The Environmental Protection Agency (EPA) appeals the district court's finding that the EPA's order requiring Asarco, Inc. to install a sampling station in the 1000 foot stack at Asarco's copper smelter was arbitrary and capricious. The case requires us to consider the extent to which the district court may go outside the administrative record to review agency action. Although we find that the district court went too far in its consideration of evidence not in the administrative record, we find, based on our own review of the administrative record, that the EPA's order was arbitrary and capricious, and remand the matter for further consideration by the agency.

Asarco is a copper producer that owns and operates a copper smelter in Hayden, Arizona. In 1974, Asarco constructed and commenced operation of a 1000 foot stack, which emits into the atmospheres gases and particulate matter that enter and mix in the stack from two separate flues or ducts. One flue carries gases and particulate matter from the Smelter's roaster and reverberatory furnaces. The other carries gases and particulate matter1 from its converters. Asarco has sampling facilities on each flue to test emissions passing through the flues. The sampling station in the roaster and reverberatory flue is approximately 175 feet from the stack; the sampling station in the converter flue is about 1000 feet from the stack. Although there is a porthole at the 400-foot level of the stack for gas sampling, the stack has no sampling facilities to test particulate matters. Consequently, Asarco cannot measure the amount of particulate matter that exists after the gases from the two ducts are mixed in the stack.

Pursuant to the Clean Air Act, 42 U.S.C. § 7401-7626,2 the state of Arizona adopted a plan to regulate the allowable discharges of particulate matter into the atmosphere from air pollution sources such as Asarco. The EPA Administrator disapproved the portion of the Arizona implementation plan as it related to the control of particulate matter in the Phoenix-Tucson Intrastate Air Quality Control Region, of which Asarco's Hayden facility was a part. The EPA then adopted a substitute "process weight" regulation, prescribing limitations on the amount of particulate emissions from process sources such as those of the Hayden facility. See 40 C.F.R. § 52.126(b). The regulations also incorporated the test methods and procedures that were to be followed by operators of these process sources, i. e., Methods 1, 2, 3, 4, and 5 contained in Appendix A to 40 C.F.R. Part 60. The EPA's replacement regulation was to remain in effect until Arizona adopted an implementation plan approved by the EPA.

In 1973, the EPA first notified Asarco of the air quality standards and specified dates by which it should be in compliance. Asarco contended that it was unable to comply because it was unable to operate at full capacity, as required by EPA test regulations. The EPA reiterated its position that testing must be done at full capacity. After this exchange of correspondence between the EPA and Asarco, the EPA notified Asarco in April 1975 that it was in violation of EPA regulations because of Asarco's failure to submit performance test data and to demonstrate compliance.

Pursuant to § 113(a)(4) of the Clean Air Act, 42 U.S.C. § 7413(a)(4), a conference was held in Phoenix on May 1975 at which representatives of the EPA, Asarco, and the state of Arizona attended to discuss Asarco's non-compliance. At the conference, Asarco agreed to conduct performance tests at the two sampling facilities in the flues while operating at full capacity. The Arizona representatives remarked that it believed that chemical reactions were occurring in the stack by the mixture of the gases from the two flues, and that Asarco should install a sampling fixture in the stack to measure the mixture. Asarco's attorney objected to the state's position. He argued that the purpose of the proceeding was to discuss Asarco's compliance with the testing requirements, and that Asarco was unprepared to "battle" on the issue of where the testing was to take place. The parties appeared to agree that Asarco would be given the opportunity to supply data on the need for testing in the stack before any action on the matter would be taken. Accordingly, on June 17, 1975, L. G. Cahill, Asarco's plant manager, sent a letter to the EPA discussing the possibility that chemical reactions might occur in the stack and restating Asarco's position that accurate tests could be conducted at the existing sampling sites.

On June 19, 1975 the EPA issued an order finding Asarco in violation of § 114(a)(1) of the Clean Air Act, 42 U.S.C. § 7414(a)(1), for failing to conduct the required performance testing for particulates emanating from the two flues. The order contained no mention that stack testing would be required. The EPA made no response to Cahill's letter.

