Opinion by Judge O’SCANNLAIN.
Concurrence by Judge BERZON.
OPINION
O’SCANNLAIN, Circuit Judge:
We must decide whether the Environmental Protection Agency’s regional haze regulations for the State of Montana lawfully prescribe emission limits at certain power plants.
I
Petitioner PPL Montana operates and partially owns coal-fired and hydroelectric power plants in Montana, including the Colstrip' Steam Electric Generating Station (“Colstrip”) and the J.E. Corette Steam Electric Station" (“Corette”). Petitioners National Parks Conservation Association, Montana Environmental Information Center, and Sierra Club (collectively, “NPCA”) are nonprofit conservation organizations whose members enjoy wilderness areas impacted by EPA’s regional haze regulations for the State of Montana. Both petitioners are dissatisfied with such regulations. PPL Montana argues, in essence, that they are too stringent; NPCA argues, to the contrary, that they do not do enough to remedy visibility impairment caused by regional haze in various relevant wilderness areas.
A
Regional haze is “visibility impairment caused by geographically dispersed sources emitting fine particles and their precursors into, the air.” Am. Corn Growers Ass’n v. EPA 291 F.3d 1, 3 (D.C.Cir.2002) (per curiam) (citing Regional Haze Regulations, 64 Fed.Reg. 35,714 (July 1, 1999) (codified at 40 C.F.R. Pt. 51)). Congress enacted §§ 169A and 169B of the Clean Air Act (the “CAA” or the “Act”) to address the problem of regional haze. Id. at 3-4; see Clean Air Act Amendments of 1977, Pub.L. No. 95-95, § 128, 91 Stat. 685, 742 (current version at 42 U.S.C. § 7491); Clean Air Act Amendments, Pub L. No. 101-549, § 816, 104 Stat. 2695 (1990) (current version at 42 U.S.C. § 7492). These provisions establish as a national goal the “prevention of any future, and the remedying of any existing, impairment of visibility in mandatory Class I areas which impairment results from man-made air pollution.”1 42 U.S.C. § 7491(a)(1).
The Act imposes several requirements on States and on EPA relevant to this case. First, the Act requires EPA to promulgate regulations to “assure ... reasonable progress toward meeting the national goal” of regional haze reduction. 42 U.S.C. § 7491(a)(4). Second, the Act invites each State to submit to EPA a “State Implementation Plan” (“SIP”) setting forth emission limits and other measures necessary to make reasonable progress toward the national visibility goal. See 42 U.S.C. §§ 7410(a), 7491(b)(2). If, like Montana, a State chooses not to submit such a plan, the Act requires EPA to [1139]*1139produce a “Federal Implementation Plan” (“FIP”) for that State. See 42 U.S.C. § 7410(c)(1)(A).
The Act further provides that all implementation plans must require installation of the “best available retrofit technology” (“BART”) to reduce emissions from certain emission sources that were operational between 1962 and 1977 (“BART-eligible sources”). See 42 U.S.C. § 7491(b)(2), (g). Five statutory factors determine which type of emissions-reducing technology constitutes BART for such sources:
(a) the costs of compliance;
(b)'the energy and non-air quality environmental impacts of compliance;
(c) any existing pollution control technology at a source;
(d) the remaining useful life of the emission source; and
(e) the degree of visibility improvement anticipated!.]
42 U.S.C. § 7491(g)(2).
Pursuant to the Act, EPA promulgated its Regional Haze Regulations (the “Regulations”), which asked certain States, including Montana, to analyze sources of emissions within the State and to develop a plan to eliminate all man-made visibility impacts by 2064. See 64 Fed.Reg. at 35, 714; 40 C.F.R. § 51.308. The Regulations require any implementation plan to include (1) “reasonable, progress goals”; (2) a calculation of baseline and natural visibility conditions; (3) a long-term strategy for achieving “reasonable progress goals”; and (4) additional monitoring of emission sources in Class I federal areas. See 40 C.F.R. § 51.308(d)(1)-(4). After the D.C. Circuit vacated the provisions of the Regulations relating to BART determinations, see Am. Corn Growers, 291 F.3d at 6, EPA promulgated new BART regulations in its Regional Haze Regulations and Guidelines for Best Available Retrofit Technology (BART) Determinations (the “2005 Regulations”), which revised the text of the earlier Regulations. See 70 Fed.Reg. 39,-104 (July 6, 2005).
