United States v. NCR Corporation

Summarize Case

Related Cases

In the United States Court of Appeals For the Seventh Circuit ____________________ Nos. 13-2436 & 13-2441 UNITED STATES OF AMERICA, Plaintiff-Appellee, v. P. H. GLATFELTER COMPANY and NCR CORPORATION, Defendants-Appellants. ____________________ Appeal from the United States District Court for the Eastern District of Wisconsin. No. 1:10-cv-00910-WCG — William C. Griesbach, Chief Judge. ____________________ ARGUED FEBRUARY 28, 2014 — DECIDED SEPTEMBER 25, 2014 ____________________ Before WOOD, Chief Judge, and KANNE and TINDER, Circuit Judges. TINDER, Circuit Judge. Today we issue two decisions relat- ed to the cleanup of the Lower Fox River and Green Bay Su- perfund Site in northeastern Wisconsin. This decision ad- dresses a claim brought by the United States to enforce a 2007 unilateral administrative order issued by the Environ- mental Protection Agency (EPA) under § 106 of the Com- 2 Nos. 13-2436 & 13-2441 prehensive Environmental Response, Compensation, and Liability Act of 1980 (CERCLA), 42 U.S.C. § 9606. 1 EPA’s 2007 order directs several potentially responsible parties (PRPs) to clean up the portion of the Site downstream of Lit- tle Lake Butte des Morts. Some of the issues relevant to the government’s claim to enforce EPA’s order were resolved on summary judgment, then the district court held a bench trial. After trial, the court ruled in favor of the government and entered a declaratory judgment and permanent injunction requiring the PRPs to comply with EPA’s order. Four PRPs appealed—NCR Corporation, P.H. Glatfelter Company, Menasha Corporation, and WTM I Company— and their appeals were consolidated. However, shortly after oral argument, the government lodged in the district court a proposed consent decree that would resolve its claims against Menasha and WTM. Thus, we have deconsolidated the appeals of those two PRPs, and we will resolve only the appeals of NCR and Glatfelter in this opinion. For the rea- sons that follow, we affirm in part and reverse in part. I. BACKGROUND The Superfund Site at issue encompasses both the Lower Fox River and Green Bay. The history of the Site is more ful- ly described in our contemporaneously issued opinion in NCR Corp. v. George A. Whiting Paper Co., No. 13-2447 (7th Cir. Sept. 25, 2014). For now, suffice it to say that several pa- per mills discharged wastewater containing polychlorinated 1 In the same action, the United States and the State of Wisconsin have brought claims to recover response costs and natural resource damages un- der § 107 of the CERCLA, 42 U.S.C. § 9607. Those claims are pending in the district court and are not at issue here. Nos. 13-2436 & 13-2441 3 biphenyls (PCBs) into the River from the mid-1950s through the 1970s, and since 1998, the Site has been the subject of massive remedial efforts conducted pursuant to CERCLA. The Lower Fox River begins at the outlet of Lake Winne- bago and flows northeast for approximately 39 miles before it enters Green Bay. In 1998, EPA began working with the Wisconsin Department of Natural Resources (WDNR) to de- velop a remedial plan for the Site. As part of that plan, the Site was divided into five geographic sections, or “operable units,” which are used “when phased analysis and response is necessary or appropriate given the size or complexity of the site.” 40 C.F.R. § 300.430(a)(1)(ii)(A). Operable Unit 1 or “OU1” runs from the outlet of Lake Winnebago to the Ap- pleton Dam, a stretch of the River also known as Little Lake Butte des Morts; OU2 runs from the Appleton Dam to the Little Rapids Dam; OU3 runs from the Little Rapids Dam to the De Pere Dam; OU4 runs from the De Pere Dam to the mouth of the River at Green Bay; and OU5 is Green Bay it- self. Only OU2–OU5 are at issue here, as the cleanup of OU1 was litigated separately and carried out pursuant to a con- sent decree. The Site and its operable units are shown in the figure below, which is taken from the agencies’ 2007 record of decision amendment. 4 Nos. 13-2436 & 13-2441 NCR and Glatfelter are PRPs under CERCLA § 107(a), 42 U.S.C. § 9607(a), because they or their corporate predeces- sors formerly owned and operated paper mills that dis- charged wastewater containing PCBs into the River. NCR is responsible for two mills that produced carbonless copy pa- per using an emulsion containing PCBs and then discharged PCB-contaminated wastewater into OU2. Glatfelter is re- sponsible for a mill that recycled scraps of carbonless copy Nos. 13-2436 & 13-2441 5 paper unusable by the original manufacturer and then dis- charged PCB-contaminated wastewater into OU1. In 2002, EPA and WDNR issued a record of decision (ROD) that selected a remedy for OU1–OU2. That remedy called for the dredging of approximately 784,000 cubic yards of sediment in OU1 but was limited to monitored natural recovery in OU2, with the exception of some dredging in Deposit DD at the downstream end of OU2 that would be undertaken as part of the OU3 remedy. In 2003, EPA and WDNR issued a second ROD, which selected a remedy for OU3–OU5. That remedy called for the dredging of approxi- mately 9,000 cubic yards of sediment in Deposit DD at the downstream end of OU2, approximately 586,800 cubic yards of sediment in OU3, and approximately 5,880,000 cubic yards of sediment in OU4. For OU5, the remedy was limited to monitored natural recovery, with the exception of some dredging near the mouth of the River. In the years that followed, Glatfelter and other OU1 PRPs agreed to perform the necessary remedial design and action in OU1. Meanwhile, NCR and one other PRP agreed to per- form the remedial design work for OU2–OU5. In 2007, based on information obtained during the full-scale remediation activities in OU1 and the remedial design work for OU2– OU5, the agencies decided to amend the ROD for OU2–OU5. The agencies determined that the all-dredging remedy they previously selected would not sufficiently reduce PCB con- centrations in OU2–OU5 and that an approach that utilized capping or sand covering in some areas would cure that de- ficiency. The agencies also determined that capping and sand covering would be less expensive than dredging and that a remedy incorporating those methods would therefore 6 Nos. 13-2436 & 13-2441 be more cost effective. Ultimately, the agencies adopted a hybrid remedy, which maintained dredging as the default approach but allowed for capping and sand covering where certain design criteria were met. In November 2007, EPA issued a unilateral administra- tive order pursuant to CERCLA § 106(a), 42 U.S.C. § 9606(a), directing the PRPs to conduct the cleanup required by the ROD amendment for OU2–OU5. Thereafter, NCR led the remedial efforts in OU2 and OU3 and conducted a signifi- cant amount of remedial action in OU4. Even so, NCR main- tained that it should not be responsible for all of the cleanup costs. Thus, in 2008, it filed an action seeking contribution from the other PRPs. In response, the other PRPs filed coun- terclaims seeking contribution from NCR. Near the end of 2009, the district court ruled against NCR on its claim for contribution, and in 2011, it ruled in favor of the other PRPs on their counterclaims, holding that NCR was required to reimburse them for their response costs. Meanwhile, in 2010, the agencies determined that they had significantly underestimated the costs associated with the cleanup of OU2–OU5. Thus, they published an explana- tion of significant differences, which adjusted their estimat- ed total project costs for OU2–OU5 by about 62 percent, from about $432 million to about $701 million. 