therefore defer to the agency’s
    expertise if we can discern “a rational connection between the
    facts found and the choice made.”52 Even when “[a]n agency
    has engaged in line-drawing determinations” we review
    primarily to determine whether the result is “‘patently
    unreasonable’ or run[s] counter to the evidence before the
    agency.”53
    B. Standing
    We briefly consider standing, and find it is satisfied
    here. The Sierra Club asserts that it holds standing based on the
    interests of its members.54 It describes the particular impact
    increased pollution will have on the health and enjoyment of
    its members, who prioritize outdoor activities (and,
    presumably, good health). To have standing in an
    environmental case, “[t]he relevant showing . . . is not injury
    to the environment but injury to the plaintiff.”55 Standing exists
    where “(1) [the plaintiff] . . . has suffered an ‘injury in fact’
    that is (a) concrete and particularized and (b) actual or
    imminent, not conjectural or hypothetical; (2) the injury is
    49
    Natural Res. Def. Council, Inc. v. E.P.A., 
    790 F.2d 289
    ,
    309 (3d Cir. 1986).
    50
    Pa. Fed’n of Sportsmen’s Clubs, Inc. v. Kempthorne, 
    497 F.3d 337
    , 347 (3d Cir. 2007) (quoting Nat’l Labor Relations
    Bd. v. Brown, 
    380 U.S. 278
    , 291 (1965)).
    51
    Motor Vehicle Mfrs. Ass’n, 
    463 U.S. at 43
    .
    52
    Prometheus Radio Project v. FCC, 
    373 F.3d 372
    , 389–90
    (3d Cir. 2004), as amended (June 3, 2016) (quoting State
    Farm Mut. Auto. Ins. Co., 
    463 U.S. at 43
    ).
    53
    Id. at 390 (quoting Sinclair Broad. Grp. Inc. v. FCC, 
    284 F.3d 148
    , 162 (D.C. Cir. 2002).
    54
    Pet. Br. at 16-18.
    55
    Friends of the Earth, Inc. v. Laidlaw Envtl. Servs. (TOC),
    Inc., 
    528 U.S. 167
    , 181 (2000).
    12
    fairly traceable to the challenged action of the defendant; and
    (3) it is likely, as opposed to merely speculative, that the injury
    will be redressed by a favorable decision.”56 When, as here, an
    association sues on behalf of its membership, it may do so
    when “the interests at stake are germane to the organization's
    purpose, and neither the claim asserted nor the relief requested
    requires the participation of individual members in the
    lawsuit.”57
    The Sierra Club’s claim here is sufficiently
    particularized to satisfy the Supreme Court’s test from
    Laidlaw.58 There, the Court accepted statements similar to the
    ones Petitioner makes here that the direct interests of its
    members were at stake. The Court held that this imminent harm
    was distinct from the theoretical future interest in visiting
    remote destinations threatened with environmental harm
    rejected as insufficient in Lujan.59 The Sierra Club describes
    the negative impact of ozone pollution on the health and
    recreational opportunities available to its members.60 When, as
    here, the complaint demonstrates such an actual harm that will
    “directly affect[] those affiants’ recreational, aesthetic, and
    economic interests,” standing is satisfied.61
    III.
    The Sierra Club challenges the EPA’s approval of the
    Pennsylvania SIP on the three grounds described in their public
    comments above. We consider each in turn: the selection of a
    0.12 NOx/MMBtu pollution limit instead of a stricter
    requirement; the 600-degree temperature threshold; and the
    56
    
    Id. at 180-81
    .
    57
    
    Id.
     at 181 (citing Hunt v. Washington State Apple Advert.
    Comm'n, 
    432 U.S. 333
    , 343 (1977)).
    58
    
    528 U.S. at
    183–84.
    59
    Lujan v. Defs. of Wildlife, 
    504 U.S. 555
    , 564 (1992).
    60
    JA332 (discussing health consequences which particularly
    affect children and those with asthma, and environmental
    damage including visible injury to trees and vegetation).
    61
    