EPA also published its Guidelines for BART Determinations Under the Regional Haze Rule (the “Guidelines”), 40 C.F.R. Pt. 51, App. Y (Sept. 6, 2005), prescribing five steps for application of the five statutory BART factors:
(Step 1) Identify all available retrofit control technologies;
(Step 2) Eliminate technically infeasible options;
(Step 3) Evaluate the control effectiveness of remaining control technologies;
(Step 4) Evaluate impacts (identified in § 7491(g)(2), see 40 C.F.R. Pt. 51, App. Y, § IV.D.4, 70 Fed.Reg. at 39, 166) and document the results;
(Step 5) Evaluate visibility impacts (measured in “deciviews,” see 40 C.F.R. § 51.301).2
Id. App. Y § IV.D.
B
In 2006, the Montana Department of Environmental Quality notified EPA that it did not intend to produce a SIP triggering EPA’s obligation to produce a FIP for the State of Montana: See 42 U.S.C. § 7410(c)(1)(A). EPA published a proposed FIP for Montana on April 20, 2012 (the “Proposed Rule”). See Approval and Promulgation of Implementation Plans; State of Montana, 77 Fed.Reg. 23,988 (Apr. 20, 2012).
[1140]*1140The Proposed Rule required petitioner PPL Montana to take various actions to reduce emissions of two compounds — nitrogen oxide (“NOx”) and sulfur dioxide (“S02”) — at two power plants it partially owns and operates, Colstrip and Corette. First, PPL Montana was required to implement several new technologies at Units 1 and 2 of the four-unit Colstrip station. To reduce NOx emissions to a 30-day rolling average of 0.15 lb/mmBtu, the Proposed Rule required PPL Montana to install two new technologies — separated overfire air (“SOFA”) and selective non-catalytic reduction (“SNCR”) at Colstrip Units 1 and 2. Id. at 24,027, 24,035. To reduce S02 emissions to a.30-day rolling average of 0.08 lb/mmBtu, the Proposed Rule required PPL Montana to implement two additional new technologies at Colstrip Units 1 and 2 — lime injection and a fourth “scrubber.” Id. at 24,028, 24,035. The Proposed Rule did not require PPL Montana to implement new technologies at Colstrip Units 3 and 4.
Second, the Proposed Rule required PPL Montana to limit NOx and S02 emissions at the Corette station. The Proposed Rule imposed 30-day average rolling emission limits of 0.40 lb/mmBtu for NOx and 0.70 lb/mmBtu for S02. See id. at 24,042, 24,046. It required PPL Montana to achieve such emissions using current technology; unlike at Colstrip Units 1 and 2, the Proposed Rule does not require installation of new technology at Corette. Id. at 24,043, 24,047.
Both PPL Montana and NPCA commented on the Proposed Rule. At the conclusion of the notice and comment period, EPA issued its final FIP for Montana on September 18, 2012. See Approval and Promulgation of Implementation Plans; State of Montana (the “Final Rule” or the “Rule”), 77 Fed.Reg. 57,864 (Sept. 18, 2012) (codified at 40 C.F.R. § 52.1396). The Final Rule implemented the Proposed Rule in almost all respects relevant to this appeal.3
C
PPL Montana and NPCA both filed petitions for review of EPA’s BART determinations at Colstrip and Corette, which petitions were consolidated for purposes of this appeal. PPL Montana contends that the emissions limits set forth in the Rule for NOx and S02 at Colstrip Units 1 and 2 — as well as the BART determinations underlying such limits — are essentially too stringent. It argues that EPA failed reasonably to explain the Rule’s BART selections and that EPA’s visibility model does not reasonably anticipate visibility improvement as a result of the Rule’s requirements. PPL Montana also challenges the emissions set forth in the Rule for Corette.
NPCA also challenges the Rule’s emissions limits and BART determinations at Colstrip Units 1 and 2 and Corette, contending essentially that such limits are not stringent enough. According to NPCA, EPA’s decision not to require installation of more advanced technology at these locations was unexplained, arbitrary, and capricious. Moreover, according to NPCA, EPA’s decision not to require installation of any new technologies at Colstrip Units 3 and 4 fails to satisfy the requirement that, the Rule make reasonable progress toward visibility improvement.
