WALD, Circuit Judge:

The National Lime Association (NLA), representing ninety percent of this country's commercial producers of lime and lime hydrate (the industry), challenges the new source performance standards (NSPS) for lime manufacturing plants issued by the Environmental Protection Agency (EPA, Administrator or Agency) under § 111 of the Clean Air Act (the Act), 42 U.S.C. § 7411 (Supp. I 1977). The standards limit the mass of particulate that may be emitted in the exhaust gas from all lime-hydrating and from certain lime-manufacturing facilities and limit the permitted visibility of exhaust gas emissions from some facilities manufacturing lime. We find inadequate support in the administrative record for the standards promulgated and therefore remand to the Administrator.

I. RELEVANT PARTICULARS OF THE LIMESTONE INDUSTRY

A. The Industry

In sheer size and weight of production, the limestone industry ranks among the largest in this country. Limestone production in the United States ranks second only to sand and gravel in commodity tonnage and exceeds petroleum, coal and iron ore in volume produced. Limestone deposits can be found beneath an estimated fifteen to twenty percent of the surface of the United States and occur in every state. Total national production approximates twenty-two million tons annually and derives from plants in over forty states.

The recent development of two important industrial uses for lime has ensured the continuing growth of production despite a decline in agricultural use. The industry is capital-intensive with declining employment, but because so many other industrial processes depend on the use of lime, any decline in production would have "a large multiplier effect on U.S. employment."

B. The Production of Lime From Limestone

The process by which commercially valuable lime is produced is relatively simple. Limestone is quarried, crushed, sized and fed into a kiln where it is subjected to high temperatures (1100°C/2000°F). By a process known as "calcination," the heating ("burning") of limestone produces quicklime, a soft, porous, highly reactive material commonly used in industry. As might be expected, the process generates a substantial quantity of dust, or in the language of the Agency, particulate matter, sufficiently lightweight to be carried off in the hot exhaust gas and emitted from the kiln. The particulate matter thus released is composed of partially burned limestone, raw limestone feed, deadburned lime and quicklime. Typically, the process also releases sulfur dioxide (SO[2]).

Almost ninety percent of total United States lime production is processed in rotary kilns. Uncontrolled emissions from rotary kilns have been reported to run from 150 to 200 pounds per ton of lime produced, roughly five percent of the feed poundage and nine percent of the produce. A typical lime plant producing 500 tons per day from a rotary kiln, conforming to typical state pollution-control standards, emits about 150 megagrams (165 tons) of particulate matter per year. Rotary kilns produce a greater volume of particulate emissions than the formerly widely used vertical kilns but they are also the only kilns which can retain product quality while burning coal, a fuel on which the industry has become increasingly dependent.

C. The Production of Hydrated Lime

A comparatively small amount (ten percent) of all lime produced is further processed into hydrated or slaked lime. This is done by adding water to lime and introducing the mixture into an agitated hydrator. An exothermic reaction occurs and a fluffy, dry, white powder, known as hydrated lime, is the result. Particulate matter is carried off in the steamy exhaust emitted from the hydration process.

D. Emissions Control in the Production of Lime

Rotary kilns here and abroad have employed several different methods of emissions controls including the fabric filter baghouse, the electrostatic precipitator (ESP), the high energy scrubber, and the gravel bed filter. One survey showed that of eighty-five domestic rotary kilns, twenty-four percent used a baghouse, thirty-one percent used a high energy scrubber and eight percent used an ESP. However, use of the baghouse method is increasing because this method requires less energy and does not itself create additional problems of pollution control.

EPA has identified baghouses, ESPs and scrubbers as "best systems" of emissions control for rotary lime kilns.

Baghouses

The operation of baghouses and electrostatic precipitators was briefly explained in our initial review of EPA's performance standards for portland cement plants, Portland Cement Association v. Ruckelshaus, 486 F.2d 375, 390-91 (D.C. Cir. 1973), cert. denied, 417 U.S. 921, 94 S.Ct. 2628, 41 L.Ed.2d 226 (1974) [hereinafter cited as Portland Cement I]. The baghouse method employs fabric filters ("bags"), situated within an enclosed area (a "house"), to remove particulate from the kiln exhaust gas which is channeled through the house.

As the exhaust gas passes through, a dust cake forms on the filters. The cake itself improves filtration efficiency, but from time to time the filters must be cleaned. This is done by forcing a reverse gas flow through the fabric, thus releasing the cake for disposal.

EPA acknowledges that fabric filter effectiveness is primarily a function of kiln exhaust particle size distribution, fabric type, fabric age and maintenance history.

Electrostatic Precipitators

Under this method, "dust particles are charged [by discharge electrodes] and pass through an electrical field [collector plates] of the opposite charge, thus causing the dust to be precipitated out of the exhaust gas . . .." Portland Cement I, 486 F.2d at 390. Two basic criteria must be met before an ESP can be utilized: (1) the suspended particle must be able to accept an electric charge; and (2) the particle must then pass through an electric field of sufficient strength to ensure removal of the particulate from the gas stream at the desired efficiency.

Precipitability is a function of the chemical composition of the dust particles, and will vary with the different kinds of material that make up the kiln exhaust dust (limestone, quicklime, fly ash, calcium sulfate, etc.). Assuming precipitability, the two main factors influencing the efficiency of a precipitator are the gas velocity and treatment time. The ESP method experiences a relatively low collection efficiency on submicron particles.

Although most particles collected by an ESP fall by gravity into waiting hoppers, programmed rapping of the electrodes is also required to keep the collector plates and discharge electrodes clean. As with the baghouse method, the dust collected is dry and may be disposed of in a variety of ways. A high level of maintenance skill is needed to keep an ESP in operation at design conditions.

Scrubbers

Scrubbers operate on the principle that wet particles are easier to control than dry. High pressure (or high energy) scrubbers of the type EPA considers capable of meeting the promulgated standards are those which because of their design increase the likelihood of contact between particle and water.

The most common high pressure drop scrubber used for controlling emissions from rotary lime kilns is the venturi scrubber. This scrubber operates by accelerating the velocity of the exhaust gas through a narrow venturi-shaped throat, where it is then brought into contact at great force with a spray of water. The particles thus dampened coalesce to form a slurry that can then be collected by a comparatively simple water-gas separation device. The separated gas is then released into the atmosphere.

The efficiency of particulate removal is a direct function of energy input, measured by pressure drop across the venturi throat. Gas-water contact in the venturi scrubber is so thorough that even submicron particles are removed. Although low pressure drop scrubbers use less energy than high pressure drop scrubbers, even a low efficiency scrubber requires more energy than either the baghouse or the ESP. The slurry which is the by-product of scrubber use is deposited in ponds, where the collected particulate settles out from the scrubbing water. The "clean" scrubbing water is then reused. Under present law settling ponds must be located so that they do not receive excessive rainwater run-off, causing overflow into local navigable waters.