2 Shortly after the district court held that NCR was re- quired to reimburse the other PRPs for their response costs, 2 The 2007 ROD amendment estimated total project costs at about $390 million, in 2005 U. S. Dollars. In the 2010 explanation of significant differ- ences, the agencies adjusted that number to 2009 U. S. Dollars, resulting in an estimate of about $432 million, so as to provide a proper basis for com- parison with their revised cost estimates. Nos. 13-2436 & 13-2441 7 NCR decided that it would no longer comply with EPA’s 2007 order. NCR cut its remediation work in half during 2011, and it refused to commit to perform any work in 2012. This action ensued, in which the United States sought pre- liminary and permanent injunctive relief, along with a de- claratory judgment, requiring NCR and the other PRPs to comply with EPA’s order. In 2012, the district court entered a preliminary injunc- tion against NCR, requiring it to complete the remediation work scheduled for that year, and we affirmed. United States v. NCR Corp., 688 F.3d 833 (7th Cir. 2012). Thereafter, the dis- trict court resolved two issues on summary judgment. First, it upheld the remedy selected by EPA and WDNR. Second, it held that Glatfelter and the other OU1 PRPs were liable for downstream cleanup costs. Then, in December 2012, the dis- trict court held an eleven-day bench trial, and a few months later, it ruled in favor of the government and entered a de- claratory judgment and permanent injunction requiring the nonsettling PRPs to comply with EPA’s 2007 order. On appeal, NCR and Glatfelter ask us to vacate the de- claratory judgment and permanent injunction. They attack the district court’s summary judgment rulings, as well as its findings of fact and conclusions of law following the bench trial. For the reasons that follow, we hold that the district court erred in its consideration of NCR’s divisibility defense and in its decision to enter a permanent injunction but cor- rectly resolved the other issues before it. Therefore, we af- firm in part and reverse in part. 8 Nos. 13-2436 & 13-2441 II. SUMMARY JUDGMENT RULINGS The district court’s summary judgment rulings came af- ter the parties filed cross-motions for summary judgment on the propriety of the remedy, and the United States moved for summary judgment on the OU1 Defendants’ liability. “As with any summary judgment motion, we review cross- motions for summary judgment ‘construing all facts, and drawing all reasonable inferences from those facts, in favor of the non-moving party.’” Laskin v. Siegel, 728 F.3d 731, 734 (7th Cir. 2013) (quoting Wis. Cent., Ltd. v. Shannon, 539 F.3d 751, 756 (7th Cir. 2008)). Summary judgment is appropriate if “there is no genuine dispute as to any material fact and the movant is entitled to judgment as a matter of law.” Fed. R. Civ. P. 56(a). A. PROPRIETY OF THE REMEDY When the government brings an action under 42 U.S.C. § 9606 to enforce an administrative cleanup order, CERCLA allows for judicial review of EPA’s decision in selecting the response action. 42 U.S.C. § 9613(h)(2). However, the court must “uphold [EPA’s] decision in selecting the response ac- tion unless the objecting party can demonstrate, on the ad- ministrative record, that the decision was arbitrary and ca- pricious or otherwise not in accordance with law.” Id. § 9613(j)(2). EPA’s remedy selection for the Site is reflected in the 2007 ROD amendment and the 2010 explanation of sig- nificant differences. For different reasons, both Glatfelter and NCR challenged that remedy on summary judgment, and they reassert their arguments on appeal. Nos. 13-2436 & 13-2441 9 1. EPA and WDNR’s Cooperative Agreement First, Glatfelter argues that EPA’s decision in selecting the response action was not in accordance with law because EPA delegated the task of conducting a remedial investiga- tion and feasibility study to WDNR without a valid coopera- tive agreement. CERCLA allows for the delegation of re- sponsibility for remedial action as follows: A State or political subdivision thereof or Indi- an tribe may apply to [EPA] to carry out ac- tions authorized in this section. If [EPA] de- termines that the State or political subdivision or Indian tribe has the capability to carry out any or all of such actions in accordance with the criteria and priorities established pursuant to section 9605(a)(8) of this title and to carry out related enforcement actions, [EPA] may en- ter into a contract or cooperative agreement with the State or political subdivision or Indian tribe to carry out such actions. [EPA] shall make a determination regarding such an appli- cation within 90 days after [EPA] receives the application. 3 42 U.S.C. § 9604(d)(1)(A). Glatfelter argues that a state has no authority to “carry out the actions authorized” by CERCLA without a cooperative agreement and therefore the lack of a cooperative agreement between EPA and WDNR would un- dermine the selected remedy for the Site. 3 The President has delegated to EPA the authority to undertake response action under CLERCLA; thus, where the statute refers to the President, we substitute EPA. See Exec. Order No. 12,580, 52 Fed. Reg. 2923 (Jan. 23, 1987). 10 Nos. 13-2436 & 13-2441 Glatfelter’s argument is academic, however, because EPA and WDNR did in fact enter into a cooperative agreement regarding the remedial investigation and feasibility study for the Site, and the government filed authenticated copies of that agreement in the district court. Glatfelter complains that the cooperative agreement was not included in the adminis- trative record, but nothing in CERCLA or the national con- tingency plan requires that cooperative agreements be in- cluded in the administrative record. The administrative rec- ord need only include “the documents that form the basis for the selection of a response action.” 40 C.F.R. § 300.800(a). Typically, that includes the following: (1) Documents containing factual information, data and analysis of the factual information, and data that may form a basis for the se- lection of a response action. … (2) Guidance documents, technical literature, and site-specific policy memoranda that may form a basis for the selection of the re- sponse action. … (3) Documents received, published, or made available to the public under § 300.815 for remedial actions, or § 300.820 for removal actions. … (4) Decision documents. … (5) Enforcement orders. … ; and (6) An index of the documents included in the administrative record file. … Nos. 13-2436 & 13-2441 11 Id. § 300.810(a). Notably, this list contains no mention of cooperative agreements. Instead, it emphasizes documenta- tion of the substantive factors that play a role in the selection of a remedy. As a result, we find no support for Glatfelter’s argument that the failure to include the cooperative agree- ment in the administrative record undermines the selected remedy. It is true that as a general matter “judicial review of any issues concerning the adequacy of any response action taken or ordered by [EPA] shall be limited to the administrative record.” 