D
EPA’s implementation of the regional haze plan is governed by Section 307(d) of [1141]*1141the Act, 42 U.S.C. § 7607(d). The statute requires that EPA explain the basis for its decisions, including underlying factual bases, methods of analysis, and legal and policy considerations. Id. § 7607(d)(6)(A). EPA must also respond to the comments, criticism, and new data submitted during the comment period. Id.
When we review an agency action “involv[ing] primarily issues of fact,” and where “analysis of the relevant documents ‘requires a high level of technical expertise,’ we must defer to ‘the informed discretion of the responsible federal agencies.’ ” Marsh v. Oregon Natural Res. Council, 490 U.S. 360, 377, 109 S.Ct. 1851, 104 L.Ed.2d 377 (1989) (quoting Kleppe v. Sierra Club, 427 U.S. 390, 412, 96 S.Ct. 2718, 49 L.Ed.2d 576 (1976)). However, we do not defer to EPA actions that are “arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law,” or that exceed EPA’s statutory jurisdiction. 42 U.S.C. §§ 7607(D)(9)(A), (C). EPA’s actions must be reasoned; EPA acts in an arbitrary and capricious manner if it fails to consider an important aspect of a decision or if its explanation contradicts the evidence before it. Motor Vehicle Mfrs. Ass’n of U.S., Inc. v. State Farm Mut. Auto. Ins. Co., 463 U.S. 29, 43, 103 S.Ct. 2856, 77 L.Ed.2d 443 (1983). EPA’s actions must also be consistent; an internally inconsistent analysis is arbitrary and capricious. Gen. Chem. Corp. v. United States, 817 F.2d 844, 857 (D.C.Cir.1987) (per curiam).
II
PPL Montana and NPCA each contest the Rule’s BART determinations and prescribed limits for NOx and S02 emissions at Colstrip Units 1 and 2 — and, for similar reasons, Corette.
The heart of the dispute about EPA’s BART selection at Colstrip Units 1 and 2 is its determination that use of SNCR in addition to SOFA is cost-effective, and that use of selective catalytic reduction (“SCR”) — a more aggressive technology— in addition to SOFA is not. Both parties urge that EPA’s cost-effectiveness analysis fails appropriately to consider the costs of compliance and degree of visibility impairment, as required by the Act, see 42 U.S.C. § 7491(g)(2), at Step Three of EPA’s BART analysis under the Guidelines, see 40 C.F.R. Pt. 51, App. Y § IV.D.
The Rule requires PPL Montana to reduce NOx emissions at Colstrip Units 1 and 2 to 0.15 lb/mmBtu over a 30-day rolling average. See 40 C.F.R.. § 52.1396(c)(1). EPA concluded that such reduction could be achieved by installing both SOFA and SNCR technologies. See Final Rule, 77 Fed.Reg. at 57,866. NPCA contends that EPA offered insufficient justification for its rejection of SCR as BART in favor of SOFA and SNCR together, and argues that PPL Montana should be required to install SCR instead of SNCR, and in addition to SOFA, at all four Col-strip units. PPL Montana makes a related point, but on the other side of the argument, contending that EPA offered insufficient justification for requiring SOFA and SNCR as BART instead of SOFA alone — a remedy PPL Montana does not oppose. EPA responds that the BART determination is discretionary and that, it considered all of the relevant factors in reaching its conclusions.
The parties also challenge several more minor aspects of EPA’s cost-effectiveness analysis. PPL Montana argues that EPA’s use of the. dollar-per-ton metric for balancing cost and visibility benefit was improper. NPCA, for its part, maintains [1142]*1142that EPA underestimated the emissions baseline at Colstrip Units 1 and 2, and that it miscalculated the cost of implementing SCR at these units.
EPA identified the costs of the various technologies for NOx reduction at Colstrip Units 1 and 2 as follows:
[[Image here]]
See Proposed Rule, Tables 66-74, 89-97, 77 Fed.Reg. at 24,024-27, 24,032-34.
