The National Marine Fisheries Service (NMFS) authorized several states to kill California sea lions under section 120 of the Marine Mammal Protection Act (MMPA), which allows the intentional lethal taking of individually identifiable pinnipeds. Plaintiffs filed action for declaratory and injunctive relief against Defendants. The Court held that NMFS 1) did not adequately explain its finding that sea lions were having a “significant negative impact” on the decline or recovery of listed salmonid populations; and 2) NMFS did not adequately explain why a California sea lion predation rate of 1 percent would have a significant negative impact on the decline or recovery of these salmonid populations. Therefore, the agency's action was “arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law.” under the Administrative Procedure Act.
FISHER, Circuit Judge:
*1 In March 2008, the National Marine Fisheries Service (NMFS) authorized the states of Oregon, Washington and Idaho to kill up to 85 California sea lions annually at Bonneville Dam. NMFS made the decision under section 120 of the Marine Mammal Protection Act (MMPA), which allows “the intentional lethal taking of individually identifiable pinnipeds which are having a significant negative impact on the decline or recovery of salmonid fishery stocks” that have been listed as threatened or endangered under the Endangered Species Act (ESA). 16 U.S.C. § 1389(b)(1). We must decide whether the agency's action was “arbitrary” or “capricious” within the meaning of the Administrative Procedure Act (APA), as well as whether the agency violated the National Environmental Policy Act (NEPA) by failing to prepare an environmental impact statement.
I. Factual Background
Like seals and walruses, California sea lions are pinnipeds-marine mammals having fin-like flippers for locomotion. The Bonneville Dam is on the Columbia River, which serves as a migration path for a number of ESA-listed salmonid populations, including five salmon and steelhead populations at issue here: the Upper Columbia River Spring run of Chinook salmon, the Snake River Spring/Summer run of Chinook salmon, the Snake River Basin population group of steelhead, the Middle Columbia River population group of steelhead and the Lower Columbia River population group of steelhead. Each of these populations is listed as threatened or endangered under the ESA. See Final Listing Determinations for 10 Distinct Population Segments of West Coast Steelhead, 71 Fed.Reg. 834, 859-60 (Jan. 5, 2006); Final Listing Determinations for 16 ESUs of West Coast Salmon, and Final 4(d) Protective Regulations for Threatened Salmonid ESUs, 70 Fed.Reg. 37,160, 37,193 (June 28, 2005).
Before 2001, few California sea lions were observed feeding in the area of the dam. In recent years, however, sea lion predation has become more prevalent. Since 2002, the U.S. Army Corps of Engineers has observed sea lion predation of salmonids in the area below the dam each year from January to May, when sea lions are present. The Corps has observed, among other things, the number of pinnipeds present, the number of salmonids consumed and the proportion of all salmonids passing the dam that are taken by pinnipeds foraging in the area:
2002 2003 2004 2005 2006 2007
Number of California sea lions 30 106 101 80 72 69
observed during the year
Estimated salmonid predation 1,010 2,329 3,533 2,920 3,023 3,859
(fish taken) by pinnipeds based
on the Corps' observations
Estimated pinniped predation of 0.4 1.1 1.9 3.4 2.8 4.2
salmonids as a percentage of
salmonid run size (%)1
U.S. Dep't of Commerce, Nat'l Oceanic & Atmospheric Admin., Nat'l Marine Fisheries Serv., Decision Memorandum Authorizing the States of Oregon, Washington and Idaho to Lethally Remove California Sea Lions at Bonneville Dam under Section 120 of the Marine Mammal Protection Act (Mar. 12, 2008). Under the Corps' estimates, California sea lions kill between 0.4 and 4.2 percent of migrating salmonid each year, although the Corps considers these to be minimum estimates because not all predation events are observed.2
*2 NMFS has concluded that the actual number of salmonids consumed by California sea lions “is certainly larger than the numbers actually observed, since not all sea lions are observed nor are all predation events.” Pinniped Removal Authority, 73 Fed Reg. 15,483, 15,485 (Mar. 24, 2008). Accordingly, NMFS calculated the potential consumption of salmonids based on the average number of California sea lions actually observed (86) and their bioenergetic needs. See id. Applying this formula, NMFS estimated that 86 California sea lions at the dam can consume up to 17,458 salmonids annually, of which up to 6,003 salmonids would be listed spring Chinook and up to 611 would be listed steelhead. See id. “Using the observed minimum rate of predation averaged over 2005-2007, and the estimated maximum potential predation rate, yields predation rates ranging from 3.6 percent to 12.6 percent for listed spring Chinook and 3.6 percent to 22.1 percent for listed steelhead.” Id.
Sea lions are only one source of salmonid mortality on the Columbia River. Fisheries and federal power system dams are also major contributors to mortality among listed salmonids. Consistent with the ESA, NMFS manages these other sources of mortality through a variety of recovery plans. Under these plans, commercial, recreational and tribal fisheries are authorized to take between 5.5 and 17 percent of listed salmonids, depending on the size of the run in any particular year. The dam system takes a comparable number of salmonids. Over the past several years, NMFS, the Corps and other federal agencies have issued a series of environmental and biological assessments concluding that those fishery- and dam-related takes have minimal adverse impacts on the viability of listed salmonid populations in the Columbia River. Plaintiffs contend that those assessments are incompatible with NMFS's conclusion here, that California sea lion predation causing lesser mortality among the listed salmonid populations is having a significant negative impact on the populations' decline or recovery.
In November 2006, the states of Washington, Oregon and Idaho applied to NMFS for authorization to lethally remove California sea lions from the Bonneville Dam area under section 120 of the MMPA, which “authorize[s] the intentional lethal taking of individually identifiable pinnipeds which are having a significant negative impact on the decline or recovery of salmonid fishery stocks which ... have been listed as threatened ... or endangered species under the[ESA].” 16 U.S.C. § 1389(b)(1). In accordance with the MMPA, NMFS appointed an 18-member task force to evaluate the application. See id. § 1389(c)(1). In November 2007, the task force delivered its formal recommendations to NMFS. Seventeen members concluded that California sea lions at Bonneville Dam were having a “significant negative impact on the decline or recovery of salmonid fishery stocks” within the meaning of the MMPA and recommended approving the states' application. The Humane Society, a plaintiff in this action, was the sole member of the task force to dissent from that recommendation.
