The opinion of the court was delivered by: Easterbrook, Chief Judge.
Petitions for Review of Orders of the Environmental Protection Agency
Before EASTERBROOK, Chief Judge, and EVANS and WILLIAMS, Circuit Judges.
Two sections of the Clean Air Act provide that neither national nor state officials may make any changes that cause air quality to deteriorate in parts of the country that have yet to attain the required standard. See sections 110(l) and 193, codified at 42 U.S.C. §§ 7410(l), 7515. See also Environmental Defense v. Duke Energy Corp., 549 U.S. 561 (2007), which discusses the new source review (NSR) program against the background of the prevention-of-significant-deterioration (PSD or anti-backsliding) requirement. In 2002 the Environmental Protection Agency changed the rules that determine when polluters need permits in order to modify existing facilities-and, if they need permits, what restrictions they carry. 67 Fed. Reg. 80,186 (Dec. 31, 2002). These new rules were challenged as violations of §§ 7410(l) and 7515, among other statutes, but the D.C. Circuit concluded that the new rules are rational and consistent with the Act. New York v. EPA, 413 F.3d 3 (D.C. Cir. 2005). Along the way, the court deemed unripe an argument that the agency's new approach actually would lead to more emissions. 413 F.3d at 43. The EPA's models project that the new approach will have neutral or beneficial effects on aggregate emissions; whether that is true, or instead backsliding occurs, depends on data rather than lawyers' arguments, the court stated.
Businesses have argued that the new approach is too strict rather than too lax. Our decision in United States v. Cinergy Corp., 623 F.3d 455 (7th Cir. 2010), illustrates how one aspect of the current approach substantially limited the allowable emissions from aging coal-fired power plants in need of major repairs-so much so that a district court ordered an entire plant shut down and emissions from other plants reduced. We reversed that decision, holding that the new approach does not apply to the repair work in question, but our opinion and the district court's decision in Cinergy show how the new approach can limit air pollution.
The Natural Resources Defense Council and the Sierra Club, however, are convinced that the 2002 revisions will make pollution worse. Wisconsin's latest implementation plan includes features from the 2002 federal regulations. The EPA approved Wisconsin's plan. 72 Fed. Reg. 19,829 (Apr. 20, 2007); 73 Fed. Reg. 76,560 (Dec. 17, 2008); 75 Fed. Reg. 10,415 (Mar. 8, 2010) (denying petition for reconsideration). The NRDC and the Sierra Club have filed petitions for review, which repeat arguments that failed to persuade the D.C. Circuit in New York. This proceeding is the first challenge to a specific state plan that implements the 2002 changes; one other is pending in the Sixth Circuit, and perhaps more are in the works.
Three aspects of the 2002 revisions were contested in New York and again here. First, the 2002 approach deter-mines whether a modification requires a permit (and, if so, what controls are necessary) by comparing actual emissions in the past with projected actual emissions in the future-rather than by comparing the maximum potential emissions before the modification against maximum potential emissions after. The 2002 rules also change the period in which these are compared, asking about emissions over the course of a year rather than emissions per hour. (The old approach favored repairs that increased the number of hours a plant could operate while leaving emissions per hour unchanged; the new approach treats more hours at the same level per hour just like it treats holding constant the number of hours but emitting more per hour.) It was the actual-to-projected-actual comparison that affected the electric utility in Cinergy.
Second, the 2002 approach allows polluters to select two years from a ten-year baseline to measure their past emissions; the former approach used a shorter baseline. Third, the 2002 approach allows a plantwide applicability limitations (PAL) calculation, similar to the "bubble" sustained in Chevron U.S.A. Inc. v. Natural Resources Defense Council, Inc., 467 U.S. 837 (1984). Under PAL a polluter can choose to treat its entire plant as the source of emissions, which allows it to increase emissions from one machine or process as long as it makes offsetting reductions elsewhere in the plant. Under the PAL program, a polluter needs a permit to make modifications or repairs only when they will increase emissions from the plant as a whole.
The NRDC and the Sierra Club observe that all three of these changes could lead to more pollution. Some, such as the actual-to-projected-actual comparison, could do this by making new permits so onerous that a firm will choose to run an old plant into the ground without repairs, avoiding a need to get a permit for modifications, even though old plants generally are dirtier than new ones (or newly repaired ones). Others, such as the ten-year baseline, could allow firms to choose their two highest-emissions years from the decade, making it seem as if their pollution has gone down even though today's emissions are higher than the average from the decade (or higher than the amount computed under the shorter baseline under the former rules). Even the plantwide limit could increase pollution, if emitters can use reductions they would have made anyway to offset increases elsewhere in a plant, increases that would have required their own permits under pre-2002 rules.
These are not new arguments. They were presented to the D.C. Circuit in New York, and that court sustained the rules. See New York, 413 F.3d at 21--38. The D.C. Circuit observed that the EPA was well aware of these possibilities and took them into account when evaluating the net effects of the 2002 revisions. The agency's models projected that the 2002 changes would either reduce pollution or have no net effect. As the D.C. Circuit saw things, scientific estimates, plus the presumption that an agency's estimates are rational, see FCC v. National Citizens Committee for Broadcasting, 436 U.S. 775 (1978), beat lawyers' conjectures.
In the current litigation, NRDC and the Sierra Club observe that the EPA is relying on the same models that it used in 2002 and has no better reason now than it did then to think that a concrete plan, such as Wisconsin's, will curtail emissions. But if the EPA is in the same position as 2002 (and 2005), so are petitioners. The lineup is still: models on one side, lawyers' talk on the other. As in 2002 and 2005, the models supply substantial evidence for the EPA's decision and show that it is neither arbitrary nor capricious.
At oral argument, counsel for petitioners and the EPA told us that about half of the states have changed their implementation plans to track the 2002 revisions. We asked what effect these changes have had on aggre-gate emissions: neither side knew, and we have not found any published studies on the subject. Counsel did know, and debated the significance of, how many permits have been issued in several of the states under one or another feature of the 2002 approach. But the statutes concern the quantity of emissions, not the quantity of permits.
Petitioners rely on a prediction that the staff of Wisconsin's Department of Natural Resources made in 2003, a prediction that the state agency has since repudiated. They complain that the EPA did not analyze this prediction adequately. But the way to test the EPA's models is not to compare them with someone else's ballpark numbers. (The state agency's staff did not have a model-at least, did not reveal one-but made what seems to have been a back-of-the-envelope calculation. The 2003 prediction comprises eight PowerPoint slides that give conclusions without any supporting rationale, plus two pages of tables that do not even hint at the method of their derivation.) The way to test a model is to compare its projection against real outcomes. Alternatively one might validate a model by "retrodiction"-using the model to "predict" past events. The two-in-ten rule, for example, might allow a business to increase average emissions, but does it? So far, we have no answer to that question, either from actual experience in adopting states or through efforts to test a model by retrodiction.
The experience in the states that have implemented the 2002 revisions may vindicate the EPA or may refute it; as long as the judiciary remains behind the veil of ignorance, it must accept the EPA's projections. An agency that clings to predictions rather than performing readily available tests may run into trouble. See Bechtel v. FCC, 10 F.3d 875 (D.C. Cir. 1993). At some point, preferring predictions over facts is no longer rational. But challengers who fail to put data in the administrative record-likely because ...