As agreed, Asarco conducted further performance testing at the sampling facilities in the two flues on June 30, 1975. In an internal memorandum, the EPA concluded that the test data was inadequate to demonstrate compliance with the emissions standards contained in 40 C.F.R. 52.126(b). It added that it was concerned that the testing took place in the ducts rather than the stack because of the particulate reactions that would be expected to occur because of the interaction of the two exhaust systems.

In September 1975, Asarco again conducted compliance testing for particulate emissions. Upon receipt of the test results, the EPA internally concluded that, with a few reservations, the procedures had complied with EPA regulations. The memo noted that source tests should be conducted in the stack rather than the flues.

In February 1976, the Acurex Corporation, an independent consulting firm, completed a study for the EPA on sampling at the Hayden smelter. The study was prepared by its employee, James Steiner, who visited the plant for several hours on February 6, 1976. Steiner concluded the present sampling sites were inadequate and that it would be possible to make measurements at the existing locations only if there were expensive modifications of those facilities. He recommended that Asarco construct the proper sampling facilities on their 1000 foot stack to insure measurement accuracy. Steiner's report was not furnished to Asarco.

On April 20, 1976, the EPA advised Asarco that it was required to install sampling facilities in the 1000 foot stack so that source tests could be conducted in accordance with EPA test methods. It specified guidelines for the construction of the facility, and required that Asarco conduct performance tests in the stack within 150 days.

Asarco responded by letter ten days later. It stated that installation of sampling facilities in the stack, while technically feasible, was not economically feasible because it would cost in excess of $300,000. Asarco also asserted that the present sampling facilities, as evidenced by three series of tests (including the June and September tests), demonstrated compliance with EPA testing requirements.

On June 2, 1976 the EPA reiterated its position that a sampling site must be constructed within the stack. Several days later, representatives of the EPA and Asarco met to discuss the proposed monitoring requirements. At the meeting, Asarco provided the EPA with technical comments in support of its position that stack testing was unnecessary because the amount of particulate matter formed in the stack was indeterminable. Asarco provided the EPA with additional technical data a few days later, and expressed its willingness to modify the converter flue and conduct new tests in both flues. The EPA issued an order of violation on July 22, 1976, citing Asarco's failure to install a sampling site in the stack and conduct adequate performance tests. A conference was held between the EPA and Asarco to discuss the monitoring requirements, and Asarco reiterated its position that, although it was willing to rectify the deficiencies in its present compliance testing, a sampling facility in the stack was unnecessary. On September 15, 1976 the EPA issued a revised order requiring construction of stack testing facilities, and informed Asarco that non-compliance could lead to potential civil or criminal liability.

Asarco then filed its present action, claiming that the EPA's finding of violation and its order and revised order were invalid as arbitrary and capricious insofar as it required Asarco to construct stack testing facilities at its Hayden smelter. It also requested a stay of enforcement pending judicial review. The EPA moved to dismiss the action on the ground that the district court did not have jurisdiction to conduct preenforcement review. When this motion was denied, the EPA counterclaimed for enforcement.

The district court then conducted a four-day hearing on the merits of Asarco's complaint. Numerous experts testified during the hearing about the necessity for the EPA's stack testing requirement. The administrative record was also admitted. The district court found that the evidence did not support the conclusion that significant chemical reactions occurred after combination of the two gas streams and held that the orders requiring stack-testing were arbitrary and capricious.

CONSIDERATION OF EVIDENCE OUTSIDE THE ADMINISTRATIVE RECORD

The EPA disputes the district court's finding that its order was arbitrary and capricious. First, however, it contends that the district court exceeded the scope of its authority by hearing expert testimony instead of confining itself to review of the administrative record. Asarco responds by arguing that the district court did not conduct a de novo review of the EPA's action,3 and, in any event, that the EPA waived its objection to the scope of the review by its acquiescence in the procedure adopted by the district court.

Courts have apparently split over the issue whether an agency can waive the requirement that a district court must not conduct a de novo review of agency adjudicatory action. Compare Cooperative Services, Inc. v. United States Department of Housing and Urban Development, 562 F.2d 1292, 1295 (D.C.Cir.1977) (allowing agency to stipulate to de novo review) with Independent Meat Packers Association v. Butz, 526 F.2d 228, 239 n.30 (8th Cir. 1975), cert. denied, 424 U.S. 966, 96 S. Ct. 1461, 47 L. Ed. 2d 733 (1976) (parties cannot consent to de novo review because the district court is not empowered to conduct one. We need not reach this issue, because we find that the EPA did not waive its right to object to the manner in which the district court conducted proceedings.