*3 Once the task force completed its work, NMFS addressed the merits of the application. First, to comply with NEPA, NMFS prepared an environmental assessment. See Nat'l Marine Fisheries Serv., Nw. Region, Final Environmental Assessment (Mar. 12, 2008). The final environmental assessment resulted in a finding of no significant impact under NEPA, concluding that approval of the states' application would not significantly affect the quality of the human environment. See 42 U.S.C. § 4332(2)(C). Accordingly, the agency determined that preparation of an environmental impact statement was not required.
Second, NMFS issued a decision partially approving the states' MMPA application. See Pinniped Removal Authority, 73 Fed.Reg. 15,483. NMFS adopted a two-part test for determining whether “individually identifiable pinnipeds” were having a “significant negative impact on the decline or recovery” of at-risk salmonids. Id. at 15,484. NMFS would first determine “whether pinnipeds collectively are having a significant negative impact on listed salmonids” and then determine “which pinnipeds are significant contributors to that impact and therefore may be authorized for removal.” Id.
NMFS found that both parts of the test were satisfied. First, NMFS found that California sea lions collectively were having a significant negative impact on the decline or recovery of the listed salmonid populations, based on three factors: (1) “[t]he predation is measurable, growing, and could continue to increase if not addressed”; (2) “[t]he level of adult salmonid mortality is sufficiently large to have a measurable effect on the numbers of listed adult salmonids contributing to the productivity of the affected [populations]”; and (3) “[t]he mortality rate for listed salmonids is comparable to mortality rates from other sources that have led to corrective action under the ESA.” Id. at 15,485.
Second, NMFS identified the individual sea lions that were significant contributors to the impact-the animals that could be lethally removed. The significant contributors were those California sea lions that both have identifiable physical characteristics, such as natural features or brands, and can be classified as predatory. See id. at 15,486. To be deemed predatory, sea lions must meet three criteria: (1) they have been observed eating salmonids in the observation area below Bonneville Dam between January 1 and May 31 of any year, (2) they have been observed in the observation area on a total of any five days between January 1 and May 31 of any year and (3) they have been sighted in the observation area after having been subjected to active nonlethal deterrence. See id.
NMFS authorized the states to kill California sea lions meeting these criteria for an initial period of five years, with the possibility of a renewal for an additional five years. See id. at 15,487-88. NMFS limited the number of sea lions that could be killed, however, to the lesser of either 85 sea lions per year or, and particularly relevant to our analysis, “the number required to reduce the observed predation rate to 1 percent of the salmonid run at Bonneville Dam.” Id. at 15,486. The 1 percent predation target, the agency said, “is not equivalent to a finding that a one percent predation rate represents a quantitative level of salmonid predation that is ‘significant’ under section 120, and that less than one percent would no longer be significant.” Id. “Rather, it is an independent limit on the numbers of sea lions that can be lethally removed to address the predation problem and is intended to balance the policy value of protecting all pinnipeds, as expressed in the MMPA, against the policy value of recovering threatened and endangered species, as expressed in the ESA.” Id.
II. Procedural Background
*4 On March 24, 2008, the same day NMFS published notice of its action in the Federal Register, Humane Society of the United States, Wild Fish Conservancy and two individuals (collectively, “plaintiffs”) filed this action for declaratory and injunctive relief against the Secretary of Commerce and two NMFS officials (collectively, “defendants”), alleging that NMFS's decision violates section 120 of the MMPA, as well as NEPA. The Washington State Department of Fish and Wildlife and the State of Oregon Department of Fish and Wildlife intervened as defendants.
Plaintiffs in due course filed a motion for summary judgment. Defendants filed cross-motions for summary judgment, as well as a motion to strike the environmental assessments and biological assessment relating to the fisheries and dams to the extent they were not already part of the administrative record. The district court denied plaintiffs' motion for summary judgment, granted defendants' cross-motions for summary judgment and granted in part defendants' motion to strike. The court entered judgment, and plaintiffs timely appealed. In February 2009, a motions panel of this court denied plaintiffs' motion for a stay of NMFS's decision pending appeal. See Humane Soc'y v. Gutierrez, 558 F.3d 896 (9th Cir.2009) (order). We now proceed to the merits of plaintiffs' claims.
STANDARDS OF REVIEW
We review de novo a district court's grant or denial of summary judgment. LaVine v. Blaine Sch. Dist., 257 F.3d 981, 987 (9th Cir.2001). Section 120 of the MMPA does not contain a separate provision for judicial review. Accordingly, our review of NMFS's actions under the MMPA is governed by the judicial review provisions of the APA, 5 U.S.C. §§ 701-06. See City of Sausalito v. O'Neill, 386 F.3d 1186, 1205 (9th Cir.2004); see also Ctr. for Biological Diversity v. Kempthorne, 588 F.3d 701, 707 (9th Cir.2009). The APA also governs our review of an agency's compliance with NEPA, see Ocean Advocates v. U.S. Army Corps of Eng'rs, 402 F.3d 846, 858 (9th Cir.2005), including an agency's decision, based on an environmental assessment, not to prepare an environmental impact statement, High Sierra Hikers Ass'n v. Blackwell, 390 F.3d 630, 640 (9th Cir.2004). Under the APA, agency decisions may be set aside if “arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law.” 5 U.S.C. § 706(2)(A). We discuss the APA standard in greater detail below. A district court's decision to exclude extra-record evidence when reviewing an agency's decision is reviewed for an abuse of discretion. See Partridge v. Reich, 141 F.3d 920, 923 (9th Cir.1998); Great Basin Mine Watch v. Hankins, 456 F.3d 955, 975 (9th Cir.2006) (NEPA).
I. Marine Mammal Protection Act Claim
A. Administrative Procedure Act
Plaintiffs contend that NMFS's application of the MMPA is “arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law.” 5 U.S.C. § 706(2)(A). We agree with their contention, at least to the extent that we conclude that NMFS has not satisfactorily explained the basis of its decision.
*5 Under the APA, “the agency must examine the relevant data and articulate a satisfactory explanation for its action including a ‘rational connection between the facts found and the choice made.’ “ Motor Vehicle Mfrs. Ass'n of U.S., Inc. v. State Farm Mut. Auto. Ins. Co., 463 U.S. 29, 43 (1983) (quoting Burlington Truck Lines v. United States, 371 U.S. 156, 168 (1962)). “The reviewing court should not attempt itself to make up for [an agency's] deficiencies: ‘We may not supply a reasoned basis for the agency's action that the agency itself has not given.’ “ Id. (quoting SEC v. Chenery Corp., 332 U.S. 194, 196 (1947)).