    528 U.S. at 184
    . Cf. Sierra Club v. E.P.A., 
    754 F.3d 995
    ,
    1000-02 (D.C. Cir. 2014) (finding the Sierra Club lacked
    standing when it alleged merely speculative harm from
    potential future changes to clear air policies).
    13
    lack of a reporting requirement on power plant inlet
    temperatures. While the reasoning supporting each element is
    questionable individually, joined together they are decidedly
    worse than the sum of their parts. Working in tandem, these
    three elements spawn a pernicious loophole centered on the
    600-degree threshold, one which would easily permit operators
    to exceed the 0.12 NOx/MMBtu cap.
    A. 0.12 NOx/MMBtu Limit
    The Pennsylvania Department of Environmental
    Protection (DEP) has freely admitted that its requested 0.12
    NOx/MMBtu pollution limit “is not the lowest achievable
    emissions rate (LAER) for this technology.”62 It correctly notes
    that satisfying RACT is not the same thing as meeting LAER.63
    This is obvious since LAER applies only to new sources,
    whereas RACT is a technology forcing standard for existing
    polluters.64 Thus, DEP ignores the obvious possibility that an
    existing source can both fall short of LAER and also fail to
    satisfy RACT.
    Nonetheless, the EPA and DEP argue that RACT need
    not be the best possible emissions limit; it must be the best limit
    that is also economically and technically achievable for plant
    operators. That general proposition is certainly correct.
    However, it fails when applied to this record. Here, DEP itself
    analyzed the pollution reduction capabilities of five different
    plants and found that three are already capable of meeting the
    0.12 lb/MMBtu limits with little or no changes.65 One, Homer
    City, has committed to updating its SCR systems to meet that
    standard.66 But a fourth, Cheswick, is currently operating well
    above the limit (from 0.21 to 0.22 lb/MMBtu).67 0.12
    lb/MMBtu was selected as it represents the average pollution
    output of the three plants that are already compliant over the
    62
    JA247
    63
    
    Id.
    64
    Nat. Resources Def. Council, Inc. v. Thomas, 
    838 F.2d 1224
    , 1236 (D.C. Cir. 1988) (noting that LAER applies to
    “new sources in nonattainment areas” whereas RACT
    governs “existing sources in nonattainment areas”).
    65
    JA325.
    66
    
    Id.
    67
    JA329.
    14
    last five years.68 Suffice to say, an average of the current
    emissions being generated by existing systems, will not usually
    be sufficient to satisfy the RACT standard.69
    The EPA seeks to explain this 0.12 lb/MMBtu choice
    by observing that power needs fluctuate between winter and
    summer, so using data from only part of the year is not
    necessarily reflective of the overall pollution possible on a
    yearly basis. But this simply ignores the fact that by averaging
    current emissions over five years, the 0.12 lb/MMBtu result
    already accounts for seasonal variation.70
    Moreover, the DEP data in the record appears to simply
    average the net emissions rate of each plant, without
    considering these facilities’ widely varying actual output of
    pollutants. For example, the DEP data purports to give Unit 2
    at both Keystone (0.09 lb/MMBtu emissions) and Montour
    (0.15 lb/MMBtu) equal weighting in its average (for a 30-day
    average of 0.12 lb/MMBtu between the two Units). But this
    record suggests that Keystone Unit 2 is normally operating at
    higher capacity than Montour Unit 2—in June 2003, for
    instance, Keystone Unit 2 produced 128.8 tons of NOx
    emissions, whereas Montour Unit 2 produced only 74.9 tons.71
    An equal weighting thus distorts the real-world emissions
    rates. In the example above, the correctly weighted average
    rate would have been 0.112 lb/MMBtu, below the
    unconvincing 0.12 lb/MMBtu proposal.
    Even more disquieting, the EPA ignores its own Air
    Markets Program Data showing that all five Pennsylvania
    power plants noted above have actually achieved much greater
    reductions than 0.12 NOx/MMBtu during at least 150 months
    that Sierra Club identifies (0.07 NOx/MMBtu or lower average
    in those months).72 The EPA claims that other unspecified
    plants may perform worse than these five, but since this
    68
    
    Id.
    69
    See Navistar Intern. Transp. Corp., 
    941 F.2d at 1343
    (reiterating that RACT must be, “the lowest emission
    limitation that a particular source is capable of meeting . . .”).
    70
    JA329.
    71
    JA337.
    72
    JA336-39.
    15
    limitation applies only to SCR-equipped plants, the five plants
    under consideration (Cheswick, Montour, Keystone, Bruce
    Mansfield, and Homer City) represent the vast majority of the
    regulated sites in Pennsylvania.73 The EPA fails to specify
    whether or how these five units are unrepresentative of
    Pennsylvania’s SCR-equipped coal-fired plants, aside from
    speculating that they may be.
    Indeed, the EPA itself acknowledges that the
    Pennsylvania plants are capable of achieving better than 0.12
    NOx/MMBtu. It relies on data from 2011, 2015, and 2016,
    when, on average, the year-long average pollution across
    Pennsylvania power plants equipped with SCR was 0.107
    lb/MMBtu.74 While the EPA uses this data to support the
    marginally higher 0.12 lb/MMBtu standard, it provides no
    explanation for setting RACT at a higher level than 0.107
    lb/MMBtu. It has already been shown that the latter lower
    level is readily achievable on a yearly average.
    Furthermore, the Sierra Club demonstrates that the
    EPA’s baseline data from 2011, 2015, and 2016 is not
    necessarily reflective of the lowest possible emissions
    averages. In fact, those non-consecutive chosen years happen
    to be among the highest emission years in recent records.75 For
    example, the modern Seward Generation plant in New
    Florence, PA, which touts itself as the largest waste coal-fired
    power plant in the world and was constructed in 2004,76
    produced its highest emission count on record in 2011, and
    nearly equaled that in 2015.77 By contrast, emissions were
    lower by 10% in 2012 (0.092 instead of 0.103 in 2011).78
    73
    Or. Arg. Trans. at 5-6. Only 14% of large Pennsylvania
    plants are equipped with SCR. JA053. It was also revealed at
    oral argument that this limited number has since fallen
    further: the Bruce Mansfield plant was decommissioned in
    2019. Or. Arg. Trans. at 5.
    74
    JA55 n.26.
    75
    JA340.
    76
    Seward Generation Power Plant,
    https://robindale.energy/seward-generation.
    77
    JA340.
    78
    