Asarco argues that the EPA failed to preserve its objection because it entered into various stipulations about the manner in which the review hearing would be conducted and did not object during the hearing when evidence outside the record was introduced. Counsel for the EPA, however, made clear at the outset of the hearing the EPA's position on the scope of review:

I would like to clarify for the record that the first sentence in Paragraph 2 is our precise position in terms of the scope of review,4 and the second sentence concerning the consideration of additional facts, this sentence is in response to judge Fitzgerald's ruling on the motion to dismiss that additional evidence would be taken.

We would like our precise position to be stated for the record, that the administrative record only would be the basis for review of the reasonableness of the monitoring requirements.

Once the EPA made its position clear, it would have been futile for it to continue to raise objections throughout the course of the proceedings. See Independent Meat Packers Association, 526 F.2d at 239 n.30. Although the EPA may have made a few statements in the course of its discussion with the court inconsistent with its present position, it is plain that the EPA clearly communicated its position to the court and that the court rejected it. The EPA, therefore, has not waived its right to object here.

The parties do not dispute that the standard of review of EPA's order is whether the EPA's action was "arbitrary and capricious." See Citizens to Preserve Overton Park v. Volpe, 401 U.S. 402, 91 S. Ct. 814, 28 L. Ed. 2d 136 (1971). They vigorously disagree, however, about the scope of review, i. e., the extent of the district court's inquiry in applying the arbitrary and capricious standard. The EPA contends that the district court erred by not limiting itself to the administrative record. Asarco contends that the district court acted properly in also considering expert testimony and documents produced for the first time at the district court hearing.

The analysis of the proper scope of review begins with the Supreme Court's decision in Citizens to Preserve Overton Park v. Volpe, 401 U.S. 402, 91 S. Ct. 814, 28 L. Ed. 2d 136 (1971). In Overton Park, a citizens' organization challenged the Secretary of Transportation's approval, without findings, of a highway through a park under a statute forbidding such a highway if a feasible alternative route exists. Relying on affidavits prepared by the parties, the lower courts held that there was no basis for determining that the Secretary had exceeded his authority. The Court remanded the case to the district court. The appropriate standard of review was "arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law." Id. at 413-14, 91 S.Ct. at 822. Under this standard, de novo review of the agency's action was not warranted. Id. at 415, 91 S.Ct. at 823. De novo review is authorized only where the action is adjudicatory and agency fact-finding procedures are inadequate, or where issues not before the agency are raised in a proceeding to enforce nonadjudicatory agency action.

Nevertheless, the Court held that the reviewing court is required to engage in a "substantial inquiry," i. e., "a thorough, probing, in-depth review." Id. Pursuant to this inquiry, the reviewing court must first consider whether the Secretary acted within the scope of his authority. Next, to determine whether the decision was arbitrary or capricious, the court must consider whether the decision was "based on a consideration of the relevant factors and whether there has been a clear error in judgment." Id. Although this factual inquiry is to be "searching and careful" the ultimate standard of review is narrow. "The court is not empowered to substitute its judgment for that of the agency." Id. at 416, 91 S.Ct. at 824. Finally, the court must inquire whether the Secretary followed the necessary procedural requirements.

Because the lower courts had relied on the litigation affidavits and the administrative record was not before the Court, the Court remanded to the district court for "plenary review" of the Secretary's decision. Id. at 420, 91 S.Ct. at 825. The district court's review was to be based "on the full administrative record that was before the Secretary at the time he made his decision." Id. at 420, 91 S.Ct. at 825. Because the bare record might not disclose the factors considered by the Secretary, the district court could require some testimony from the administrative officials who participated in the decision explaining their action. Alternatively, the Secretary could prepare formal findings as a means of providing adequate explanation.