42 U.S.C. § 9613(j)(1). But “[o]therwise applicable principles of administrative law shall govern whether any supplemental materials may be considered by the court.” Id. Those principles allow a reviewing court to consider materi- als outside the administrative record “when it is necessary to create a record without which the challenge to the agency’s action cannot be evaluated.” USA Grp. Loan Servs., Inc. v. Ri- ley, 82 F.3d 708, 715 (7th Cir. 1996). Here, it would be impossible to evaluate Glatfelter’s chal- lenge to the selected remedy without considering the docu- ments that the government contends make up the coopera- tive agreement between EPA and WDNR. Thus, the district court properly considered those documents. Moreover, those documents show that EPA and WDNR entered into a valid cooperative agreement, which granted WDNR authori- ty to conduct a remedial investigation and feasibility study for the Site. As a result, EPA’s reliance on WDNR to com- plete those tasks was in accordance with the national contin- gency plan and does not provide a basis for reversal. 12 Nos. 13-2436 & 13-2441 2. Substantive Review of the Remedy Next, Glatfelter argues that the district court failed to undertake any substantive review of the selected remedy, instead focusing on the process by which the selection was made. We agree that to determine whether an agency’s deci- sion was arbitrary or capricious, the reviewing court “must go beyond the agency’s procedures to include the substan- tive reasonableness of its decision.” James Madison Ltd. v. Ludwig, 82 F.3d 1085, 1098 (D.C. Cir. 1996). This is because “reasonable procedures alone cannot absolve a court from making a ‘thorough, probing, in-depth review’ to determine if the agency has considered the relevant factors or commit- ted a clear error of judgment.” Id. (quoting Citizens to Pre- serve Overton Park, Inc. v. Volpe, 401 U.S. 402, 416 (1971)). But “the reasonableness of the agency’s procedures is relevant to the court’s inquiry,” James Madison Ltd., 82 F.3d at 1098, and although the district court stated that “arbitrary and capri- cious are terms that describe the manner of remedy selection more than they do the result,” it did in fact engage in sub- stantive review of the selected remedy. Moreover, “[e]ven if the district court here based its decision on the strength of the process alone—which we do not believe is the case—our de novo review of the record satisfies us that the agency’s conclusions were not arbitrary.” Id. Under the arbitrary-and-capricious standard of review, an agency’s decision must be upheld unless it has relied on factors which Congress had not intended it to consider, entirely failed to con- sider an important aspect of the problem, of- fered an explanation for its decision that runs counter to the evidence before the agency, or is Nos. 13-2436 & 13-2441 13 so implausible that it could not be ascribed to a difference in view or the product of agency ex- pertise. Nat’l Ass’n of Home Builders v. Defenders of Wildlife, 551 U.S. 644, 658 (2007) (quoting Motor Vehicle Mfrs. Ass’n v. State Farm Mut. Auto. Ins. Co., 463 U.S. 29, 43 (1983)) (internal quo- tation marks omitted). The district court concluded that the agencies’ decision to maintain a preference for dredging in the amended remedy was rationally related to the facts before them. In particular, the court noted that dredging represents a more permanent solution because it actually removes PCBs from the Site, while capping and sand covering merely contain PCB- contaminated sediment. Moreover, capping and sand cover- ing require long-term monitoring to ensure their effective- ness, and they are susceptible to failure during catastrophic events like floods. Ultimately, the district court concluded that the agencies acted rationally by adopting “a mild pref- erence for the benefits of dredging and viewed these as be- ing worth their added expense.” We agree. Most of the attacks Glatfelter levies against the agencies’ remedy selection have to do with the 2003 all-dredging rem- edy, which of course has been abandoned. Glatfelter argues that the problems with that remedy also infect the amended remedy, because in 2007 the agencies failed to reconsider all of the alternatives they had considered in 2003, even though by 2007 they had obtained more accurate information on the cost and effectiveness of the various remedial approaches. More specifically, Glatfelter contends that the agencies 14 Nos. 13-2436 & 13-2441 should have reconsidered an all-capping remedy, which would have been shown to be more cost-effective. However, as the government points out, an all-capping remedy was never on the table. In the 2003 ROD, the agen- cies explained that capping is limited by “Site-specific condi- tions such as water depth, average river current, river cur- rent under flood conditions, wave energy, ice scour, and boat traffic.” Thus, the agencies never considered an all- capping alternative, as Glatfelter suggests. Instead, they con- sidered an alternative that would involve “capping to the maximum extent practicable with dredging in areas where capping is not appropriate.” Under the criteria developed for the 2003 ROD, this would have resulted in capping only about 2.25 million cubic yards of contaminated sediment and dredging the rest of the contaminated sediment. In contrast, the 2007 ROD amendment called for capping (or sand cover- ing) about 3.5 million cubic yards of contaminated sediment. In other words, the alternative that Glatfelter contends the agencies should have reconsidered in 2007 actually called for less capping and more dredging than the remedy the agencies adopted. There is no indication that Glatfelter wanted the agencies to reconsider a more expensive remedy, so its argument makes little sense. Moreover, it was not arbi- trary or capricious for the agencies to refuse to consider a pure capping remedy, because such a remedy was infeasible. As a result, Glatfelter has failed to show that the selected remedy should be overturned. 3. The 2010 Explanation of Significant Differences NCR argues that the selected remedy is not in accordance with law because the agencies failed to amend the ROD Nos. 13-2436 & 13-2441 15 when their cost estimates increased dramatically in 2010. The district court rejected this argument, holding that the agencies complied with the national contingency plan in publishing an explanation of significant differences rather than amending the ROD in 2010. Sometimes new information arises during the remedial design or action phases of a CERCLA cleanup, and the rem- edy selected in the ROD must be altered. Under such cir- cumstances, the national contingency plan gives the lead agency two options. 40 C.F.R. § 300.435(c)(2). If “the differ- ences in the remedial or enforcement action … significantly change but do not fundamentally alter the remedy selected in the ROD with respect to scope, performance, or cost,” then the lead agency may simply publish “an explanation of significant differences.” Id. § 300.435(c)(2)(i). However, “if the differences in the remedial or enforcement action … fun- damentally alter the basic features of the selected remedy with respect to scope, performance, or cost,” then the lead agency must propose an amendment to the ROD. Id. § 300.435(c)(2)(ii). Amending the ROD is a more cumber- some process that involves receiving and responding to pub- lic comments on the proposed changes. Id. The explanation of significant differences that EPA and WDNR published in 2010 left intact the remedial approach outlined in the 2007 ROD amendment but revised the cost estimates associated with that approach. Specifically, the to- tal cost of the remedial action in OU2–OU5, which was orig- inally estimated at about $432 million, was now expected to be about $701 million, an increase of roughly 62 percent. NCR argues that such a drastic increase in cost “fundamen- tally alter[s] the basic features of the selected remedy with 16 Nos. 13-2436 & 13-2441 respect to ... cost” and that as a result, EPA and WDNR were required to propose an