Here, we hold that NMFS has not offered a satisfactory explanation for its action. First, the agency has not adequately explained its finding that sea lions are having a “significant negative impact” on the decline or recovery of listed salmonid populations given earlier factual findings by NMFS that fisheries that cause similar or greater mortality among these populations are not having significant negative impacts. Second, the agency has not adequately explained why a California sea lion predation rate of 1 percent would have a significant negative impact on the decline or recovery of these salmonid populations. These procedural errors require us to direct the district court to vacate NMFS's decision and remand to the agency to reconsider the action or provide a fuller explanation.
In four environmental assessments prepared between 2003 and 2007, NMFS concluded that fishery takes comparable to (or greater than) takes by California sea lions would not have a significant adverse effect on the survival or recovery of many of the same listed salmonid populations involved here. In 2003, NMFS prepared, under NEPA, an environmental assessment of a fisheries plan submitted by Oregon and Washington for the Lower Columbia River and concluded that the plan, which would result in the taking of up to 4 percent of steelhead in one area, would “adversely affect[ ]” ESA-listed salmon and steelhead in the Lower Columbia populations but “will not appreciably reduce the likelihood of survival and recovery of Lower Columbia River steelhead, chinook salmon, and chum salmon in the wild” (emphasis added). In April 2005, NMFS prepared a NEPA environmental assessment with respect to a comprehensive plan for the management of fisheries in the Columbia River. The plan, based on an agreement between NMFS, the states of Oregon and Washington and several Indian tribes, allows fisheries to take between 5.5 and 17 percent of listed Columbia River salmonids annually, depending on the size of the run. NMFS found that implementation of the decision would be expected to result in “minimal adverse effects on Listed Salmonid [populations] in the Columbia River Basin,” and that “[c]umulative impacts from NMFS's Proposed Action would be minor if at all measurable ” (emphasis added). In January 2007, NMFS conducted a NEPA environmental assessment addressing a fisheries plan submitted by Oregon and Washington for the Middle Columbia River. The agency found that the plan, which would result in the taking of up to “10 percent of the annual abundance of natural-origin adult and juvenile steelhead” (emphasis added), would not appreciably reduce the likelihood of survival and recovery of salmon and steelhead listed under the ESA. Finally, in March 2007, NMFS conducted a NEPA environmental assessment of a plan to provide limited ocean fishing of Klamath River fall Chinook salmon. NMFS found that the plan, which permitted fishing of 10 percent of Klamath River fall Chinook salmon, would not “jeopardize the long-term productivity” of the Klamath River fall Chinook population and would result in “no significant adverse effects to the environment.”
*6 NMFS has not adequately explained its finding that sea lion predation is having a significant negative impact on salmonid decline or recovery in light of its positive environmental assessments of harvest plans having greater mortality impacts. The absence of an explanation is particularly concerning with respect to the 2005 fishery environmental assessment. In that assessment, NMFS found that a plan providing for fisheries to take between 5.5 and 17 percent of listed salmonids annually, depending on run size, would be expected to result in “minimal adverse effects on Listed Salmonid [populations] in the Columbia River Basin,” and that the “[c]umulative impacts from NMFS's Proposed Action would be minor if at all measurable.” Those findings are in apparent conflict with NMFS's finding in this case that sea lions responsible for less or comparable salmonid mortality have a “significant negative impact” on the decline or recovery of these same populations, yet the agency has not offered a rationale to explain the disparate findings. Cf. FCC v. Fox Television Stations, Inc., 129 S.Ct. 1800, 1811 (2009) (explaining that an agency must offer “a reasoned explanation” when its current course “rests upon factual findings that contradict those which underlay” a previous course). Without an adequate explanation, we are precluded from undertaking meaningful judicial review. Divergent factual findings with respect to seemingly comparable causes of salmonid mortality raise questions as to whether the agency is fulfilling its statutory mandates impartially and competently. A satisfactory explanation is therefore required.
We cannot gloss over the absence of a cogent explanation by the agency by relying on the post hoc rationalizations offered by defendants in their appellate briefs. Defendants' briefs offer several explanations designed to reconcile NMFS's findings: that “[t]he facile percentage-based comparisons that [plaintiffs' brief] offers oversimplify and, in several instances, mischaracterize the complex facts addressed in those [earlier] analyses”; that the 2005 fishery take environmental assessment is not comparable to the MMPA determination here because the former “cover[s] all fishing on over 280 miles of the Columbia River, Snake, and Clearwater rivers” whereas “the decision at issue here involves mortality from a single source at a single location”; that the fishery plans reviewed under the 2003 Lower Columbia River environmental assessment “completely prohibited the retention of any unmarked wild steelhead,” limiting mortality to “incidental injuries associated with catch and release of listed steelhead”; that the fishery plans under review in the 2007 Middle Columbia River fishery environmental assessment “completely prohibit[ed] the retention of wild listed salmonids”; and that, “unlike fishing, NMFS lacks the ability to regulate sea lion predation from year to year so as to reduce the effects in years when salmon and steelhead runs are low.” These distinctions might be valid, but with one exception they are raised for the first time in defendants' briefs and were not mentioned by NMFS in the decision under review.3 They are therefore not part of our review. See Nw. Envtl. Def. Ctr. v. Bonneville Power Admin., 477 F.3d 668, 688 (9th Cir.2007) (“[W]e ‘may not accept appellate counsel's post hoc rationalizations for agency action.’ “ (quoting Burlington Truck Lines, 371 U.S. at 168)). Defendants' post hoc explanations serve only to underscore the absence of an adequate explanation in the administrative record itself.