    Id.
    16
    Further, as the Sierra Club stresses, those levels were
    achieved voluntarily. They were not the result of any coercive
    RACT regulations. Absent any regulatory pressure
    whatsoever, the plants were able to operate at limits lower than
    those in Pennsylvania’s SIP.79 Thus, the EPA’s apparent
    supposition that a limit below 0.12 lb/MMBtu was not feasible
    is not only unsupported by this record, but actually
    contradicted by it.
    The agency suggests that even if the five plants cited
    could achieve a lower limit, other unspecified facilities might
    not be able to. This position is even less compelling after it was
    revealed at oral argument that the SCR-equipped facilities in
    the state are primarily just the five plants under consideration.80
    The EPA’s reliance on the supposed inability of other unnamed
    facilities, for which it has provided no data or details, is not
    helpful. Moreover, even if a RACT standard was unachievable
    by a particular older plant, that plant could take advantage of a
    “source-specific RACT,” which eases requirements for a
    specific facility that is unable to comply with pollution
    controls.81
    Finally, the EPA’s technical report on the Pennsylvania
    plan for emissions does provide some guidance for expected
    emissions levels from coal-fired boilers.82 However, the fact
    that this data is based on EPA research dating back to 1994 is
    relegated to an accompanying footnote, in which the agency
    quietly concedes, “[i]t is possible that further technological
    advancements may have been proven to result in lower NOx
    emissions levels than those reported [here].”83
    Given this concession by the agency that technological
    advances may allow for a more environmentally friendly
    standard than the one approved here, reliance on a study that is
    more than two-and-a-half decades old is neither a persuasive
    79
    