amendment to the ROD. In the 2010 explanation of significant differences, EPA and WDNR explained their decision to forgo another ROD amendment as follows: As set forth in the EPA guidance document en- titled, “A Guide to Developing and Documenting Cost Estimates During the Feasibility Study,” EPA 540-R-00-002 OSWER 9355.0-75 (July 2000), the expected accuracy range of a cost estimate for a detailed analysis of remedial action alterna- tives is -30% to +50%. As the current estimated cost of the OU 2 - 5 remedial action is 62% greater than the original estimate, it is nearly within EPA’s expected accuracy range for the cost of a remedial action and represents a “sig- nificant” but not “fundamental” change from the 2007 ROD Amendment. As NCR points out, however, the guidance document cited by the agencies contemplates that a “detailed analysis of al- ternatives” will be completed without the benefit of remedi- al design work. This was the case with the 2003 ROD, which explicitly cited the -30% to +50% accuracy range for its cost estimates. But the 2007 ROD amendment was issued after substantial remedial design work had been completed. Thus, we might expect its cost estimates to be more accurate. In fact, EPA also thought the 2007 ROD amendment’s cost estimates would be more accurate. Its response to com- ments questioning the accuracy of those estimates included the following statement: Nos. 13-2436 & 13-2441 17 The cost estimates for the ROD Amendment should be more reliable than the cost estimates in the 2003 ROD because new estimates are based on substantially more engineering anal- ysis and a much larger number of sediment samples (i.e., 10,000 sediment samples versus 1,700 sediments samples prior to the 2003 ROD). Thus, EPA represented to the public that the 2007 ROD amendment would be more accurate than the 2003 ROD, which identified a cost uncertainty range of -30% to +50%. We do not know how accurate the 2007 cost estimates were expected to be, because neither the 2007 ROD amendment nor the basis of design report on which it relied identified a range of uncertainty for total project costs. However, we would expect the range to be smaller than -30% to +50% but larger than -10% to +15%, which, according to the EPA guid- ance document cited by the agencies, is the range expected to be achieved after the remedial design process is complete. As a result, we reject the government’s argument that a cost increase of 62 percent is only marginally outside the range of uncertainty associated with the cost estimates in the 2007 ROD amendment. But even so, another ROD amend- ment was not necessarily required. We must still determine whether “the differences in the remedial or enforcement ac- tion … fundamentally alter[ed] the basic features of the se- lected remedy with respect to scope, performance, or cost.” 40 C.F.R. § 300.435(c)(2)(ii). This standard is ambiguous as it relates to the change at issue, i.e., an increase in cost unac- companied by any change in the remedial approach. As a 18 Nos. 13-2436 & 13-2441 result, EPA’s interpretation of the regulation, which it promulgated, is relevant. Indeed, an agency’s interpretation of its own regulation is controlling unless it is “plainly erroneous or inconsistent with the regulation,” or there is “reason to suspect that the interpretation does not reflect the agency’s fair and consid- ered judgment on the matter in question.” Auer v. Robbins, 519 U.S. 452, 461–62 (1997) (citation and internal quotation marks omitted). “This might occur when the agency’s inter- pretation conflicts with a prior interpretation, or when it ap- pears that the interpretation is nothing more than a conven- ient litigating position or a post hoc rationalizatio[n] ad- vanced by an agency seeking to defend past agency action against attack.” Christopher v. SmithKline Beecham Corp., 132 S. Ct. 2156, 2166–67 (2012) (citations and internal quotation marks omitted). The preamble to the national contingency plan makes EPA’s position clear: “EPA believes that the appropriate threshold for amending a ROD is when a fundamentally dif- ferent approach to managing hazardous wastes at a site is proposed.” 55 Fed. Reg. 8666, 8771 (March 8, 1990). In con- trast, “[s]ignificant changes to a remedy are generally incre- mental changes to a component of a remedy that do not fun- damentally alter the overall remedial approach.” Id. at 8772. Moreover, EPA’s position is that a change in cost alone does not fundamentally alter the remedial approach and thus does not require a ROD amendment. See id. (“Where [a] new requirement would affect a basic feature of the remedy, such as timing or cost, but not fundamentally alter the remedy specified in the ROD (i.e., change the selected technology), Nos. 13-2436 & 13-2441 19 the lead agency would need to issue an explanation of signif- icant differences announcing the change.”). In other words, EPA takes the view that to “fundamen- tally alter the basic features of the selected remedy with re- spect to … cost” is to choose a different remedial approach that costs more, not to stick with the same remedial approach but decide that it will cost more than previously thought. Because of the national contingency plan’s emphasis on “fundamentally alter[ing] the basic features of the selected remedy,” we cannot say that this interpretation is “plainly erroneous or inconsistent with the regulation.” Auer, 519 U.S. at 461 (internal quotation marks omitted). Moreover, EPA adopted this interpretation in an attempt “to develop an administrative process which balances the public’s continuing need for information about, and input into, post-ROD remedial action decisions, with the lead agency’s need to move forward expeditiously with design and implementation of the remedy after fundamental deci- sions have been made in the ROD.” 55 Fed. Reg. at 8773. And EPA’s approach seems to us an eminently reasonable way to balance these competing interests. Thus, we have no “reason to suspect that the interpretation does not reflect the agency’s fair and considered judgment on the matter in question.” Auer, 519 U.S. at 462. Consequently, EPA’s inter- pretation of its regulation is binding, and the agencies were not required to amend the ROD in 2010. Having disposed of each of Appellants’ arguments on the propriety of the remedy, we will affirm the district court’s entry of summary judgment in favor of the government on that issue. 20 Nos. 13-2436 & 13-2441 B. GLATFELTER’S LIABILITY Next, Glatfelter challenges the district court’s entry of summary judgment on the issue of its liability for response costs in OU4, where remediation work is ongoing. Section 107(a) of CERCLA imposes strict liability for response costs upon four classes of responsible parties. 