Full opinion text
ORDER The full court has been advised of the petitions for rehearing en banc. A judge of the court requested a vote on en banc rehearing. The majority of the active judges have voted to deny rehearing the matter en banc. Fed. R.App. P. 35(f). The panel has voted to amend its opinion and to deny appellees’ petitions for rehearing with the following amendments. The opinion filed March 16, 2007 and amended September 4, 2007, published at 502 F.3d 781 (9th Cir.2007), is hereby further amended as follows: 1.On page 790, add a new footnote 5 after <When the trucks carrying D-D ... large tanks by hoses. > reading: Shell was deeply involved in the delivery process: The district court found that Shell determined and arranged for the means and methods of delivery of D-D to the Arvin plant and detailed loading and unloading procedures. It also found that the trucking companies with which Shell contracted for delivery did the transfers for most of the relevant period. It was only in the early 1980s that Shell dictated that B & B personnel should instead do the unloading. 2.Renumber footnote 7 on page 792 as footnote 8 and replace its text with <0.9 acres 4.7 acres = 0.191 (19.1%). >. 3. In the last paragraph on page 792, replace <In accord with this purpose, CERCLA is a “super-strict” liability statutes with <In accord with this purpose, CERCLA creates a system of strict liability. >. 4. In the first paragraph on page 793, replace with. 5. In the first full paragraph on page 794, replace < other indications of federal common law> with < other sources of federal common law>. 6. Replace <the “super-strict” nature of> with <the reach of> in the second-to-last sentence of footnote 16 on page 795. 7. Replace the paragraph beginning on page 795 and continuing to page 796 with: The question, then, is what the uniform federal law should be. Once again, all the circuits that have addressed this question have followed Chenu-Dyne, holding that the appropriate starting point for a common law rule of apportionment applicable to CERCLA cases is Section 433A of the Restatement of Torts. See Hercules, 247 F.3d at 716 & n. 9, 717 (noting that courts support the divisibility doctrine as borrowed from the Restatement); Bell Petroleum, 3 F.3d at 895 (relying on the Restatement); Chem Dyne, 572 F.Supp. at 810 (establishing this method). We agree that harm may be apportioned when “there exists a reasonable basis for divisibility” of a single harm or when several “distinct harms” are present. Hercules, 247 F.3d at 717. Because CERCLA’s statutory liability scheme differs from the common law in important respects, however, our sister circuits have recognized that its principles must be somewhat modified to fit the CERCLA context. See, e.g., Bell Petroleum, 3 F.3d at 902 (“Restatement principles must be adapted, where necessary, to implement congressional intent with respect to liability under the unique statutory scheme of CERCLA.”); Hercules, 247 F.3d at 717 (The Restatement is “the starting point .... [but] only to the extent that it is compatible with the provisions of CERCLA.”).We concur in this general conclusion and acknowledge, in particular, that there are two areas where the Restatement approach is a somewhat poor fit and requires slight modifications to ensure that its approach comports with the liability and remediation scheme of CERC-LA. First, as we describe infra, there are important distinctions between causation as conceived in the Restatement and causation in the context of CERC-LA. Unlike the Restatement’s common law causation, CERCLA affixes liability based upon its PRP provisions, which define classes of liable parties based upon a party’s statutorily-defined nexus to the contaminated site. And second, the concept of “harm” in the Restatement as actual injury does not correspond easily to CERCLA’s priorities. Contamination and the cost of remediation are both relevant for the “harm” analysis under CERCLA. Finally, we recognize that the Restatement’s emphasis on objective considerations to determine whether apportionment is justified in a given case comports with CERCLA’s strict liability scheme. Equitable considerations may play a role in a later contribution action amongst liable parties, but not in an action such as this where the only relevant issue is whether there is a reasonable basis, founded in record evidence, to apportion damages amongst defendants. 8.In the first full paragraph on page 796, replace <§ 433A(l)(b) and thereby incorporated a modified concept of causation. > with <§ 433A(l)(b).>. 9. Add <traditional> before ccausation concepts are useful > in the paragraph beginning on page 796 and continuing to page 797. 10. In the second full paragraph on page 797, replace <We therefore adjust .... time of the disposal > with: The statute thus departs from Restatement principles by abjuring the traditional “causation” principles in favor of a nexus concept defined by its PRP provisions. Where, as here, the pertinent PRP status is as landowner, the landowner can establish divisibility by demonstrating a reasonable basis for concluding that a certain proportion of the contamination did not originate on the portion of the facility that the landowner owned at the time of the disposal. 11. Add the following text after the last full paragraph on page 798: In so holding, we begin from the fundamental difference between apportionment and contribution. Apportionment, which is the relevant question in this case, looks to whether defendants may avoid joint and several liability by establishing a fixed amount of damage for which they are liable. Section 433A of the Restatement speaks to this issue. Contribution is a distinct concept. If there is insufficient evidence to support apportionment, jointly and severally liable PRPs may still seek to recover from each other in a later contribution action. See, e.g. Restatement (ThiRd) of ToRts § 23(a) (2000) (“When two or more persons are or may be liable for the same harm and one of them discharges the liability of another by settlement or discharge of judgment, the person discharging the liability is entitled to recover contribution from the other, unless the other previously had a valid settlement and release from the plaintiff.”). As we explain below, CERCLA permits equitable considerations to be taken into account in a contribution action amongst PRPs, but CERCLA’s strict liability scheme does not permit equitable considerations to have any bearing in an action to determine whether defendants have presented sufficient evidence to apportion liability. 