    Id.
    80
    Or. Arg. Trans. at 5-6.
    81
    JA68-69.
    82
    JA55.
    83
    Id. at n.25.
    17
    nor reasonable basis for adopting the higher standard it is
    urging upon us now.84
    Indeed, nearby states’ data demonstrate the progress
    that has been made in the intervening decades. Maryland, for
    instance, requires its SCR-equipped plants to adhere to a 0.09
    NOx/MMBtu limit.85 Further, as noted above, Maryland
    indicates that 0.12 NOx/MMBtu was “nearly 60% higher” than
    the emissions limits Pennsylvania’s plants “have achieved in
    the past.”86
    In response, the EPA contends that Pennsylvania’s
    limits are comparable to some other states. They may be, and
    we do not suggest that Pennsylvania must achieve the absolute
    lowest level of emissions that is technologically possible for
    the approved limit to satisfy RACT. However, there is an
    important distinction that the agency’s assertion ignores. The
    0.12 lb/MMBtu limit applies only to SCR-equipped plants—in
    other words, the most technologically advanced facilities in the
    state.87 It is therefore simply not reasonable for the EPA to
    attempt to justify its approval of the higher rate by comparing
    it with the general pollution limitations in other states. Those
    other states are regulating all their coal-fired plants, rather than
    separating out their SCR-equipped facilities. This makes
    Pennsylvania’s lax limit of 0.12 lb/MMBtu even less justifiable
    on this record. As we have noted, those more efficient limits
    apply only to a handful of its most modern plants. Thus, the
    EPA’s approval of Pennsylvania’s more lenient pollution
    levels cannot be explained by the fact that that some
    unspecified “other” plants may not be able to attain a lower
    limit. This is particularly true since the EPA’s own data clearly
    establishes the contrary for nearly all of the SCR-equipped
    plants in the state, and does so in years with higher than
    average emissions.
    As the EPA’s own data demonstrates, a lower limit
    would clearly have been practicable. We realize, of course, that
    “mere” data alone may not require a particular result. However,
    84
    JA10.
    85
    JA355.
    86
    Id.
    87
    JA56.
    18
    the EPA is able neither to offer a reasonable justification for
    failing to require a stricter standard, nor to justify the standard
    it endorsed. That standard represents a mere acceptance of the
    status quo.
    We confronted a similar challenge to the EPA’s
    approval of a Pennsylvania SIP proposal (brought by a
    coalition of environmental groups, including the Sierra Club)
    in 2015.88 There, the underlying data and analysis conducted
    by Pennsylvania were clearly inadequate. For example, it
    failed to consider the cumulative impact of multiple pollution
    sources, rather than each individually. Moreover, even the EPA
    agreed that the Commonwealth’s proposal contained “systemic
    deficiencies,” and that it should have done more review.89
    Despite that very appropriate concession, the agency
    nevertheless approved the SIP. It claimed to have done so
    because the conclusions were at worst harmless error.90
    However, approving the proposed standards without sufficient
    explanation was not “harmless” at all. Rather, we stated that
    the error of relying on an admittedly faulty analysis was a
    bridge too far: “the EPA has identified a host of problems with
    Pennsylvania's . . . analysis. What it has not done, however, is
    provide a sufficient explanation as to why it overlooked these
    problems and approved Pennsylvania’s SIP.”91 Despite the
    deference appropriately afforded to agency determinations of
    these issues, we found it necessary to remand “[b]ecause we,
    as a reviewing court, need an agency to show its work before
    we can accept its conclusions . . .”92
    The same is true here. Without any basis for justifying
    their selection of a 0.12 lb/MMBtu limit, and evidence that a
    lower limit was practicable, Respondents struggle to meet even
    our deferential standard of review. Our deference to agency
    expertise is not a blank check allowing the EPA to act
    arbitrarily, nor does it give the EPA a pass on having to explain
    how its proposed limit is consistent with RACT. Here, the EPA
    88
    Natl. Parks Conserv. Ass’n v. E.P.A., 
    803 F.3d 151
     (3d Cir.
    2015).
    89
    
    Id.
     at 165–67.
    90
    
    Id. at 166
    .
    91
    
    Id. at 167
    .
    92
    
    Id.
    19
    has failed to adequately explain how and why a 0.12 lb/MMBtu
    limit is permissible on this record. Although it has offered
    vague allusions to the inability of unspecified plants to meet a
    lower standard, the agency has deprived us of the ability to
    review its decision by “show[ing] its work.”
    93 B. 600
    -Degree          Temperature
    Threshold
    The EPA’s questionable reasoning above is badly
    compounded by the agency’s simultaneous approval of a 600-
    degree threshold for SCR use. Taken together with the
    reporting requirements discussed below, approval of this
    regulatory scheme was clearly arbitrary and capricious.
    Regarding the threshold, neither the EPA nor DEP can
    explain why it is necessary at all. It is not a common
    exemption.94 Moreover, even assuming such a temperature
    threshold were reasonable, the record does not support the
    conclusion that 600 degrees Fahrenheit is the proper limit. The
    EPA instead justifies the threshold by observing that SCR
    controls become increasingly less effective at lower
    temperatures.95 We readily accept that nearly all chemical
    reactions occur with greater efficiency at higher
    temperatures—this is merely a rote application of the
    Arrhenius Equation (which maintains that “reaction rates
    depend on the . . . temperature”96). However, that does not by
    itself justify approving higher pollution levels at cooler
    operating temperatures because the ill effects of running SCR
    at low temperatures are temporary or easily reversed.97 For
    instance, a higher rate of accumulation of the by-products
    (primarily ammonium bisulfate) at an increased rate at lower
    temperatures can be readily reversed when the plant reverts to
    93
    