42 U.S.C. § 9607(a). With respect to Glatfelter, the relevant class is any former owner or operator of a facility at which hazardous substanc- es were disposed of and “from which there is a release, or a threatened release which causes the incurrence of response costs, of a hazardous substance.” Id. This is because Glat- felter is the corporate successor to Bergstrom Paper Compa- ny, which formerly operated a paper recycling mill in Neenah, Wisconsin, and discharged PCB-contaminated wastewater into Little Lake Butte des Morts. The question we must decide is whether the government established Glatfelter’s liability for response costs in OU4, even though the Bergstrom Mill was located at the upstream end of OU1. In this regard, Glatfelter argues that the gov- ernment should have been required to prove that PCB dis- charges from the Bergstrom Mill made their way into OU4 in sufficient quantities to “cause[] the incurrence of response costs” there. The district court disagreed, holding that there need be no “nexus” between a given de- fendant’s release and a specific response cost incurred—it is enough that (a) the defendant released a pollutant and (b) response costs were incurred to clean up “a” release. If the de- fendant truly released a minimal amount, that speaks not to its own liability (for which there Nos. 13-2436 & 13-2441 21 is no de minimis defense) but to whether that li- ability is divisible. On appeal, Glatfelter argues that the district court improper- ly relieved the government of its burden of proof on causa- tion by instead requiring Glatfelter to disprove causation in relation to its divisibility defense. Section 107(a) of CERCLA is ambiguous as to whether proof of a causal relationship between the incurrence of re- sponse costs and an actual (as opposed to threatened) release of a hazardous substance is required to establish liability. Again, the statute imposes liability upon those responsible for a facility “from which there is a release, or a threatened release which causes the incurrence of response costs, of a hazardous substance.” Id. § 9607(a). [T]he phrase “from which there is a release ...” omits a comma after “threatened release”— suggesting that the clause is restrictive—while simultaneously using the word “which”— suggesting that the clause is nonrestrictive. That is, the missing comma implies that a per- son can be held liable for an actual release even when that release does not “cause the incur- rence of response costs,” while the word “which” suggests just the opposite. Asarco LLC v. Cemex, Inc., No. No. EP–12–CV–155–PRM, 2014 WL 2112121, at *11 n.19 (W.D. Tex. Mar. 31, 2014). However, we need not resolve this ambiguity today, because even if the government were required to establish a causal relation- ship between the incurrence of response costs and the actual 22 Nos. 13-2436 & 13-2441 release of PCBs from the Bergstrom Mill, it has satisfied that burden. Where Glatfelter’s argument goes astray is in its assump- tion that the government must prove all of the elements of liability in relation to each operable unit of the Site. Such a requirement is nowhere to be found in the statute. Instead, once it is established that a party is responsible for a facility “from which there is a release, or a threatened release which causes the incurrence of response costs, of a hazardous sub- stance,” that party “shall be liable for … all costs of removal or remedial action incurred by the United States Govern- ment or a State or an Indian tribe not inconsistent with the national contingency plan.” 42 U.S.C. § 9607(a). In short, even assuming that a release for which the de- fendant is responsible must have caused the incurrence of response costs, nothing in the statute limits the defendant’s liability to the response costs its release caused. Instead, the defendant is liable for all response costs “not inconsistent with the national contingency plan.” Id. The national contin- gency plan authorizes removal or remedial action at the “site,” 40 C.F.R. §§ 300.415, 300.435, and it defines “on-site” to mean “the areal extent of contamination and all suitable areas in very close proximity to the contamination necessary for implementation of the response action,” id. § 300.5. Where, as here, releases from multiple facilities contaminate an interconnected environmental system like a river, the en- tire system falls within this definition. Thus, in this case, the Site was properly defined to include the entire Lower Fox River and Green Bay, and so long as PCBs released from the Bergstrom Mill caused the incurrence of some response costs Nos. 13-2436 & 13-2441 23 within the Site, Glatfelter may be held liable for all response costs within the Site. Glatfelter does not dispute that PCB discharges from the Bergstrom Mill caused the incurrence of response costs in Little Lake Butte des Morts, which is within the Site. Yet it insists that its liability should not extend to OU4, where dis- charges from the Bergstrom Mill may not have caused the incurrence of response costs. In essence, Glatfelter wants us to treat OU4 as a separate site for which the government must establish liability. But this is inconsistent with the na- tional contingency plan, which defines an operable unit as “a discrete action that comprises an incremental step toward comprehensively addressing site problems.” Id. To put it simply, operable units are not separate sites; thus, they do not determine the extent of a party’s liability. Of course, an operable unit “may address geographical portions of a site,” id., and in some cases, a divisibility de- fense may prevail based upon those same geographic por- tions. However, the burden to prove divisibility rests on the defendant. The government need not prove each party’s lia- bility in relation to each geographic unit of a site the first in- stance. It need only prove each party’s liability as to the site as a whole. In this case, the undisputed facts establish Glat- felter’s liability as to the Lower Fox River and Green Bay Site. Glatfelter protests that holding it liable for response costs in OU4 is like holding it liable for “the Sheboygan River or the Hudson River, two other, unrelated, sediment PCB sites.” But of course, OU4 is not unrelated to OU1, nor is it a separate site; rather, OU1 and OU4 are part of the same site. Glatfelter cannot be held liable for cleanup in the Sheboygan 24 Nos. 13-2436 & 13-2441 River or the Hudson River because EPA could not define the Site to include those rivers, which are not part of “the areal extent of contamination and all suitable areas in very close proximity to the contamination necessary for implementa- tion of the response action.” Id. However, as Glatfelter con- cedes, EPA properly defined the Site to include the entire Lower Fox River. Therefore, Glatfelter may be held liable for cleanup costs there, and the district court properly granted summary judgment to the government on the issue of Glat- felter’s liability. III. TRIAL RULINGS In December 2012, the district court held an eleven-day bench trial on the government’s claim to enforce EPA’s 2007 cleanup order. The trial focused on the defendants’ divisibil- ity defenses as to OU4, the only operable unit where active remediation work is ongoing. Ultimately, the district court rejected the defendants’ divisibility defenses and entered a declaratory judgment and permanent injunction requiring the defendants to comply with EPA’s order. On appeal, NCR and Glatfelter argue that the district court wrongly rejected their divisibility defenses and that, even apart from those defenses, injunctive relief was improper. “Because the issues before this Court were adjudicated pursuant to a full bench trial, we review the district court’s conclusions of law de novo and its findings of fact for clear error.” Carpet Serv. Int’l, Inc. v. Chi. Reg’l Council of Carpenters, 698 F.3d 394, 397 (7th Cir. 2012), cert. denied, 133 S. Ct. 1856 (2013). A. DIVISIBILITY NCR and Glatfelter each presented a divisibility defense at trial, but the theories were quite different. NCR conceded Nos. 13-2436 & 13-2441 25 that it contributed to the contamination in OU4 but argued that the harm was capable of apportionment. Glatfelter, on the other hand, maintained that it did not cause any of the contamination in OU4 and that it should therefore not be li- able for any of the cleanup costs in OU4. We address each argument in turn. 