Full opinion text
Opinion for the Court filed by Circuit Judge WILLIAMS. I. Background .......................................................1231 II. Stack Height Validation: Emissions Rate Assumptions in Demonstrations ........................................................1233 A. The Control-First Dispute .......................................1233 B. Demonstrations Supporting Stack Height Increases Within the Formula .......................................................1239 1. Attacks on the formula .....................................1239 2. Attacks on the demonstration procedures ....................1239 C. The NSPS Presumption for Above-Formula Stacks ...............1240 1. Substantive objections ......................................1241 2. Procedural challenges .......................................1242 III. Stack Grandfathering Issues .....................................1243 A. Sheltering pre-October 1, 1983 Within-Formula Stack Increases from the Demonstration Requirement .................................1244 B. Automatic Credit to Formula Height for pre-January 12,1979 Stacks 1246 1. Credit up to 2.5H for pre-1979 sources showing reliance .....1247 2. Credit up to H+1.5L for pre-1979 sources not showing reliance 1248 C. EPA’s Definition of “Stack Height in Existence” ................1248 D. Application of New Demonstration Requirements to Sources that Have Completed Demonstrations ................................1249 IV. Plume Rise .......................................................1251 A. Original Design and Construction as One Stack .................1252 B. General Rule for Merged Stacks ................................1254 C. Partial Grandfathering of Stacks Merged Before July 8, 1985 ____1255 V. Miscellany .......................................................1256 A. Multi-Point Rollback ............................................1256 B. Definition of “Nearby” as Used in Demonstrations ..............1256 C. Modeling Adjustments for Complex Terrain .....................1257 Conclusion .............................................................1257 WILLIAMS, Circuit Judge: Under the Clean Air Act as amended in 1970, 42 U.S.C. §§ 7401 et seq. (1982), the Environmental Protection Agency sets national ambient air quality standards (“NAAQS”) for various pollutants. Id. § 7409. Once they are set, each state must adopt and submit to the EPA a state implementation plan (“SIP”) providing for achievement of the standards in each air quality control region. Id. § 7410(a)(1). Such plans obviously must distribute the necessary pollution cutbacks among the various pollution sources. From 1970 to this day a dispute has raged over the extent to which pollution sources may make their required contribution toward these localized clean air goals by dispersing pollution rather than by reducing their emissions. Dispersion may be either through space or time. A source may disperse its pollution through space by such devices as “tall stacks,” which carry the pollutants away from the region and from the ground levels at which satisfaction of the NAAQS is measured. It may disperse pollution over time by intermittent controls systems (“ICS”), which vary the time of discharges so as to take advantage of changes in weather conditions. Dispersion techniques vary from emission reductions in two fundamental ways. They are, at least up to a point, considerably cheaper than emissions reductions. This makes them attractive to industry and often to the states of origin. (The attraction may be particularly great where the state of origin produces high-sulphur coal.) On the other hand, reliance on such techniques increases the aggregate amounts of pollution dumped into the atmosphere. This makes them unattractive to environmentalists and to the citizens of downwind states, to which the pollution will be swept and where acid rain may result. First the courts and then Congress intervened to prevent states from allowing pollution sources to satisfy their obligations by means of dispersion. As a result, reductions in local ground-level pollution do not “count” toward satisfaction of the NAAQS to the extent that they rely on those dispersion techniques that are disapproved. While these limitations obviously relate to important goals, the system has a certain eccentricity. The ambient air quality standards are ones to be fulfilled in more than 236 specific local areas. The anxiety over dispersion stems primarily from dispersion’s impact outside the region of origin. But the means of allaying that anxiety is to disregard, for purposes of measuring contribution to local clean air, conduct which indisputably helps clean local air. Yet none of the constraints on dispersion, whether devised by courts, EPA or Congress, has forged an operating link between those constraints and the injuries inflicted by dispersion. Thus, although the parties adduce some figures as to changes over time in total atmospheric “loadings” of sulphur dioxide, these are not related to any statutory goal or to any scheme for attaining specific cutbacks. This incongruity may account for some of the logical difficulties encountered in trying to apply the statute and controlling precedents. The battle has proceeded in the agency, the courts and Congress. The latter tried to resolve the matter in 1977 by adding a new provision to the Act, § 123, 42 U.S.C. § 7423 (1982), which has not proved at all free of ambiguity. This court reviewed the EPA’s 1982 “stack height” regulations in Sierra Club v. Environmental Protection Agency, 719 F.2d 436 (D.C.Cir.1983), cert. denied, 468 U.S. 1204, 104 S.Ct. 3571, 82 L.Ed.2d 870 (1984) (“Sierra Club”). We affirmed many aspects of those regulations, invalidated two provisions, and directed the agency to reconsider other provisions on remand. In 1985 the agency promulgated a new set of regulations attempting to respond to Sierra Club. In these consolidated cases, environmental petitioners led by the Natural Resources Defense Council (and supported by an array of northeastern states) and industry petitioners (supported by an array of middle western states) challenge the amended regulations. We regret to say that we cannot find them in full compliance with § 123 as construed in Sierra Club. I. Background This court described the statutory provisions at issue in this case, together with their legislative and administrative history, in Sierra Club, 719 F.2d at 439-43, and in our earlier opinion in Alabama Power Co. v. Costle, 636 F.2d 323, 388-91 (D.C.Cir.1979). Here we confine ourselves to a brief summary. Section 110 of the Act directed the EPA Administrator to approve a SIP if it complied with the applicable procedural requirements and included (among other things) “emissions limitations ... and such other measures as may be necessary to insure attainment” of the NAAQS. 42 U.S.C. § 7410(a)(2)(B). The agency initially approved state plans that authorized the use of “tall stacks” and ICS compliance measures. Several courts found this approach illegal, reading § 110 to establish a hierarchy among control techniques. Under the hierarchial view, “other measures” qualified as “necessary” only to the extent that the SIP had exhausted the “emissions limitations” approach, i.e., only where further compliance through such limitations was “unavailable or infeasible.” NRDC v. EPA, 489 F.2d 390, 410 (5th Cir.1974), rev’d on other issues sub nom., Train v. NRDC, 421 U.S. 60, 95 S.Ct. 1470, 43 L.Ed.2d 731 (1975); Big Rivers Electric Corp. v. EPA, 523 F.2d 16, 21-22 (6th Cir.1975) (involving ICS), cert. denied, 425 U.S. 934, 96 S.Ct. 1663, 48 L.Ed.2d 175 (1976); Kennecott Copper Corp. v. Train, 526 F.2d 1149, 1153-54 (9th Cir.1975) (involving ICS and tall stacks), cert. denied, 425 U.S. 935, 96 S.Ct. 1665, 48 L.Ed.2d 176 (1976). (We refer to these cases collectively as the “NRDC trilogy.”) Round I to the environmentalists. In 1976 EPA issued guidelines seeking to implement the Fifth Circuit’s view of § 110. 