E. Emissions Control in the Production of Hydrated Lime

Hydration emissions have been shown to be most effectively controlled by wet scrubbers and they are the only system of emission reduction considered by EPA for lime hydrators.

The most common type of scrubber used on lime hydrators is the wetted fan type with centrifugal separation. In this scrubber water is sprayed into the center of a draft fan where it is forced to mix with the exhaust gas. More water is sprayed just after the fan into the duct carrying this gas-water mixture. The dust laden slurry water is then removed from the cleaned gas stream by centrifugal separation and the "scrubbed" gas is then vented to the atmosphere.

Slurry water is returned immediately to the hydrator for reuse; the hydration process requires the addition of water and the captured dust seems to contribute to, rather than interfere with, the production of hydrate. Recycling the slurry water eliminates the settling ponds and waste sludge disposal problems usually associated with particulate scrubbers.

II. PROCEDURAL HISTORY

Section 111 of the Clean Air Act, formerly 42 U.S.C. § 1857c-6 (1976) (repealed 1977), now 42 U.S.C. § 7411 (Supp. I 1977), authorizes the Administrator to limit the air pollutants that can lawfully be emitted from newly constructed or modified plants. This the Administrator can do by promulgating new source performance standards requiring new or modified plants to meet standards which can be met through application of the best system of emission reduction (considering costs) which has been "adequately demonstrated." The purpose is to assure that new or modified plants will not create significant new air pollution problems.

On May 3, 1977, EPA added lime manufacturing plants to the list of sources that "may contribute significantly to air pollution which causes or contributes to the endangerment of public health or welfare" pursuant to section 111(b) of the Clean Air Act, 42 U.S.C. § 1857c-6(b)(1)(A) (1976) (repealed 1977). 42 Fed.Reg. 22510 (1977). At the same time, EPA proposed NSPS for lime plants. 42 Fed.Reg. 22506 (1977). The information underlying both actions was contained in the SSEIS.

Although lime plants were determined to be sources of nitrogen oxides, carbon monoxide and sulfur dioxide as well as particulates, standards of performance were proposed and ultimately promulgated only with respect of particulate matter. Furthermore, of the various types of kilns that may be used in the calcination of limestone, only rotary kilns are regulated by the standards.

The kiln standards limit emissions to 0.15 kilogram of particulate matter per megagram of limestone feed (0.3 pound per ton) and ten percent "opacity." The owner or operator of an affected facility is required by the regulations to monitor continuously the opacity of emissions. Where the scrubber method is used for control, both the opacity standard and the opacity monitoring requirement are waived, and the pressure drop and liquid supply pressure of the scrubber must be monitored instead.

The standard proposed and promulgated for lime hydrators limits emissions to 0.075 kilogram of particulate matter per megagram of lime feed (0.15 pound per ton). No opacity standard was set. The hydrator standard requires that the electric current and the liquid supply pressure of the scrubbers used to control emissions be monitored continuously.

The standards promulgated for particulate emissions are considerably stricter than the average applicable state regulations already in effect. Plants conforming to the NSPS here would — in the case of rotary kilns — be required to emit less than one-third the particulate permitted under average state regulations and — in the case of hydrators — less than one-sixth the particulate permitted by these regulations. See SSEIS 4-15.

Evidently, EPA had engaged in a dialogue with the NLA concerning the anticipated NSPS for at least a year before the standards were proposed. After publication of the proposed standards on May 3, 1977, EPA received additional written comments both from the NLA and from others and on June 16, 1977 held a public meeting to "provide an opportunity for oral presentations and comments on the standards."

Final responses to some of the comments received were issued in a final support statement document in October 1977. The final notice of rulemaking was published March 7, 1978. Except for two minor changes the final standards did not differ from those proposed ten months earlier. A petition for review was timely filed in this court, the exclusive court of review of new source performance standards.

III. PREVIOUS REVIEW UNDER SECTION 111

As amended in 1977, section 111 of the Clean Air Act requires the Administrator to prescribe standards of performance for new statutory sources that reflect

the degree of emission limitation and the percentage reduction achievable through the application of the best technological system of continuous emission reduction which (taking into consideration the cost of achieving such emission reduction, any nonair quality health and environmental impact and energy requirements), the Administrator determines has been adequately demonstrated. . . .

42 U.S.C. § 7411(a) (Supp. I 1977). As the court of exclusive review for NSPS, we have examined section 111 standards on several prior occasions. Portland Cement Association v. Ruckelshaus, 486 F.2d 375 (D.C. Cir. 1973), cert. denied, 417 U.S. 921, 94 S.Ct. 2628, 41 L.Ed.2d 226 (1974) ( Portland Cement I) (Portland cement plants); Essex Chemical Corp. v. Ruckelshaus, 486 F.2d 427 (D.C. Cir. 1973), cert. denied, 416 U.S. 969, 94 S.Ct. 1991, 40 L.Ed.2d 558 (1974) [hereinafter cited as Essex Chemical] (sulfuric acid plants and coal-fired steam generators); National Asphalt Paving Association v. Train, 539 F.2d 775 (D.C. Cir. 1976) (asphalt concrete plants) [hereinafter cited as Nat'l Asphalt]; Portland Cement Association v. Train, 513 F.2d 506 (D.C. Cir.), cert. denied, 423 U.S. 1025, 96 S.Ct. 469, 46 L.Ed.2d 399 (1975) [hereinafter cited as Portland Cement II].

These decisions, viewed independently, have established a rigorous standard of review under section 111. We have not deviated from the approach applied to the first NSPS to reach this court. In that case, Portland Cement I, we acknowledged that

[w]hile we remain diffident in approaching problems of this technical complexity, . . . the necessity to review agency decisions, if it is to be more than a meaningless exercise, requires enough steeping in technical matters to determine whether the agency "has exercised a reasoned discretion." . . . We cannot substitute our judgment for that of the agency, but it is our duty to consider whether "the decision was based on a consideration of the relevant factors and whether there has been a clear error of judgment." . . . Ultimately, we believe, that the cause of a clean environment is best served by reasoned decision-making.

486 F.2d at 402 (citations omitted).

In Essex Chemical we reiterated this concept of the court's role in examining the basis for section 111 standards:

The judgment of the Administrator is to be weighted against his statutory function and limitations, the record searched to determine if indeed his decisions and reasons therefor are themselves reasoned, and at that point our function terminates. Our expertise is not in setting standards for emission control but in determining if the standards as set are the result of reasoned decisionmaking. Yet even this limited function requires that we foray into the technical world to the extent necessary to ascertain if the Administrator's decision is reasoned. While we must bow to the acknowledged expertise of the Administrator in matters technical we should not automatically succumb thereto, overwhelmed as it were by the utter "scientificity" of the expedition.