12. Beginning on page 798 and continuing to page 799, replace <Although CERCLA is not explicit on this issue.... whittle their liability to zero:> with: Section 9613(f), added to CERCLA in 1986, provides for contribution actions among PRPs once liability to the plaintiffs has been determined. That provision describes, quite generally, the considerations applicable in a contribution action for determining whether one PRP can collect from another a portion of the costs for which it has been held liable: “In resolving contribution claims, the court may allocate response costs among liable parties using such equitable factors as the court determines are appropriate.” § 9613(f) (emphasis added). In contrast, CERCLA contains no provision explicitly providing for initial apportionment of liability. And § 433A(l)(b) of the Restatement and the appended commentary concerning divisibility are silent as to equitable considerations. As noted, this circuit heretofore has not addressed divisibility analysis. But the implication from our cases deciding § 9613(f) contribution issues is that the proper time to focus on such factors is at the contribution phase, not the liability-phase. See, e.g., Carson Harbor Vill., 270 F.3d at 871 (noting that the “contribution provision aims to avoid a variety of scenarios by which a comparatively innocent PRP might be on the hook for the entirety of a large cleanup bill”); Pinal Creek, 118 F.3d at 1301 (“A PRP’s contribution liability will correspond to that party’s equitable share of the total liability and will not be joint and several”). Other circuits have been careful to delineate the difference between the equitable considerations pertinent to contribution under § 9613(f) and the objective considerations pertinent to divisibility under § 9607(a). See Hercules, 247 F.3d at 718; Township of Brighton, 153 F.3d at 318; Bell Petroleum, 3 F.3d at 901. As the Sixth Circuit has noted, divisibility analysis, unless carefully limited, has the potential to eviscerate the strict liability principles of CERCLA entirely, “because defendants who can show that the harm is divisible, and that they are not responsible for any of the harm” could whittle their liability to zero. 13.In the paragraph beginning on page 799 and continuing to page 800, replace <While it may seem unfair ... liability scheme. > with <While it may seem unfair to hold a partial owner liable for all of the contamination cleanup costs, that perceived unfairness is the result of CERC-LA’s expansive statutory liability scheme. >. 14. In the second full paragraph on page 801, strike <, and the original ... from the Railroads >. 15. In the second full paragraph on page 801, replace <perfect information> with < adequate information >. 16. In the paragraph beginning on page 801 and continuing to page 802, re- place <are in no respect traceable to> with <did not originate on>. 17. Renumber footnote 27 on page 802 as footnote 29 and replace <We do not go as far _ present in this case.> with <We do not agree with Judge Moore in this regard. >. 18. Add a new footnote 30 on page 803 following <the single Arvin facility> reading: In its discussion of the Railroads’ apportioned liability, the district court found that B & B used the Railroad parcel as part of its total agricultural-chemical operations, and it earlier cited approvingly to documents describing the land as “an integral part of the adjacent farm chemical distribution facility.” The dissent from denial of rehearing en banc therefore wrongly characterizes our description of the site as “appellate fact-finding.” See Dissent at 2914-16 [520 F.3d 958-960]. 19.In the first full paragraph on page 804, replace < excluded in its calculations > with < excluded from its calculations >. 20. Renumber footnote 32 on page 811 as footnote 35, strike following cinclude the $1.3 million>, and add <Because we therefore do not hold Shell hable for this Dow product, the dissent from denial of rehearing en banc is wrong when it claims we err by holding Shell liable for “contamination from products Shell did not sell to B & B.” Dissent at 2921-22 n.22 [520 F.3d 962-963], > after < calculation of Shell’s liability. > 21. Renumber all footnotes in sequential order. No further petitions for rehearing or rehearing en banc will be entertained. The mandate shall issue in due course. OPINION BERZON, Circuit Judge: A now-defunct company, Brown & Bryant, Inc. (B & B), owned and operated a facility at which toxic chemicals were stored and distributed. Part of the land on which the chemical operation was located was owned by two railroad companies (the Railroads), and some of the chemicals used by B & B were supplied and delivered to the facility by Shell Oil Company (Shell). Because toxic chemicals remaining at the facility threatened groundwater and may continue to do so in the future, the United States Environmental Protection Agency (EPA) and the State of California’s Department of Toxic Substances Control (DTSC) spent a considerable amount of money to clean up the site and may need to spend more in the future. The two agencies sought to recover these response costs under the Comprehensive Environmental Response, Compensation, and Liability Act, 42 U.S.C. §§ 9601 9675, (CERCLA), but the district court held the Railroads and Shell liable for only a minor portion of the total cleanup costs. B & B was defunct by that time, and so could not contribute to the cleanup costs. The agencies were thus left holding the bag for a great deal of money. Seeking to hold the Railroads and Shell jointly and severally liable for the entire judgment, the agencies appeal. Shell cross-appeals, claiming that it was not an “arranger” under CERCLA, § 9607(a)(3), and therefore is not a party on whom any cleanup liability can be imposed. We reverse the portion of the judgment that declined to impose full joint and several liability on the Railroads and Shell and affirm the portion of the judgment that imposed liability on Shell as an arranger. I. Background Beginning in 1960, B & B operated an agricultural chemical storage and distribution facility in Arvin, California on a 3.8-acre parcel of land (the B & B parcel). In 1975, B & B’s agricultural chemical distribution business outgrew that parcel, and B & B began leasing a 0.9-acre parcel of land adjacent to its own parcel. The 0.9-acre parcel (the Railroad parcel) was jointly owned by the Railroads — Atchison, Topeka & Santa Fe Railroad Co., the predecessor in interest to Burlington Northern & Santa Fe Railway Co., and Southern Pacific Transportation Co., the predecessor in interest to Union Pacific Transportation Co. B & B used the Railroad parcel principally to park fertilizer rigs. The Railroad parcel comprised the western portion of the Arvin site. Directly to the east of the Railroad parcel sat B & B’s warehouse. The Railroad parcel, like the rest of the Arvin site, was graded toward a drainage pond on the B & B parcel. B & B used the Railroad parcel as an integral part of its overall agricultural chemical facility. From its facility B & B sold local growers agricultural chemical products produced by various manufacturers. In particular, B & B purchased, received delivery of, stored on the Arvin site, and distributed two Shell-produced agricultural chemicals: the soil fumigants D-D and Nemagon. D-D and Nemagon— members of a class of chemicals called nematocides — are designed to kill nematodes, microscopic worms that attack the roots of crops. Nematocides work by penetrating the soil and then dispersing. B & B also stored on the Arvin site dinitro (dinoseb) weed killer, purchased from Dow Chemical Company. During the 1960s and 1970s, Shell strongly encouraged its customers, including B & B, to purchase D-D in bulk, a policy requiring customers to maintain large storage tanks. Shell delivered the bulk D-D to B & B “FOB Destination” via common carrier trucks. When the trucks carrying D-D arrived at the Arvin facility, the contents of the trucks were transferred to B & B’s large tanks by hoses. The process was quite messy, with frequent spills. To apply D-D to growers’ fields, B & B used rigs loaded with the chemical. The rigs were stored on the Railroad parcel, as were bulk containers of dinoseb and, occasionally, empty fertilizer cans. Chemicals also reached the Railroad parcel through water flow