The Rule offers essentially no reasoning behind EPA’s selection of SOFA and SNCR together i as opposed either to SOFA alone or SOFA and SCR together, as BART to reduce NOx emissions at Col-strip Units 1 and 2. With respect to the decision to require SNCR in addition to SOFA, EPA asserted in the Proposed Rule only that the cost of SOFA and SNCR together “is justified when the visibility improvement is considered.” See Proposed Rule, 77 Fed.Reg. at 24,027. In response to PPL Montana’s comment requesting that EPA explain this assertion, EPA noted that the Proposed Rule discloses the cost for both SOFA alone and SOFA combined with SNCR, and that it “selected SNCR as BART in consideration of these costs, all of which were presented to the public in our proposed rule.” Final Rule, 77 Fed.Reg. at 57,886. But PPL Montana’s objection is not that EPA failed to disclose the cost differential between implementing SOFA alone and SOFA along with SNCR. It contends -instead that EPA failed to explain the reasons why the improvement achieved through SOFA alone was insufficient, but the improvement achieved through S.OFA and SNCR together was deemed both sufficient and cost-effective.
With respect to SCR, EPA simply asserted that the cost of SOFA and SCR together ($3,195/ton at Colstrip Unit 1) was “not justified by the visibility improvement of .404 deeiviews,” but that “[t]he lower cost of SOFA + SNCR ($1,564/ton) is justified when the visibility improvement [of .264 deeiviews] is considered.” Proposed Rule, 77 Fed.Reg. at 24,027. EPA maintains that it did not use a set cost-effectiveness or improvement threshold to disqualify SCR, and we do not suggest that it must do so. But absent any explanation at all of how EPA determines cost-effectiveness, it is impossible for the Rule’s reader to determine why EPA ruled SOFA, and SNCR in and ruled SCR out.
EPA acknowledged that the Regional Haze Rule does not prevent it from implementing what it called a “bright line” rule for cost-effectiveness, but contends that its regulations do not require it to do so. See Final Rule, 77 Fed.Reg. at 57,872. To be sure, the Act and the Regulations do not specifically require that EPA explain its cost-effectiveness decisions through use of a “bright line” rule. But the law does require EPA to “cogently explain why it has exercised its discretion in a given manner.” See State Farm, 463 U.S. at 48, 103 S.Ct. 2856; Greater Yellowstone Coalition, Inc. v. Servheen, 665 F.3d 1015, 1030 (9th Cir.2011) (requiring “a rational connection [1143]*1143between the data before [the agency] and its conclusion”); Nw. Envtl. Def. Ctr. v. Bonneville Power Admin., 477 F.3d 668, 691 (9th Cir.2007). EPA’s unexplained assertions that the combination of SOFA and SNCR is cost-effective, but that SOFA alone is insufficient, and that the combination of SOFA and SCR is not cost-effective, are unsupported by any explained reasoning. These assertions leave the Rule’s reader wondering what metric, if any, EPA used to determine BART, or if EPA employed no metric, why not. Therefore, we conclude that EPA’s BART determination for NOx emissions at Colstrip Units 1 and 2 is arbitrary and capricious.
EPA’s responses to petitioners’ more minor challenges to its cost-effectiveness analysis make clear that it is capable of the required rational explanation.
First, NPCA challenges EPA’s selection of the years 2008-2010 as the emissions baseline period for calculating BART at Colstrip Units 1 and 2. It contends .that EPA underestimated the potential benefit of using SCR by first underestimating the existing emissions baseline. EPA’s BART Guidelines require it to choose a representative baseline period that supplies “a realistic depiction of anticipated annual emissions for the source.” BART Guidelines, 70 Fed.Reg. at 39,167. EPA chose the period 2008-2010 on the reasoning that Colstrip had installed additional combustion controls in 2007, • reducing its emissions. See id. NPCA argues that EPA should have chosen earlier periods because PPL Montana is not required to maintain the rate of emissions achieved between 2008-2010, describing the changes at Col-strip as “unenforceable.” But EPA offered a reasoned response to NPCA’s comment about its choice — that Colstrip had achieved reduced emissions using technology it has no plans to deactivate — and NPCA has suggested no reason to believe that PPL Montana would change course and remove the additional combustion controls it had already installed. EPA gave a reasoned, rational response to this argument, a response to which a reviewing court defers. See Latino Issues Forum v. EPA, 558 F.3d 936, 941 (9th Cir.2009).
Second, PPL Montana objects that EPA’s reliance on a dollars-per-ton metric to calculate cost effectiveness is inappropriate because it does not actually measure the improvement achieved in visibility. But EPA responded that it has previously defined “cost effectiveness” as “tons of pollutant emissions removed” compared with “annualized control costs.” 2005 Regulations, 70 Fed.Reg. at 39,167. Indeed, PPL Montana essentially abandoned this argument in reply after EPA pointed to its regulation in its response brief.