*7 Defendants also object to the very premise that an agency is obligated to address apparent inconsistencies such as those at issue here. They point out that the previous environmental assessments addressed the impact of fisheries under NEPA, whereas the present action assessed the impact of pinnipeds under the MMPA. They contend that the MMPA action here cannot reasonably be construed as a “swerve” from “prior precedent.” We agree with defendants' argument up to a point: NMFS's MMPA action here is not a swerve from prior precedent, as the courts have applied that principle in administrative law cases.4 But an agency's duty to explain cogently the bases of its decisions is not limited to circumstances in which the agency departs directly from an earlier path. An agency “must examine the relevant data and articulate a satisfactory explanation for its action including a ‘rational connection between the facts found and the choice made.’ “ State Farm, 463 U.S. at 43 (emphasis added) (quoting Burlington Truck Lines, 371 U.S. at 168). In the circumstances of this case, NMFS's factual findings in the earlier environmental assessments appear to be “relevant data,” such that it was incumbent on the agency to offer a “satisfactory explanation” for its decision in light of the earlier findings.5
NMFS cannot avoid its duty to confront these inconsistencies by blinding itself to them. We do not suggest that an agency has a duty to identify and address any potential tension between current and earlier factual determinations in marginally related administrative actions. But in this case the agency's seemingly inconsistent approach to, on the one hand, fishery and hydropower activities, which are deemed not to be significant obstacles to the recovery of listed salmonid populations, and, on the other hand, sea lion predation, which is deemed to be a significant barrier to salmonid recovery, has occupied the center of this controversy from the start. The issue surfaced prominently in the task force proceedings, see generally Minority Report, Final Report and Recommendations of the Marine Mammal Protection Act, Section 120 Pinniped-Fishery Interaction Task Force: Columbia River (Nov. 5, 2007), and has been raised repeatedly and forcefully by the Marine Mammal Commission, which is a federal entity possessing expertise on issues relating to the protection of marine mammals, see 16 U.S.C. § 1402, throughout the administrative decisional process, see, e.g., Letter from Timothy J. Ragen, Executive Director, Marine Mammal Commission, to Donna Darm, Assistant Regional Administrator, Protected Resources Division, NMFS, at 3 (Apr. 2, 2007) (“[T]o put the estimated level of pinniped predation on listed stocks in context, it should be compared to other sources of mortality, including the various forms of human-related take.”); Letter from Timothy J. Ragen to D. Robert Lohn, Regional Administrator, Northwest Region, NMFS, at 1 (Nov. 23, 2007) (recommending that NMFS “compare the estimated level of removals of ESA-listed salmonids by pinnipeds with authorized levels of incidental and directed take from other sources and explain why some sources are considered significant while others are not”); id. at 3 (same); id. at 5 (noting that “the comparison of pinniped predation with authorized levels of takes from other sources is an area largely glossed over by the Task Force,” and recommending that NMFS “provide[ ] clear explanations to support any determinations that some are significant while others are not”); id. at 6 (noting that “this issue is at the heart of the controversy over pinniped predation” and advising NMFS to “provide a rationale to support its decisions on how to reduce a significant take level when multiple risk factors are involved”); cf. H.R.Rep. No. 103-439, at 47 (1994) (House Committee Report on the Marine Mammal Protection Act Amendments of 1994, Pub.L. No. 103-238, which added section 120 to the MMPA) (“[T]he Committee recognizes that a variety of factors may be contributing to the declines of these stocks, and intends that the current levels of protection afforded to seals and sea lions under the Act should not be lifted without first giving careful consideration to other reasons for the decline, and to all other available alternatives for mitigation.”). Under these circumstances, NMFS was required to provide some explanation for the apparent inconsistencies.
*8 We also conclude that further explanation is required for NMFS's conclusion that California sea lion predation greater than 1 percent would have a significant negative impact on the decline or recovery or the listed salmonid populations.6
NMFS said only that the 1 percent target “is intended to balance the policy value of protecting all pinnipeds, as expressed in the MMPA, against the policy value of recovering threatened and endangered species, as expressed in the ESA.” Pinniped Removal Authority, 73 Fed.Reg. at 15,486. This may be a worthy public policy goal, but the agency's explanation does not help us to understand why this level of predation amounts to a “significant negative impact” or how this level of removal is related to the decline or recovery of listed salmonids. Without that explanation, we cannot ascertain whether NMFS has complied with its statutory mandate under the MMPA.
In this respect we once again echo the concerns of the Marine Mammal Commission, which repeatedly emphasized to NMFS the need to “identify the level at which predation of salmonids by pinnipeds no longer would be considered significant,” because “the taking authority should lapse once predation is reduced to a level where it no longer is having a significant impact.” Letter from Timothy J. Ragen to D. Robert Lohn, Regional Administrator, Northwest Region, NMFS, at 1, 6 (Nov. 23, 2007); see also Letter from Timothy J. Ragen, Executive Director, Marine Mammal Commission, to Donna Darm, Assistant Regional Administrator, Protected Resources Division, NMFS, at 2 (Apr. 2, 2007) (imploring the task force to “take a hard look at the justification for the number of any lethal removals that it recommends be authorized”); Letter from Timothy J. Ragen, Executive Director, Marine Mammal Commission, to D. Robert Lohn, Regional Administrator, Northwest Region, NMFS, at 1 (Nov. 23, 2007) (recommending that NMFS “identify the level at which predation of salmonids by pinnipeds no longer would be considered significant and adopt that level as the goal of any authorized removal program”); id. at 6 (urging NMFS to determine “the level at which the impact would cease to be significant,” at which point the taking authority would “lapse”); id. (criticizing the task force for adopting a 1 percent predation target without engaging “in any quantitative analyses to support” it); id. at 10 (indicating that some members of the task force “appeared to be driven more by the pragmatic goal of designing the authorization they thought most likely to resolve the pinniped-fishery conflict than by whether that authorization would satisfy the requirements of section 120” of the MMPA); Letter from Timothy J. Ragen, Executive Director, Marine Mammal Commission, to Garth Griffin, NMFS, at 3 (Feb. 19, 2008) (stating that “a justification should be provided for establishing [the 1 percent] level of predation as ... a threshold” at which predation “would no longer be considered significant”); Letter from Timothy J. Ragen, Executive Director, Marine Mammal Commission, to Robert Lohn, Regional Administrator, Northwest Regional Office, NMFS, at 2 (Feb. 25, 2008) (“There are two issues about which the Service should be particularly clear in its rationale. The first is the basis for determining the extent to which predation must be reduced to promote conservation and recovery of the salmonid stocks.... The second is the manner and rationale by which the Service is, in effect, allocating allowable salmonid mortality among different sources of mortality.”). The finding that predation at the 1 percent level is significant is not adequately explained.