1. NCR In NCR’s prior appeal, we affirmed the district court’s entry of a preliminary injunction because, at that stage, NCR had failed to show that the harm was capable of apportion- ment. In doing so, we relied on the principle that courts should not apportion responsibility for a single harm among multiple sufficient causes. United States v. NCR Corp., 688 F.3d at 839. In applying this principle, we agreed with the Ninth Circuit that the harm in CERCLA cases is properly characterized as the “contamination traceable to each de- fendant.” Id. at 841 (quoting United States v. Burlington N. & Santa Fe Ry. Co., 520 F.3d 918, 939 (9th Cir. 2008)) (internal quotation marks omitted). We further noted that while “cleanup costs, on their own, are not exactly equal to harm,” United States v. NCR Corp., 688 F.3d at 840, they “may some- times be a relevant factor for courts to use to determine the level of contamination, and thus the level of harm, caused by each polluter,” id. at 841. The sparse record before us in the prior appeal indicated that the contamination (and consequently the harm) in the Lower Fox River was binary in nature: PCB concentrations above 1.0 ppm were harmful, but PCB concentrations below 1.0 ppm were not. See id. at 839 (“EPA has set a maximum safety threshold of 1.0 ppm of PCB. Anything above that amount is dangerous to human life and requires remedia- 26 Nos. 13-2436 & 13-2441 tion.”); id. at 841 (“[C]ontamination occurs whenever PCBs pass a threshold level (thereby triggering remedial require- ments).”). Because NCR failed to refute the proposition that its discharges were sufficient by themselves to cause the con- tamination in OU4 (i.e, to cause PCB concentrations in OU4 to surpass the 1.0 ppm threshold), we agreed with the dis- trict court that the harm was not capable of apportionment. Id. at 839. After the case was tried, the district court continued to treat the harm as binary, but the evidence presented at trial shows that this was an oversimplification. The 1.0 ppm re- medial action level is not quite the line of demarcation it previously appeared to be. Indeed, it is not even EPA’s re- medial goal. Instead, EPA seeks to achieve a surface- weighted average concentration (SWAC) of 0.25 ppm throughout OU4,4 and it has determined that this can be achieved by undertaking some form of remediation wherev- er PCB concentrations exceed 1.0 ppm. Thus, sediment with a PCB concentration of 0.99 ppm will be left alone, not be- cause it is uncontaminated, but because it is insufficiently contaminated to push the SWAC for the operable unit as a whole above 0.25 ppm, at least after other areas with higher concentrations have undergone remediation. EPA has identified SWAC as the relevant metric for re- medial purposes because SWAC drives the ultimate harm with which EPA is concerned, i.e., the harm to human health and the environment. A particular SWAC leads to a particu- lar concentration of PCBs in fish tissue, which presents a 4 Surface-weighted average concentration measures the average concentra- tion of PCBs in the top 10 centimeters of sediment over the surface area of an entire operable unit. Nos. 13-2436 & 13-2441 27 particular risk that humans or piscivorous animals will con- tract a disease or experience other adverse health effects. And because SWAC drives the ultimate harm, it is also the appropriate measure of contamination for our purpose, i.e., determining whether the harm is divisible. But even EPA’s target SWAC of 0.25 ppm does not amount to a boundary between harmfulness and geniality. EPA’s “sediment quality thresholds,” which identify SWAC levels that pose acceptable risks, are typically lower than 0.25 ppm. Consequently, those thresholds will not be achieved through active remediation efforts. Instead, they will be achieved through a process of natural recovery that will take many years after the active remediation work is complete. For example, EPA estimates that 20 years of natu- ral recovery will be required to achieve acceptable noncan- cerous health risks to recreational anglers from walleye con- sumption. The corresponding time required to achieve an acceptable cancer risk is estimated at 45 years. And even then, some risk of harm will remain. EPA estimates that more than 100 years of natural recovery will be required to achieve PCB concentrations at which no adverse health ef- fects have been observed. As a result, the harm resulting from PCB contamination in the Lower Fox River cannot be characterized as binary. PCB concentrations below the 1.0 ppm remedial action level and even the 0.25 ppm target SWAC still pose a threat to human health and the environment. Even in areas where no remediation is required, higher PCB concentrations contrib- ute more to the risk of harm and require a longer period of natural recovery to achieve an acceptable risk. Moreover, not all concentrations above EPA’s remedial thresholds are 28 Nos. 13-2436 & 13-2441 equally harmful; the risk of harm increases with concentra- tion even at high levels. The continuous (as opposed to binary) nature of PCB contamination in the Lower Fox River leads us to reexamine EPA’s remediation rules to determine whether remediation costs are still a useful approximation of the contamination caused by each party. As with the contamination, the district court thought that remediation costs resembled an on/off switch: sediment with PCB concentrations below 1.0 ppm would impose no remediation costs, while sediment with PCB concentrations above 1.0 ppm would always impose about the same remediation costs. We think the district court got this wrong as well. In fact, remediation costs increase with the degree of contamination above 1.0 ppm. As a result, remediation costs are still a useful approximation of the de- gree of contamination caused by each party. The default remedial approach is dredging, but if certain criteria are met, an alternative approach such as capping (with various cap thicknesses) or sand covering may be used. Dredging is the most expensive approach, sand cover- ing is the cheapest, and capping lies somewhere in between (with thicker caps being more costly). Location-specific de- sign requirements do come into play; for example, a cap cannot be used if the required depth of the navigation chan- nel cannot be maintained. But all else being equal, higher PCB concentrations nearer the surface are more likely to re- quire dredging, while lower concentrations at greater depths are more likely to be eligible for capping or even sand cover- ing. Thus, the cost of the remedial approach in a particular area is positively correlated with the level of contamination Nos. 13-2436 & 13-2441 29 near the surface of that area, which contributes to the opera- ble unit’s SWAC, and consequently, the harm. In sum, neither the PCB contamination in OU4 nor the associated remediation costs are binary in nature. However, remediation costs are still a useful approximation of the de- gree of contamination, because both remediation costs and the relevant measure of contamination (SWAC) are positive- ly correlated with the concentration of PCBs near the sur- face, even for concentrations that exceed the remedial action level of 1.0 ppm. As a result, we think the harm would be theoretically capable of apportionment if NCR could show the extent to which it contributed to PCB concentrations in OU4. And if NCR cleared that hurdle, we think a reasonable basis for apportionment could be found in the remediation costs necessitated by each party. Two of NCR’s experts, Dr. John Connolly and Philip Si- mon, estimated