In Camp v. Pitts, 411 U.S. 138, 93 S. Ct. 1241, 36 L. Ed. 2d 106 (1973), the Court reiterated that, where the district court reviews agency action under the arbitrary and capricious standard, it is not entitled to conduct a de novo hearing. Pitts involved review of the Comptroller of the Currency's decision to deny a national bank charter. The district court upheld the Comptroller's action upon a review of the administrative record, but the Court of Appeals remanded to the district court for de novo review because the Comptroller's ruling was not stated with sufficient clarity to allow appellate review.

The Supreme Court found this procedure unwarranted. The focal point for judicial review is the administrative record already in existence, "not some new record made initially in the reviewing court." Id. at 142, 93 S.Ct. at 1244. Instructions requiring the parties to make an evidentiary record in the district court suggests that the administrative record is to be put aside. If additional explanation is necessary, the proper course is not to hold a de novo hearing but to obtain, through affidavit or testimony, further explanation from the agency. If the finding, once explained, is not sustainable on the administrative record, then the Comptroller's action must be vacated and remanded to him for further consideration. This court has applied these principles.

In Proietti v. Levi, 530 F.2d 836 (9th Cir. 1976), a case involving review of the Attorney General's decision to deny a request by Master Sergeant Proietti for legal defense in a state court tort action, the district court conducted a de novo hearing and found the final decision of the Attorney General to be arbitrary and capricious. The court vacated, holding that the district court should have "focused entirely on the administrative record compiled by the Attorney General." Id. at 838. If the district court found that the administrative record did not sustain the action, the remedy was not to take additional evidence, but instead to remand the matter for reconsideration by the agency.

This court further enlightened the role of the court conducting review of agency action in Bunker Hill Co. v. EPA, 572 F.2d 1286 (9th Cir. 1977). There, the court reviewed the EPA's decision to reject portions of a state implementation plan under the Clean Air Act and substitute its own. Because of the highly technical nature of the subject matter, the court decided that it was not "straightjacketed" to the original record and requested the parties to provide supplementary materials to aid the court in its understanding. These new materials were merely "explanatory" of the original record and contained no new rationalizations. Id. at 1292. See also Association of Pacific Fisheries v. Environmental Protection Agency, 615 F.2d 794, 811, (9th Cir. 1980).

A number of rules governing the scope of judicial review of agency action emerge from these cases. Predominant is the rule that agency action must be examined by scrutinizing the administrative record at the time the agency made its decision. The Supreme Court recognized in Overton Park, however, that even where the agency has employed adequate fact-finding procedures, the courts may find it necessary to go outside the agency record to evaluate agency action properly. The Court contemplated that any additional material should be explanatory in nature, such as requiring the involved administrative officials to demonstrate the basis for their action. A satisfactory explanation of agency action is essential for adequate judicial review, because the focus of judicial review is not on the wisdom of the agency's decision, but on whether the process employed by the agency to reach its decision took into consideration all the relevant factors.

The same cases make clear that judicial consideration of evidence relevant to the substantive merits of the agency action but not included in the administrative record raises fundamentally different concerns. When a reviewing court considers evidence that was not before the agency, it inevitably leads the reviewing court to substitute its judgment for that of the agency. This is true even if such judicial review is not strictly de novo in the sense that the court also considers the administrative record. Nevertheless, as we noted in Bunker Hill, it is both unrealistic and unwise to "straightjacket" the reviewing court with the administrative record. It will often be impossible, especially when highly technical matters are involved, for the court to determine whether the agency took into consideration all relevant factors unless it looks outside the record to determine what matters the agency should have considered but did not. The court cannot adequately discharge its duty to engage in a "substantial inquiry" if it is required to take the agency's word that it considered all relevant matters.

We think these conflicting considerations can be satisfactorily reconciled. If the reviewing court finds it necessary to go outside the administrative record, it should consider evidence relevant to the substantive merits of the agency action only for background information, as in Bunker Hill, or for the limited purposes of ascertaining whether the agency considered all the relevant factors or fully explicated its course of conduct or grounds of decision. See Association of Pacific Fisheries, supra. Consideration of the evidence to determine the correctness or wisdom of the agency's decision is not permitted, even if the court has also examined the administrative record. If the court determines that the agency's course of inquiry was insufficient or inadequate, it should remand the matter to the agency for further consideration and not compensate for the agency's dereliction by undertaking its own inquiry into the merits.