*9 For each of the foregoing reasons, we hold that NMFS's explanation is incomplete and inadequate to permit meaningful judicial review. On the current record, NMFS has not explained its significance determination in light of seemingly inconsistent factual determinations in earlier environmental assessments of fishery impacts. Nor has the agency properly explained the basis of its determination that California sea lion predation of salmonids is significant at the 1 percent level. The agency's action is therefore “arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law.” 5 U.S.C. § 706(2)(A). Accordingly, we direct the district court to vacate NMFS's decision approving the states' MMPA application and remand to NMFS to afford the agency the opportunity either to articulate a reasoned explanation for its action or to adopt a different action with a reasoned explanation that supports it. See Local Joint Exec. Bd. v. NLRB, 309 F.3d 578, 585 (9th Cir.2002).7
In so holding, we do not impose an undue burden on NMFS on remand. The APA requires only a “cogent explanation.” Nw. Envtl. Def. Ctr., 477 F.3d at 691. We recognize the challenges NMFS faces in addressing salmonid conservation and recovery in the Columbia River, the efforts it has taken to address multiple sources of mortality and the practical difficulties presented by uncertainties and changing conditions on the ground. We also recognize that sea lion predation is a serious and potentially significant problem in this location, and that Congress, in enacting section 120 of the MMPA, has authorized NMFS to give priority to ESA-listed salmonid populations over MMPA-protected pinnipeds under specific circumstances. As judges, our limited role is to ensure that NMFS has properly determined that those specific circumstances exist. To do so, we require an explanation from the agency that enables meaningful judicial review. We conclude that a remand is necessary in this case to permit us to fulfill our function.
B. Chevron Analysis
We are not persuaded by plaintiffs' argument that NMFS's interpretation of the MMPA is impermissible or unreasonable under the Chevron framework. See Chevron, U.S.A., Inc. v. Natural Res. Def. Council, Inc., 467 U.S. 837 (1984).8 Chevron establishes a two-step framework for reviewing agency interpretations of statutes they administer. Under the first step, we determine “whether Congress has directly spoken to the precise question at issue. If the intent of Congress is clear,” then we “must give effect to the unambiguously expressed intent of Congress .” Id. at 842-43. Under step two, “if the statute is silent or ambiguous with respect to the specific issue, the question for the court is whether the agency's answer is based on a permissible construction of the statute.” Id. at 843. “If a statute is ambiguous, and if the implementing agency's construction is reasonable, Chevron requires a federal court to accept the agency's construction of the statute, even if the agency's reading differs from what the court believes is the best statutory interpretation.” Nat'l Cable & Telecomm. Ass'n v. Brand X Internet Servs., 545 U.S. 967, 980 (2005).
*10 Here, plaintiffs challenge NMFS's interpretation of the MMPA's requirement that “individually identifiable pinnipeds” are having a significant negative impact on the decline or recovery of listed salmonid populations. Following the suggestion of the Marine Mammal Commission, NMFS adopted a two-part interpretation for this requirement, in which it would first determine whether California sea lions collectively were having a significant negative impact on listed salmonids and would next determine which sea lions were significant contributors to that impact and therefore eligible for lethal removal. Plaintiffs contend that the agency's interpretation is contrary to the plain language and the legislative history of the statute, which, they argue, require a finding that an individual pinniped is having a significant negative impact on fishery stocks, not whether sea lions in the aggregate are having the requisite impact. The legislative history lends some support to plaintiffs' position, see S.Rep. No. 103-220, at 524 (1994) (“The Secretary would be authorized to allow the lethal removal of a nuisance pinniped if it is identified as habitually exhibiting dangerous or damaging behavior that cannot be deterred by other means.” (emphasis added)), but the language and purpose of the statute as a whole do not preclude the agency's two-part interpretation. See 16 U.S.C. § 1389(b)(1) (authorizing “the intentional lethal taking of individually identifiable pinnipeds which are having a significant negative impact” (emphasis added)), (b)(2) (requiring applicants to specify “a means of identifying the individual pinniped or pinnipeds ” sought to be removed (emphasis added)), (d)(3) (requiring the agency to consider “the extent to which such pinnipeds are causing undue injury or impact to, or imbalance with, other species in the ecosystem, including fish populations” (emphasis added)). We therefore cannot say that the agency's interpretation is an unreasonable one.
Plaintiffs also challenge NMFS's interpretation of the phrase “significant negative impact,” arguing that the three factors relied on by the agency to make a finding of significance amount to an impermissible interpretation. Because we conclude that NMFS's action is inadequately explained and must be remanded under the APA for reasons we have explained above, we need not resolve this question here. See Shay v. FEC, 414 F.3d 76, 97 (D.C.Cir.2005) (“[W]e need not decide whether these ... rules represent altogether impermissible interpretations ...-the Chevron step two inquiry-because in any event the [agency] has given no rational justification for them, as required by the APA's arbitrary and capricious standard.”) (citation omitted).
II. National Environmental Policy Act Claim
We next address plaintiffs' argument that NMFS violated NEPA by failing to prepare an environmental impact statement (EIS). Under NEPA, a federal agency must prepare an EIS for any “major Federal actions significantly affecting the quality of the human environment .” 42 U.S.C. § 4332(2)(C). An agency may first prepare an environmental assessment (EA) to determine whether the effects of a proposed action are significant. If the EA establishes that the action may have significant environmental impacts, the agency must prepare an EIS. If the agency makes a finding on the basis of an EA that the action will have no significant impact (FONSI), no EIS is required. See 40 C.F.R. § 1501.4. Plaintiffs contend an EIS was required in this case on two bases, which we discuss in turn.
A. Beneficial Impact
*11 Plaintiffs first argue that NMFS's finding of significance under the MMPA in essence compels a finding of significance under NEPA. According to plaintiffs, NMFS's determination under the MMPA that sea lions are having a “significant negative impact on the decline or recovery” of listed salmonid populations necessarily implies that the environmental benefits of authorizing the lethal removal of sea lions will have a significant positive impact on these salmonid populations. They contend that this significant beneficial environmental impact triggers the duty to prepare an EIS under NEPA.