from the B & B parcel. In 1978, after a windstorm destroyed the bulk D-D storage tank used to store Shell D-D, B & B began using converted stainless steel milk trailers to store the bulk DD. The chemical, which is highly corrosive and eats through steel, can cause leakage in steel tanks only a few years old. B & B kept these leak-prone tanks all over the Arvin facility, including on the Railroad parcel. D-D, when it leaks, evaporates quickly if exposed to air but is highly soluble in water. When D-D infiltrates the ground, it moves through the soil by molecular diffusion, dispersing in all directions. A slight pull from gravity, however, makes the chemical a bit more likely to flow downward into groundwater than laterally through the soil. Dinoseb, similarly, tends to move to the groundwater table if there is water movement in that direction. No toxic chemicals can reach the groundwater level currently used as a source of drinking water because of an impermeable layer of soil. The next highest level, however, is a potential source of drinking water, and contamination can reach that level. After more than twenty years of leakage and dissemination of hazardous materials, the DTSC in 1983 found B & B in violation of several hazardous waste laws. The EPA investigated separately and found evidence of substantial soil and groundwater contamination at B & B’s Arvin facility. The EPA and DTSC (the Governments) began to remedy the contamination pursuant to their cleanup authority under CERCLA, incurring substantial remediation costs. In 1991, the EPA ordered the Railroads to take specific preventative steps on the Railroad parcel, including installing groundwater monitoring wells. None of the contamination requiring immediate remediation was on the Railroad parcel. In 1992, the Railroads filed an action against B & B and certain of its principals for contribution for costs incurred in the EPA-ordered cleanup. Four years later, the Governments each filed CERCLA actions against B & B, the Railroads, and Shell for reimbursement of their investigation and cleanup costs. The district court consolidated the three cases and, after a twenty-seven day bench trial, issued an exceedingly detailed 185-page Findings of Fact and Conclusions of Law, thereafter slightly amended. The district court found the Railroads liable as owners of the Arvin facility and as persons who “at the time of disposal of any hazardous substance owned or operated any facility at which such hazardous substances were disposed of.” § 9607(a)(1), (2). Shell was held liable as a “person who ... arranged for disposal ... of hazardous substances.” § 9607(a)(3). Turning to whether the Railroads and Shell were liable for all or only a portion of the cleanup costs, the district court found that the harm to the Arvin site was capable of apportionment and proceeded to apportion it. The Railroads and Shell had, by acknowledging no liability at all, taken what the district court termed a “ ‘scorched earth,’ all-or-nothing approach to liability,” and so provided little assistance on the apportionment issue. The district court nonetheless proceeded to “perform the equitable apportionment analysis demanded by the circumstances of the case.” For the Railroads, the court multiplied three proportions: (1) the percentage of the overall site that was owned by the Railroads, 19.1%; (2) the percentage of time that the Railroads leased the parcel in relation to B & B’s total operations, 45%; and (3) the fraction of hazardous products attributable to the Railroad parcel, 66%. This calculation resulted in a determination of 6% liability. Then, to account for any “calculation errors,” the district court assumed 50% error and raised the Railroads’ proportion of the total liability to 9%. For Shell, the district court approximated the percentages of leakage from various activities attributable to Shell and multiplied them together to set Shell’s proportion of the total liability at 6%. Shell was also assigned, in the contribution action, 6% of the costs incurred by the Railroads in their cleanup effort. DTSC and the EPA timely appealed the district court’s judgment. Shell timely cross-appealed the finding that it was liable as an “arranger” under CERCLA. II. Standards of Liability Under CERCLA CERCLA was enacted in 1980 to provide for effective responses to health and environmental threats posed by hazardous waste sites. See generally Mardan Corp. v. C.G.C. Music, Ltd., 804 F.2d 1454, 1455 (9th Cir.1986). Under CERCLA, state and federal governments can first begin the cleanup of toxic areas, see § 9604(a)-(d), and then sue potentially responsible parties (PRPs) for reimbursement, see § 9607(a). A key purpose of this scheme is “shifting] the cost of cleaning up environmental harm from the taxpayers to the parties who benefited from the disposal of the wastes that caused the harm.” EPA v. Sequa Corp. (In the Matter of Bell Petroleum Servs., Inc.), 3 F.3d 889, 897 (5th Cir.1993) (citing United States v. Chem-Dyne Corp., 572 F.Supp. 802, 805-06 (S.D.Ohio 1983)). In accord with this purpose, CERCLA creates a system of strict liability. Under its provisions, parties can be liable for cleaning up toxic chemicals if they fit into one or more of the four PRP categories set out in § 9607(a): (1) the owner and operator of ... a facility, (2) any person who at the time of disposal of any hazardous substance owned or operated any facility at which such hazardous substances were disposed of, (3) any person who by contract, agreement, or otherwise arranged for disposal or treatment, or arranged with a transporter for transport for disposal or treatment, of hazardous substances owned or possessed by such person ..., and (4) any person who accepts or accepted any hazardous substances for transport to disposal or treatment facilities.... A “facility” is defined in § 9601(9)(B) as “any site or area where a hazardous substance has been deposited, stored, disposed of, or placed, or otherwise come to be located; but does not include any consumer product in consumer use or any vessel.” The statute’s basic liability provision, in turn, provides that “subject only to the defenses set forth in subsection (b) of this section [PRPs shall be liable for]— (A) all costs of removal or remedial action incurred by the United States Government or a State or an Indian tribe not inconsistent with the national contingency plan -” § 9607(a). Thus, PRPs , can be responsible for the costs of cleaning up hazardous waste sites without any finding that they were negligent or that they caused the contamination, unless they can establish the third-party defense set out in § 9607(b)(3). A. Validity of Apportionment CERCLA does not address the question whether, as between PRPs who are liable for cleanup costs, liability is joint and several — meaning that each PRP responsible for all cleanup costs at a facility is liable for such costs — or severable — meaning that cleanup costs at a single facility can be apportioned among PRPs on some basis. In this circuit, liability is joint and several when the harm is indivisible. Fireman’s Fund Ins. Co. v. City of Lodi, 302 F.3d 928, 945 (9th Cir.2002); see also Carson Harbor Vill., Ltd. v. Unocal Corp., 270 F.3d 863, 871 (9th Cir.2001) (en banc). Thus, a defendant “may be held fully liable for the entire clean-up costs at a site despite the fact that the defendant PRP was in fact responsible for only a fraction of the contamination.” Fireman’s