41 Fed.Reg. 7450 (February 11, 1976) (“1976 Guidelines”). The guidelines employed a variety of distinctions that continue to haunt the area. First, they focused on the control credit that a source could receive for pollution control through stack height. Thus they did not purport to restrict a source’s actual employment of a tall stack, but simply limited the extent to which the actual reduction in ground-level pollution achieved by such a stack would count towards compliance with the NAAQS. Id. at 7451. Second, they drew a distinction between stacks that were equal to or less than 2lk times the height of the facility (the “2.5H” formula) and stacks that were taller, favoring within-formula stacks. Id. Third, they looked relatively askance at increases in stack height as opposed to stack heights attained in original construction, presumably because the former were more likely than the latter to reflect a purpose to avoid emission reduction costs rather than adherence to conventional engineering practice. Id. Fourth, they employed “grandfathering” concepts both with respect to increases and original construction stacks. For example, they treated the stacks of sources that received construction permits before the Fifth Circuit’s decision more favorably than those of sources initiating construction later. Id. This is not the place to describe the rather complex pattern that emerged from all these elements. We will return to aspects of the 1976 Guidelines as we go along. For the moment, we note that credit for control through stacks was to be unlimited if the source applied “the best available control technology” (“BACT”), even for the least favored vintage of stack and for stacks of increased height. Id. at 7451-52. Stacks initiated before the Fifth Circuit’s NRDC decision were grandfathered, up to whatever figure the 2.5H formula might produce. Id. In 1977 Congress stepped in with § 123, moved at least in part by concern that the 1976 Guidelines were too lax. See Sierra Club, 719 F.2d at 440. The section in essence elaborates on § 110’s references to “other methods,” and, to a degree, subordinates the use of tall stacks and other dispersion techniques to emission controls. It creates a new concept, “good engineering practice” (“GEP”), with a complex definition. The methods and distinctions used in the 1976 guidelines reappear, with important variations. Like the earlier guidelines, § 123 addresses only the issue of credit for pollution reduction through dispersion techniques, explicitly stating that the Administrator is not to prohibit increases or restrict stack height. § 123(c), 42 U.S.C. § 7423(c). Further, it builds in at least some significance for the 2.5H formula, specifying that GEP height must not exceed that formula unless the source owner demonstrates to the Administrator that such greater height is “necessary” as that term is used in the GEP definition. Third, it employs grandfathering, but in a more limited sense than did the 1976 guidelines, protecting only stacks “in existence” or dispersion techniques “implemented” before the 1970 amendments (December 31, 1970). § 123(a), 42 U.S.C. § 7423(a). § 123’s structure differs from that of the 1976 Guidelines in that it draws no explicit distinction between increases in stack height and original construction. Although § 123(c) directed the EPA to issue regulations implementing its provisions by February 7, 1978, they did not emerge in final form until February 8, 1982. 47 Fed.Reg. 5864. The Sierra Club and the Natural Resources Defense Council challenged the regulations in this court under 42 U.S.C. § 7607(b) and prevailed in part. We will not here summarize the holdings of Sierra Club, as a summary would be unnecessary for the cognoscenti, meaningless for others. The important aspects of the decision appear throughout this opinion as we address the many current issues. The upshot was a remand to the agency, with directions to promulgate new final regulations “within six months from the date of issuance of [the] court’s mandate,” 719 F.2d at 470. The mandate issued on July 18, 1984, after the denial of certiorari by the Supreme Court, 468 U.S. 1204, 104 S.Ct. 3571, 82 L.Ed.2d 870. The agency later secured a postponement from the court, and promulgated the final regulations on June 27, 1985. 50 Fed.Reg. 27,892 (1985). Environmental and industry petitioners challenge the revised regulations in this court under 42 U.S.C. § 7607(b)(1), which directs review here for “nationally applicable regulations.” Thus, 17 years after the' 1970 amendments and 10 years after the enactment of § 123, we again address the permissible scope of reliance on dispersion techniques. II. Stack Height Validation: Emissions Rate Assumptions in Demonstrations A. The Control-First Dispute The 1982 Regulations allowed sources to build stacks to formula height and to increase existing stacks to formula height at will. But a source seeking credit for above-formula stacks was required to demonstrate (by fluid modeling or field studies) that the extra height was “necessary” within the meaning of § 123’s definition of GEP (“height necessary to insure that emissions ... do not result in excessive [downwash-induced] concentrations of any air pollutant____”). 47 Fed.Reg. 5865/2 (1982). The agency defined “excessive concentrations” in terms of a “relativist” test: downwash-induced pollutant concentrations were excessive if they exceeded maximum non-downwash concentrations by 40 percent or more. Id. at 5869/1. Though the 40 percent figure is not in itself at issue, a word of explanation is in order. In preparing the 1982 regulations EPA found that downwash increased ground-level pollution concentration levels by about 40 percent where a source’s stack was at formula height (i.e., 2.5H). Sierra Club, 719 F.2d at 446. As Congress had recognized the 2.5H formula as indicative of traditional engineering practice and therefore presumptively sound, the agency reasoned that any downwash-induced pollution increase exceeding what a formula-height stack would normally produce should be regarded as excessive. Id. In Sierra Club, environmental petitioners attacked the relativist test and prevailed. They argued that Congress was content to give credit to stacks only to the extent that their height was necessary to protect human health, so that a downwash-induced concentration could be “excessive” only if it were health-threatening. 