OPINION
HURWITZ, Circuit Judge:
*1 Before us is a challenge to a decision of the United States Fish and Wildlife Service (“Service”) reversing its previous decision that the Pacific walrus qualified for listing as an endangered or threatened species under the Endangered Species Act of 1973 (“ESA”). We find that the Service did not sufficiently explain why it changed its prior position. We therefore reverse the district court's grant of summary judgment to the Service.
I
A
The ESA is “the most comprehensive legislation for the preservation of endangered species ever enacted by any nation.” Tenn. Valley Auth. v. Hill, 437 U.S. 153, 180, 98 S.Ct. 2279, 57 L.Ed.2d 117 (1978). The Act directs the Secretary of the Interior to protect any “threatened” or “endangered” species. 16 U.S.C. § 1533. A species is endangered if it is “in danger of extinction throughout all or a significant portion of its range,” and threatened if it is “likely to become an endangered species within the foreseeable future throughout all or a significant portion of its range.” 16 U.S.C. § 1532(6), (20). The Service interprets “foreseeable future” to mean the period through which it can reliably determine the threats to a species and the likely consequences. U.S. Dep't of the Interior, Office of the Solicitor, Memorandum on the Meaning of “Foreseeable Future” in Section 3(20) of the Endangered Species Act, No. M-37021, at 13 (Jan. 16, 2009).
*2 The Secretary must maintain a list of species that qualify for protection. 16 U.S.C. § 1533(c). A species qualifies if it is threatened or endangered by: “(A) the present or threatened destruction, modification, or curtailment of its habitat or range; (B) overutilization for commercial, recreational, scientific, or educational purposes; (C) disease or predation; (D) the inadequacy of existing regulatory mechanisms; or (E) other natural or manmade factors affecting its continued existence.” Id. § 1533(a)(1). The Secretary must make listing determinations “solely on the basis of the best scientific and commercial data available,” and “after conducting a review of the status of the species.” Id. § 1533(b)(1)(A).
Any “interested person” may petition the Secretary to list a species. Id. § 1533(b)(3)(A). Upon receiving a petition, the Secretary must determine whether it presents sufficient information to suggest that listing may be warranted. Id. If so, the Secretary must review the species’ status and issue a “12-month finding” that listing is either (1) warranted, (2) not warranted, or (3) warranted but precluded by higher priority listing actions. Id. § 1533(b)(3)(B). Species in the third category become listing candidates, and their status is reviewed annually pending a final “warranted” or “not warranted” finding. Id. § 1533(b)(3)(C)(i). The Secretary has delegated authority to administer the Act with respect to certain species, including the Pacific walrus, to the Service. See 50 C.F.R. § 402.01(b).
B
In 2008, the Center for Biological Diversity (“Center”) petitioned the Service to list the Pacific walrus as threatened or endangered, citing the claimed effects of climate change on its habitat. Endangered & Threatened Wildlife and Plants; 12-Month Finding on a Petition to List the Pacific Walrus as Endangered or Threatened, 76 Fed. Reg. 7,634, 7,634 (Feb. 10, 2011). The Pacific walrus, one of three walrus subspecies, is the largest pinniped (fin or flipper-footed marine mammal) in the Arctic. Id. at 7,635. Its habitat mostly spans the shallow continental shelf waters of the Bering and Chukchi Seas, where its prey (mostly benthic invertebrates, like clams) is abundant. Id. at 7,636.
To perform essential life functions, Pacific walruses must regularly “haul out” of the water, usually in groups onto broken “pack ice.” Id. at 7,635. The ice provides a platform from which to hunt prey and travel to new feeding grounds; habitat on which to breed, birth, and nurse young; a substrate on which to make long-distance movements; and a means of avoiding predators. Id. at 7,635, 7,637.
Reflecting their dependence on sea ice, the Pacific walrus's distribution varies in response to changes in sea-ice cover. Id. at 7,636. Walruses typically spend the winter in the Bering Sea. Id. As sea ice in the Bering Sea deteriorates in the spring, most walruses migrate north to feeding areas in the Chukchi Sea, where sea ice has historically remained year-round. Id. Some, however, remain in the Bering Sea year-round, coming onto land in “coastal haulouts” when sea ice is unavailable. Id.