the percentages of PCB mass attributable to each party in OU4. A third expert, John Butler, took these mass percentages and multiplied them by the actual concen- trations of PCBs in OU4, which he obtained from a database of core samples, to determine the concentrations of PCBs at- tributable to each party. Butler then used these concentra- tions to calculate the remediation costs caused by each party using EPA’s remediation rules. The district court thoroughly critiqued the mass- percentage estimates provided by Simon and Dr. Connolly, and we agree that those estimates likely understated NCR’s contribution to the PCBs in OU4. However, Butler also ran his analysis using the higher estimates provided by Georgia- Pacific’s expert, Dr. John Wolfe. The district court failed to explain why Dr. Wolfe’s mass-percentage estimates were 30 Nos. 13-2436 & 13-2441 unreliable. Moreover, apart from its assumption that the PCB contamination in the Lower Fox River is binary in na- ture, the district court levied no criticism at Butler’s applica- tion of the mass-percentage estimates he used. There may be reasons to find that Dr. Wolfe’s mass-percentage estimates are unreliable, and there may be reasons to find that Butler’s use of those estimates was unsound, but we will not under- take such factfinding in the first instance. Therefore, we will reverse the district court’s decision on NCR’s divisibility de- fense and remand for further proceedings. 2. Glatfelter Although PCB discharges from the Bergstrom Mill un- disputedly made their way into OU4, Glatfelter has made no attempt to quantify what percentage of the contamination in OU4 it may have caused. Instead, even in relation to its di- visibility defense, Glatfelter insists that it caused none of the contamination in OU4. Because the burden to prove divisi- bility rests on the defendant, Glatfelter is playing an all-or- nothing game, and we agree with the district court that it has lost. Glatfelter’s theory of divisibility relied almost entirely upon the expert opinion of Dr. Victor Magar. As part of his analysis, Dr. Magar estimated that only 14,000 kg were dis- charged into the river from the Bergstrom Mill. In contrast, government estimates ranged from 128,000 to 188,000 kg. To reach his much lower estimate, Dr. Magar first measured the mass of the sludge in the mill’s adjacent landfill, as well as the PCB concentrations in that sludge. He then used compa- ny records to establish the efficiency with which solids were removed from Bergstrom’s wastewater and deposited in the landfill, which allowed him to estimate the mass of solids Nos. 13-2436 & 13-2441 31 that remained in Bergstrom’s wastewater and were dis- charged into the river. Then, assuming that the discharged solids had the same concentration of PCBs as the solids de- posited in the landfill, he calculated the mass of PCBs dis- charged into the river. Relying on other experts, the district court soundly criti- cized Dr. Magar’s PCB mass estimate for the Bergstrom Mill. As the court noted, the clarifier used to remove solids from Bergstrom’s wastewater would have allowed larger solids to settle out and be removed while leaving smaller particles suspended in the wastewater. Because PCBs adsorb (attach) to the surface of solids, and because smaller particles have higher surface-to-mass ratios, PCB concentrations in masses of smaller particles tend to be higher. Thus, Dr. Magar’s as- sumption that the PCB concentrations in Bergstrom’s landfill were equal to the concentrations in the solids it discharged into the River was unsound. The district court’s finding on this point was not clearly erroneous. Therefore, we accept the fact that Dr. Magar greatly understated the mass of PCBs discharged into river by Bergstrom, for which Glatfelter is now responsible. Next, Dr. Magar opined that PCBs from the Bergstrom Mill would not have reached OU4 in concentrations above the 1.0 ppm threshold because (1) PCBs that deposited in the downstream part of OU1 generally did so in concentrations below 1.0 ppm, (2) the downstream part of OU1 was similar to OU4 in terms of its depositional properties, and (3) PCB concentrations decline as contaminated sediment moves downstream and mixes with clean sediment. Thus, Dr. Magar thought that PCBs from the Bergstrom Mill 32 Nos. 13-2436 & 13-2441 would have been diluted to concentrations far below 1.0 ppm by the time they reached OU4. However, the district court soundly criticized this aspect of Dr. Magar’s opinion as well. Relying on other experts, the court found that lower OU1 was not nearly as depositional as Dr. Magar claimed. Thus, in the district court’s words, “high-concentration PCB deposits were absent in lower OU1 not because PCB concentrations had already petered out, but because the PCBs simply were not permanently depositing in that part of the river.” This finding was not clearly erro- neous. Thus, we accept the fact that Dr. Magar greatly un- derestimated the concentrations at which PCBs from the Bergstrom Mill would have entered OU4. Consequently, Glatfelter failed to prove that it was not a sufficient cause of at least some of the contamination in OU4, and this alone is enough for us to affirm the district court’s decision on its divisibility defense. But the district court did not stop there. The court also found that even if Glatfelter had proved that it was not a sufficient cause of contamination in OU4, it failed to prove that it was not a necessary cause of such contamination. Glatfelter argues that it could not possi- bly have been a necessary but insufficient cause of contami- nation in OU4 and that the district court only reached this conclusion because it failed to grasp how PCB concentra- tions are calculated. We disagree. Dr. Magar opined that if PCBs from the Bergstrom Mill made their way into OU4 in concentrations less than the re- medial action level of 1.0 ppm, they could not have been a necessary cause of the cleanup in OU4 because, as he put it, “This is not an additive process. This would be an averaging process.” Nos. 13-2436 & 13-2441 33 For example, if 30 mg PCBs in 1 kg of suspend- ed solids (30 mg/kg PCBs) were mixed with 1 mg PCBs in 1 kg of suspended solids (1 mg/kg PCBs). The resulting mixture would contain 31 mg PCBs and 2 kg of suspended soils. The mix- ture concentration, which is equal to the mass of PCBs divided by the mass of suspended soils, would be equal to 31 mg PCB/ 2 kg sus- pended soils or 15.5 mg/kg. Thus, the concen- tration is the weighted average of the two sources. Expert Report of Victor Magar, PhD, PE, at 18-19, Sep. 28, 2012. In Dr. Magar’s opinion, this averaging process means that PCBs entering OU4 at a concentration below 1.0 ppm and mixing with other PCB-contaminated solids could never push the combined PCB concentration above 1.0 ppm. We have no qualms with the general proposition that when two masses of PCB-contaminated solids mix, their combined concentration is the mass-weighted average of their separate concentrations. It does not follow, however, that removing the PCBs that came in at a lower concentra- tion increases the average concentration. This is because the solids to which that lower concentration of PCBs would have attached still join the mix, only now they are PCB-free, resulting in a lower average concentration. Dr. Magar’s mis- take was ignoring the clean solids that would remain after removing the lower concentration of PCBs. Perhaps this is best illustrated through an example. In his expert report, Dr. Magar expressed the averaging concept 34 Nos. 13-2436 & 13-2441 through the following