719 F.2d at 447. The 40 percent relativist test of course had no direct connection with any health threat. The relativist-absolute dispute appeared to the court in Sierra Club to dissolve into the question whether Congress meant in § 123 to codify a traditional engineering formula or to create a health-and-welfare-based stack height standard. The court found that the statute and its legislative history “disclose[d] sharply conflicting signals,” and concluded, after reviewing the question in detail, that “Congress [probably] thought traditional engineering practice and protection of health were the same thing.” Id. at 448. But the court also found evidence in the legislative history that Congress recognized that a choice between the two standards would be necessary if traditional engineering practice dictated a height greater than that necessary to protect human health. Id. The court concluded that “meeting air quality standards was primary in [Congress’s] mind and that good engineering practice was merely a way to do so.” Id. Finding it unlikely that the 40 percent standard would identify “an absolute pollutant concentration that is dangerous to health,” the court remanded to the agency with instructions to “develop a standard directly responsive to the concern for health and welfare that motivated Congress to establish the down-wash exception.” Id. at 450. The mandate to develop an absolute test revealed an issue that did not exist under the relativist one. Ground-level concentrations are obviously a function not only of stack height and other elements mentioned in § 123’s GEP definition, but also of the emissions emerging at the top of the stack. Once “excessive” concentrations are defined in absolute terms, the stack height “necessary ” to avoid those concentrations on the ground will obviously vary with a source’s actual emissions. Thus Sierra Club opened a gap in § 123’s GEP definition, the gap expressed in the bracketed and emphasized clause below: “height necessary [given a specified emissions level ] to insure that emissions from the stack do not result in excessive concentrations....” Although the parties disagree as to how much the assumed emissions rate affects any computation of credit-worthy stack height, they agree on the direction of the impact: high assumed emissions rates entail relatively generous stack credits (and thus relatively high permissible emissions rates), low assumptions the opposite. EPA’s choice of a baseline emissions rate in the 1985 regulations has varied with the particular contexts presented by the rule-making. The critical decisions have related to the “demonstrations” (consisting of field studies or fluid modeling demonstrations) that the regulations require of sources in some circumstances. First, sources seeking credit for any stack height increase after original construction must (unless the stack is grandfathered) demonstrate compliance with § 123’s GEP definition, even though the height for which credit is sought in mthin EPA’s formula. 40 C.F. R. § 51.1(kk)(2). Here, EPA has specified that the baseline must be “the emission rate specified by any applicable State implementation plan (or, in the absence of such a limit, the actual emissions rate).” Id. NRDC argues that instead the baseline must be that emissions rate which would result from the source’s using all “available methods.” NRDC Brief at 22 (emphasis added). The parties have dubbed NRDC’s contention the “control-first” approach. EPA also requires a demonstration for any stack' height above that resulting from its formula (unless the stack is grandfathered). 40 C.F.R. § 51.1(kk)(1). Here it uses as the baseline the rate provided in the “new source performance standards” (“NSPS”) promulgated for new power plants under § 111 of the Act, 42 U.S.C. § 7411, unless the source can show that the NSPS rate is unfeasible. This conditional NSPS standard is, of course, a variant of control-first; it is vigorously attacked by industry. Finally, the baseline emissions rate is relevant to EPA’s validation of its formula, a matter that is indirectly at issue here. That validation was based in part on studies from various power plants, the cleanest having an emissions rate of 4.65 pounds of sulphur dioxide per million British thermal units (Btu). J.A. 834. This compares with an NSPS emissions limit of 1.2 pounds per million Btu for plants built between 1971 and 1978, 40 C.F.R. § 60.43(a)(2), and thus obviously does not fit the control-first model. In this part we address the general question whether § 123 requires use of control-first and conclude that it does not. Later sections face the baseline emissions rate problem in the specific contexts already mentioned. We first address EPA’s contention that NRDC’s control-first claims are barred by res judicata, as it failed to raise the argument in Sierra Club. Of course where res judicata (claim preclusion) applies, it bars relitigation not only as to all matters which were determined in the previous litigation, but also as to all matters that might have been determined. Tutt v. Doby, 459 F.2d 1195, 1197 (D.C.Cir.1972). Moreover, enforcement of the 60-day time limits imposed by the statute providing for review, 42 U.S.C. § 7607(b)(1), requires that issues raised by an initial set of rules be raised within that time limit, not saved for use against the rules that may emerge from a remand. Neither of these barriers applies here. The issue of the proper baseline emissions rate became ripe only after the Sierra Club court remanded for application of an absolute test. While the environmentalists’ briefs in Sierra Club might have mentioned the issue, they could hardly have induced this court to pass on it before EPA had a chance to do so. They are not, therefore, precluded from raising the matter in the present litigation. On the merits, we start by noting the scope of our review. If, using “traditional tools of statutory construction,” we can discern Congress’s intentions in regard to baseline emissions rates, obviously we must give effect to Congress’s will. NLRB v. United Food & Commercial Workers Union, Local 23, — U.S. -, -, 108 S.Ct. 413, 421, 98 L.Ed.2d 429 (1987) (citations omitted); Chevron U.S.A., Inc. v. NRDC, 467 U.S. 837, 842-43, 104 S.Ct. 2778, 2781-82, 81 L.Ed.2d 694 (1984). If we find that “the statute is silent or ambiguous” on the issue, however, we must defer to the EPA’s interpretation if it is based upon a “permissible” statutory construction. NLRB, — U.S. at -, 108 S.Ct. at 421, quoting Chevron, 467 U.S. at 843, 844, 104 S.Ct. at 2782, 2783. Before evaluating the various appeals to the dictionary and to legislative history, we think it suitable to note that the view which NRDC says Congress adopted carries drastic implications. Control-first would require each source to assume, for purposes of its stack height credit demonstration, the lowest achievable emissions rate, determined on either a generic or an individualized basis. Although the record does not allow us to infer exactly the impact of the baseline emissions rate on the emissions rate that would emerge (after the stack height credit were calculated and then used to determine the permissible emissions), all parties agree that that impact is substantial. Indeed, that is what the fight is all about. If Congress in § 123 prescribed the use of such a baseline emissions rate, with all its implications for ultimate emissions ceilings, it did so in a remarkably cryptic way. In the Clean Air Act Congress has formulated standards for development of emissions limits in a variety of contexts. For example, it provided for generic emissions limits for new sources in § 111, 42 U.S.C. § 7411 (1982), spelling out at length the criteria for calculation of the limits. In a variety of other contexts, it has provided —at length — for determination of the lowest feasible emissions level on a source-by-source basis. See, e.g., 42 U.S.C. §§ 7475(a)(4), 7479(3) (mandating use of the “best