C
In February 2011, after completing a species status assessment, the Service issued a 45-page decision (the “2011 Decision”), with citations to supporting studies and data, finding that listing of the Pacific walrus was warranted. Id. at 7,634, 7,674. Examining the statutory listing factors, the Service identified a number of threats to the Pacific walrus, including sea-ice loss through 2100, that would “cause substantial losses of abundance and an anticipated population decline ... that will continue in the foreseeable future.” Id. at 7,642, 7,674.
First, the 2011 Decision found the loss of sea-ice habitat threatened the Pacific walrus. Id. at 7,653–54. Climate models projected a substantial decline in Bering Sea ice for all months of the year by the late twenty-first century, and an increasingly long ice-free season in the Chukchi Sea. Id. at 7,643–44. Without sea ice, walruses would be required to seek refuge on land, congregating at coastal haulouts. Id. at 7,645. The walrus population would therefore become more concentrated, leading both to localized prey depletion and increased energy expenditure as walruses traveled greater distances to reach prey. Id. at 7,646–49. This would, in turn, cause decreased body condition, calf abandonment, and decreased survival. Id. The walruses would also experience increased mortality and injuries from disturbance-caused stampedes on land, and increased exposure to predation and hunting. Id.
*3 Second, the 2011 Decision found that subsistence hunting threatened the Pacific walrus. Id. at 7,654, 7,658. The Service estimated average harvests between 2000 and 2008 at 5,285 walruses per year and predicted they would “continue at similar levels” in the foreseeable future. Id. at 7,655, 7,658. Although those levels were sustainable under the current walrus population, the Service predicted they would continue “independent of future walrus population trends.” Id. at 7,658. So, if the walrus population declined for the other predicted reasons, harvests would become unsustainable. Id.
Third, the 2011 Decision found that existing regulatory mechanisms to reduce or limit greenhouse gas emissions to stem sea-ice loss or ensure that harvests decrease at a level commensurate to predicted population declines “are currently inadequate to address [the] threats” of sea-ice loss and subsistence harvest. Id. at 7,674; see also 7,660–65.1
Although the Pacific walrus qualified for listing, the need to prioritize more urgent listing actions led the Service to conclude that listing was presently precluded. Id. at 7,634, 7,676. So, the Service added the Pacific walrus to a list of candidate species, id. at 7,676, and reviewed its status annually through 2016, each time again finding listing warranted but precluded, see, e.g., 81 Fed. Reg. 87,246, 87,256 (Dec. 2, 2016).
D
A settlement of an earlier lawsuit by the Center required the Service to submit a proposed rule or not-warranted finding for the Pacific walrus by September 30, 2017. Endangered and Threatened Wildlife & Plants; 12-Month Findings on Petitions to List 25 Species as Endangered or Threatened Species, 82 Fed. Reg. 46,618, 46,642 (Oct. 5, 2017). So, in May 2017, the Service completed a final species status assessment (“Assessment”). The Assessment “identif[ied] vital needs of the species and evaluat[ed] the current and future conditions affecting those needs.” The Assessment expressly did “not constitute a decision document”; it was intended only to inform the Service's listing decision and “form[ed] the scientific basis from which the [Service] will draw conclusions and make a decision.”
The Assessment concluded that “environmental changes over the last several years such as sea-ice loss and associated stressors are impacting Pacific walruses, but that other stressors that were identified in 2011 have declined in magnitude.” The review team believed that “Pacific walruses are adapted to living in a dynamic environment and have demonstrated the ability to adjust their distribution and habitat use patterns in response to recent shifting patterns of sea ice.” The team acknowledged, however, that the species’ ability “to adapt to or cope with increasing stress in the future is uncertain.”
The Assessment did not offer a comparison between its findings and those in the 2011 Decision, only mentioning the 2011 process on limited occasions, and indicated uncertainty in several critical conclusions. For example, the review team cited evidence that the impact of sea-ice loss might be less severe than previously forecast. But, the team also recognized that sea-ice loss will continue to impose stressors on the Pacific walrus, among them “increasingly negative effects on the population manifested through increased energy expenditure and disturbance related mortality events.” “These factors suggest that increased stress on abundance of the Pacific walrus population will likely result in future population declines,” although “the magnitude of the decline is unknown.” The team similarly cited evidence of declining harvests but recognized its inability to reliably predict future hauls.