486 F.2d at 434. The search for reasoned decisionmaking in a world of technical expertise must continue if judicial review is to have any meaning in the statutory scheme.

Section 111 requires that the emissions control system considered able to meet the standard be "adequately demonstrated" and the standard itself "achievable." 42 U.S.C. § 7411(a) (Supp. I 1977). We have in the past remanded section 111 standards for the "seeming refusal of the agency to respond to what seem to be legitimate problems with the methodology of the [ ] tests," Portland Cement I, 486 F.2d at 392; and the limited relevance and reliability of the tests relied upon in support of the standard. Id. at 396, 401. In Essex Chemical as well as Portland Cement I we expressed concern that the standards set might not have been achievable in periods of abnormal operation, e. g., during the "startup, shutdown and [equipment] malfunction" periods that occur in plant operation; and we remanded for further consideration of this issue. Portland Cement I at 398-99; Essex Chemical, 486 F.2d at 433. We have also questioned the significance of tests conducted for purposes of standard development under conditions different from those specified by the regulations for enforcement. Essex Chemical at 436. In analogous review proceedings under other sections of the Clean Air Act and under the Federal Water Pollution Control Act, 33 U.S.C. § 1251 et seq. (1976), this court and other courts have evinced a similarly rigorous approach.

However, we think it serves little purpose to elaborate on the standard of review as applied before we explain how, under the general approach required by statute and our earlier decisions, we have evaluated petitioner's and respondents' contentions.

The issue presented here is primarily one of the adequacy of EPA's test data on which the industry standards are based. NLA disagrees with EPA's conclusion that the standards are achievable under the "best technological system of continuous emission reduction which . . . the Administrator determines has been adequately demonstrated." Specifically, NLA claims that the test data underlying the development of the standards do not support the Administrator's conclusion that the promulgated emission levels are in fact "achievable" on a continuous basis. Promulgation of standards based upon inadequate proof of achievability would defy the Administrative Procedure Act's mandate against action that is "arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law." 5 U.S.C. § 706 (1976).

IV. ASSESSMENT OF THE OBJECTIONS RAISED BY THE INDUSTRY

Our review has led us to conclude that the record does not support the "achievability" of the promulgated standards for the industry as a whole. This conclusion is a cumulative one, resulting from our assessment of the many points raised by the industry at the administrative level and in this court; no one point made is so cogent that remand would necessarily have followed on that basis alone. In the analysis that follows, common threads will be discerned in our discussions of individual points. Chief among these common threads is a concern that the Agency consider the representativeness for the industry as a whole of the tested plants on which it relies, at least where its central argument is that the standard is achievable because it has been achieved (at the tested plants). The Agency's failure to consider the representativeness — along various relevant parameters — of the data relied upon is the primary reason for our remand. The locus of administrative burdens of going forward or of persuasion may shift in the course of a rulemaking proceeding, but we think an initial burden of promulgating and explaining a non-arbitrary, non-capricious rule rests with the Agency and we think that by failing to explain how the standard proposed is achievable under the range of relevant conditions which may affect the emissions to be regulated, the Agency has not satisfied this initial burden.

Bearing this initial burden will involve first, identifying and verifying as relevant or irrelevant specific variable conditions that may contribute substantially to the amount of emissions, or otherwise affect the efficiency of the emissions control systems considered. And second, where test results are relied upon, it should involve the selection or use of test results in a manner which provides some assurance of the achievability of the standard for the industry as a whole, given the range of variable factors found relevant to the standards' achievability.

EPA itself acknowledged in this case that "standards of performance . . . must . . . meet these conditions for all variations of operating conditions being considered anywhere in the country." SSEIS 2-6 (emphasis supplied). As set forth in the standards support statement, EPA's guidelines require data to be assessed with consideration of the "representativeness" of the source tested, including the "feedstock, operation, size and age" of the source. SSEIS at 2-7. Furthermore, the record strongly suggests other factors that may affect the particulate emissions from lime plants. Yet at no point does EPA evaluate the relevance or irrelevance of such factors to regulable emissions; nor does the Agency explain how such factors might have been taken into account in choosing test plant sites or in analyzing the data from the sites it chose.

The critical question presented here is whether the regulated industry, through its trade association, should have borne the entire burden of demonstrating the unreliability for the industry as a whole of the conclusions drawn by the EPA. In this connection we are candidly troubled by the industry's failure to respond, at a crucial juncture in the standards development process, to the Agency's invitation to submit data supporting a fundamental industry objection to the achievability of the standard. We would have expected the industry to have been eager to supply supporting data for its position, assuming the "cost" of obtaining such data were less than the "cost" of compliance with a standard that was argued to be unachievable on any reliably repetitive basis for the industry as a whole. We cannot help but wonder if the industry's failure to supply such data means that the data available or obtained would not be favorable to the industry's position. Nevertheless we remand because we think, on balance, EPA must affirmatively show that its standard reflects consideration of the range of relevant variables that may affect emissions in different plants.

The showing we require does not mean that EPA must perform repeated tests on every plant operating within its regulatory jurisdiction. It does, however, mean that due consideration must be given to the possible impact on emissions of recognized variations in operations and some rationale offered for the achievability of the promulgated standard given the tests conducted and the relevant variables identified. To facilitate public comment, we think this rationale should have appeared in the Agency's initial standards support statement.

We must remand to the Agency for a more adequate explanation or, if necessary, for supplementary data to justify the standard in terms of the "representativeness" of the sources tested. The specific doubts generated by our review of the record in light of the lime industry's attack on the standard are more fully explained below.

A. The Particulate Emission Standards 1. Rotary Kilns

EPA tested emissions at six plants before it proposed its mass emission standard for rotary lime kilns. These six plants were selected for testing on the basis of visits to thirty-nine plants, during which the visibility of emissions was observed and information obtained on the emissions control systems employed. The thirty-nine plants were themselves selected because they had been identified as effectively controlled after a review of the literature and contact with industry representatives. SSEIS A-1. The results of the tests of one plant (Plant A) which could not meet the proposed standard were excluded from consideration because the plant was thought not to represent best technology. From what we can gather from the record, three plants were able to meet the standard consistently.

Our doubts about the representativeness of the data relied upon are grouped under three subheadings below: Variations in Quantity of Particulate Generated in the Kiln; Variations in Controllability of Particulate Generated; and Explanation of Discarded Data from Plants A and F. Under the subheading Variations in Quantity of Particulate Generated in the Kiln, we discuss the possible impact on the standard's achievability of composite dust levels generated by the tested plants and two factors (feedstock variations and gas velocity) that may contribute to composite dust levels. Under the subheading Variations in Controllability of Particulate Generated, we discuss two factors — apart from sheer quantity of dust — that may affect emissions control: coal usage and particulate size. Finally under the subheading Explanation of Discarded Data from Plants A and F, we discuss the EPA's handling of the results of two tested plants that were unable to meet the standards proposed.

a. Variations in Quantity of Particulate Generated in the Kiln

That the quantity of dust produced in the kilns would affect the controllability of emissions and the achievability of the standards does not seem an unreasonable expectation. The Agency, however, appears to have taken conflicting positions on the reasonableness of this expectation and perhaps as a consequence has devoted inadequate attention to several variables which EPA's own documents and the industry suggest may affect the volume of dust produced in different kilns.