available control technology” (“BACT”) for new sources in attainment areas, and defining the parameters of the standard); 42 U.S.C. §§ 7502(b)(2), 7501(3) (requiring “lowest achievable emissions rate” (“LAER”) for new sources in nonat-tainment areas, and defining LAER); 42 U.S.C. § 7502(b)(3) (requiring “reasonably available control technology” (“RACT”) for existing sources in nonattainment areas); 42 U.S.C. § 7491(g) (requiring “best available retrofit technology” (“BART”) for pre-1971 sources impairing visibility in national parks and certain other clean areas). NRDC in essence contends that in § 123 Congress mandated agency pursuit of a similar strategy, but without providing a word of guidance. All NRDC’s arguments must be assessed in light of one’s estimate of the plausibility of any such scenario. We think it quite unlikely. NRDC argues that its control-first interpretation is consistent with the plain meaning of the statutory language and is clearly supported by relevant legislative history. It argues further that even if the statute were ambiguous, EPA’s interpretation is internally inconsistent and therefore invalid. NRDC contends that the control-first approach follows inexorably from the plain meaning of the word “necessary.” Webster’s Third New World Dictionary defines “necessary” to mean essential or “indispensable.” Id. at 1511. It argues that this definition compels a finding that stack height cannot be “necessary” to control ground level pollution unless the source has employed all feasible emission-control alternatives for reaching the desired levels under downwash conditions. NRDC Brief at 22. But courts have frequently interpreted the word “necessary” to mean less than absolutely essential, FTC v. Rockefeller, 591 F.2d 182, 188 (2d Cir.1979); 9 to 5 Organization for Women Office Workers v. Board of Governors, 721 F.2d 1, 10 (1st Cir.1983), and have explicitly found that a measure may be “necessary” even though acceptable alternatives have not been exhausted. In FTC v. Rockefeller, supra, for example, the court said that “the word ‘necessary’ is not always used in its most rigid sense.” 591 F.2d at 188. Specifically, it found that a subpoena could be “necessary” to an FTC investigation even though the Commission had not pursued “reasonably available alternatives.” Id.; cf. Chrisner v. Complete Auto Transit, Inc., 645 F.2d 1251, 1261-62 (6th Cir.1981) (statutory term “reasonably necessary” is not absolute and does not require complete absence of alternatives). As these courts have recognized, meaning varies with context; Webster’s definition by no means tells us that Congress intended that the baseline emissions rate — an issue that surfaced only after this court’s Sierra Club decision — must be premised on the source’s having employed all available emissions controls. NRDC asserts that the legislative history clearly demonstrates Congress’s intent to adopt a control-first posture. The House Committee which drafted § 123 stated that it was “intended to ratify the general thrust, if not the specific holdings, of the three U.S. courts of appeals which have considered the issue of the permissibility of use of intermittent controls, tall stacks, and other dispersion enhancement techniques.” House Report at 91-92. The cases cited by the House Report are, of course, the familiar ones of the NRDC trilogy, and each found that “other measures” for complying with the NAAQS are “necessary,” for purposes of § 110(a)(2)(B) review, only when further limits on emissions were infeasible. We do not believe, however, that legislative history endorsing the “general thrust” of the NRDC trilogy demonstrates that Congress in § 123 itself clearly resolved the exact degree of hierarchy appropriate in ranking control above dispersion. Just as the Sierra Club court did not anticipate the present issue, we see no evidence that the courts involved in the NRDC trilogy anticipated it. Once one recognizes —as Congress indisputably recognized in § 123 — that dispersion through stacks can play a quite legitimate role in protecting health and welfare from downwash, it is not by any means obvious that in calculating the appropriate role of stacks one must assume that reduction efforts have been pushed to the point of infeasibility. Section 123’s concept of GEP stack height not only acknowledges a legitimate role for stacks, but endorses engineering practices developed at a time when emissions were largely unregulated. In fact, Congress gave those practices a specific endorsement in authorizing the EPA to rely on the 2.5H formula in granting credit. See § 123(c) (Administrator is only obligated to conduct demonstrations for above-formula stacks). It is true that this court decided in Sierra Club that in any potential clash between defining GEP as that height necessary to avoid downwash-induced effects on human health and welfare, and defining it as 2.5H, the former must prevail. 719 F.2d at 448. But that decision did not undermine the statutory assumption that sources might legitimately rely on stacks aimed at protecting health and welfare from emisssions rates such as have prevailed on plants built before Congress’s vigorous 1970 intervention into clean air regulation. Thus it appears to us that the cases of the NRDC trilogy simply do not speak to the issue of calculating GEP height, because those courts never even considered the possibility that sources could rely on stacks except when driven to do so by the infeasibility of all alternatives. Legislative history endorsing their “general thrust” therefore cannot answer new questions posed by § 123 itself: “Necessary” as used in the GEP definition of § 123 cannot be answered by reference to the trilogy’s construction of the word in § 110(a)(2)(B). NRDC further points to legislative history that in its view suggests that Congress believed credit should be given only for stack heights needed by “well controlled” sources. In discussing its reasons for granting sources credit for GEP stack height, the House Committee explained that for many years, good air quality management has meant building a stack sufficiently tall to offset aerodynamic down-wash created by structures in the immediate vicinity of the stack. Without some provision for stack height, a plume released downwind of such structure might become engulfed by turbulent eddys [sic] within the wake of the structure. When this occurs even the plume from a well-controlled source may cause air quality standards (or other requirements) to be violated. House Report at 93, U.S.Code Cong. & Admin.News 1977, p. 1171 (emphasis added). We find NRDC’s reading strained. It seems clear to us that the Committee was merely justifying its decision to grant credit up to GEP stack height with the observation that even well controlled sources may need to rely on the dispersion engendered by a GEP height stack to avoid excessive local ground-level pollution concentration. Finally, NRDC points to a number of references expressing disapproval of the use of the dispersion effects of “tall stacks” to meet the NAAQS. See, e.g., House Report at 93 (“the courts have determined that the 1970 act prohibited tall stacks as a final compliance method”); Conference Report, H.R.Rep. No. 564, 95th Cong., 1st Sess. 144 (1977), U.S.Code Cong. & Admin.News 1977, pp. 1077, 1171, 1524 (“Tall stacks are not a means of emission limitation under the Clean Air Act of 1970.”). But we noted in Alabama Power v. Costle that “tall stack” is “a term that really covers a too-tall stack. 