Fund, 302 F.3d at 945. We have also referred in general terms to the possibility of apportioning liability. See id. (noting the use of “federal common law principles” of apportionment); Carson Harbor Vill, 270 F.3d at 871 (stating that once liability has been found, “the defendant may avoid joint and several liability by establishing that it caused only a divisible portion of the harm”). Yet, in none of our cases has there been an actual dispute regarding whether liability should be apportioned among the liable PRPs. This case squarely presents that question. To determine whether the district court was correct to apportion liability in this case, we thus must address, initially, the general propriety of severability. In line with every circuit that has addressed the issue, we hold that apportionment is available at the liability stage. In so ruling we rely, as have the other circuits that have analyzed the issue, on a seminal case decided in 1983 in the Southern District of Ohio, Chem-Dyne. After reviewing the evolution of the statute, Chem-Dyne concluded that liability under § 9607(a) may be joint and several even though the statute does not expressly so provide. Chem-Dyne, 572 F.Supp. at 810. Suggesting that Congress intended to leave the matter to the usual common law rules, adjusted to CERCLA as necessary, Chem-Dyne held that courts should look to the Restatement (Second) of Torts, as well as to other sources of federal common law, for the principles of joint and several liability applicable under CERCLA. See id. at 809-10. Later, circuit court cases endorsed this approach. See Chem-Nuclear Sys., Inc. v. Bush, 292 F.3d 254, 259-60 (D.C.Cir.2002); United States v. Hercules, Inc., 247 F.3d 706, 717 (8th Cir.2001); United States v. Township of Brighton, 153 F.3d 307, 318 (6th Cir.1998); Bell Petroleum, 3 F.3d at 895-96; United States v. Alcan Aluminum Corp. (Alcan-PAS), 990 F.2d 711, 721-22 (2d Cir.1993); United States v. Alcan Aluminum Corp. (Alcan-Butler), 964 F.2d 252, 268-69 (3d Cir.1992); United States v. Monsanto Co., 858 F.2d 160, 171-72 (4th Cir.1988). As Chem-Dyne persuasively recounts, the history of § 107(a) of CERCLA, 42 U.S.C. § 9607(a), indicates that although Congress declined to mandate joint and several liability, it did not intend by doing so “a rejection of joint and sever-able liability.” Chemr-Dyne, 572 F.Supp. at 808. Instead, recognizing the difficulties inherent “ ‘in prescribing in statutory terms liability standards which will be applicable in individual cases,’ ” id. at 806 (quoting 126 CONG. REC. S14964 (Nov. 24, 1980) (remarks of Sen. Randolph)), Congress meant “to have the scope of liability determined under common law principles, where a court performing a case by case evaluation of the complex factual scenarios associated ... will assess the propriety of applying joint and several liability on an individual basis,” id. at 808. We agree with this account of Congress’s intent and hold that apportionment can be appropriate under CERCLA. B. Standards for Apportionment Because we hold that apportionment is available at the liability stage in CERCLA cases, we must determine the appropriate standards for determining when apportionment is available and, when it is, how to ascertain the proper division of damages among defendants. Again, we draw on the experience of our sister circuits. The circuits that have addressed these questions have looked to common law principles of tort in general, and the Restatement in particular, for guidance as to when and how to impose joint and several liability under § 9607(a). We agree that this approach is proper and adopt it here. We also follow Chem-Dyne and all of the courts of appeals that have addressed the question in holding that the resulting standard must be a uniform federal rule. See, e.g., Aviall Servs., Inc. v. Cooper Indus., Inc., 312 F.3d 677, 684 (5th Cir.2002) (holding that apportionment of CERCLA liability “is ... a matter of federal common law”), reversed on other grounds by 543 U.S. 157, 125 S.Ct. 577, 160 L.Ed.2d 548 (2004); United States v. Burlington N.R. Co., 200 F.3d 679, 697 (10th Cir.1999) (same); Township of Brighton, 153 F.3d at 329 (same); Monsanto Co., 858 F.2d at 172 (same). As Chem-Dyne noted, the legislative history of CERCLA supports such an approach, as does its policy favoring national uniformity so as to discourage “illegal dumping in states with lax liability laws.” Chem-Dyne, 572 F.Supp. at 809. The question, then, is what the uniform federal law should be. Once again, all the circuits that have addressed this question have followed Chentr-Dyne, holding that the appropriate starting point for a common law rule of apportionment applicable to CERCLA cases is Section 433A of the Restatement of Torts. See Hercules, 247 F.3d at 716 & n. 9, 717 (noting that courts support the divisibility doctrine as borrowed from the Restatement); Bell Petroleum, 3 F.3d at 895 (relying on the Restatement); Chem-Dyne, 572 F.Supp. at 810 (establishing this method). We agree that harm may be apportioned when “there exists a reasonable basis for divisibility” of a single harm or when several “distinct harms” are present. Hercules, 247 F.3d at 717. Because CERCLA’s statutory liability scheme differs from the common law in important respects, however, our sister circuits have recognized that its principles must be somewhat modified to fit the CERCLA context. See, e.g., Bell Petroleum, 3 F.3d at 902 (“Restatement principles must be adapted, where necessary, to implement congressional intent with respect to liability under the unique statutory scheme of CERCLA.”); Hercules, 247 F.3d at 717 (The Restatement is “the starting point .... [but] only to the extent that it is compatible with the provisions of CERCLA.”).We concur in this general conclusion and acknowledge, in particular, that there are two areas where the Restatement approach is a somewhat poor fit and requires slight modifications to ensure that its approach comports with the liability and remediation scheme of CERCLA. First, as we describe infra, there are important distinctions between causation as conceived in the Restatement and causation in the context of CERCLA. Unlike the Restatement’s common law causation, CERCLA affixes liability based upon its PRP provisions, which define classes of liable parties based upon a party’s statutorily-defined nexus to the contaminated site. And second, the concept of “harm” in the Restatement as actual injury does not correspond easily to CERCLA’s priorities. Contamination and the cost of remediation are both relevant for the “harm” analysis under CERCLA. Finally, we recognize that the Restatement’s emphasis on objective considerations to determine whether apportionment is justified in a given case comports with CERCLA’s strict liability scheme. Equitable considerations may play a role in a later contribution action amongst liable parties, but not in an action such as this where the only relevant issue is whether there is a reasonable basis, founded in record evidence, to apportion damages amongst defendants. 