(1) Feedstock Variations

For example, the record suggests that the size and chemical composition of the limestone feedstock used will affect the amount of dust produced.

The MRI Report, prepared for EPA as a prelude to proposal of the particulate emission standards and an important background document considered in developing the proposed standards, stated that "[d]usting in the kiln with the resulting generation of particulate emissions is reportedly a function of the limestone raw material, the rate of rotation of the kiln, and the velocity of the gases in the kiln." R. 8, 2 (emphasis supplied). This suggests to us that some analysis should have been performed or tests conducted which took into account significant variations in limestone feed, or other variables relevant to dust generation.

The same theme was struck by NLA's comments on the proposed standard: "No consideration has been given by EPA to variations produced in stone size or preparation, or to the physical characteristics of the stone feed and lime produced, with the resultant variations in the quantity of flue dust to be handled." R. 103, 10.

The EPA did note in its SSEIS that "[r]otary kilns can handle a range of stone feed sizes between 1/4 inch and 2 1/2 inches," SSEIS 3-6, and that larger feed size generally results in lower dusting in the kiln. See SSEIS 3-14. The Agency also acknowledged that the grade and composition of limestone varies widely across the country. SSEIS 3-1. However, no data on stone size are included by the EPA in the summary data on plants tested (SSEIS App. C) and little information concerning the chemical composition of the feed used at the tested plants is provided. The feedstock at two of the plants tested is characterized as "high calcium lime" (Plants E and F, SSEIS C-6-C-7, and at two other plants as "dolomitic limestone" (Plant C, SSEIS C-3) or "dolomitic stone" (Plant D, SSEIS C-4). The feedstock at one plant (Plant B) is not described at all. We are, more importantly, left in the dark about which kinds of limestone can be expected to produce the greatest volume of emission dust and what, if any, processing adjustments can be expected of producers using particular kinds of feed in order to achieve the standard proposed. For all we know, the six plants tested could be using kinds and sizes of feed which are representative of only a small segment of the industry spectrum. If that were true the plants may not be "representative" and the regulation might not be "achievable" by the industry as a whole.

(2) Gas Velocity and Operation Levels

According to the MRI Report, quoted above, dust generation is in part a function of gas velocity in the kiln. Gas velocity appears in turn to depend on several factors, including the percentage of capacity at which the kiln is operating. The MRI Report stated that kiln gas velocity has "the most [apparent] effect [on dust generation] when the kiln is operated close to 100 percent of design capacity," and noted that in one plant studied an increase in production — from 100% to 135% of design capacity — resulted in double the rate of emissions where a reduction from 100% to 75% resulted in only an eight percent reduction. R. 8, 2-3. This seems to mean that at levels close to or exceeding capacity, gas velocity and consequently dust generation increases at a faster rate than at lower levels of production. Thus the level of capacity at which the plant was operating at the time of sampling and the gas velocity would appear relevant to the representativeness of the test data.

Both in this court and at the administrative level the industry has addressed the possible atypicality of the production level of some of the test plants, which it alleges were not tested at full capacity and in doing so, it has echoed a concern expressed by this court in an earlier case.

Data on the production level and air flow rate (velocity) at the tested plants were included in the support document filed in this case. These data indicate that the two baghouse-controlled kilns relied upon (Plants B and E) were operating at 111% and 91% of rated capacity, respectively. Plant A, also baghouse-controlled, but whose test results were rejected as unsatisfactory, operated at 92% capacity during the tests. One of the two ESP-controlled kilns (Plant C) was operating at 97% of design capacity and the other (Plant D) was tested at capacity, slightly over capacity, and 86% capacity, but achieved the standard consistently only when operated at less than capacity. The scrubber-controlled kiln, whose test results were discounted in the formulation of the standards, was operated at 95% of design capacity during the tests. Stack effluent flow rate (gas velocity) for the tested plants ranged from a high of 180,000 ACFM for Plant C to a low of 48,100 ACFM for Plant E.

Having stated that much, however, the Agency did not explain how the range of test results fully takes account of any significant differences in operating conditions in the industry. The support document is totally devoid of analysis of the relevance or irrelevance of operating level or gas velocity to the achievability of the standard, notwithstanding assertions in the EPA's own contracted-for report that gas velocity bears upon dust generation rates.

(3) Dust Levels at the Tested Plants

The SSEIS asserts, without explaining how the conclusion was reached, that Kilns A, B and E each generated dust at a rate of twenty-two to twenty-five percent (pounds of dust collected per pound of lime produced), higher rates than the average rate of dust generation at the eleven plants for which data were submitted by the NLA. The NLA data, however, indicate a much greater range in dust generation levels than that suggested by the EPA's test plant figures. R. 103, 13, App. 72 (figures ranging from low of six percent to high of thirty-five percent of lime produced).

As laypersons it seems entirely logical to us to suppose that dust generation levels would directly affect emissions controllability, viz., the higher the dust generation, the more difficult the achievability of the standard by the technological control device. But the exact relationship between volume of dust generated and the efficiency of the emissions control systems is never clearly stated or explained by the Agency. Instead, the Agency sends us several mixed signals.

On the one hand, the Agency suggests both directly and indirectly that more dust means a more difficult control problem. The direct suggestion is made in the Agency's rationale for the standard, which states that the two baghouse-controlled test kilns generated "higher [dust levels] than the industry reported average and therefore represent difficult control situations" SSEIS 8-17 (emphasis supplied). The indirect suggestion is made by the standard itself, which permits higher levels of emissions when larger quantities of feed are being burned, a circumstance under which the production of more dust would be expected.

On the other hand, the Agency asserts that the amount of dust generated is irrelevant to the efficiency of at least one control method and therefore to the achievability of the standard. In correspondence with the NLA antedating the standard's proposal, EPA stated, "It is generally accepted that outlet dust concentrations from baghouses vary only slightly with changing inlet dust concentrations." R. 71, 2, App. 57. This statement finds some support in the MRI Report which notes at one point:

The general opinion among the manufacturers of emissions control equipment was that all four types of control systems would be equally tolerant of process upsets leading to short-term heavy dust loadings. In fact, as the dust loadings increase, within a certain limit, the emissions removal efficiency of some of the systems will reportedly increase.