As a threshold matter, plaintiffs' argument appears to raise an issue of first impression in this circuit: whether NEPA requires an agency to prepare an EIS when an action has a significant beneficial impact but no significant adverse impact on the environment.9 This is a question we need not resolve, however, because even if solely beneficial impacts trigger an EIS, the record does not demonstrate a significant beneficial impact on the human environment in this instance. First, just because NMFS has concluded that sea lions are having a significant negative impact on listed salmonid populations does not mean that the agency has also determined that the removal action authorized here will have a significant positive impact on these same populations. Second, even if NMFS concluded that its action would have a “significant” positive impact on the fish populations involved, that would not necessarily translate into a finding of a significant effect on the quality of the human environment, as required by NEPA: although both statutes speak of significance, the legal standards under the MMPA and NEPA are distinct.10
B. Adverse Impacts
In the alternative, plaintiffs contend that NMFS should have prepared an EIS based on significant adverse impacts. They rely on three theories of adverse impact, none of which is persuasive.
First, plaintiffs contend that an EIS was required based on the controversial and uncertain nature of the action. See 40 C.F.R. § 1508.27(b)(4) (in deciding whether an action has a significant impact, the agency should consider”[t]he degree to which the effects on the quality of the human environment are likely to be highly controversial”); id. § 1508.27(b)(5) (the agency also should consider “[t]he degree to which the possible effects on the human environment are highly uncertain or involve unique or unknown risks”). As plaintiffs correctly point out, the lethal removal of sea lions at Bonneville Dam has been the subject of some controversy. Most significantly, the Marine Mammal Commission criticized aspects of NMFS's decisionmaking. NMFS, however, acted reasonably in concluding that the Commission's concerns did not make the agency's action highly controversial within the meaning of the NEPA regulations. “The term ‘controversial’ refers ‘to cases where a substantial dispute exists as to the size, nature, or effect of the major federal action rather than to the existence of opposition to a use.’ “ Found. for N. Am. Wild Sheep v. U.S. Dep't of Agric., 681 F.2d 1172, 1182 (9th Cir.1982). “A substantial dispute exists when evidence ... casts serious doubt upon the reasonableness of an agency's conclusions.” Nat'l Parks & Conservation Ass'n v. Babbitt, 241 F.3d 722, 736 (9th Cir.2001) (emphasis added). Here, the Commission criticized some aspects of NMFS's draft EA, but it did not disagree with the EA's primary conclusions that an EIS was not required, that sea lions are having a significant impact under section 120 and that some sort of lethal removal is reasonable.
*12 Second, plaintiffs contend an EIS was required based on the action's potentially deadly consequences for Steller sea lions that frequent the Bonneville Dam area. Plaintiffs contend that Steller sea lions, which, unlike California sea lions, are listed as threatened under the ESA, could be killed mistakenly because NMFS's decision authorizes shooting sea lions that are in the water, where Steller sea lions can be hard to distinguish from California sea lions. See 40 C.F.R. § 1508.27(b)(9) (in deciding the question of significance under NEPA, an agency should consider “[t]he degree to which the action may adversely affect an endangered or threatened species”). Plaintiffs' argument is unpersuasive because NMFS adopted a series of safeguards to ensure that only targeted sea lions would be killed, and concluded that, “[b]ecause of these requirements, it is highly unlikely that a marksman would shoot any sea lion other than one on the list of predatory sea lions.” The record does not demonstrate that this conclusion is in error.
Third, plaintiffs contend an EIS was required based on the impacts to local wildlife viewing opportunities if sea lions are removed from the Bonneville Dam area. Plaintiffs correctly observe that the lethal removal program will reduce or eliminate sea lion viewing opportunities in the vicinity of the dam. But their argument is unpersuasive, in part because the NEPA regulations do not treat wildlife viewing opportunities as a major factor in deciding whether an EIS is required. See 40 C.F.R. § 1508.27(b)(8). Nor does the record contain substantial evidence showing that the dam is a popular site for sea lion viewing. We therefore reject this argument.
III. Defendants' Motion to Strike
Plaintiffs also contend that the district court abused its discretion by granting defendants' motion to strike NMFS's previous, fishery-related environmental assessments to the extent they were not included in the administrative record. As a general matter, judicial review of agency decisions is limited to the record considered by the agency in making its decision. See, e.g., Camp v.. Pitts, 411 U.S. 138, 142 (1973). A court may consider extra-record documents, however, “if necessary to determine ‘whether the agency has considered all relevant factors and has explained its decision.’ “ Sw. Ctr. for Biological Diversity v. U.S. Forest Serv., 100 F.3d 1443, 1450 (9th Cir.1996) (quoting Inland Empire Pub. Lands Council v. Glickman, 88 F.3d 697, 703 (9th Cir.1996)). As we have discussed, NMFS should have explained the apparent factual inconsistency between its significance finding here and its previous fishery assessments. Thus, supplementing the administrative record to include those earlier assessments is “necessary to determine ‘whether the agency has considered all relevant factors and has explained its decision.’ “ Id. We therefore vacate the order granting defendants' motion to strike.
IV. NMFS's Use of Bioenergetic Modeling
*13 Finally, plaintiffs challenge NMFS's use of bioenergetic modeling to supplement the Corps' observation-based estimates of sea lion predation. Plaintiffs have not demonstrated that NMFS's estimates are arbitrary or capricious under the APA. We therefore uphold the agency's use of bioenergetic modeling.
As we have explained, the Corps observed pinniped predation at the Bonneville Dam between 2002 and 2007 and produced predation estimates based on those actual observations. NMFS determined that the Corps' observation-based method underestimated the number of fish killed by sea lions at the dam and therefore supplemented the Corps' figures with estimates produced through a bioenergetic consumption model.11 Although plaintiffs raise a number of objections to NMFS's reliance on the bioenergetic estimates to augment the Corps' predation rates, their arguments are unpersuasive.
First, there is no meaningful dispute that the Corps' observation-based estimates lead to some undercounting. Thus, NMFS's conclusion that “[t]he actual number of salmonids consumed is certainly larger than the numbers” reported by the Corps is not unreasonable. The Marine Mammal Commission agreed with the agency's conclusion that the Corps understated predation, concurring in NMFS's conclusion “that the number of pinnipeds present at the dam likely is greater than the number observed.” NMFS's decision to look beyond the Corps' observation-based estimates was therefore reasonable. See Lands Council v. McNair, 537 F.3d 981, 993-94 (9th Cir.2008) (en banc) (“[A]s non-scientists, we decline to impose bright-line rules on the [agency] regarding particular means that it must take in every case to show us that it has met the [statutory] requirements. Rather, we hold that the [agency] must support its conclusions that a project meets the requirements of the [statute] ... with studies that the agency, in its expertise, deems reliable.”), abrogated in part on other grounds by Winter v. Natural Res. Def. Council, Inc., 129 S.Ct. 365 (2008).