    Id.
     The agency should also have specifically analyzed the
    30-day averaging that Pennsylvania’s plan entails, which is
    longer than neighboring states’ averaging periods during the
    ozone season. Response Br. at 32-33.
    94
    According to the record, other states have not requested
    such a threshold in their SIP proposals. JA343.
    95
    JA54.
    96
    JOHN C. KOTZ, ET AL., CHEMISTRY AND CHEMICAL
    REACTIVITY 696 (7th ed. 2009).
    97
    JA343.
    20
    normal temperatures during the daytime, or peak hours.98 The
    EPA failed to consider the practicality of such remedies to
    lower temperature operation, despite the fact, as discussed
    below, that SCR controls routinely operate well below 600
    degrees.99
    DEP attempts to bolster the threshold requirement by
    pointing to a response it gave during the public comment
    period.100 However, while that response defends a 600-degree
    limit, it is contrary to record evidence. The DEP had noted:
    “While the minimum operating temperature
    varies depending on the type of SCR system,
    typically for the SCR to function at its target
    efficiency rate and optimize the control of NOx
    emissions, the temperature of the EGU flue gas
    entering the SCR must be no less than 600°F.
    When the EGU flue gas temperature falls below
    600°F, less efficient NOx emission reduction
    occurs along with increased ammonia slip and
    increased potential for air heater fouling leading
    to unscheduled outages.”101
    The EPA relies upon this to argue “Pennsylvania
    determined that SCR controls cannot result in lower NOx
    emissions rates when operating below the temperature
    threshold.”102 Yet, in making that argument, the EPA fails to
    explain why operating below SCR’s “target efficiency rate,” as
    DEP explains would occur below 600 degrees, means that SCR
    “cannot result in lower NOx emissions.” That inference simply
    does not naturally follow, and the agency impermissibly leaves
    us to puzzle how it reached its conclusion.
    98
    
    Id.
     (citing EPRI, Investigation of Catalyst Deactivation
    from Operation Below the Minimum Operating Temperature
    (abstract), Product ID: 1023928 (Sept. 11, 2012), available at
    https://www.epri.com/#/pages/product/1023928 [link not
    functioning as of 8/14/2020]).
    99
    JA343.
    100
    Intervener Br. at 36-39.
    101
    JA283.
    102
    Response Br. at 35 (emphasis added).
    21
    Further, the EPA’s own data demonstrates that DEP’s
    argument, and the EPA’s conclusion drawn from it, is
    incorrect. In selecting the threshold, the EPA found 600
    degrees “consistent with the technical limitations of the
    SCR.”103 However, in the same paragraph of its technical
    summary, the EPA admitted that SCR-based pollution controls
    are used in temperatures ranging from 315 to 698 degrees
    Fahrenheit.104 The EPA also offers a telling chart of SCR peak
    performance, which demonstrates that as the reaction
    temperature declines, the efficiency of NOx reduction also
    falls.105 Although it can operate at over 90% efficiency at 750
    degrees, SCR is reduced to approximately 77.5% efficiency at
    600 degrees and 62.5% efficiency at 550 degrees.106 Yet the
    agency fails to provide any information to support the crucial
    next step of its reasoning: whether and how 77.5% compares
    in terms of economic and technical viability with 62.5%
    efficiency.
    This Court is left to infer that whereas 77.5% efficiency
    is worthwhile, at 62.5% efficiency, SCR controls no longer add
    value to the pollution control process. That is not an impossible
    conclusion, but there is simply no evidence, nor any effort to
    supply such evidence, in this record. Further, it directly
    contradicts the agency’s suggestion that operating below 600
    degrees “cannot result in lower NOx emissions rates.”107
    While this shows that efficiency does decline at an
    increasing rate below 600 degrees, the same is true of 650
    degrees, and 700 degrees. Further, nothing in the record before
    the agency demonstrates why the 65% efficiency at 550
    degrees is inadequate or, in the terms of RACT, economically
    or technologically infeasible. The EPA’s case here boils down
    to showing that there is a general connection between setting a
    limit and lower NOx removal efficiency, which the record
    103
    JA54.
    104
    
    Id.
    105
    Response Br. at 37 (SCR System NOx Removal Versus
    Temperature).
    106
    
    Id.
    107
    
    Id. at 35
    .
    22
    supports, and then asking us to trust their “technical judgment”
    as to the proper limit.108
    While we defer to the agency’s expertise, the agency’s
    decisions must nevertheless be rational and supported by
    record evidence. Here, we have only the agency’s generic
    reliance on “technical expertise” to explain why 600 degrees
    was a rational cutoff.109 Indeed, the EPA points out that 700-
    750 degrees is the optimal operating temperature of most SCR
    systems.110 Thus, if the 600-degree cutoff was based upon the
    efficiencies of higher burning temperatures, the selection of the
    cooler, and thus less efficient and “dirtier,” 600 degree
    threshold would be an even greater mystery. We would have
    hoped (and the law requires) that the agency would rely upon
    its technical expertise to justify and explain this decision, not
    to simply adopt it via ipse dixit authority. Moreover, it is
    important to note that the agency neglected to explain why
    Pennsylvania in particular needs this threshold when
    neighboring states have neither requested nor received one.
    Without explaining specifically how the 600-degree threshold
    satisfies RACT, the EPA accepted an arbitrary number
    proposed by PA DEP.
    108
    