*4 The team also explained that “[s]tressors acting on a species and the species response to those stressors are not, in general, equally predictable.” Even recent “observed responses have changed as Pacific walruses have adapted their behaviors,” so “it is likely that further changes in Pacific walrus behaviors will occur as they refine responses, as stressors intensify, or new stressors emerge.” The team had “less confidence in [its] ability to predict the potential behavioral and physiological adaptations of Pacific walruses, and the resulting consequences for reproduction and survival under the sea-ice conditions projected for 2100 because of the extensive time between now and 2100.”
E
In October 2017, after reviewing the Assessment, the Service issued a terse 3-page final decision that the Pacific walrus no longer qualified as a threatened species (“2017 Decision”). Id. at 46,642. The 2017 Decision referred to the 2011 Decision only in its procedural history. See id.
Like the 2011 Decision, the 2017 Decision identified the primary threat to the Pacific walrus as the loss of sea-ice habitat, and the resulting increase in use of coastal haulouts. Id. at 46,644. But, in contrast to the earlier decision, the 2017 Decision did not discuss each statutory listing factor in detail and cited few supporting studies. Instead, it simply incorporated the Assessment by reference. See id. The 2017 Decision concluded that, although the Pacific walrus will see a future reduction in available sea ice,
resulting in reduced resiliency and redundancy, we are unable to reliably predict the magnitude of the effect and the behavioral response of the Pacific walrus to this change, and we therefore do not have reliable information showing that the magnitude of this change could be sufficient to put the subspecies in danger of extinction now or in the foreseeable future.
Id. The 2017 Decision also recharacterized the “foreseeable future”:
While we have high certainty that sea-ice availability will decline as a result of climate change, we have less certainty, particularly further into the future, about the magnitude of effect that climate change will have on the full suite of environmental conditions ... or how the species will respond to those changes. We find that beyond 2060 the conclusions concerning the impacts of the effects of climate change and other stressors on the Pacific walrus population are based on speculation, rather than reliable prediction.
Id. at 46,643.
As to the stressors on the Pacific walrus resulting from increased use of coastal haulouts, the Service stated:
Although Pacific walruses prefer sea ice habitat, they also use land habitat during the summer and fall, but likely not without tradeoffs related to energetic costs and other risks of using coastal haulouts (e.g., trampling events, predation, and disease). Nonetheless, if land habitat proves to be comparable in quality to ice habitat, including access to foraging sites, then it is likely that their habitat needs will be met. If land habitat is inferior to ice habitat for Pacific walruses in summer and fall, then survival and recruitment of Pacific walruses will likely decline and population-level effects would occur. However, while it is likely that the increased use of land habitat will have some negative effects on the population, the magnitude of effect is uncertain given the demonstrated ability of Pacific walruses to change their behavior or adapt to greater use of land.
Id. As to the stress from subsistence hunting, the 2017 Decision noted only that “harvest levels have ... decreased.” Id.
II
The Center filed this action in 2018, alleging that the 2017 Decision violated the Administrative Procedure Act (“APA”) and the ESA. In particular, the Center argued that the Service violated the APA by failing to sufficiently explain its change in position from the 2011 Decision.
*5 The district granted summary judgment to the Service. Ctr. for Biological Diversity v. Bernhardt, No. 3:18-cv-00064-SLG, 2019 WL 4725124, at *14 (D. Alaska Sept. 26, 2019). This timely appeal followed.2
III
We review the district court's grant of summary judgment de novo and must determine whether the Service's decision was “arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law.” Alaska Oil & Gas Ass'n v. Pritzker, 840 F.3d 671, 675 (9th Cir. 2016) (quoting 5 U.S.C. § 706(2)). Although our review is “narrow,” we conduct a “searching and careful” consideration of the agency's decision-making process. Japanese Vill., LLC v. Fed. Transit Admin., 843 F.3d 445, 453 (9th Cir. 2016) (cleaned up). We must determine whether the “agency considered the relevant factors and articulated a rational connection between the facts found and the choices made.” Alaska Oil & Gas Ass'n v. Pritzker, 840 F.3d at 675 (cleaned up). If the agency did not meet its burden, we “should not attempt ... to make up for such deficiencies” and “may not supply a reasoned basis for the agency's action that the agency itself has not given.” Motor Vehicle Mfrs. Ass'n of U.S., Inc. v. State Farm Mut. Auto. Ins. Co., 463 U.S. 29, 43, 103 S.Ct. 2856, 77 L.Ed.2d 443 (1983) (cleaned up).