636 F.2d 323, 389 (D.C.Cir.1979) (emphasis added). Thus the references merely reiterate the Congress’s intent to deny stack height credit beyond the height dictated by GEP; they do not reflect any legislative decisions on how to calculate GEP. Far from finding a clear Congressional intent to adopt NRDC’s control-first strategy, we find the statute’s use of the term “necessary” to be completely ambiguous. We find no evidence in the statute or the legislative history that Congress ever thought through the question of how to determine GEP or formulated any view on the “control-first” approach. In view of Congress’s endorsement of the historic practice of using stacks to protect health from downwash-induced pollution, we think the agency, in adopting existing or SIP-required emissions rates as the baseline for demonstrations to support within-formula height increases, gave the statute a quite reasonable interpretation. NRDC asserts that even if we find the statute to be silent as to the appropriate emission level assumption, we should reject the agency’s approach as internally inconsistent and therefore arbitrary and capricious. See 5 U.S.C. § 706(2)(A). As noted above, while EPA authorizes use of existing or SIP-required emissions as the baseline for justifying within-formula increases, 40 C.F.R. § 51.1(kk)(2), it requires an assumption of NSPS emissions for demonstrations needed to justify above-formula stack heights, 40 C.F.R. § 51.1(kk)(l). NRDC claims that as NSPS is defined in terms of technological and economic feasibility, the latter provision reflects the agency’s partial acceptance of the control-first approach. The agency’s treatment of within-formula increases, it says, is hopelessly inconsistent with its view of above-formula stacks. NRDC Brief at 28-29. The EPA explained in the rulemaking that it rejected use of existing emissions in the above-formula context for fear of impermissible circularity — “to the extent that [a source’s preexisting emissions] limit relied on greater than formula height, it would amount to using a tall stack to justify itself.” See 50 Fed.Reg. 27,898/2. But we think that § 123 left the agency free to regard this circularity as permissible in the within-formula context, impermissible in the above-formula context. EPA explained that its use of NSPS (the stringent standard applicable to new sources) for above-formula stacks conformed to Congress’s expectation that the “credit for stacks above formula height ... be granted only in rare cases.” 50 Fed.Reg. 27,898/1. NRDC is certainly correct when it notes that the circularity problem that attends the use of existing emissions levels for above-formula demonstration purposes also applies to within-formula demonstrations. Where the formula has overstated the stack height necessary to avoid excessive ground level concentrations, the SIP or existing emissions levels may reflect the benefit of the erroneous assumption, so that its use in a within-formula stack height demonstration will inflate the “necessary” stack height. The problem, however, is plainly less severe than in the above-formula height context. The initial stack height credit was by definition within-formula, thus limiting the source’s existing emissions rate, and thus the extent to which the laxity of the process may inflict potential damage. The reliance on the formula in the initial computation establishes an outer bound on the degree of error. This is quite different from the risks entailed by use of existing rates in the above-formula context, precisely where both this court and Congress itself have warned the agency to extend credit only with “utmost caution,” Sierra Club, 719 F.2d at 450. Accordingly, we reject NRDC’s contentions that § 123 prohibits use of existing emissions rates in within-formula demonstrations and that the discrepancy between the baseline assumptions for above- and within-formula demonstrations renders the latter arbitrary and capricious. B. Demonstrations Supporting Stack Height Increases Within the Formula. 1. Attacks on the formula. In the Sierra Club litigation the environmental petitioners did not directly challenge the accuracy of the stack height formula. (By then the 2.5H formula had evolved into H + 1.5L, where L refers to the lesser of the height or width of any structure near the stack. Some references are to formulas in the plural because the old 2.5H persisted as to some grandfathered stacks.) They did, however, attack the agency’s failure to require demonstrations to justify stack height credit in either of two special cases: (1) when a source raised a preexisting stack, and (2) when federal, state or local authorities believed the formula had overstated the necessary stack height. The EPA in turn defended the omission of such demonstration requirements largely on the ground of the formula’s accuracy. 719 F.2d at 456. The court found this faith unsupported by the record, as it was explicitly based on the agency’s erroneous reía-tivistic conception of “excessive concentrations.” Id. at 458. The court therefore held that the agency had not considered whether the formulas were an accurate enough measure, in light of the construction of “excessive” concentrations as related to health and welfare, to justify dispensing with a demonstration requirement in the two special cases. It remanded for that reconsideration. Id. Thus, despite the absence of direct attack, the Sierra Club decision invited reconsideration of the formula, by suggesting to EPA that superi- or validation of the formula was an alternative to adopting demonstration requirements in the areas specifically found to be vulnerable. On remand, the agency provided demonstration requirements not only for the two circumstances specifically disputed in the preceding litigation, 40 C.F.R. §§ 51.-l(kk)(2) (stack height increase demonstration), 51.1(kk)(3) (governmental authority instigated demonstration), but also for sources with porous structures or buildings whose shapes are aerodynamically smoother than the simple structures on which the formulae were based. 50 Fed.Reg. 27,900/2 (explaining the reach of 40 C.F.R. § 51.1(kk)(3)). Given these demonstration requirements, nothing in our Sierra Club opinion required EPA to reevaluate the accuracy of its formula. Accordingly, consideration of deficiencies in the formula is barred by res judicata (and by the time limits of 42 U.S.C. § 7607(b)(1)), unless we find that the demonstration procedures chosen by EPA are insufficient to fulfill the statutory purposes. We now turn to that issue. 2. Attacks on the demonstration procedures. In Sierra Club this court found it proper to assume that large plants of the sort at issue here would have been built in accordance with “good air quality management” practices, or, effectively, GEP as the term is used in § 123. 719 F.2d at 459. The corollary of this was an assumption that post-construction stack increases were not justified by any need to correct downwash-induced dangers to health and welfare. The court said that this assumption could be rebutted in individual cases “only by a reliable indicator” of the height needed for that purpose. Id. EPA responded by imposing demonstration requirements. Under the 1985 regulations, sources seeking credit for height increases must show, through fluid modeling or wind tunnel demonstrations, that the increase is necessary to avoid downwash that would otherwise exceed at least one of several health- or welfare-related criteria (applicable NAAQS, “prevention of significant deterioration” standards covering areas in full attainment of the NAAQS under § 7475(a)(4), or levels amounting to a local nuisance). 