R. 8, 10 (emphasis supplied). However, the MRI Report does not indicate whether long term heavy dust loading or extremely heavy short term dust loading would impair the efficiency of the control system; nor does the report indicate what manufacturers consider to be a heavy or short term dust loading; nor does it indicate on what basis the manufacturers' opinion is predicated.

Our examination of the record thus yields a conflict: while in one breath EPA appears to acknowledge the relevance of dust generation levels to the proposed standard, in another breath the relevance is denied. In our view, the conflict is not adequately explained, nor is the industrywide achievability of the standard adequately justified, in light of the acknowledged possibility that heavy dusting creates a more difficult control problem. From what appears in the record, both variations in dust volume produced and its contributing factors received inadequate attention from the Agency in the development and explanation of this standard.

b. Variations in Controllability of Particulate Generated

The record points to other variables which were also given short shrift in the stated rationale: the use of coal to fuel the kiln (as it relates to controllability of emissions); and variations in size of emitted particles. The record strongly supports the relevance of coal usage to the efficiency of at least the ESP control method and it also suggests a relationship between particle size and the efficiency of both the ESP and the baghouse control method. Nothing indicates how — if at all — variations in these factors were considered in proposing an "achievable" standard.

(1) Coal Usage

It is clear that the trend in the industry is not only toward coal, but toward high sulfur coal, as other energy sources become scarcer. EPA estimates that by 1986, fifty percent of the lime plant new capacity will have high sulfur coal as the only fuel available. SSEIS 3-5. One-half of all coal used will be between one and four percent sulfur content; the average, as high as three percent. SSEIS 6-6. Moreover, conversion to coal is expected to be a major "modification" that will bring old plants into the regulatory orbit under section 111. SSEIS 5-2-5-3, 8-23. Finally, Congress was especially concerned in passing the 1977 Clean Air Act Amendments that the increased use of coal enter into the Agency's regulatory approach.

However, the impact of high sulfur coal usage on the controllability of particulate emissions under any of the three "best" emissions control systems was not clearly or closely examined by EPA in the development of this standard.

With respect to the ESP system, for example, EPA acknowledged that "precipitability [or efficiency of the ESP method] is a function of the chemical composition of the dust particles and will vary with the different kinds of material that make up the kiln exhaust dust (limestone, quicklime, flyash, calcium sulfate, etc.)." SSEIS 4-6. However, neither of the two ESP plants burned coal, the burning of which will affect the chemical composition of the dust and hence the "precipitability" of emissions.

The support document acknowledged:

The tests that were performed on the ESP-controlled kilns are not indicative of normal operation since the current trend in the lime manufacturing industry is toward the use of coal as fuel and the kilns that were tested were fired by oil and natural gas. It is expected that this use of coal would produce a more difficult control problem. However, with proper design of the ESP, it is EPA's judgment that the system could easily meet the level of the proposed standard.

SSEIS 8-12. EPA does not, however, explain the basis for its optimistic judgment that an ESP could meet the standard on a coal burning kiln. Although other factors may affect the chemical composition and hence the precipitability of emissions, EPA's failure adequately to consider the impact of coal usage is a particularly obvious omission.

In still other ways the critical influence of coal, particularly high sulfur coal, was not adequately taken into account. For example, EPA acknowledges that conversion to coal will "cause an increase in particulate emissions in the kiln." SSEIS 5-3. Indeed, three (baghouse-controlled) coal burning kilns were characterized by EPA as "most representative" because they burned coal. SSEIS 8-17. However, of these three "most representative" plants, only two could meet the standard. The insensitivity of the baghouse control method "to small changes in the inlet [dust] loading," SSEIS 5-3, was thought by EPA sufficient to compensate for increased emissions caused by conversion to coal when this method is used, id., but little attention was devoted to this topic.

In addition, the record reflects little consideration of the impact of variations in the sulfur content of coal used. For example, the sulfur content at the coal burning plants tested was considerably smaller than the average projected sulfur content (3 percent) for all new lime plants in the near future. Plant B used 0.6 percent sulfur coal and Plant E used 0.92 percent sulfur coal. Plant F (which failed) used 1.86 percent sulfur coal and Plant A (which also failed) used 1.3 percent.

It is certainly plausible that the use of high sulfur coal will result in a greater increase in uncontrolled or difficult-to-control particulate emissions. (The standards support statement suggests that sulfur content may affect particulate weight. SSEIS D-7.) Yet EPA did not state whether the one coal-converted plant which showed no increase in controlled emissions used high or low sulfur coal.

These little bits of information about the impact of coal usage on the controllability of particulate emissions are left for us to piece together. This obvious and important trend at least deserves to be discussed in a coherent fashion.

Given the high emphasis in the 1977 Clean Air Act Amendments on coal — especially high sulfur coal — as the fuel of choice, we think the effect on emissions of this fuel's use should have been specifically examined and a rationale offered to demonstrate the standard's "achievability" — under any of the best methods of emissions control — when high sulfur coal is burned.

(2) Particle Size

Although there is (a) considerable evidence in the record that the efficiency of available control technology varies with emitted particle size and (b) that lime dust particle size varies regionally (probably due to feedstock variation), the EPA (c) undertook no analysis of the impact of particle size distribution on the achievability of its standard. Each of these points is discussed under separate subheadings below.

(a) The relationship of particle size to efficiency of control methods

That particle size affects the efficiency of at least two of the three "best" technological control systems seems clear.

With respect to the baghouse method of emissions control, the support statement itself states that "[f]abric filter effectiveness is primarily a function of kiln exhaust particle size distribution, fabric type, fabric age and maintenance history." SSEIS 4-2 (emphasis supplied). Indeed, in response to comments submitted in another case (the asphalt concrete NSPS), EPA recognized that "[p]articulate matter which is spherical in shape, has an average fineness below 5 microns, and is slippery and smooth will decrease the performance of a baghouse . . .." EPA, Background Information for New Source Performance Standards, EPA 450/2-74-003, 122 (1974).

With respect to the ESP method, EPA acknowledges that ESPs experience a "relatively low collecting efficiency on submicron particles." SSEIS 4-7. Furthermore, EPA has made a similar acknowledgment with respect to both the ESP and the baghouse method on remand from this court's decision in Portland Cement I. In a document prepared in response to the remand, EPA stated: "These collectors, fabric filters or electrostatic precipitators, are more effective in removing coarse particles than fine particles." EPA, Response to Remand Ordered by U.S. Court of Appeals for the District of Columbia in Portland Cement Association v. Ruckelshaus ( 486 F.2d 375, [D.C. Cir.] June 29, 1973), EPA 450/274-023, 113 (1945) [hereinafter cited as EPA, Response to Remand].