When an agency changes its position, it must (1) “display[ ] awareness that it is changing position,” (2) show “the new policy is permissible under the statute,” (3) “believe[ ]” the new policy is better, and (4) provide “good reasons” for the new policy. Org. Vill. of Kake v. U.S. Dep't of Agric., 795 F.3d 956, 966 (9th Cir. 2015) (quoting FCC v. Fox Television Stations, Inc., 556 U.S. 502, 515–16, 129 S.Ct. 1800, 173 L.Ed.2d 738 (2009)). Moreover, if a “new policy rests upon factual findings that contradict those which underlay its prior policy,” the agency must provide “a reasoned explanation ... for disregarding facts and circumstances that underlay or were engendered by the prior policy.” Fox, 556 U.S. at 515–16, 129 S.Ct. 1800. This framework applies to a change in position on whether a species warrants protection under the ESA. Ctr. for Biological Diversity v. Zinke, 900 F.3d 1053, 1070 (9th Cir. 2018).
IV
The Center first argues that our review is limited to the four corners of the 3-page 2017 Decision. We disagree. Our obligation is to discern whether the agency offered a reasoned explanation for its change in position. This does not call for a formalistic approach. If an agency's decision document explains the reasons for a change in position, we may be able to adequately review that decision by looking to see whether the record supports the reasons offered. See Alaska Oil & Gas Ass'n v. Jewell, 815 F.3d 544, 563 (9th Cir. 2016) (rejecting a “one document” requirement where the statute does not prevent an agency “from referencing other publicly-available documents in support of its justifications”). Similarly, if a published decision incorporates by reference a separate, fully reasoned document explaining why the agency changed positions, that may suffice.
*6 We are, however, limited to the reasons given by the agency for its action. Zinke, 900 F.3d at 1069. “[A] policy change violates the APA if the agency ignores or countermands its earlier factual findings without reasoned explanation for doing so.” Organized Vill. of Kake, 795 F.3d at 966. And an agency must provide its “reasoned explanation” in a form that can adequately be examined on judicial review, not simply present arguments in its briefing how the decision might have been reached. See Zinke, 900 F.3d at 1069 (“FWS cannot rely on its briefing in this case to explain why the 2014 Finding relied on the Leary study rather than the DeHaan study. The explanation must be evidenced from the listing decision itself.”); Greater Yellowstone Coal., Inc. v. Servheen, 665 F.3d 1015, 1027 n.4 (9th Cir. 2011) (“If the Service relied on this study in making its determination, it did not adequately connect the dots in the Rule such that its path may reasonably be discerned.”) (cleaned up).
The essential flaw in the 2017 Decision is its failure to offer more than a cursory explanation of why the findings underlying its 2011 Decision no longer apply. If, as is the case here, the agency's “new policy rests upon factual findings that contradict those which underlay its prior policy,” a sufficiently detailed justification is required. Fox, 556 U.S. at 515, 129 S.Ct. 1800. “In such cases it is not that further justification is demanded by the mere fact of policy change; but that a reasoned explanation is needed for disregarding facts and circumstances that underlay or were engendered by the prior policy.” Id. at 515–16, 129 S.Ct. 1800.
The 2011 Decision, 45 pages in length, contained specific findings, replete with citations to scientific studies and data, that detailed the multiple stressors facing the Pacific walrus and explained why those findings justified listing. See generally 76 Fed. Reg. at 7,634–79. The 2017 Decision, by contrast, is a spartan document, simply containing a general summary of the threats facing the Pacific walrus and the agency's new uncertainty on the imminence and seriousness of those threats. See 82 Fed. Reg. at 46,642–44. Because the 2017 Decision inherently rejects the specific findings underlying the 2011 Decision, more is needed.