1. Causation Section 433A of the Restatement allows for apportionment of damages where “(a) there are distinct harms,[] or (b) there is a reasonable basis for determining the contribution of each cause to a single harm.” RESTATEMENT (SECOND) OF TORTS § 433A(1) (1965) (emphasis added). CERCLA, however, does not require causation as a prerequisite to liability (except with regard to the third-party defense, see § 9607(b), not at issue here). Nonetheless, most of the leading cases on joint and several liability under CERCLA have addressed divisibility under § 433A(1)(b). See, e.g., Bell Petroleum, 3 F.3d at 902-03; Monsanto, 858 F.2d at 172; Chem-Dyne, 572 F.Supp. at 810. Notably, these cases often dealt with simpler facts than those we confront. Chemr-Dyne, for instance, assumed a case quite different from this one. There, the court stated that “[t]ypically ... there will be numerous hazardous substance generators or transporters who have disposed of wastes at a particular site.” 572 F.Supp. at 810. It was in that context — that is, where the question was apportionment among defendants who all disposed of wastes themselves — that Chem-Dyne determined that courts could follow the divisibility principles of the Restatement and remain true to CERCLA. In a situation in which the several defendants are all polluters themselves, divisibility under the Restatement standard is indeed a relatively straightforward analysis, and one in which traditional causation concepts are useful. If the court can estimate with some confidence the amount of waste that each defendant disposed of and has a basis for determining that the extent of contamination of the site is proportional to the amount of waste disposed of, then the Restatement approach to apportionment works nicely. The situation here is different. The three “responsible” parties are: the now-insolvent majority owner and operator of the site; the mostly absentee landlord of a portion of the site; and a seller of chemicals shipped to and stored at the site. Each party had an entirely different role in the contamination process, with overlapping effects, and not all “caused” contamination in any meaningful sense. Most notably, PRP status premised on ownership of a facility does not require any involvement in the disposal of hazardous substances. Thus, to speak of a PRP “causing” contamination of its land simply by owning land on which someone else disposes of hazardous wastes is to indulge in metaphor. At the same time, to allow CERCLA defendants, especially landowner PRPs, to prove through traditional causation analysis that they were not entirely liable would be to undermine the premise on which the statute designated them as PRPs to begin with. CERCLA requires a connection — for example, that the PRP be a landowner “at the time of disposal,” see § 9607(a)(2) — but no further causation. The statute thus departs from Restatement principles by abjuring the traditional “causation” principles in favor of a nexus concept defined by its PRP provisions. Where, as here, the pertinent PRP status is as landowner, the landowner can establish divisibility by demonstrating a reasonable basis for concluding that a certain proportion of the contamination did not originate on the portion of the facility that the landowner owned at the time of the disposal. The arranger nexus is more straightforward, with a focus not on ownership of the facility but rather on the relevant, arranged disposals in light of other contamination at the facility. 2. Harm A second difficulty that results from relying on tort principles in a scheme not based on tort law concerns the application of the term “harm,” used in the Restatement, as applied to CERCLA. See RESTATEMENT (SECOND) OF TORTS § 433A. The CERCLA cost recovery section does not focus on “harm,” but rather on “costs of removal or remedial action” and “necessary costs of response.” § 9607(a). Thus, when applying the Restatement in the context of CERCLA, the question becomes: What is the “harm” that we are attempting to divide? There are three possible kinds of “harm” in actions for remediation costs under CERCLA: the initial disposal, the resulting contamination, and the costs of remediating the contamination. Actual injury to individuals or to property, the usual “harms” in a tort suit, are not a pertinent consideration; the statute is concerned with averting future injury by remediating contamination, not with compensation for past injuries. If the harm were the disposal, then divisibility based on volume of discharge by operators or by parcel would always make sense, because disposal occurs in specific amounts at specific places. If the harm were contamination, then some attempt would have to be made either to justify a direct correlation between disposal and contamination under the specific circumstances or to separate out the leakage that remained as contamination from leakage that either evaporated, was adequately diluted, or for other reasons did not remain on the property in toxic form. If the harm is the cost of remediation, then divisibility would have to be based on the pro rata cost of cleaning up each defendant’s contribution to the contamination. That pro rata cost will sometimes differ from the proportion of contamination caused by each defendant, because the cost of removing contamination can vary with geographical considerations, degree of toxicity, the means of extraction used for different toxic substances, or other factors. In light of a CERCLA liability suit’s central purpose — recovering the cost of eradicating eontamination-we conclude that it is most useful for purposes of determining divisibility to view the “harm” under CERCLA as the contamination traceable to each defendant. Disposal itself is not the focus of the statute, unless it results in contamination. And the cost of cleaning up the contamination is most analogous to the damages recovered in a tort suit, not to the injury on which liability is based. S. Equity Because this case is one in which the harms are not distinct, apportionment must be under Restatement § 433A(l)(b) if it is to be allowed at all. That is, there must be a reasonable basis for determining the contribution of each PRP to the harm. While nothing in the statute directly addresses the question whether equitable factors are appropriate for purposes of apportioning liability among joint tortfea-sors, all the other circuits that have addressed the issue have held that they are not. We again follow their lead. In so holding, we begin from the fundamental difference between apportionment and contribution. Apportionment, which is the relevant question in this case, looks to whether defendants may avoid joint and several liability by establishing a fixed amount of damage for which they are liable. Section 433A of the Restatement speaks to this issue. Contribution is a distinct concept. If there is insufficient evidence to support apportionment, jointly and severally liable PRPs may still seek to recover from each other in a later contribution action. See, e.g. RESTATEMENT (THIRD) OF TORTS § 23(a) (2000) (“When two or more persons are or may be liable for the same harm and one of them discharges the liability of another by settlement or discharge of judgment, the person discharging the liability is entitled to recover contribution from the other, unless the other previously had a valid settlement and release from the plaintiff.”)