40 C.F.R. § 51.1(kk)(2). As noted above, a source is to assume for the purposes of the demonstration an emission level equivalent to the applicable SIP, or, if no SIP applies, to its actual emissions rate. Id. NRDC argues that these emissions rate assumptions undermine the demonstrations, denying them the reliability demanded by Sierra Club. As we have noted, an assumption of existing or SIP-required emissions rates plainly gives the demonstrations a certain circularity: the now existing or required emissions rate will have been based on a given stack height, which will then be used to justify a stack height. NRDC Brief at 29. NRDC argues that instead EPA was required to assume the best achievable emissions rate. Id. at 22. We have already rejected NRDC’s claims that failure to employ control-first assumptions directly violates § 123 or is arbitrary and capricious in light of EPA’s own condemnation of circularity and partial adoption of control-first for above-formula demonstrations. The sole question currently before us, then, is whether EPA’s use of existing rates in this context is so defective as to fall short of Sierra Club's reliability requirement. We note at this point that the agency’s failure to establish perfect logic in support of its emissions baseline decision may well stem from the nature of § 123 itself. While its goal is the reduction of overall pollution loadings, at least in part with a view to protect against acid rain in regions distant from the sources, it operates solely on the means by which sources meet national goals for local health and welfare. It would be startling if implementation of this process did not involve a few logical imperfections. There appears, in fact, to be no completely logical basis on which to select a baseline rate for any demonstration. If EPA were to use NSPS for all demonstrations, for example, demonstrations would in a sense underpredict the appropriate stack height for sources with higher emissions rates. Assumption of a single high emissions rate, conversely, would overpredict the appropriate stack height for cleaner plants. But selecting each plant’s existing permitted rate is subject to the circularity objection. In this world of imperfections, we think EPA’s choice reasonable. This is particularly so in light of Congress’s having obviously contemplated reliance on a historic notion of “good engineering practice,” a notion developed during an era of relatively primitive emissions controls. We do not find EPA’s methodology for the conduct of within-formula demonstrations arbitrary or capricious. C. The NSPS Presumption for Above-Formula Stacks. In developing a baseline emissions rate for demonstrations to justify above-formula stacks, EPA initially proposed that sources assume “either (1) the existing, approved emission limit; (2) any applicable technology-based emission limit, such as the new source performance standards (NSPS); or (3) the emission limit that would result from the use of GEP formula stack height, whichever is applicable to the source being modeled.” 49 Fed.Reg. 44,882/1. The last phrase clearly indicates that NSPS would be used only for plants to which it applied by virtue of § 111. The final rule was dramatically different. It required each source to assume NSPS emission levels, or, if it could show those to be infeasible, the lowest achievable levels. 40 C.F.R. § 51.1(kk)(l). Industry petitioners strenuously object to this conditional-NSPS assumption on both substantive and procedural grounds. 1. Substantive objections. Petitioners raise three distinct challenges to the agency’s decision to adopt the NSPS presumption. First, they argue that because Congress did not prescribe the use of a technology-based emissions limit for GEP fluid modeling demonstrations, the Administrator lacks the authority to mandate its use. Alabama Power Brief at 19. As we have already noted in our discussion of NRDC’s control-first argument, Congress imposed technology-based emission limitations— NSPS, BACT, LAER, RACT and BART — in a variety of situations. Two of these, BART and RACT, govern pre-1971 sources. 42 U.S.C. §§ 7502(b), 7491(b)(2)(A). Industry petitioners would have us infer from the contrast between those express conferrals of authority, and the absence of any such reference here, that Congress denied EPA the authority to assume such an emissions rate. We find the attempt of industry to bar control-first here no stronger than NRDC’s effort to require it in the within-formula context. As we noted in discussing NRDC’s theory, the record raised considerable doubt whether anyone in Congress even recognized the issue. The silence alone seems to support neither a requirement nor a prohibition. What Congress did in § 123 was to grant broad discretion to the agency, requiring owners of above-2.5H stacks to demonstrate the necessity for the higher stacks “to the satisfaction of the Administrator.” 42 U.S.C. § 7423(c). In Sierra Club we read the section to mean above-2.5H credit should be granted only with the “utmost caution,” 719 F.2d at 447, which the selection of the NSPS baseline seems to reflect. Second, industry petitioners assert that the use of the NSPS presumption only for above -formula stack height demonstrations will unfairly prejudice sources located in mountainous terrain, since it is in such areas that above-formula stacks are most likely to be found. Industry petitioners argue that this contravenes the will of Congress. Alabama Power Brief at 34. In Sierra Club, however, we found a congressional recognition “that the tall stacks provision would have a disproportionately heavy impact on polluters in mountain areas.” 719 F.2d at 455. In fact, the court found an affirmative intent to “discourage utilities from locating in hilly terrain, because such locations tend to require very tall stacks, leading to greater dispersion of pollutants.” Id. at 445. The court adopted that construction in the context of rejecting EPA’s claim that it could consider “plume impaction” in computing excessive concentrations. Id. at 452-56. As that decision applied even to plants constructed before the adoption of § 123, whose owners were obviously not free to respond to its “discouraging]” influence, Sierra Club’s interpretation of congressional non-solicitude for plants in hilly terrain was a strong one. Any disadvantages inflicted on such plants by EPA’s choice of the NSPS baseline fit readily within our prior reading of the law. Finally, the industry petitioners assert that in order to use the NSPS presumption, EPA must be able to point to substantial evidence that it is attainable by most of the affected sources. But as EPA allows any source to use a higher emissions rate when NSPS is infeasible, there is no need for any sort of generic demonstration that it is normally feasible. Nor was it improper for EPA to place the burden of showing infeasibility on the source owner, rather then assuming the burden of showing feasibility. Congress appears to have intended that above-formula stack height be approved only in “rare circumstances.” House Report at 93, U.S.Code Cong. & Admin.News 1977, p. 1171. Cf. Sierra Club, 719 F.2d at 450 (“utmost caution” to be exercised in granting above-formula credit). EPA’s location of the burden is thus rationally related to the purposes of the statute and well within the Administrator’s discretion. 