Third, NPCA objects that EPA improperly calculated the cost of installing SCR at Colstrip Units 1 and 2, leading it to reject SCR as BART. It challenges EPA’s calculation of the capital cost, its choice of interest rate, and the lifespan and maintenance costs assumed in EPA’s calculation. EPA provided a reasoned response to NPCA’s comments on these questions. See Final Rule, 77 Fed.Reg. at 57,888-89. NPCA maintains that EPA should have looked to different data sources to determine these costs. But it offers a reason that EPA should look to different underlying cost data in only one example case. It contends EPA should have chosen a 5% interest rate instead of a 7% interest rate in calculating these costs to comport with the requirements set forth in the EPA Air Pollution Cost Manual. But that manual instructs industry actors that they may wish to use a different interest rate for their own calculations. As EPA explained, see Final Rule, 77 Fed.Reg. at 57,888-89, EPA’s Manual clearly contemplates that EPA will use the interest rate set by the [1144]*1144OMB, as it properly did here, in pursuing its own calculations. See id.
Both petitioners dispute the rationality of EPA’s selection of a fourth scrubber as BART for S02 emissions control at Col-strip Units 1 and 2 for essentially the same reasons they dispute EPA’s NOx BART determinations. EPA determined the costs of the various technologies for reducing S02 emissions at Colstrip as follows:
[[Image here]]
See Proposed Rule, Tables 78-87, 102-111, 77 Fed.Reg. at 24,028-29, 24,036-37.
PPL Montana contends that EPA underestimated the cost of installing a fourth scrubber at Colstrip Units 1 and 2 and failed sufficiently to explain its cost-effectiveness determination for this requirement. NPCA, to the contrary, contends that EPA should have required installation of replacement advanced scrubbers at Col-strip, not just the introduction of an additional scrubber. EPA responds that it appropriately explained its calculation of the costs involved in adding an additional scrubber and that it appropriately decided not to require an entirely new system of scrubbers, as opposed only to adding one additional scrubber.
EPA’s cost-effectiveness analysis with respect to S02 emissions at Colstrip Units 1 and 2 suffers the same defect as its cost-effectiveness with respect to NOx emissions. EPA’s S02 analysis is an improvement in one respect: EPA acknowledged the incremental cost of adding an additional scrubber vessel ($2,410/ton at Colstrip Unit 2, see Final Rule, 77 Fed.Reg. at 57,892), and noted that adding a scrubber vessel would incrementally improve visibility from 0.225 deciview (using lime injection alone) to 0.280 deciview (using lime injection and an additional scrubber vessel). See id. But EPA then concluded that-it “continuéis] to find that the cost is reasonable given the visibility benefits[.]” Id. Although in this case EPA acknowledged the comment concerning incremental cost-effectiveness, it nevertheless still failed to explain what makes a cost reasonable in light of potential visibility benefits.
The Rule’s S02 emission limits for Col-strip Units 1 and 2 force PPL Montana to spend an additional $25,000,000 in capital costs and an additional $2,210,000 annually to achieve a visibility improvement of 0.055 deciviews at the sampled location. See id. This improvement very well may be necessary and ultimately cost-effective. But EPA has supplied no reasons justifying that determination. Based on its explanation, the Rule’s reader is left to wonder what rationale EPA used to determine cost-effectiveness. Again, the law requires a reasoned answer to that question. See State Farm, 463 U.S. at 48, 103 S.Ct. 2856; Servheen, 665 F.3d at 1028; Nw. Envtl. Def. Ctr., 477 F.3d at 691.
By contrast, EPA thoroughly and rationally explained its response to NPCA’s objection on this point. NPCA contends that EPA should have considered requiring complete replacement of the existing [1145]*1145control systems for S02 emissions at Col-strip Units 1 and 2. EPA explained its decision not to do so: its BART Guidelines recommend constructing a new system when a current control system achieves “less than 50 percent removal efficiencies.” Final Rule, 77 Fed.Reg. at 57,892 (quoting 2005 Regulations, 70 Fed.Reg. at 39,171) (internal quotation marks omitted). The current S02 emissions control system at Colstrip Units 1 and 2 achieves removal efficiencies exceeding 50 percent, see Proposed Rule, 77 Fed.Reg. at 24,028, and therefore EPA is not required to consider replacement technology. See Final Rule, 77 Fed.Reg. at 57,892.