- As we explain below, CERCLA permits equitable considerations to be taken into account in a contribution action amongst PRPs, but CERCLA’s strict liability scheme does not permit equitable considerations to have any bearing in an action to determine whether defendants have presented sufficient evidence to apportion liability. Section 9613(f), added to CERCLA in 1986, provides for contribution actions among PRPs once liability to the plaintiffs has been determined. That provision describes, quite generally, the considerations applicable in a contribution action for determining whether one PRP can collect from another a portion of the costs for which it has been held liable: “In resolving contribution claims, the court may allocate response costs among liable parties using such equitable factors as the court determines are appropriate.” § 9613(f) (emphasis added). In contrast, CERCLA contains no provision explicitly providing for initial apportionment of liability. And § 433A(1)(b) of the Restatement and the appended commentary concerning divisibility are silent as to equitable considerations. As noted, this circuit heretofore has not addressed divisibility analysis. But the implication from our cases deciding § 9613(f) contribution issues is that the proper time to focus on such factors is at the contribution phase, not the liability phase. See, e.g., Carson Harbor Vill., 270 F.3d at 871 (noting that the “contribution provision aims to avoid a variety of scenarios by which a comparatively innocent PRP might be on the hook for the entirety of a large cleanup bill”); Pinal Creek, 118 F.3d at 1301 (“A PRP’s contribution liability will correspond to that party’s equitable share of the total liability and will not be joint and several.”). Other circuits have been careful to delineate the difference between the equitable considerations pertinent to contribution under § 9613(f) and the objective considerations pertinent to divisibility under § 9607(a). See Hercules, 247 F.3d at 718; Township of Brighton, 153 F.3d at 318; Bell Petroleum, 3 F.3d at 901. As the Sixth Circuit has noted, divisibility analysis, unless carefully limited, has the potential to eviscerate the strict liability principles of CERCLA entirely, “because defendants who can show that the harm is divisible, and that they are not responsible for any of the harm” could whittle their liability to zero. Township of Brighton, 153 F.3d at 318. Additionally, as Township of Brighton also noted in rejecting a fairness-based approach, divisibility analysis is not an invitation to “split the difference” and come up with a “compromise amount.” Id. at 319. While it may seem unfair to hold a partial owner liable for all of the contamination cleanup costs, that perceived unfairness is the result of CERCLA’s expansive statutory liability scheme. Assuring fairness among PRPs is the proper subject of the contribution stage, not of apportionment at the liability stage. See United States v. Rohm & Haas Co., 2 F.3d 1265, 1280-81 (3d Cir.1993), overruled on other grounds by United States v. E.I. DuPont De Nemours & Co., 432 F.3d 161, 162-63 (3d Cir.2005) (en banc). At the liability stage, CERCLA simply assigns liability to statutorily responsible parties so as to assure that, as between those with some connection to the contamination&emdash;and who have, it may be assumed, benefited from the contamination-causing process&emdash;and those with none, such as the taxpayers. Any court-created structure that would allow PRPs to whittle their share to little or nothing and leave the taxpayers holding the bag may seem more equitable to some PRPs but would violate the basic structure of the CERCLA statutory scheme. Because of such concerns, courts have generally refrained from using an equity-based allocation analysis, so as not to weaken further the strict liability principle basic to CERCLA. We agree that while joint and several liability need not be universally applied, see Bell Petroleum, 3 F.3d at 897, the inquiry as to whether such liability is appropriate must focus strictly on whether there is a reasonable basis for apportionment, see, e.g., id. at 901-04. Consequently, in an action under § 9607(a), a court is not to look to equitable considerations, such as relative fault, in determining whether liability is to be joint and several or apportioned. III. Analysis of Railroads’ and Shell’s CERCLA Liability We now proceed to apply these fairly straightforward principles to the circumstances of this case. Here, the Railroads were found to be PRPs under § 9607(a)(2), as the owners of a “facility at which ... hazardous substances were disposed of,” and Shell was found to be a PRP under § 9607(a)(3), as a person who “arranged for disposal ... of hazardous substances owned or possessed by such person.” The first question we address is whether the Railroads and Shell are liable for all the cleanup costs at the Arvin site, or, as the district court held, only some of them. The second question, addressed later, is whether Shell is liable for any of the harm, as an “arranger.” A. Apportionment of Liability 1. Standard of Review and Burden of Proof Because we have not heretofore faced a CERCLA apportionment issue directly, there is no Ninth Circuit precedent concerning the standard of appellate review for such an issue. Three circuits have addressed the question, and two separate approaches have emerged. The Fifth and Eighth Circuits look first to whether there is a reasonable basis for apportioning the harm, an inquiry they consider a question of law reviewed de novo. See Hercules, 247 F.3d at 718-19; Bell Petroleum, 3 F.3d at 896, 902. These two circuits then examine, as a question of fact reviewed under the clearly erroneous standard, precisely how damages are to be divided. See Hercules, 247 F.3d at 718 (holding that “actual apportionment” of damages is a question of fact); Bell Petroleum, 3 F.3d at 896 (same). In contrast, the Sixth Circuit considers divisibility as a whole a factual matter of causation, reviewed entirely under the clearly erroneous standard. Township of Brighton, 153 F.3d at 318 n. 13. This view, however, disregards a distinction between conceptual divisibility and actual allocation that we find both persuasive and useful. The latter inquiry can involve the resolution of credibility issues and of conflicting evidence, while the former ordinarily does not. We believe the most appropriate approach, and the one we therefore adopt here, is the one adopted in Hercules and Bell Petroleum, with a refinement suggested by Judge Parker’s dissent in Bell Petroleum. Judge Parker thought that the majority confused the distinction between the “legal burden that the single harm at issue caused is of a type capable of apportionment, and the factual burden of proving the amount of harm attributable to a particular party.” Bell Petroleum, 3 F.3d at 909 (Parker, J., concurring in part and dissenting in part). We are not sure that there was any such confusion. Rather, an aspect of clear error review is the legal determination whether the party with the burden of proof met that burden; if the party did not and the district court nonetheless ruled for it, then the district court clearly erred. See Lloyd v. Schlag, 884 F.2d 409, 415 (9th Cir.1989) (reviewing “whether the district court committed clear error by holding that [plaintiff] had not met his burden of proof’). Thus, although the harm may be capable of apportionment, the harm may not actually be apportionable in the particular case as a factual matter, given the evidence produced, because the party advocating apportionment has not come forward with the minimum showing needed to meet its burden of proof as to the proper division of liability. We therefore proceed as follows: We inquire, first, whether the particular harm at issue in the case is theoretically capable of apportionment — i.e., whether it could ever be apportioned or whether it is, by nature, too unified for apportionment. That question is one of law, reviewed de novo. Cf. Taisho Marine & Fire Ins. Co. v. M/V Sea-Land Endurance, 815 F.2d 1270, 1274 (9th Cir.1987). Second, we review for clear error whether the defendant submitted evidence sufficient to establish a reasonable basis for the apportionment of liability, taking into account that the burden of proof is on the party seeking allocation, as well as the district court’s actual division of liability. There is no dispute here on the first, purely legal question — whether the harm is capable of apportionment. See Bell Petroleum, 3 F.3d at 896; Chem-Dyne, 572 F.Supp. at 810. Some of the contamination on the B & B site occurred before the Railroads’ parcel became part of the facility. Only some of the toxic substances were stored on the Railroads’ parcel, and only some of the water on the facility washed over the Railroads’ site. As to Shell, only some of the toxic substances spilled on the facility were sold by the company. The different toxic substances vary in their likelihood to leak and in the manner and speed in which they disseminate in ground water. So, conceptually, the contamination traceable to the Railroads and Shell, with adequate information, would be allocable, as would be the cost of cleaning up that contamination. The questions, then, are whether the district court clearly erred in finding that the Railroads and Shell established a “reasonable basis” for apportionment, Bell Petroleum, 3 F.3d at 901, and whether, having so found, the district court properly apportioned the harm. We recognize that the district court at one point stated that the Railroads failed to “meet their burden of proof’ as to divisibility. But its overall ruling was necessarily to the contrary, as the court also stated that it “independently found [in the record] a reasonable basis for apportionment in spite of the parties^] presentations.” Thus, while the district court rejected both defendants’ theories as to divisibility, it used record evidence it found persuasive to determine apportionment. Whether the district court was correct in this regard is, as we have noted, part of the review of the factual decision regarding apportionment, discussed hereafter. The burden of proof issue thus melds with the merits of the apportionment issue, rather than barring us from considering it. 2. The Railroads As we have established, if apportionment is to be allowed under the Restatement approach, there must be a reasonable basis for calculating the connection between the Railroads’ PRP status and the relevant harms. Again, the harm we consider is the contamination on the Arvin site. Where, as for the Railroads, the PRPs’ responsibility under the statute derives solely from their status as landowner, the PRPs can establish divisibility by demonstrating that discrete portions of the contamination did not originate on land they owned at the time of the toxic disposal. Here, the district court’s severability analysis — after 191 pages of an amended opinion that included over 80 pages of factual findings — ultimately relied on the simplest of considerations: percentages of land area, time of ownership, and types of hazardous products. Although we do not fault the district court’s factfinding — its numbers are mostly correct — its legal conclusion that these three factors alone suffice to support apportionment cannot stand. We address each factor below to show why. a. Land Area The only court of appeals case that has fully addressed divisibility of landowner liability takes a relatively strict approach to apportionment on the basis of land area. In United States v. Rohm and Haas Co., 2 F.3d 1265 (3d Cir.1993), the most analogous CERCLA divisibility case to this one, the Third Circuit held, as do we, that “simply showing that one owns only a portion of the facility in question is [not] sufficient to warrant apportionment.” Id. at 1280. Like this case, Rohm and Haas concerned a landowner PRP and changes in landownership over time. Although the Third Circuit’s divisibility analysis is fairly cursory, its reluctance to apportion landowner liability on the basis of land boundaries is informative. Rohm and Haas indicates that the mere percentage of land owned by one PRP relative to the entire facility cannot alone be a basis for apportionment, as it does not provide a minimally reliable basis for tracing the proportion of leakage, contamination, or cleanup costs associated with the entire parcel. Contrary to Rohm and Haas, the district court’s analysis gave star billing to the percentage of land ownership, even in a unified facility. We agree with Rohm and Haas that this approach, seemingly straightforward though it is, fails in most circumstances to comport with the “reasonable basis” test, as the facts of this case illustrate. The Arvin site was a single facility. CERCLA premises landowner liability on ownership of a facility, not on ownership of a certain parcel of land that is part of a facility. The operations on the site were dynamic, with fertilizer rigs stored on the Railroad parcel and filled up on the B & B parcel. Empty pesticide cans were stored on the Railroad parcel before they were crushed and disposed of. After the 1978 windstorm, tanks were stored all over the facility, including on the Railroad parcel. A simple calculation of land ownership does not capture any data that reflect this dynamic, unitary operation of the single Arvin facility. In addition, the synergistic use of different parts of the Arvin site makes division based on percentage of land ownership particularly untenable. The record shows that B & B leased the Railroad parcel to accommodate its expanding operations. The Railroad parcel added an unquantifiable and perhaps exponential amount to B & B’s soil contamination. Were the Railroad parcel not part of the facility, there would have been less overall storage capacity. One can assume that a smaller amount of toxic chemicals would have been delivered to, and spilled on, the Ar-vin site. The fertilizer rigs, for example, were stored almost exclusively on the Railroad parcel. Had that parcel not been available, less fertilizer might have been delivered to-and leaked onto-the Arvin parcel. , As these descriptions suggest, nothing in the record supports a conclusion that the leakage of contaminants that ended up on the B & B parcel occurred on each parcel in proportion to its size. Instead, given the circumstances of this case, more pertinent comparisons would be the proportion of the amount of chemicals stored, poured from one container to another, or spilled on each parcel. For example, were adequate records kept, it would be possible to estimate the amount of leakage attributable to activities on the Railroad parcel, how that leakage traveled to and contaminated the soil and groundwater under the Arvin parcel, and the cost of cleaning up that contamination. But none of this data is in the record. It may well be that such information is, as a practical matter, not available for periods long in the past, when future environmental cleanup was not contemplated. Unlike records concerning the amount of toxic chemicals produced by a given operator of a facility, records that separate out, with any precision, the amount of toxic chemicals stored on one part of a facility as opposed to another would have had little utility to B & B, the operator