    Id. at 38
    . By contrast, the EPA is on firmer ground when
    the Sierra Club argues that the EPA has failed to justify the
    30-day averaging period by which emissions are produced.
    Pet. Br. at 33. There the EPA did consider the averaging
    periods in neighboring states and concluded that
    Pennsylvania’s fleet of coal-fired plants and the nature of the
    shorter averaging periods elsewhere (active only during ozone
    season) did not require a change to the Pennsylvania SIP.
    Response Br. at 32; JA77-78. This careful analysis further
    highlights how much was missing from the agency’s
    approach elsewhere.
    109
    Response Br. at 38.
    110
    JA11 (“NOx removal efficiency decreases more drastically
    when temperatures are lower than the optimal operating
    range,” which is “700° to 750° F”). The EPA also explicitly
    acknowledged in its final approval of the plan that “neither
    Pennsylvania nor EPA explained in detail why the minimum
    SCR temperature exemption . . . is adequate for RACT.” 
    Id.
    23
    The U.S. Court of Appeals for the D.C. Circuit has
    previously rejected similar efforts by the EPA to proceed based
    upon such a “hunch.”111 In Sierra Club v. EPA, the EPA set a
    130-ppm carbon monoxide output limit on certain industrial
    boilers. That was based on data suggesting that at less than 150
    ppm, the benefits of further reduction were negligible.112
    However, the data the EPA used to reach that determination
    had previously been rejected by the agency as unreliable. Thus,
    the D.C. Circuit Court held that the proposed standard was
    arbitrary and capricious. The EPA had relied on the data to
    conclude that further reductions would not be helpful.113 The
    court explained, “even if EPA had grounds to conclude that
    there is some nonzero level of CO emissions that marks a point
    below which [pollution] emissions cannot be further reduced,
    it offered no basis for identifying 130 ppm as that level.”114
    Likewise, here the data could support establishing a
    threshold limit, but the agency fails to establish why 600
    degrees suffices whereas 550 or 650 degrees does not. Unlike
    the flawed underlying data rejected by the D.C. Circuit, here
    the EPA fails to provide any uncontradicted data to support its
    threshold limitation. Instead, it summarily concludes that were
    Pennsylvania to institute a 550-degree limit, “additional NOx
    reductions achieved would be marginal.”115
    The threshold limit also serves to eviscerate the
    agency’s already flawed 0.12 lb NOx/MMBtu limit discussed
    above. The evidence here establishes that since the SIP took
    effect, the Cheswick power plant appears to have begun using
    the 600-degree threshold as a loophole to avoid activating its
    SCR system.116 At night, when power needs drop and the plant
    can lower its operating temperatures, the emissions at
    Cheswick have not declined commensurately. On the contrary,
    they have soared—up to 0.35 lb NOx/MMBtu—more than
    111
    Sierra Club v. E.P.A., 
    884 F.3d 1185
    , 1198 (D.C. Cir.
    2018).
    112
    
    Id. at 1195
    .
    113
    
    Id.
     (“EPA relied on the same data it had elsewhere
    decisively characterized as untrustworthy.”).
    114
    
    Id. at 1197
    .
    115
    Response Br. at 39; JA12.
    116
    JA342.
    24
    three times the daytime emissions. This resulted in the entire
    plant’s average emissions substantially increasing.117 We
    cannot ignore the fact that this increase in polluting emissions
    only started after implementation of the new SIP standard was
    approved.118
    The EPA tacitly acknowledged the existence of this
    loophole in its public response to the Sierra Club’s concerns.
    There, the agency tried to argue that there was no loophole
    because the plants must still cap emissions at the higher, non-
    SCR limit of 0.16 lb-0.40 lb/MMBtu, depending on the boiler
    type.119 The EPA argues it is thus not a “loophole” but the
    natural result of plants lowering their operating temperatures at
    night due to reduced demand.120 That argument collapses upon
    itself. It is, in fact, describing what can only be defined as a
    loophole. It should be obvious that a loophole need not allow
    plants to escape regulation entirely. Nevertheless, the
    temperature threshold permits plants to evade the lower SCR
    limits of the Pennsylvania SIP. The EPA is thus flatly wrong
    when it concludes that allegations the threshold constitutes a
    loophole are not “supported in theory or fact.”121 Thus, we
    must remand in order for the agency to develop an alternative
    federal implementation plan.122
    C. Reporting Requirement
    Finally, the proposed SIP’s lack of a tangible reporting
    requirement is the third chink in the plan’s shoddy enforcement
    regime. In light of this deficiency, the temperature threshold
    exception discussed above becomes even more pernicious.
    Operating in tandem, these two provisions allow plant
    operators to evade the above 0.12 lb/MMBtu limits altogether
    by claiming they were operating below 600 degrees at the time
    117
    JA341-42.
    118
    See JA344-45 (demonstrating the pattern of increased
    emissions after SIP was implemented).
    119
    JA12.
    120
    Response Br. at 45.
    121
    