This kind of explanation — not elaborate, but rational, and thus adequate — is missing from EPA’s conclusion that PPL Montana must install a fourth scrubber at Colstrip Units 1 and 2. EPA’s determination of BART to control S02 emissions suffers the same failure of rational explanation as its BART determination for NOx emissions. Because the rule offers no reasoned explanation to support its requirement of a fourth scrubber at Colstrip Units 1 and 2, we conclude that such requirement is arbitrary and capricious.
PPL Montana also, relatedly, contends that EPA’s BART determinations at Col-strip Units 1 and 2 are arbitrary and capricious because they are inconsistent with EPA’s Corette analysis, which does not require additional controls at that station, see Final Rule, 77 Fed.Reg. at 57,893. PPL Montana points out that EPA rejected implementation of SOFA at Corette, but found similar costs for SOFA installation justified at Colstrip Units 1 and 2. See supra § II.A.1. NPCA, on the other hand, argues that EPA’s conclusion that additional controls at Corette were not cost-effective was arbitrary and capricious.
The entirety of EPA’s cost-effectiveness reasoning with respect to Corette follows: “We have weighed costs against the anticipated visibility impacts for Cor-ette. Any of the control options would have a positive impact on visibility; however, the cost of controls is not justified by the visibility improvement.” Proposed Rule, 77 Fed.Reg. at 24,043. Again, this reasoning fails to reveal to a reader how EPA determined that the cost of controls were not justified. Moreover, PPL Montana is correct that this reasoning appears inconsistent with EPA’s analysis for Col-strip Units 1 and 2. At those units, EPA concluded that a cost-per-ton rate of approximately $1,500 for NOx emissions controls was justified. See supra § II.A.1. Yet at Corette, EPA concluded that a cost-per-ton rate of $1,487 did not justify the potential emissions reductions. See Proposed Rule, 77 Fed.Reg. at 24,040-43. Corette, moreover, is closer to Class I areas than Colstrip, and emissions controls there could have improved visibility at least as much as they were set to control visibility at Colstrip. See id. The seeming inconsistency in EPA’s BART determinations at Colstrip Units 1 and 2 and Corette is, absent explanation, “the hallmark of arbitrary action.” Sierra Club v. EPA 719 F.2d 436, 459 (D.C.Cir.1983); see also Gen. Chem. Corp. v. United States, 817 F.2d 844, 846 (D.C.Cir.1987) (per curiam) (finding analysis arbitrary and capricious because it was “internally inconsistent and inadequately explained”).
Finally, PPL Montana objects, on two grounds, to EPA’s use of the CALPUFF4 [1146]*1146visibility model in determining BART at Colstrip Units 1 and 2. PPL Montana first objects that the maximum potential incremental visibility benefit of SNCR is below the range- of perceptibility and falls within the model’s margin of error, meaning such improvement cannot be “reasonably ... anticipated” as required by the Act. See 42 U.S.C. § 7491(g)(2). It also challenges the application of CALPUFF to the emissions sources at issue because of their distance from Class I areas, contending the model is inaccurate at such great distances.
EPA responds that, because the CAL-PUFF model was approved in the Guidelines, EPA need not defend its every application. See Final Rule, 77 Fed.Reg. at 57,867. EPA also responds that the Regional Haze Rule itself anticipated and rejected PPL Montana’s argument, reasoning that “[flailing to consider less-than-perceptible contributions to visibility impairment would ignore the CAA’s intent to have BART requirements apply to sources that contribute to, as well as cause, such impairment.” Final Rule, 77 Fed.Reg. at 57,883 (quoting 2005 Regulations, 70 Fed. Reg. at 39,129). EPA further responds that it “has acknowledged that there is uncertainty in the CALPUFF model predicted visibility impacts,” and “the CAL-PUFF. model can both underpredict and overpredict visibility impacts,” but that “CALPUFF ... is a reasonable application to determine whether such a facility [with an impact exceeding 2 or 3 deci-views] can reasonably be anticipated to cause or contribute to any impairment of visibility.” Final Rule, 77 Fed.Reg. at 57,-868 (quoting 2005 Regulations, 77 Fed. Reg. 39,123).