Second, plaintiffs have not demonstrated that the bioenergetic models used by the agency produced unreliable estimates. Plaintiffs rely on a letter in which the Marine Mammal Commission questioned some of the agency's early estimates based on bioenergetic modeling. Plaintiffs have not shown, however, that the Commission criticized the agency's final bioenergetic calculations. Nor have they shown that the Commission asserted that modeling was an improper method of helping to estimate actual sea lion predation. Plaintiffs also rely on a declaration submitted by their expert, Dr. Andrew Trites, that challenges some of the assumptions of NMFS's bioenergetic consumption model. Dr. Trites' submission, however, demonstrates only that several of the agency's factual assumptions are subject to some dispute, not that the agency's assumptions were arbitrary or capricious. See Marsh v. Oregon Natural Res. Council, 490 U.S. 360, 378 (1989) (“When specialists express conflicting views, an agency must have discretion to rely on the reasonable opinions of its own qualified experts even if, as an original matter, a court might find contrary views more persuasive.”); see also Baltimore Gas & Elec. Co. v. Natural Res. Def. Council, 462 U.S. 87, 103 (1983) (when an agency is “making predictions, within its area of special expertise ..., as opposed to simple findings of fact, a reviewing court must generally be at its most deferential”).
*14 Furthermore, in measuring sea lion predation, NMFS relied primarily on the Corps' estimates and only secondarily on the bioenergetic consumption estimates. See Pinniped Removal Authority, 73 Fed.Reg. at 15,485, 15,486. NMFS used the bioenergetic data to supplement rather than supplant the Corps' estimates.
Finally, we observe that the bioenergetic consumption estimates do not appear to have been material to NMFS's decision. NMFS determined that sea lion predation was “significant” under the MMPA even at the mortality levels supplied by the Corps. See id. at 15,485. The agency's decision thus would have been the same even if the Corps' estimates were not supplemented by the bioenergetic consumption estimates.
We affirm summary judgment in favor of defendants on plaintiffs' NEPA claim. We reverse summary judgment on plaintiffs' MMPA claim and remand to the district court with instructions to vacate the decision of NMFS and remand to NMFS. We vacate the district court's order granting defendants' motion to strike. Each party shall bear its own costs on appeal.
AFFIRMED IN PART, REVERSED IN PART, REMANDED.
* The Honorable Jeremy D. Fogel, United States District Judge for the Northern District of California, sitting by designation.
1 The last two rows in the table report data for all pinnipeds, not just California sea lions. As a practical matter, however, the numbers approximate predation by California sea lions because they are responsible for 99 percent of salmonid mortality caused by all pinnipeds collectively.
2 Although the total number of individual California sea lions observed during the year declined after 2004, the overall level of predation has increased because sea lions are staying at the dam for longer periods of time. The average number of days individual sea lions were present increased from 7.6 days in 2004 to 20.3 days in 2007.
3 The only one of these rationales mentioned in NMFS's decision is that fishery takes can be reduced to adjust for a smaller run size, whereas sea lion predation cannot be. See Pinniped Removal Authority, 73 Fed.Reg. at 15,485 (“In contrast to a managed harvest regime, which can reduce mortality in response to decreased run sizes, pinniped predation has the potential to increase even when run sizes are depressed, magnifying the impact.”). We agree with defendants that this distinction could help reconcile the apparent inconsistency between the finding that sea lion predation has a significant negative impact on the decline or recovery of listed salmonid populations and the finding that comparable fishery activities do not. NMFS's decision, however, did not adequately develop this possible rationale, so we cannot say that it resolves the apparent inconsistency. Furthermore, even at 5.5 percent, fishery takes appear to be roughly comparable to sea lion predation rates.
4 We therefore reject plaintiffs' argument that the agency's decision in this case constitutes an unexplained “swerve” from “prior precedent.” Plaintiffs are correct that an agency has a duty to explain a departure from precedent. See Atchison, Topeka & Santa Fe Ry. Co. v. Wichita Bd. of Trade, 412 U.S. 800, 808 (1973) (plurality opinion) (describing an “agency's duty to explain its departure from prior norms” and holding that when an agency departs from prior norms, its reasons “must be clearly set forth so that the reviewing court may understand the basis of the agency's action and so may judge the consistency of that action with the agency's mandate”). This principle indisputably applies when an agency rescinds an existing rule, see, e.g., State Farm, 463 U.S. at 41-42 (holding that “an agency changing its course by rescinding a rule is obligated to supply a reasoned analysis”), applies a legal standard inconsistently, see, e.g., W. States Petroleum Ass'n v. EPA, 87 F.3d 280, 285 (9th Cir.1996), or departs from longstanding practice “without supplying a reasoned analysis for its change of course,” Nw. Envtl. Def. Ctr., 477 F.3d at 690. In addition, at least one district court in this circuit has applied this principle to an agency's unexplained departure from an earlier factual finding. See Defenders of Wildlife v. Hall, 565 F.Supp.2d 1160, 1170 (D.Mont.2008) (holding that the agency acted arbitrarily when it failed to explain its departure from an earlier determination that genetic exchange between wolves located in three distinct recovery areas would be a precondition to delisting the northern Rocky Mountain gray wolf under the ESA). But plaintiffs' attempt to apply this principle here extends it too far. The factual findings contained in NMFS's earlier environmental assessments, which the agency completed under NEPA rather than the MMPA, and which analyzed the effects of fishery activities rather than sea lion predation, do not constitute “precedents” within the meaning of the Atchison line of authority. Hall is distinguishable. There, the U.S. Fish and Wildlife Service's factual determination that genetic exchange was unnecessary for the health of the wolf population directly contradicted its earlier determination that such exchange was required. Here, by contrast, NMFS's factual findings are at most potentially inconsistent, not directly contradictory. NMFS's action therefore cannot be faulted on plaintiffs' theory that the agency has “swerved from prior precedent” without explanation.