2. Procedural challenges. As described above, EPA initially outlined a scheme through which each source would assume its “applicable” emissions rate for purposes of above-formula demonstrations: (1) sources subject to technology-based emission limits would assume those rates; (2) sources not subject to such limits would assume their existing, approved SIP limits; and (3) sources not subject to either of the above would assume the limit that would result if they were to operate with the stack height credit that the formula would produce. 49 Fed.Reg. 44,882/1. Less than two weeks before promulgating the final regulations, the agency informed industry representatives of its decision to adopt instead a uniform (but conditional) NSPS presumption. Because time was short, industry representatives were only able to respond with two short letters strongly urging reconsideration of the new rule. J.A. 1483, 1491 (letters from the Utility Air Regulatory Group (“UARG”) and American Electric Power Company). Industry petitioners now assert that this abrupt shift denied them the opportunity to comment afforded by § 4 of the Administrative Procedure Act, 5 U.S.C. § 553. In the preamble to its Final Rule, the agency sought to undermine this claim by characterizing its 1984 proposal as presenting three distinct alternatives, rather than a coherent three-part scheme. 50 Fed.Reg. 27,898/2. This is quite disingenuous. Nothing in the initial formulation suggested that EPA intended to adopt one of the three rates for universal use. And where EPA was offering alternatives from which it intended to make a choice in its final rule, it said so. See, e.g., 49 Fed.Reg. 44,881/1 (proposing and soliciting comments on two alternatives for the definition of “excessive concentrations”); 44,884/1 (proposing and soliciting comments upon three approaches for modeling “nearby” terrain features). The EPA can obviously promulgate a final regulation that differs in some respects from its proposed regulation. We recognized in International Harvester Co. v. Ruckelshaus, 478 F.2d 615, 632 n. 51 (D.C.Cir.1973), that “a contrary rule would lead to the absurdity that ... the agency can learn from the comments on its proposals only at the peril of starting a new procedural round of commentary.” Thus, this court has held under both the APA and the Clean Air Act that the agency’s final rule must only be a “logical outgrowth” of its proposed rule. Small Refiner Lead Phase-Down Task Force v. EPA, 705 F.2d 506, 547 (D.C.Cir.1983) (citing circuit precedent on this matter). We think the agency’s ultimate choice qualifies — if barely — as a logical outgrowth of the original proposal. Obviously the germ of NSPS was there, as one of the possibly-applicable technology-based limits. Moreover, the primary concern that drove EPA away from allowing sources to use existing SIP limits (and thus towards NSPS) was obvious at an early stage: NRDC’s constantly asserted control-first theme. See 51 Fed.Reg. 27,898/2 (agency rejects use of existing SIP emissions limits because “to the extent that [a source’s limit] relied on greater than formula height, it would be using a tall stack to justify itself”). To be sure, EPA never explained in the administrative proceeding why it rejected use of the emissions rate that would flow from use of the formula. In fact, in its Response to Comments the agency defended this option against environmentalist charges that it was incurably circular. J.A. 314-15. Here the agency’s lawyer explained that the formula emissions level assumption was inappropriately strict. EPA Brief in Ohio Power, Nos. 86-1331, 86-1362, at 28-29. However valid this critique may be, it is little help in showing the agency’s intellectual path to the NSPS choice. However, as neither the industry petitioners nor anyone else in this litigation advocates that choice, it can hardly have been so attractive that its disappearance came as a shock. Further, the public comments raised the possibility of adopting a single, technology-based limit. The New York State Attorney General’s Office suggested an NSPS assumption for all demonstrations. J.A. 434. This gave industry participants a clear opportunity to shoot the idea down. NRDC attacked each of the three limits set forth in the 1984 proposal, J.A. 829-33, and advocated use of the rate that would result from use of “the maximum degree of control available to the source.” Id. at 832 n. 6. Though the target raised by this contention was broader than NSPS, it certainly gave industry critics an opportunity both to shore up the non-NSPS components of the original proposal and to attack any form of control-first. Nor was industry free to discount these proposals merely because they came from parties favoring a control-first reading of the statute, see Alabama Power Reply Brief at 15; there was a clearly foreseeable risk that EPA would reject the environmentalists’ reading of the law but proceed to adopt control-first as a matter of choice. This, in essence, is what it did in a limited sphere. Finally, EPA’s warning of the NSPS threat, communicated two weeks before promulgation, gave industry petitioners at least a limited opportunity to focus a direct attack on NSPS. Though severely pressed, they managed to file objections 7-10 days before the final regulations were signed. J.A. 1483, 1491. Although this case stretches the concept of “logical outgrowth” to its limits, we think it does not reach the breaking point. The NSPS assumption appears to have emerged from the agency’s notice and comment process, as the agency responded to others’ comments by stripping away the components of the original proposal that it concluded were more vulnerable. Of course, our affirmance of EPA on this point does not require it permanently to resist useful suggestions or critiques that may emerge. III. Stack Grandfathering Issues The agency has in several cases grandfathered stacks and in one instance rejected an industry request for grandfathering treatment. NRDC attacks several elements of the grandfathering as too generous; several firms (including one heavy buyer of electricity) attack the rejection of their claim. The decisions at issue are as follows: 1. EPA limited its demonstration requirement for within-formula stack height increases to “sources seeking credit after October 11, 1983 [the date of our decision in Sierra Club ].” 40 C.F.R. § 51.1(kk)(2). (The preamble makes clear that the date refers to the time owners raised the stack, not the time they “sought” credit. 50 Fed.Reg. 27,899/2.) 2. EPA affords credit up to 2.5H, free of any demonstration requirements (including such as might be imposed by state or local air pollution authorities), for any stack in existence on January 12, 1979, the date when EPA first proposed the more sophisticated formula, H + 1.5L. Pursuant to our remand in Sierra Club, this is limited to cases where the owner or operator produces evidence of having relied on the 2.5H formula in establishing its emission limitation. 40 C.F.R. §§ 51.1(ii)(2)(i). EPA also affords credit up to H + 1.5L for any stack in existence on January 12,1979, regardless of reliance. Id. at § 51.1(ii)(2)(ii). 