formula, where C is PCB concentra- (𝐶1 × 𝑇𝑆1 ) + (𝐶2 × 𝑇𝑆2 ) tion in mg/kg and TS is the total solid mass in kg: 𝐷𝑜𝑤𝑛𝑠𝑡𝑟𝑒𝑎𝑚 𝐶𝑜𝑛𝑐𝑒𝑛𝑡𝑟𝑎𝑡𝑖𝑜𝑛 = 𝑇𝑆1 + 𝑇𝑆2 Let us suppose that Party 1 discharges enough PCBs to cre- ate a concentration of 1.5 ppm (mg/kg) in 1.0 kg of sediment, and Party 2 discharges enough PCBs to create a concentra- tion of 0.5 ppm (mg/kg) in 1.0 kg of sediment. When these two masses combine, Dr. Magar’s formula yields the follow- (𝐶1 × 𝑇𝑆1 ) + (𝐶2 × 𝑇𝑆2 ) (1.5 × 1.0) + (0.5 × 1.0) ing concentration: = = 1.0 𝑝𝑝𝑚 𝑇𝑆1 + 𝑇𝑆2 1.0 + 1.0 Thus, there are just enough PCBs to require remedial action. Now, let us assume that Party 2 discharged no PCBs, but the sediment to which Party 2’s PCBs would have attached still combines with the sediment to which Party 1’s PCBs have attached. Dr. Magar’s formula then yields the following con- (𝐶1 × 𝑇𝑆1 ) + (𝐶2 × 𝑇𝑆2 ) (1.5 × 1.0) + (0.0 × 1.0) centration: = = 0.75 𝑝𝑝𝑚 𝑇𝑆1 + 𝑇𝑆2 1.0 + 1.0 Thus, remediation is required if Party 2’s PCBs are included, but remediation is not required if Party 2’s PCBs are exclud- ed, even though Party 2’s PCBs entered the mix at a concen- tration below the 1.0 ppm remedial action level. This goes to show that Dr. Magar’s conclusion that it is impossible to have necessary but insufficient causes does not follow from the principle on which he relied, i.e., that concentrations av- erage rather than add. In sum, Glatfelter failed to prove that the PCB discharges for which it is responsible were not a sufficient, or at least a Nos. 13-2436 & 13-2441 35 necessary, cause of at least some of the contamination in OU4. Therefore, the district court correctly ruled against Glatfelter on its all-or-nothing divisibility defense. B. INJUNCTIVE RELIEF Finally, NCR and Glatfelter attack the district court’s de- cision to enter a permanent injunction. We are persuaded by Glatfelter’s argument that permanent injunctive relief is an inappropriate mechanism to enforce an administrative order under § 106(b) of CERCLA, 42 U.S.C. § 9606(b). Therefore, we vacate the permanent injunction. In United States v. Ottati & Goss, Inc., 900 F.2d 429, 433 (1st Cir. 1990), then-Judge Breyer explained that there are four statutory paths EPA might pursue to achieve a CERCLA cleanup. Two of those paths warrant discussion here, and both arise under § 106 of CERCLA, 42 U.S.C. § 9606. The first is found in the first sentence of § 106(a), which provides as follows: [W]hen [EPA] determines that there may be an imminent and substantial endangerment to the public health or welfare or the environment because of an actual or threatened release of a hazardous substance from a facility, [it] may require the Attorney General of the United States to secure such relief as may be necessary to abate such danger or threat, and the district court of the United States in the district in which the threat occurs shall have jurisdiction to grant such relief as the public interest and the equities of the case may require. 36 Nos. 13-2436 & 13-2441 42 U.S.C. § 9606(a). This provision allows EPA to seek an in- junction requiring PRPs to take remedial action in “an emer- gency situation, where the agency has not yet had time to compile a thorough record and to issue an” administrative order. Ottati & Goss, 900 F.2d at 433. And it explicitly makes relevant “the public interest and the equities,” which indi- cates that the traditional elements of injunctive relief must be established. See Old Republic Ins. Co. v. Emp’rs Reinsurance Corp., 144 F.3d 1077, 1081 (7th Cir. 1998) (listing the elements that must be proved to obtain permanent injunctive relief, which include considerations of equity and the public inter- est). In less urgent situations, EPA may proceed under the second sentence of § 106(a), which allows it to follow the necessary administrative procedures and issue “such orders as may be necessary to protect public health and welfare and the environment.” 42 U.S.C. § 9606(a). Then, if the PRPs fail to comply with such an order, the government may bring an action under § 106(b)(1), which provides as follows: Any person who, without sufficient cause, willfully violates, or fails or refuses to comply with, any order of [EPA] under subsection (a) of this section may, in an action brought in the appropriate United States district court to en- force such order, be fined not more than $25,000 for each day in which such violation occurs or such failure to comply continues. Id. § 9606(b)(1). In such an action, the court may review the selected remedy. Id. § 9613(h)(2). However, its review is lim- ited to the administrative record, and EPA’s decision must Nos. 13-2436 & 13-2441 37 be upheld unless it “was arbitrary and capricious or other- wise not in accordance with law.” Id. § 9613(j). Thus, unlike an action under the first sentence of § 106(a), equitable con- siderations play no part in an action to enforce an adminis- trative order under § 106(b). Instead, “[w]hen the EPA asks a court ... to enforce a lawful (nonarbitrary) EPA order, the court must enforce it.” Ottati & Goss, 900 F.2d at 434. As a result, if courts were to undertake the traditional analysis for injunctive relief in deciding whether to enforce an EPA order, they would inject equitable considerations where they do not belong. Moreover, as Glatfelter points out, the entry of an injunction that simply orders PRPs to comply with a complex cleanup order issued by EPA may run afoul of Federal Rule of Civil Procedure 65(d)(1)(C), which requires that every injunction “describe in reasonable detail—and not by referring to the complaint or other docu- ment—the act or acts restrained or required.” Accordingly, permanent injunctive relief is incongruous with the nature and purpose of an action to enforce an administrative clean- up order under CERCLA § 106(b). 5 Further, permanent injunctive relief is unnecessary as a means of enforcing an administrative cleanup order, as the statute already provides for civil penalties of $25,000 per day that a PRP fails to comply with such an order “without suffi- cient cause.” Nothing we have said prevents the government from seeking declaratory relief to establish that a PRP lacks sufficient cause for noncompliance, such as the arbitrariness 5In contrast, preliminary injunctive relief may be appropriate in an action to enforce an EPA cleanup order, as it was in this case. Equitable principles are an appropriate consideration when deciding whether to require PRPs to undertake remedial action pending review of an EPA cleanup order. 38 Nos. 13-2436 & 13-2441 of the selected remedy or a defense to liability. And if the government obtains such declaratory relief, the PRP’s obli- gation to comply with the administrative order or face civil penalties will be established. In such circumstances, adding an injunction to the mix does little more than enjoin the de- fendant to obey the law, a practice we have criticized. See, e.g., EEOC v. AutoZone, Inc., 707 F.3d 824, 841–42 (7th Cir. 2013). For all of these reasons, we hold that the permanent injunction entered by the district court in this case was im- proper and must be vacated. IV. CONCLUSION The district court’s summary judgment rulings on the propriety of the remedy and Glatfelter’s liability are AFFIRMED. The permanent injunction entered by the district court is VACATED. The declaratory judgment entered by the district court is AFFIRMED as to Glatfelter, but VACATED as to NCR, and Case No. 13-2441 is REMANDED for the district court to reconsider NCR’s divisibility defense consistent with this opinion. Circuit Rule 36 shall not apply on remand.