5 Plaintiffs also fault NMFS for failing to address an August 2007 biological assessment (BA) of the effects of the Federal Columbia River Power System (FCRPS) on salmonid species listed under the ESA. The FCRPS kills approximately 7.7 percent of listed juvenile salmonids, 9.9 percent of adult listed spring-run Chinook salmon and 16.8 percent of listed adult Snake River steelhead, but the BA concluded “that the net effects of the proposed actions, including the existence and operations of the dams with the proposed mitigation, meet or exceed the objectives of doing no harm and contributing to recovery with respect to the [listed salmonid populations] affected by the operation of the FCRPS” (emphasis added). Plaintiffs argue with some force that the factual findings in the 2007 dam take BA are at odds with the finding of significant negative impact in this case. The dam take BA, however, was compiled by the U.S. Army Corps of Engineers, the Bonneville Power Administration and the Bureau of Reclamation within the U.S. Department of Interior, not NMFS. The unexplained inconsistencies between NMFS's finding in this case and the findings in the dam take BA thus do not raise the same level of concern as apparent inconsistencies among NMFS's own environmental assessments. We do not, however, suggest that NMFS is necessarily free to ignore the dam take BA: to comply with the APA and provide a basis for meaningful judicial review, an agency must examine the “relevant data” and articulate a rational connection between the facts found and the choice made. State Farm, 463 U.S. at 43. The dam take BA may fall within the category of relevant data.
6 This finding is implicit in NMFS's decision to authorize lethal removal up to the point at which a 1 percent predation level is achieved. The MMPA authorizes the lethal removal only of “individually identifiable pinnipeds which are having a significant negative impact on the decline or recovery of salmonid fishery stocks.” 16 U.S.C. § 1389(b)(1). Thus, by authorizing lethal removal up to the 1 percent level, NMFS has impliedly found that this level of predation would have the requisite adverse impact.
7 In rare circumstances, when we deem it advisable that the agency action remain in force until the action can be reconsidered or replaced, we will remand without vacating the agency's action. See Heartland Reg'l Med. Ctr. v. Sebelius, 566 F.3d 193, 198 (D .C.Cir.2009) (explaining that special circumstances may justify remand without vacatur “[w]hen an agency may be able readily to cure a defect in its explanation of a decision”); Idaho Farm Bureau Fed'n v. Babbitt, 58 F.3d 1392, 1405-06 (9th Cir.1995) (holding that the court may remand without vacatur “when equity demands”). In this case, the government has not specifically requested that we remand without vacatur, and it is not otherwise apparent that the circumstances call for doing so. Accordingly, the appropriate remedy here is to direct the district court to vacate NMFS's action and remand to the agency.
8 As a threshold matter, all parties agree that Chevron deference applies to NMFS's interpretation of the MMPA in this case. We therefore assume without deciding that Chevron deference applies. We observe, however, that in some circumstances we have suggested that agencies' one-time statutory interpretations, if lacking in precedential force with respect to future agency actions, may not warrant this deference. See High Sierra Hikers, 390 F.3d at 648 (holding that the U.S. Forest Service's interpretation of the Wilderness Act was not entitled to Chevron deference because the agency “was not acting with the force of law” when it was “granting permits,” an action that would not have “precedential value for subsequent parties”). Given the parties' agreement that Chevron governs, we have no occasion to decide whether High Sierra Hikers would preclude Chevron deference here.
9 We reject defendants' argument that our decision in Bering Strait Citizens for Responsible Resource Development v. U.S. Army Corps of Engineers, 524 F.3d 938, 956 (9th Cir.2008), addressed that issue. Bering Strait involved a permit issued by the Corps for a major gold-mining project near Nome, Alaska. We observed that the project would “favorably affect[ ] parts of the Nome area that suffered environmental damage from previously unconstrained resource development” and held that “the Corps was not required to prepare an EIS” because the project had “no significant detrimental effect on the environment in and near Nome.” Id. at 957. The plaintiffs in Bering Strait did not contend that an EIS was required based on beneficial environmental effects, so we did not squarely address whether significant beneficial effects alone would trigger an EIS. Bering Strait therefore left this issue unresolved.
Other circuits are divided on this question. Compare Sierra Club v. Froehlke, 816 F.2d 205, 211 n.3 (5th Cir.1987) (“[Because] NEPA is concerned with accurate and informed decisionmaking as a general matter [, a]n environmental report that erroneously depicts positive environmental consequences poses as significant an obstacle to informed decisionmaking as one that inadequately assesses adverse circumstances.”), Natural Res. Def. Council, Inc. v. Herrington, 768 F.2d 1355, 1431 (D.C.Cir.1985) (stating in dictum that “both beneficial and adverse effects on the environment can be significant within the meaning of NEPA, and thus require an EIS”), Nat'l Wildlife Fed'n v. Marsh, 721 F.2d 767, 783 (11th Cir.1983) (holding that post-EIS changes required a supplemental EIS where a “Mitigation Plan involves a number of proposed project changes that are likely to have a significant, though beneficial, impact on the environment”), and Envtl. Def. Fund v. Marsh, 651 F.2d 983, 993 (5th Cir.1981) (“[E]ven if the Corps was correct in deciding that the new land use will be beneficial in impact, a beneficial impact must nevertheless be discussed in an EIS, so long as it is significant. NEPA is concerned with all significant environmental effects, not merely adverse ones.”), with Friends of Fiery Gizzard v. Farmers Home Admin., 61 F.3d 501, 504-05 (6th Cir.1995) (concluding that only adverse impacts trigger the obligation to prepare an EIS). Although we do not reach the question, plaintiffs' view is consistent with the weight of circuit authority and has the virtue of reflecting the plain language of the statute.
10 We do not intend to suggest that the legal standards under the two statutes are unrelated. A significant impact on listed salmonids, as defined by the MMPA, certainly could constitute a significant effect on the human environment under NEPA. See 40 C .F.R. § 1508.27(b)(9) (providing that one factor an agency should consider in making the determination of whether to prepare an EIS under NEPA is “[t]he degree to which the action may adversely affect an endangered or threatened species”). We hold only that a finding of significance under the MMPA does not ipso facto mandate a finding of significance under NEPA.
11 NMFS inaccurately characterizes the predation numbers produced by the Corps as predation “actually observed.” Although the Corps' numbers are based on actual observations, they in fact are estimates that reflect not only observed kills but also kills estimated to have occurred during breaks in observation. The Corps' numbers are therefore estimates, though they are based on actual observations rather than bioenergetic consumption calculations.