of the facility, and none to the Railroads, the owners of the parcel. This observation is true in spades for the more directly pertinent data, such as the amount of leakage on the Railroad parcel, the amount of that leakage that flowed onto the B & B parcel, and the amount of that residue that remained as contamination under the B & B parcel when the cleanup began. So the failure to keep these records is quite understandable. But these practical considerations cannot justify a “meat-axe” approach to the divisibility issue, premised on percentages of land ownership, as a means of adjusting for the difficulties of proving divisibility with precision when PRP status is based on land ownership alone. Such an approach would be tantamount to a disagreement with the imposition of no-fault land ownership liability. Congress, however, created precisely such liability, placing the responsibility to pay for environmental cleanup on parties, such as the Railroads, that profited from the circumstances giving rise to the contamination so that the taxpayers are not left holding the tab. The risk of lack of adequate information for meaningful division of harm therefore must rest on the responsible parties, even when that information is extremely hard to come by. b. Period of Ownership Just as the district court’s land area calculations did not correspond to the harms in this case, its simple fraction based on the time that the Railroads owned the land cannot be a basis for apportionment. The fraction it chose assumes constant leakage on the facility as a whole or constant contamination traceable to the facility as a whole for each time period; no evidence suggests that to be the case. Again, if adequate information were available, it would make sense to eliminate the Railroads’ liability for the period before B & B leased the Railroad parcel. See, e.g., Rohm and Haas, 2 F.3d at 1280. The evidentiary vacuum concerning the amount of contamination traceable to the pre-lease period, however, precludes any such calculation here. c. Types of Hazardous Products While many of the district court’s calculations were factually correct but legally insufficient, its decision to assign a two-thirds fraction to represent the present types of hazardous products contains a basic factual error. All three chemicals were on the Railroad parcel at some time. There is no evidence as to which chemicals spilled on the parcel, where on the parcel they spilled, or when they spilled. Yet, there is evidence that there may well have been leakage on the Railroad parcel of DD, the chemical the district court excluded from its calculations. Given the record, the district court clearly erred in its attempt to rely on the proportion of hazardous products present on the Railroad parcel. d.Conclusion It will often be the case that a landowner PRP will not be able to prove in any detail the degree of contamination traceable to activities on its land. A landowner PRP need not be involved at all in the disposal of hazardous chemicals and so will often have no information concerning that disposal or its impact. The net result of our approach to apportionment of liability, consequently, may be that landowner PRPs, who typically have the least direct involvement in generating the contamination, will be the least able to prove divisibility. And contribution “is not a complete panacea since it frequently will be difficult for defendants to locate a sufficient number of additional, solvent parties.” O’Neil v. Picillo, 883 F.2d 176, 179 (1st Cir.1989). While the result may appear to fault a landowner PRP for failing to keep records proving the minor connection of its land to the contamination on the facility as a whole, CERCLA is not a statute concerned with allocation of fault. Instead, CERCLA seeks to distribute economic burdens. Joint and several liability, even for PRPs with a minor connection to the contaminated facility, is the norm, designed to assure, as far as possible, that some entity with connection to the contamination picks up the tab. Apportionment is the exception, available only in those circumstances in which adequate records were kept and the harm is meaningfully divisible. In sum, although most of the numbers the district court used were sufficiently exact, they bore insufficient logical connection to the pertinent question: What part of the contaminants found on the Arvin parcel were attributable to the presence of toxic substances or to activities on the Railroad parcel? We therefore reject the district court’s apportionment calculation and hold that the Railroads have failed to prove any reasonable basis for apportioning liability for the costs of remediation. 3. Shell Shell’s contribution to the contamination of the Arvin site is easier to isolate than that of the Railroads’, as it involved ascertainable pollutants entering the soil in a specific way. Shell thus had a greater prospect of succeeding on divisibility than did the Railroads, as there is some volumetric basis for comparing its contribution to the total volume of contamination on the Arvin site. Nonetheless, the evidence actually produced was insufficient to allow even a rough approximation of the contamination remaining on the facility, either directly or through the presumption that the pro rata cost of remediating contamination is likely to be equivalent to a PRP’s pro rata share of contamination. Indeed, Shell produced only evidence concerning leakage. Such leakage or disposal evidence cannot suffice in the present circumstances as a basis for apportioning the harm in question. As we have explained, contamination-as distinct from leakage-is the necessary consideration. Where there is disposal of multiple contaminants, courts have demanded a “showing [of] a relationship between waste volume, the release of hazardous substances, and the harm at the site.” Monsanto, 858 F.2d at 172. Factors such as “relative toxicity, migratory potential, and synergistic capacity of the hazardous substances” are relevant to demonstrating this relationship. Id. at 172 n. 26. Alternatively, volumetric calculations of contaminating chemicals-those remaining in the environment and requiring cleanup-could be sufficiently specific for apportionment. See Hercules, 247 F.3d at 719; Bell Petroleum, 3 F.3d at 903. But Shell provided no evidence regarding such factors. It thus failed to prove whether its leaked chemicals contaminated the soil in any specific proportion as compared to other chemicals spilled at the site. See United States v. Agway, Inc., 193 F.Supp.2d 545, 549 (N.D.N.Y.2002) (noting that defendants whose products have become commingled in the soil “face an uphill battle in attempting to demonstrate that volumetric contribution is a reasonable basis for apportioning liability of a single harm”). To fill these evidentiary gaps, the district court assumed equal contamination and cleanup cost from all the chemicals’ leakage. This methodology entirely faded to account for the possibility that leakage of one chemical might contribute to more contamination than leakage of another, because of their specific physical properties. Similarly, the cost of cleanup depends upon which contaminants are present; some contaminants are more expensive than others to extract from the soil. Moreover, even as an approximation of leakage, the district court’s calculations were too speculative to support apportionment. Chem-Nuclear is informative in this regard. In Chem