3. For purposes of the grandfathering explicitly afforded by § 123, for stacks “in existence” before December 31, 1970, EPA in the 1982 regulations defined “in existence” in terms of the start of continuous construction or entering into certain types of contracts. 40 C.F.R. § 51.1(gg). We upheld this in Sierra Club, 719 F.2d at 464-66. NRDC claims that EPA should have revised this definition in the light of later discoveries about the number of plants advantaged by the grandfathering. 4. EPA has refused to provide any grandfathering for plants that prior to the Final Rule conducted demonstrations to justify above-formula stacks, even though the demonstrations conformed entirely to the then-applicable rules. J.A. 64-65. Some general principles are applicable to all these issues. First, none of them is governed by the rule adopted in Georgetown University Hospital v. Bowen, 821 F.2d 750, 756-58, 760 (D.C.Cir.1987), generally invalidating retroactive rules. The rules there at issue would have limited reimbursements for past transactions. All that is at stake here are restrictions on plants’ future emissions. Retroactivity is involved here simply because enforcement of the demonstration requirement might impinge unfairly on source owners that made investments or other commitments in reasonable reliance on prior understandings. Second, our decision in Sierra Club observed that some of the considerations governing an agency’s duty to apply a rule retroactively were (1) whether the new rule represents an abrupt departure from well established practice or merely attempts to fill a void in an unsettled area of law, (2) the extent to which the party against whom the new rule is applied relied on the formed [sic] rule, (3) the degree of the burden which a retroactive order imposes on a party, and (4) the statutory interest in applying a new rule despite the reliance of a party on the old standard. 719 F.2d at 467 (quoting Retail, Wholesale & Department Store Union v. NLRB, 466 F.2d 380, 390 (D.C.Cir.1972)). Clearly the issue entails a balancing of the interest in prompt and complete fulfillment of statutory goals against the inequity of enforcing a new rule against persons that justifiably made investment decisions in reliance on a past rule or practice. Cf. Retail, Wholesale & Department Store Union v. NLRB, 466 F.2d at 390; Associated Gas Distributors v. FERC, 824 F.2d 981, 1040 (D.C.Cir.1987). We now turn to the specific complaints. A. Sheltering pre-October 1, 1983 Within-Formula Stack Increases from the Demonstration Requirement At the outset, EPA invites us to discard NRDC’s objection by means of an argument that to us is quite obscure: that the Sierra Club court did not insist on a retroactive demonstration requirement, or indeed on any demonstration requirement at all. EPA Brief at 34. The premise seems to be that, because EPA’s adoption of a demonstration requirement was voluntary, it need not explain its decision not to impose the requirement on pre-1983 increases. The argument does not add up. First, EPA’s adoption of a demonstration requirement for increases could be viewed as wholly voluntary only if EPA had set out thoroughly to validate its H + 1.5L formula. This it did not really purport to do, although its preamble contains some favorable — and hotly disputed — observations on the subject. In any event, the agency found, in light of this court’s reading of § 123 in Sierra Club, that a demonstration requirement was suitable. That being so, exemption of a large class of increased stacks is an important decision subject to attack as possibly arbitrary and capricious. It would fail that test if it is inconsistent with the re-troactivity analysis set forth in Sierra Club. Invoking the first factor identified by Sierra Club, NRDC argues that here there simply is no “well established practice.” A fluctuating rather than well-established practice presumably counts against grandfathering, as it undermines the actor’s claim that its reliance was legitimate. In fact, the record indicates considerable waffling by EPA. There appear to have been new policy moves in 1973, 1976, 1979,1980, and 1981 and 1982. In its 1973 Guidelines, it actively encouraged sources with short stacks to increase to GEP formula height. 38 Fed.Reg. at 25,701/2 (1973) (“The increase of stack height up to a height consistent with good engineering practice is acceptable without qualification.... For fairly level terrain, good engineering practice is normally taken to be a stack height 2V2 times the height of the facility or nearby structure.”). In 1976 it responded to the circuit courts’ disparaging treatment of its 1973 effort with considerable severity: putting aside stack increases started before the Fifth Circuit’s NRDC decision, the 1976 guidelines gave credit only for increases by sources that applied BACT (“best available control technology”). 41 Fed.Reg. at 7451/2-3 (1976). In 1979, after Congress’s adoption of § 123, the agency proposed regulations considerably less stringent, allowing sources that raised existing stacks automatic credit up to H + 1.5L, with the proviso that the EPA or a state or local control agency could order the source to justify use of the formula height by demonstrating, through fluid modeling, the existence of “an air quality problem, attributable to downwash.” 44 Fed.Reg. at 2614/1 (Jan. 12, 1979). In a 1980 “policy change,” sparked by a heightened concern “that use of its GEP formula for stack height increases [was] increasing pollutant loadings and acid rain,” the agency tightened again, announcing its intention to require fluid modeling demonstrations for all future stack height increases. 45 Fed.Reg. 42,282/1 (June 24, 1980). Eleven months later, the EPA drew back, leaving in place the scheme adopted in its 1979 proposed rules. 46 Fed.Reg. 28,650 (May 28, 1981). The agency’s 1982 Final Regulations moved further in the direction of leniency, granting automatic credit up to formula height for sources raising existing stacks, with no provision for support through demonstrations. 47 Fed.Reg. 5868/3 (1982). Of course it was this rule that the Sierra Club court remanded for reconsideration. 719 F.2d at 458-59. Clearly the legitimacy of increasing stack height in reliance on regulatory policy has varied radically from period to period. The equities for a firm increasing its stack in the 1976-79 era are slight compared to ones that increased under the 1973,1979 or 1981 policies. Besides, the policies represent a scatter rather than a clear line, reducing the equities for reliance even on the moments of lax policy. Finally, equitable claims have some tendency to degrade over time; a 15-year 1973 contract for the purchase of high-sulphur coal may have loomed large in 1976 but hardly amounts to anything in 1987. EPA’s defense of its grandfathering decision failed to focus on any of these difficulties. See 50 Fed.Reg. at 27,899/3-900/1. Indeed, EPA’s policy imposed no requirement of reliance at all, even though it was precisely that omission that persuaded this court to remand the grandfathering issue raised in Sierra Club. 719 F.2d at 468. Against the seemingly weak claims for grandfathering is the possible frustration of the statutory goal. This looks significant. The vulnerability of the formula persuaded EPA to require demonstrations. These demonstrations are impaired by the circularity problem that EPA has recognized. Yet EPA’s grandfathering rule allows most of the affected sources to escape even this modest check. Administrative problems may partly explain EPA’s generous grandfathering