EPA fails to understand PPL Montana’s first argument. PPL Montana does not object wholesale to EPA’s use of the CAL-PUFF model for visibility measurement. Instead, it argues that the results the model predicts in this case are too insignificant for the model -to measure, and therefore cannot be reasonably anticipated as the Act requires. During the notice and comment period, PPL Montana argued that the maximum incremental visibility benefit to be gained by installing SNCR at Col-strip Units 1 and 2 would be 0.085 deci-views, an incremental improvement PPL Montana contends falls within the CAL-PUFF model’s margin of error. EPA responded by explaining that its regulation permits visibility improvements to be required even when visibility impacts fall below the threshold of perceptibility. See Final Rule, 77 Fed.Reg. at 57,867. EPA further responded that its 2005 Regulations direct use of the CALPUFF model to estimate the 98th percentile of visibility impairment, rather than the highest data collected, to minimize uncertainty in its calculations. See id. at 57,868. But PPL Montana does not ask EPA to discontinue application of CALPUFF below the one-deciview perceptibility threshold, or to run the model using different data points than the ones selected; it asks how CALPUFF explains EPA’s conclusion that additional measures will lead to reasonable anticipation of visibility improvement in this case, when, PPL asserts, an improvement of 0.085 deeiview is “beyond the CALPUFF model’s ability to predict with any confidence.”
EPA’s response does not meaningfully address PPL Montana’s comment, as it must. See Columbia Falls Aluminum Co. v. EPA 139 F.3d 914, 923 (D.C.Cir.1998); Eagle-Picher Indus., Inc. v. EPA 759 F.2d 905, 922 (D.C.Cir.1985) (“[I]f ... the model is challenged, the agency must provide a full analytical defense.”).5 It is [1147]*1147no answer to respond, as EPA did, that low levels of visibility impairment must be addressed even though they are not perceptible to the human eye, or that measures have been taken to minimize the margin of error. The issue is not the perceptibility of the proposed improvements, but the model’s ability to anticipate improvements at a level allegedly within its margin of error, whether perceptible or not to the human eye. EPA simply offered no response to this objection.
EPA’s only detailed defense of the model addressed PPL Montana’s concern about distance — not the margin of error issue. See Final Rule, 77 Fed.Reg. at 57,867-68. With regard to distance, EPA noted that guidance issued by the Inter-agency Workgroup on Air Quality Modeling “provides for the use of the CALPUFF model at receptor distances of up to 200 to 300 km.” EPA then explained why it believed CALPUFF could be used “cautiously” for distances in that range, even though the puffs would be more dispersed at greater distances, and stated that the model was not suitable for “very long-range transport (300 km and beyond).” Id. at 57,868. EPA thus offered a reasoned response to PPL Montana’s challenge to the use of CALPUFF at the distances in question. That explanation did not, however, also suffice as a reasoned response regarding how CALPUFF could be relied upon to predict an improvement of as little as 0.085 deciviews when PPL offered reasons to think that doing so was outside the model’s capabilities.
The predictability concern is important because the Act requires that any visibility improvement be “reasonably ... anticipated” as a result of BART installation. 42 U.S.C. § 7491(g)(2). As the D.C. Circuit explained in vacating a portion of the Regional Haze Rule itself, it is arbitrary and capricious for EPA to force an emissions source “to spend millions of dollars for new technology that will have no appreciable effect on the haze’ in any Class I area.” Am. Corn Growers Ass’n, 291 F.3d at 7. In response to PPL Montana’s contentions that the Final Rule would do just that, by requiring PPL Montana to install SNCR at Colstrip Units 1 and 2 without sufficient assurance of any improvement at all, EPA has offered no reasoned explanation.
Ill
PPL Montana also challenges the emissions limitations EPA imposed at Corette — namely, the requirement that Corette lower its emissions levels even without installing additional technology. As explained, EPA’s determination that installation of additional technology to control emissions from Corette was not cost-effective suffers the same failure of explanation as its BART determinations at Col-strip. See supra § II.B. But the Rule also tightened the emissions limits identified for Corette in the Proposed Rule. Specifically, in response to comments (including comments made by PPL Montana), EPA altered its rule to establish an emission rate of 0.57 lb/MMbtu for S02 emissions and an emission rate of 0.35 lb/MMBtu for Nox emissions, both monitored on a 30-day rolling average. See Final Rule, 77 Fed.Reg. at 57,893-94. PPL Montana ar[1148]*1148gues that the CAA does not authorize EPA to impose emissions limits without determining BART, but EPA insists that its analysis comported with the BART Guidelines.6
EPA correctly argues that, after it found Corette already had BART technology in place, it was authorized by the Regional Haze Rule to “skip the remaining analyses in this section, including the visibility analysis in step 5.” See 2005 Regulations, 70 Fed.Reg. at 39,165. PPL Montana’s contention that EPA was nevertheless required to proceed with the remaining BART analysis is a challenge to this provision of the Regional Haze Rule itself, not properly asserted in this challenge to the Montana FIP. See 42 U.S.C. § 7607(b)(1) (challenge to rulemaking must be brought within sixty days).