    Id.
     Because these increased emissions started only after the
    SIP’s temperature threshold was implemented, it cannot be
    explained as a necessary byproduct of the plant’s nighttime
    operations.
    122
    
    42 U.S.C. § 7410
    (c)(1).
    25
    non-compliant emissions occurred. Without a record of inlet
    temperature data at the time of emission, it will be impossible
    to ascertain the validity of such claims.
    The EPA and DEP reject the notion that the reporting
    requirements are insufficient. First, they point to the text of the
    Pennsylvania SIP as incorporated into state regulation. They
    then add many additional details regarding how this statutory
    oversight will be enforced that are nowhere to be found in the
    administrative record. “Our review must . . . be based on ‘the
    administrative record [that was] already in existence’ before
    the agency, not ‘some new record made initially in the
    reviewing court’ or ‘post-hoc rationalizations’ made after the
    disputed action.”123 Thus, we can only consider the record as it
    existed when the EPA issued its approval, not any post-hoc
    justifications offered during the course of litigation.
    This means that support for the reporting regime must
    be found in the following requirements, which existed at the
    time of approval:
    The owner and operator of an air contamination
    source subject to this section . . . shall keep
    records to demonstrate compliance [with the
    SIP] in the following manner: (1) The records
    must include sufficient data and calculations
    to demonstrate that the requirements [of the SIP]
    are met. (2) Data or information required to
    determine compliance shall be recorded and
    maintained in a time frame consistent with the
    averaging period of the requirement.124
    The parties dispute whether this SIP requirement that
    “sufficient data and calculations” be recorded suffices to
    ensure that temperature records are preserved. The Sierra Club
    argues these terms are too vague to be enforceable. We agree.
    The Clean Air Act requires: “plan provisions shall include
    enforceable emission limitations, and such other control
    123
    Christ the King Manor, Inc. v. Sec. U.S. Dept. of Health
    and Human Servs., 
    730 F.3d 291
    , 305 (3d Cir. 2013) (quoting
    Rite Aid of Pa., Inc. v. Houstoun, 
    171 F.3d 842
    , 851 (3d
    Cir.1999)) (alterations in original).
    124
    
    25 Pa. Code § 129.100
    (d) (emphasis added).
    26
    measures, means or techniques . . . as well as schedules and
    timetables for compliance, as may be necessary or appropriate
    . . .”125 The Sierra Club also complains that the law allows only
    Pennsylvania to request the records;126 there is no provision for
    public insight into how the plants are operating, and therefore
    no way for interested members of the public, or more crucially,
    the EPA itself, to conduct oversight. The absence of public
    access conditions enforcement upon a given administration’s
    approach to enforcement of state environmental regulations.
    We do not believe that Congress intended to tether
    enforcement of the Clean Air Act safeguards to the vicissitudes
    of those political winds.
    In New York v. E.P.A., the Court of Appeals for the D.C.
    Circuit held that when a state proposal’s lack of sufficient
    reporting requirements prevents the EPA from conducting
    enforcement, a given regime thereby fails to comply with the
    Clean Air Act.127 There, the EPA had approved a reporting
    regime that depended on operators self-identifying a
    “reasonable possibility” of experiencing an increase in
    pollution. It left to the independent hands of those operators
    whether, in fact, they matched that description.128 Therefore,
    operators could freely increase their emissions and avoid
    detection simply by claiming that there was no “reasonable
    possibility” that they would have experienced an increase in
    pollution output. The court concluded, “EPA needs to explain
    how its recordkeeping and reporting requirements allow it to
    identify [noncompliant] sources.”129 The court also expressed
    doubt about an enforcement regime’s “reliance on state
    programs to establish minimum recordkeeping and reporting
    standards.”130 Effective regulation must not depend on the
    candor or veracity of the very entities being regulated. This is
    particularly true under the CAA, where past practices of
    125
    
    42 U.S.C. § 7502
    (c)(6); see also Reply Br. at 26 (citing
    cases that rejected the EPA’s approval of unenforceable
    standards).
    126
    