With these principles in mind, we think that the district court went too far in its consideration of evidence outside the administrative record. The "trial" conducted by the district court was four days in duration and filled seven volumes of transcript. Numerous exhibits were introduced. Asarco called seven witnesses on its behalf. Cahill, the plant manager at Asarco's Hayden smelter, primarily described the basic operation of the Hayden smelter and the history of Asarco's communications with the EPA concerning compliance with its testing requirements. Lawrence Bowerman, an EPA environmental engineer involved in the decision to require stack testing, explained the process by which the EPA arrived at the decision.

Asarco also called three of its employees who were knowledgeable about the technical aspects of particulates emitted from the Hayden smelter. The testimony of these witnesses concerned the level of emissions, the adequacies of the present facilities and their opinions as to the EPA's proposed testing requirements. This testimony was often highly technical in nature. Three outside experts also testified on behalf of Asarco. These experts provided technical opinions on the adequacy of Asarco's existing facilities, the probability of particulates forming in the stack and their amounts, and the need to conduct stack-testing.

The EPA called three witnesses. James Steiner and Ben Carpenter, two outside experts, testified why they thought the stack-testing requirement was reasonable. Steiner, who had visited the plant and submitted a report to the EPA on which the EPA relied, explained the methodology of his report as well as his general technical observations. Carpenter, who was engaged by the EPA for purposes of the litigation, provided his calculations about the amount of particulates formed in the stack and responded to the findings of the Asarco witnesses. Bowerman also testified for the EPA, stating that his opinion of the need for stack-testing had not been changed by any evidence that he had heard at trial.

Some of this testimony, particularly that of Cahill and Bowerman, could be considered under the standards we have set forth. Cahill provided background material on the operations of the Hayden smelter, and both Cahill and Bowerman elucidated the process by which the EPA reached its decision. Most of the expert testimony, however, should not have been admitted or at least not have been considered for the purpose of judging the wisdom of the EPA's stack-testing requirement. This technical testimony was plainly elicited for the purpose of determining the scientific merit of the EPA's decision. Although the testimony may have served some marginal purpose in allowing the district court to evaluate the EPA's court of inquiry, we can only conclude that the extent of the scientific inquiry undertaken at trial necessarily led the district court to substitute its judgment for that of the agency.

REASONABLENESS OF THE STACK TESTING REQUIREMENT

This shortcoming, however, does not prevent this court from reviewing the administrative record, supplemented by proper explanatory evidence established in the district court, and determining for itself the reasonableness of the EPA's action. See Proietti v. Levi, 530 F.2d 836, 838 (9th Cir. 1976) (suggesting that court of appeals could undertake review if in possession of complete administrative record). District court review of agency action is generally accorded no particular deference, because the district court, limited to the administrative record, is in no better position to review the agency than the court of appeals. See First National Bank of Fayetteville v. Smith, 508 F.2d 1371, 1374 (8th Cir. 1974), cert. denied, 421 U.S. 930, 95 S. Ct. 1655, 44 L. Ed. 2d 86 (1975). Although the court of appeals, in its discretion, could remand to the district court, we believe further delay in this case to be unnecessary.

In reviewing the EPA's stack testing requirement under the arbitrary and capricious standard, we emphasize that our attention is focused on whether the EPA considered all relevant factors in arriving at its decision. See Overton Park, 401 U.S. at 415, 91 S. Ct. at 823. The EPA must provide a reasoned basis for its action, fully explaining its course of conduct, i. e., its reasoning, analysis, and inquiry. See Office of Communication of United Church of Christ v. FCC, 590 F.2d 1062, 1069 (D.C.Cir.1978); Appalachian Power Co. v. Train, 545 F.2d 1351, 1356 (4th Cir. 1976). Guided by these principles, we conclude that the EPA arrived at its stack testing requirement in a manner that was arbitrary and capricious.