EPA also properly set emissions limits for Corette on a 30-day rolling average. The Regional Haze Rule prescribes such limits. See 2005 Regulations, 70 Fed. Reg. 39,172 (requiring emissions limits to “specify an averaging time of a 30-day rolling average”). EPA noted and addressed PPL Montana’s comment about the 30-day rolling average requirement. See Final Rule, 77 Fed.Reg. at 57,893-94. PPL Montana contended that converting current emissions limits to 30-day rolling averages would require EPA to raise the limits set forth in the Proposed Rule. EPA concluded to the contrary that, by choosing 99th percentile monthly emission rates and adding an additional margin for compliance, the 30-day. rolling averages provided in the rule comported with Corette’s current technological capabilities. See id. EPA’s reasoned disagreement on this topic with PPL Montana’s comment reflects its conclusion on a highly scientific question — the variance in emissions calculations that occurs when annualized rates are translated into thirty-day rolling averages — precisely the kind of question justifying deference to EPA’s discretion. See Nat’l Wildlife Fed’n v. U.S. Army Corps of Eng’rs, 384 F.3d 1163, 1177-78 (9th Cir.2004).
IV
NPCA contends that EPA’s decision not to require any additional emission-reducing technology, let alone installation of SCR, at Colstrip Units 3 and 4 was arbitrary and capricious because it fails to satisfy the Act’s reasonable progress requirements. Colstrip Units 3 and 4 are not subject to BART requirements because they were constructed after 1977. See 42 U.S.C. § 7491(b)(2), (g). Nevertheless, the Act still directs EPA to issue regulations assuring “reasonable progress” toward improving visibility in Class I areas. See 42 U.S.C. § 7491(a)(4). The statute requires EPA to consider several factors in determining reasonable progress: (a) costs of compliance; (b) the time necessary for compliance; (c) the energy and other environmental impacts of compliance; and (d) the remaining useful life of a source subject to reasonable progress requirements. See 42 U.S.C. § 7491(g)(1). EPA’s Regulations prescribe reasonable progress toward attaining natural visibility conditions by 2064. See 40 C.F.R. § 51.308(f).
[1149]*1149EPA responded to comments contending that the visibility benefits to be gained from SCR at Colstrip Units 3 and 4 justify the requirement for reasonable progress purposes. See Final Rule, 77 Fed.Reg. at 57,902-03. It explained that the cost of compliance is only one of the four statutory requirements for reasonable progress analysis. See id. EPA also offered a better explanation for its cost-based decision on this point than it did for its cost-effectiveness determination of BART at Colstrip and Corette. It reasoned that the visibility benefits to be gained from requiring SCR (ranging from 0.273 deciviews to 0.260 deciviews) were not sufficient considering their cost, between $4,574 and $4,607 per ton. See id. EPA contrasted this effectiveness with the reasonable progress goals it implemented in North Dakota at Antelope Valley Station — a location to which NPCA specifically urged comparison. See id. EPA explained that it had found improvements of 0.2 deciviews at each unit of that location to be cost-justified for reasonable progress purposes when the costs of such improvement ranged from $586/ton to $661/ton. See id. EPA’s comparison of the two sources provided NPCA with at least some broad metric for understanding which cost-per-ton ratios EPA will approve and, which it will not — a rational explanation why the reasonable progress options available at Colstrip Units 3 and 4 were not feasible.
V
For the reasons explained, we GRANT IN PART and DENY IN PART the petitions for review, VACATE the portions of the Rule setting emissions limits at Col-strip Units 1 and 2 and Corette and REMAND to EPA for further proceedings consistent with this opinion. Each party shall bear its own costs.7
PETITIONS FOR REVIEW GRANTED IN PART AND DENIED IN PART; VACATED AND REMANDED.