    25 Pa. Code § 129.100
    (i).
    127
    
    413 F.3d 3
    , 35 (D.C. Cir. 2005).
    128
    
    Id. at 34-35
    .
    129
    
    Id. at 34
    .
    130
    
    Id. at 35
    .
    27
    weighing economic factors have historically counseled against
    complete compliance.
    The same logic applies here. First, the EPA delegated
    recordkeeping entirely to Pennsylvania’s internal procedures.
    This delegation failed to require that records be available to the
    public. Instead, it deferred what “data and calculations” were
    “sufficient” for compliance to an operator’s interpretation.
    That “honor code” approach might suffice in the type of world
    we aspire to. But it is more than a little fanciful to base a
    regulatory enforcement regime upon such aspirations.131
    Instead, of requiring a regime that could readily be “policed”
    and enforced, the EPA endorsed an emissions regime with no
    discernible enforcement mechanism.
    Anticipating this weakness in their initial argument, the
    EPA and PA DEP offer a second line of justification. They
    provide more detail on how the regulatory regime will
    function. In its brief, DEP explains that under existing state
    law, plant operators are required by their operating permits to
    record their operating temperatures and to preserve those
    records for inspection on request of the state.132 In particular,
    the state describes how “major polluters” under Title V (a
    category into which, it assures us, all of the power plants under
    review fall) are required to maintain and make available
    suitable records for public review and inspection. The Title V
    permitting process itself incorporates the RACT requirements,
    131
    Moreover, in that aspirational world, regulation would not
    be needed to ensure compliance. As the Founders recognized:
    “[i]f men were angels, no government would be necessary.”
    THE FEDERALIST No. 51 (James Madison). Instead, by
    requiring a regime that could readily be “policed” and
    enforced, the EPA endorsed an emissions regime with no
    discernible enforcement mechanism.
    132
    Intervener Br. at 54-55 (citing 
    25 Pa. Code §§ 127.12
    (a)(3), 127.411(a)(4)(i), 127.12b(c), 127.441,
    127.442). Intervenors direct us to this and other parts of their
    brief in their 28j letter, but the further explication they
    provide was not present in the record when the EPA issued its
    approval of the SIP and is therefore not a valid basis for our
    decision making now. DEP 28(j) letter dated May 28, 2020 at
    1-2.
    28
    by making each permitee agree to keep suitable records to
    enforce RACT provisions.133
    Yet at oral argument, PA DEP conceded that
    temperature inlet data is not specifically included in these Title
    V permit records. Instead, counsel posited that “if the
    Department has the temperature data, citizens are able to
    review that.”134 This tacitly demonstrates that the data’s
    availability is speculative and dependent on the regulated
    entity. Counsel explicitly confirmed this point. When asked,
    “so sometimes it’ll be there, and sometimes it won’t,” PA
    DEP’s counsel admitted, “that’s correct, Your Honor . . . if the
    Department has records, they must be made available to the
    public.”135 It is a strange regulatory system indeed that is based
    on the good faith of the regulated entity to keep records which
    may be prejudicial to its operation and profitability.
    Moreover, the recordkeeping provisions cited by PA
    DEP do not require operators to record temperature inlet
    data.136 Intervenor’s own record cites reveal that PA DEP
    actively removed prior requirements for “how the records shall
    be maintained or in what format[.]”137 And even if this were
    not the case and the EPA had actually submitted after-the-fact
    evidence that temperature inlets were recorded, we would be
    unable to consider that evidence. We are bound to examine
    only “the administrative record . . . already in existence before
    the agency, not some new record made initially in the
    reviewing court or post-hoc rationalizations made after the
    disputed action.”138
    Because the SIP’s 600-degree threshold necessarily
    depends upon accurate temperature reporting, the EPA’s
    approval of such inadequate requirements on this record was
    133
    Intervener Br. at 55 (citing 25 Pa. Code Chapter 127,
    Subchapter G).
    134
    Oral Arg. Trans. at 42-43 (emphasis added).
    135
    
    Id. at 43
     (emphasis added).
    136
    Intervenor’s Br. at 12, 56 (citing 
    25 Pa. Code §§ 127.511
    ,
    129.100).
    137
    JA289.
    138
    Christ the King Manor, Inc., 730 F.3d at 305 (internal
    quotation marks and citation omitted).
    29
    arbitrary and capricious. Thus, the reporting requirements must
    be vacated.
    IV.
    Therefore, we will grant the petition for review and
    vacate the agency’s action on each of these three provisions of
    the Pennsylvania plan. On remand, the agency must either
    approve a revised, compliant SIP within two years or formulate
    a new federal implementation plan.139 That proposal must be
    technology forcing, in accord with the agency’s RACT
    standard,140 and lack the gaping loophole found in the
    enforcement regime proposed here.
    139
    
    42 U.S.C. § 7410
    (c)(1) (providing two years for the EPA
    to promulgate a federal implementation plan in such
    circumstances); E.P.A. v. EME Homer City Generation, L.P.,
    
    572 U.S. 489
    , 498 (2014) (acknowledging this timeline).
    140
    Memorandum from Roger Strelow, supra n.10, at 2.
    30