Thus, it seems likely that both dry-collection methods, the ESP and the baghouse, operate more efficiently when the proportion of large to small particles in the emissions is relatively high. It is therefore possible that a kiln which produces a high proportion of fine particulate may not be able to meet the standard, at least using energy-conserving dry collection methods.

(b) Regional variations in particle size

Two early studies on which EPA relies in support of its standard strongly suggest regional or temporal variations in lime particle size. First, the Study of Technical and Cost Information noted: "The size analysis of the [lime] dust being discharged from the kiln may contain as much as 30 percent below 5 microns and 10 percent below 2 microns." R. 1, 35. Second, the Vulcan Report included a table showing that in a typical rotary kiln in Ohio, 12.7% of particulate did not exceed 4.4 microns and 23.8% was smaller than 7.7 microns. Id. 20. The report also cautioned that there was "a significant percentage of `large' particles (larger than thirty-two microns) in this distribution," id. 19, and that "the various percentages associated with particle size distribution . . . may change from state to state depending on the characteristics of the respective limestone deposit." Finally, at an April 30, 1976 meeting between industry and Agency representatives, an industry spokesman made the challenge directly. According to EPA's file memorandum summarizing the meeting, the industry representative

pointed out that there are significant differences between the crystal structures of different limestones. These differences are dependent upon the limestone source and the type of limestone. He stated further that, because of the resulting variations in crystal sizes, particulate emissions could vary greatly from one facility to another (for both the kiln and hydrator). Much discussion of this point followed. Most of the industry representatives echoed this argument. They felt that in plants where the lime product had a large crystal structure, meeting the standard would be much easier than in those plants where the lime product had a small crystal structure.

R. 118, 1. The EPA's response at the meeting was noteworthy:

Mr. Goodwin [of EPA] and ISB members stated that they were not aware of these differences and that if the industry would provide EPA with data to back their claim, [the] standard would be reconsidered.

Id. At the same meting another industry representative suggested:

[r]elative to the particle size problem, . . . that some type of subcategorization of facilities might be needed. Mr. Goodwin stated that EPA would consider subcategorization if [it] receive[d] adequate evidence to show this need. He also indicated EPA's willingness to do further source testing if the lime industry representatives would suggest places they think [EPA] should test and [sic] [EPA] feel[s] additional testing would be productive.

Id.

This promising but aborted exchange dramatically illustrates our dilemma in this case. When particle size was identified as a potentially important variable, both the Agency and the industry failed to pick up the ball.

(c) EPA's lack of analysis

As far as we can tell the Agency gathered no data on particle size distribution at the tested plants or in the industry generally, either before or after the industry meeting which focused on this factor. Whether the EPA took particle size into account in developing and promulgating its proposed standard cannot be determined from this record.

Understandably, the Agency's main defense in court centers on the industry's total failure to respond positively to EPA's suggestion that the industry either suggest additional test sites or submit data on the basis of which EPA might reconsider or subcategorize the standard to conform to local variations. EPA's point is a sympathetic one, but not, we think, dispositive. EPA has a statutory duty to promulgate achievable standards. This requires that they approach that task in a systematic manner that identifies relevant variables and ensures that they are taken account of in analyzing test data. EPA's own support document recognizes particle size as a variable but enigmatically does not discuss it at any length or explain its importance in emissions control. That the industry did not assist the Agency in any meaningful way by data or even by suggestions for additional testing is certainly discouraging. But we do not think that inaction — lamentable though it may be — lifted the burden from the Agency of pursuing what appears to be a relevant variable or at the least discussing in its document why it was not considered important.

In this respect, we believe that the industry's comments, concerning particle size distribution, when viewed in light of the material contained in EPA's own support statement and in light of the background documents on which it relied, met a "threshold requirement of materiality," mandating an Agency response which was not forthcoming here.

c. Explanation of discarded Data From Plants A and F

Finally (with respect to the rotary kiln particulate emission standard), a few words should be devoted to the mysterious Plant A and the plant controlled by a low-pressure venturi scrubber (Plant F.). Test results obtained at Plant A were excluded from consideration and those obtained at Plant F were discounted (if not excluded entirely from consideration) in the rationale for the proposed standard. This was because after testing it was concluded these plants did not represent best technology. SSEIS 8-17, 8-18. At both plants the measured particulate emissions had significantly exceeded the proposed standard. The only reason — apart from the poor test result — given for the conclusion that Plant A did not represent best technology was as follows:

The Plant A baghouse is not typical of those in use in the lime industry. Large quantities of dilution air infiltrate through the corrugated asbestos siding and doors into the clean air side of the baghouse. It is unknown how this affects the performance of the baghouse, but this baghouse did not perform as well as the two other baghouses (Plants B and E) that were source tested in conjunction with this study.

SSEIS C-69.

It would appear that EPA's observation of "large quantities of dilution air" at this plant is related to its measurement of high oxygen levels in the effluent. Yet when the oxygen determination at the plant was questioned as "thermodynamically impossible" (R. 139, 7, App. 189), EPA conceded error but offered no other reason to support its conclusion that Plant A did not represent best technology. We think it incumbent upon the Agency, at least where it chooses to propose a standard on a data base as apparently limited as this one, to offer some supportable reason for its conclusion that a tested plant, chosen as likely to be well-controlled, does not represent best technology. The mere fact that its test results were unsatisfactory is not enough.

If, for unexplained reasons, one-third of the test plants initially chosen by EPA for their well-controlled systems fail to meet the standard, the conclusion is just as plausible that the standard is not achievable as that the plants chosen did not have well-controlled systems. It is up to EPA to dispel such doubts, and they have not done so here.

Of course, the fact that Plant A did not meet the proposed standard does not itself prove the standard is unachievable. However, ignoring the Plant A results merely because they were not satisfactory would suggest that the process by which the standard was promulgated was an arbitrary one. This is especially true where the results excluded are those obtained from one of only three plants tested which utilized the existing technology (baghouse) "that approximately 80 percent of the new and modified facilities subject to the proposed standards would use . . .." SSEIS 8-13.

EPA's handling of the Plant F (scrubber) results does not seem as troubling, primarily because neither the trend in the industry nor this standard favor the use of scrubbers for rotary kilns. It was, however, the only scrubber-controlled plant tested and it did not meet the standard. EPA attributed the poor results to the low pressure employed by the Plant F scrubber and hypothesized that a higher pressure scrubber could meet the standard proposed. In support of this hypothesis EPA relied upon a non-EPA-conducted test reported in the literature, although the conditions under which that test was conducted were not mentioned. SSEIS 8-12. Were the venturi scrubber projected to be in use for any sizable number of new or modified lime plants, we would be considerably less comfortable with the Agency's conclusion that "EPA['s] . . . source test . . . show that all [three control devices] are capable of meeting the particulate emission level of 0.15 kilogram per megagram . . .." SSEIS 8-12.