Under the Clean Air Act, the Administrator may issue orders requiring persons to comply with the requirements of an implementation plan. See 42 U.S.C. § 7413(a). In this case, the Administrator has attempted to gain Asarco's compliance with the process weight regulations that the EPA substituted for Arizona's implementation plan. For the purpose of determining whether a person is in violation of these standards, the Administrator may require the installation, use, and maintenance of monitoring equipment and the sampling of emissions. See 42 U.S.C. § 7414(a)(1). Clearly, the Administrator has the authority to require the installation of a sampling facility. The issue here is whether he exercised that authority in an arbitrary and capricious manner.

The EPA maintains that there are two reasons why it acted reasonably in ordering installation of a stack sampling facility. First, it contends that the existing sampling facilities contained structural deficiencies that could not be corrected without considerable expenditure. Second, there was a likelihood that new particulate emissions were being formed in the stack and that these emissions could be measured only by sampling in the stack.5 Although these factors, whether standing alone or in some combination conceivably could justify the installation of the stack, the EPA's consideration of both factors was inadequate.

First, the administrative record at the time of the EPA's final decision does not disclose a reasoned scientific basis for concluding that significant amounts of particulates would form in the stack. The administrative record shows that on several occasions Asarco failed to provide test data in compliance with EPA testing methods.6 The memoranda discussing these inadequacies focused on Asarco's failure to comply with approved testing procedures, particularly its failure to operate at full capacity during the tests. No mention is made in these reports of inherent structural defects in Asarco's existing facilities. Two of the memoranda state that testing in the stack would be ideal but provide no scientific opinion explaining why other than to state that "particulate reactions would be expected to occur." There is no other evidence in the record at the time of the EPA's decision that the EPA conducted tests or made any other scientific investigation into the possibility of particulate formation.

Although the EPA could reasonably order monitoring equipment without first performing physical tests to prove certain the existence of particulates, it should be required to undertake at least some investigation, even if only theoretical, into the likelihood of particulate formation. For example, calculations could be made on the basis of present emissions data. At best, the EPA's investigation consisted of abstract or conclusory statements that such reactions occur. In light of calculations submitted by Asarco that suggested the particulate formations would not be significant, the EPA's superficial inquiry into the possibility of particulate formation was deficient.

Nor can the EPA's order be sustained on the basis of its conclusion that there were structural defects in the existing sites. There is only one document appearing in the record that indicates that the EPA studied the possibility that there were inherent structural defects in Asarco's existing sampling facilities. This is the report prepared for the EPA by James Steiner, a consultant with the Acurex Corporation, who in his words spent "the better part of the morning" observing Asarco's Hayden plant operations. According to Steiner, it would be possible to make emissions measurements at the roaster-reverb outlet, but the accuracy of the measurements would be highly questionable and the cost of modifications would be high. The other outlet could meet EPA requirements after additional modifications at moderate cost. In light of the expenditures needed to remedy existing sampling sites, Steiner concluded that construction of sampling facilities on the stack would be justified because of the improvement in measurement accuracy. Steiner did not discuss the possibility of chemical reactions in the stack.

While the Steiner report supports the EPA's position, we think that in the circumstances of this case it is insufficient to sustain the EPA's order. First, Asarco made numerous offers to modify their present facilities. If modification costs were reasonable, this alternative should have been further explored. Yet the Steiner report only broadly concludes that modifications would be impractical. More disturbing, however, is that Asarco was never given an opportunity to respond to the findings reached by Steiner. The report furnished the principal basis for the agency's action and it was a prejudicial violation of agency law principles not to allow Asarco an opportunity to challenge its accuracy. See e. g., Bowman Transportation v. Arkansas-Best Freight System, Inc., 419 U.S. 281, 288 n. 4, 95 S. Ct. 438, 443, 42 L. Ed. 2d 447 (1974); Marathon Oil Co. v. EPA, 564 F.2d 1253, 1271 (9th Cir. 1977).

Because the EPA's order cannot be sustained on the present record, it must be remanded to the agency for further consideration. The EPA, upon reconsideration, should establish in the record the scientific and technical basis to support its conclusions that the existing sampling sites are inadequate and that particulates are formed in the stack. Asarco should be given an opportunity to respond to evidentiary material adduced in support of the EPA's position.7

VACATED and REMANDED for further proceedings consistent with this opinion.