2. Hydrators

Since EPA has already agreed to a remand of the standard for "pressure" hydrators, we consider the standard only as it relates to "atmospheric" hydrators.

EPA conducted particulate emission tests on two hydrators, both controlled by wet scrubbers. Each was tested three times. Average emissions at both plants fell below the standard.

However, in reviewing the record in light of the industry's attack, we have encountered the same problem with the hydrator standard as with the rotary kiln standard. There is record evidence substantial enough to raise a real question in our minds whether adequate account was taken of significant variables relevant to the standard's achievability.

Material submitted by the NLA at its June 1977 meeting with EPA suggests that lime hydrators (like rotary kilns) produce particles of different size and surface area.

Since the efficiency of the wet scrubber method of emissions control apparently depends on the probability that dust particles will collide with and be captured by small water droplets which are sprayed into an area through which the effluent must pass, the size and surface area of the particulates to be captured would certainly seem important. Yet the relevance of particulate size and surface area is nowhere addressed by EPA, insofar as this record reveals.

All the record reflects is that both hydrators utilized calcitic (rather than dolomitic) lime, again with no explanation of the relevance of that item of information to the achievability of the proposed standard on an industry-wide basis. Since the comments submitted by NLA in connection with the rotary kiln standard suggest that particle size in calcination is affected by the chemical composition of the material used, a similar effect might therefore be anticipated in the hydration process; but the EPA does not address this possibility either through assumptions, tests performed, data collected and reported, or analysis of results. We are asked to conclude that the projection of an achievable standard for the industry as a whole based on tests conducted at two hydrator plants using calcitic stone represents a reasoned decision, without knowing why.

Because we remand, the Agency will have the opportunity to consider the hydrator standard more fully in light of the additional material and more elaborate arguments relating to the achievability of the standard for hydrators that were first submitted by the industry when the matter was brought to this court.

B. The Opacity Standard and Continuous Monitoring Requirement 1. The Opacity Standard

"Opacity" is defined by regulation to mean "the degree to which emissions reduce the transmission of light and obscure the view of an object in the background." 40 C.F.R. § 60.2(j) (1979). EPA explains that "[t]he opacity level of visible emissions is an indication of the mass concentration of a particular pollutant" and that "[v]arious studies have shown that opacity varies directly with mass concentrations of particulate matter." SSEIS 8-19. EPA considers opacity standards to be "a necessary supplement to particulate mass emission standards" basically because "[o]pacity test methods are quicker, easier to apply, and less costly than concentration/mass tests for particulate matter." SSEIS 8-19.

The performance standards prescribed by EPA for rotary lime kilns consist of both a mass emission standard (grams of particulate emission per gram of feed) and an opacity standard (ten percent). 43 Fed.Reg. 9453 (1978). Only those kilns using dry methods of emissions control are subject to the ten percent opacity standard. As previously noted, no opacity standard was promulgated for lime hydrators (which almost never employ dry control methods) and rotary kilns using wet scrubbers have been exempted from compliance with the opacity standard.

We have considered the various arguments made by the NLA and conclude that EPA's apparent failure to consider in this case some variables which were (1) given more careful consideration in the promulgation of earlier opacity standards and (2) given inadequate consideration in the companion mass emission standard requires us to remand the opacity standard to the Administrator for additional explanation or for revision.

a. Variables Considered in the Promulgation of Earlier Opacity Standards

On remand from Portland Cement I, 486 F.2d 375, the Administrator undertook extensive reconsideration of both the opacity standard proposed for portland cement plants and the methodology (EPA's "Method 9," 40 C.F.R. Part 60, App. A (1979)) of opacity measurement. In the year and a half that followed, the methodology was revised and the standard set for portland cement plants was raised from ten to twenty percent. EPA has explained its relaxation of the portland cement standard as an effort to accommodate the complete range of available data obtained in that case, having adjusted the data for stack diameter. SSEIS II, 13. Stack diameter was thus a variable for which EPA made adjustments in the portland cement case but it was not the only variable considered in formulating that standard. It is clear that the possible impact of other variables were also taken into account on remand, including: particle size and shape, EPA Response to Remand 112; and stack gas exit velocities, id. 116.

The impact of variations in particle size and shape were also considered by EPA in evaluating an opacity standard for asphalt concrete plants. It is this difference in Agency methodology underlying the various new source opacity standards and not the difference in the standards themselves that gives us most pause. No attention to particle size and shape appears to have been given by EPA in the preparation of opacity standards for lime plants.

b. Variables Inadequately Considered in Mass Emission Standard

Opacity standards are intended to operate in tandem with mass emission standards, notwithstanding their independent enforceability. Ideally, a violation of an opacity standard should indicate a violation of a mass emission standard. See SSEIS 8-19. For this reason the Agency relies on data from the same test plants to support both the opacity and the mass emission standard; but for this reason when the representativeness of data relied upon for one standard is inadequately shown, the representativeness of data relied upon for the other standard is drawn in question.

As discussed above, the Agency failed to consider the representativeness of the particle size produced at its tested plants. This failure is particularly striking in connection with the opacity standard because variations in particle size have been given careful consideration in the development of earlier opacity standards.

We have already noted that the emissions control systems favored by the standards and by prevailing economic and technological trends may operate more efficiently when the predominant size of particulate emissions is large. As it happens, large particulate is also likely to appear less opaque. Thus, it is possible that a plant would meet both standards only because the particles emitted are uniformly large and we cannot ascertain how the plants tested here "measure up."

c. EPA's Arguments

Both in this court and at the administrative level EPA emphasizes the overwhelming extent to which the plants tested were able to meet the ten percent opacity standard. But without knowing the representativeness of the plants tested or of test conditions, we cannot say that the standard is neither arbitrary nor capricious. Certainly the fact that virtually all plants tested were able to meet the standard is an important consideration, but our doubts are sufficient, when coupled with our doubts concerning the mass emissions standard (discussed above), to remand to the Agency for amplification of the record.

EPA has committed itself to take the possibility of inaccurate opacity measurement into account in the enforcement of the standard. It has also provided a type of "variance" mechanism under which new sources which meet the mass emission standard but which cannot meet the opacity standard may petition the Administrator to establish a separate opacity standard for that facility. 40 C.F.R. § 60.11(e) (1979), 39 Fed.Reg. 39872 (1974). The variance mechanism, however, seems clearly to have been intended to be narrowly construed.

The Agency relies upon the flexibility built into the regulatory scheme to support the rationality of its standards. The wisdom of such flexibility has been applauded on earlier occasions by this court, but the statutory scheme prescribes "achievable" standards and there is a limit to the flexibility with which the Agency is or should be endowed.

We recognize the usefulness of opacity standards as an enforcement tool. Opacity can be monitored by the Agency with little advance warning or costly preparation. We also realize that "[o]pacity standards are not novel . . . opacity standards have been upheld previously by this court[] under closely analogous circumstances . . . [and that] Congress . . . has expressed concern for opacity values in measuring air pollution under the Clean Air Act. . . . Alabama Power Co. v. Costle, No. 78-1006, slip op. at 32, (D.C. Cir. Dec. 17, 1979) (Wilkey, J.), and we do not wish to imply that the Agency cannot justify their use. We remand to the Agency because on this record the reasonableness of the standard has not been demonstrated.

2. Continuous Monitoring

On the opacity monitoring requirement, the petitioner's argument is simple: there is no adequately demonstrated technology for monitoring opacity. One company operating affected facilities (Dow Chemical) commented, "We have tried several continuous monitoring systems in the past and have been unable to find an instrument that will suitably do the job and can be maintained in operation." R. 148, App. 327-28. The company cited high opacity readings attributable to instrument malfunctioning "as frequently as twice a day" and also remarked that "[t]he opacity readings [of the monitor] do not relate to the actual stack conditions as measured by visual observers."

EPA answers that the continuous monitoring data would not be used to determine compliance with the opacity standard but "to keep a check on the operation and maintenance of the control equipment," and to trigger performance checks by trained observers. Brief for Respondents at 12-13, citing SSEIS 8-24 and standard as proposed ( 42 Fed.Reg. 22506, 22509 (1977)). The Agency argues that if the equipment gives any "indication" of changed opacity it is enough to justify a continuous monitoring requirement. Brief for Respondents at 29. It dismisses the industry's contention that reliable monitoring equipment is not available to perform this limited a function and shifts the burden to the industry to show "by supporting data," SSEIS II, 13, that it is not.

EPA states that it now routinely requires continuous monitoring of opacity in new source performance standards. Brief for Respondents at 14-15. Opacity monitoring was first required and performance specifications for monitoring systems prescribed in connection with the NSPS for fossil-fuel fired steam generators and petroleum refineries. As of the date the lime standard was proposed, five other promulgated NSPS included a requirement for the continuous monitoring of opacity. Since the lime standard's proposal, at least one other standard has been promulgated that contains such a requirement. In answer to NLA's observation that no continuous opacity monitoring is required of portland cement plants, R. 139, 14 App. 196, EPA informs the court that the Agency is now reviewing the portland cement standards pursuant to § 111(b)(1)(B), 42 U.S.C. § 7411(b)(1)(B) (Supp. I 1977) to see if the same requirement should be imposed there. Brief for Respondents at 15.

EPA maintains that it has had considerable experience with the use of continuous monitoring devices and that in its experience a monitor will show if an emissions control device is being properly operated and maintained and the opacity standards met. SSEIS II, 14; SSEIS 8-24. Thus monitoring will act as a needed warning alarm when the control system is out of kilter. SSEIS 8-24.

The industry itself admits there is some value to a continuous monitoring requirement. Dow Chemical took a critical stance (adopted by NLA, Brief for Petitioner at 52) but also acknowledged that monitoring equipment "gives an indication of whether the opacity is increasing or decreasing." R. 148, 2, App. 328. Given this concession, we cannot find the continuous monitoring requirement arbitrary as an adjunct to a non-arbitrary, non-capricious opacity standard. We have today remanded the opacity standard for lime plants. If on remand an opacity standard is retained, EPA may continue to require continuous monitoring.

V. THE STANDARD OF REVIEW AS APPLIED

Our requirement that the EPA consider the representativeness of the test data relied upon in the development and justification of its standard does not presage any new or more stringent standard of judicial review. The rigorousness of the review in which this court has engaged in previous NSPS decisions — known to some as the "hard look" standard — has already been described.

In enacting the Clean Air Act Amendments of 1977, Congress expressly approved the rigorous standard of review which the courts had theretofore applied to Agency decisions under the Clean Air Act. Although the judicial review provisions of the 1977 Amendments do not apply to this rulemaking proceeding, Congress' express affirmance of the standards already developed fortifies our adherence to the learning of our earlier Clean Air Act decisions in reviewing the new source performance standards currently before us.

We think these decisions amply support our conclusion that a remand is appropriate in this case. Both decisions reviewing the NSPS and those reviewing other administrative determinations under the Clean Air Act evince a concern that variables be accounted for, that the representativeness of test conditions by ascertained, that the validity of tests be assured and the statistical significance of results determined. Collectively, these concerns have sometimes been expressed as a need for "reasoned decision-making" and sometimes as a need for adequate "methodology." However expressed, these more substantive concerns have been coupled with a requirement that assumptions be stated, that process be revealed, that the rejection of alternate theories or abandonment of alternate courses of action be explained and that the rationale for the ultimate decision be set forth in a manner which permits the public to exercise its statutory prerogative of comment and the courts to exercise their statutory responsibility upon review. The standard we apply here is neither more rigorous nor more deferential than the standard applied in these earlier cases.

Our opinion should not suggest the necessity of "ninety-five percent certainty" in all the "facts" which enter into the Agency's decision. We would require only that the Agency provide sufficient data to demonstrate a systematic approach to problems, not that it adduce vast quantities of factual data. However, where the facts pertinent to the standard's feasibility are available and easily discoverable by conventional technical means, there is somewhat less reason for so limited a data base. Nothing in the record suggests the relevant facts are not readily accessible to the Agency; the number of plants is large, use of the control methods found by the Agency to represent the "best systems" is wide-spread, and stack emission measurement techniques have been known and applied for many years.

With respect to the standard's achievability we are thus not presented with the question how much deference is owed a judgment predicated on limited evidence when additional evidence cannot be adduced or adduced in the near future. We do not depart from some of the most carefully considered and closely reasoned decisions of this court which permit an agency latitude to exercise its discretion in accordance with the remedial purposes of the controlling statute where relevant facts cannot be ascertained or are on the frontiers of scientific inquiry.

A systematic approach may not necessarily require a conclusion grounded in actual test results. We do not intend to bridle the Agency's discretion to make wellfounded assumptions even where the assumption could be replaced by valid test results, but we think first, the assumption should be stated and second, where test data could have verified the assumption, a reason for not testing or relying on such data should be given.

We recognize, for example, that the finding of facts, especially through elaborate testing, is costly and the costs of additional testing may be added by the Agency to the costs of delay in issuing the proposed rule and the sum of these costs weighed against the benefit of proposing a rule without additional data.

We leave to the Agency on remand the decision whether additional Agency-conducted testing is appropriate in this case. Data may already be available to the Administrator which would support the achievability of these standards for the industry as a whole. If so, satisfaction of the concerns we have expressed in this opinion may be a fairly simple matter.

To ensure that the Agency has engaged in reasoned decisionmaking, we remand. We have outlined our substantive misgivings; the Agency may choose the appropriate method of response.

Remanded.