The 2017 Decision's incorporation of the Assessment does not remedy the deficiencies. The Assessment does not purport to be a decision document; it provides information but does not explain the Service's reasons for its change in position in the 2017 Decision. Unlike the 2011 Decision, which arrived at specific conclusions as to each of the identified threats, the Assessment reflects substantial disagreement and uncertainty—both among the team and with respect to the relevant threats—and does not identify the agency's rationale for concluding that the specific stressors identified as problematic in the 2011 Decision no longer pose a threat to the species within the foreseeable future.
Diligently parsing the record, the district court identified reasons why the Service could have rationally decided to alter its position. Ctr. for Biological Diversity, 2019 WL 4725124, at *6–11. The Service also offers some in its appellate brief, noting that the Assessment “re-examined the very stressors that led the agency to conclude that the species warranted listing in 2011.” Had the 2017 Decision offered such explanations, they might have been sufficient to support the Service's change in position. See Zinke, 900 F.3d at 1071–72 (finding reliance on recent population data and new scientific opinions on local water temperature “a sufficient ‘reasoned explanation’ for FWS's change in position” (quoting Organized Vill. of Kake, 795 F.3d at 968)).
But, although the Assessment contains some new information, the actual decision document does not explain why this new information resulted in an about-face from the Service's 2011 conclusion that the Pacific walrus met the statutory criteria for listing. Indeed, a review of the reasons offered by the Service in its appellate briefing shows why we cannot conduct the required appellate review unless the decision document offers the agency's own reasoned explanation.
*7 First, with respect to localized prey depletion, the Service's briefing cites a study that foraging trips by tagged walruses are shorter than expected and suggests that “prey near costal haulouts may be more abundant than previous surveys estimated.” But, the Service merely cites the location of that study in the Assessment's appendix, while identifying no part of the document relevant to the agency's prior findings. Moreover, the Service did not explain why prey would not be depleted in the foreseeable future in light of predicted coastal haulout usage, as the 2011 Decision expressly found. 76 Fed. Reg. at 7,648–49.
Second, the Service's brief cites a study showing that female walruses can travel longer distances than expected to forage. The Service says that this undercuts its prior findings that female walruses might be at greater risk as they cannot travel as far to forage. Yet, the Service again just cites the study's location in the appendix to the Assessment, without explaining how the ability to travel further relates to the risk of increased energy expenditure due to increased forage trip length, which the Assessment again identifies as potentially problematic.
Third, with respect to stampede-related mortalities, the Service's brief cites the Assessment's statement “that management programs in the U.S. and Russia have been effective at reducing disturbances and haulout related mortalities in recent years.” But, the Assessment's next sentence reads: “[I]n spite of these efforts, it is likely that mortalities among younger animals at coastal haulouts will always occur where large aggregations form on land and the number of mortalities is likely a function of the duration of time spent hauled out.” The Assessment does not explain why mortalities, especially of juveniles, will no longer be an issue amid the still-expected increase in use of coastal haulouts, and why the impact, even if somewhat reduced, is no longer problematic.
Fourth, with respect to habitat loss generally, the Service points to “the potential for walruses to change behavior to adapt to the loss of sea ice.” But, the 2017 Decision expresses uncertainty as to whether the Pacific walrus will actually find land habitat to be equivalent to sea ice. See 82 Fed. Reg. at 46,643.
Finally, with respect to subsistence harvest, the Service's appellate brief cites new information in the Assessment showing that harvest levels are the “lowest” on record, which the Service argues contradicts its 2011 expectation that harvests would continue at similar levels. See 76 Fed. Reg. at 7,658. But, this does not explain, for instance, why the finding in the 2011 Decision that hunting may increase as use of coastal haulouts increases no longer applies, id., or whether this level of hunting will remain sustainable given that the Assessment still predicted a decline in the Pacific walrus population.
The agency's change in position on the scope of the foreseeable future presents a closer question, but it ultimately fares no better. In the 2011 Decision, the Service expressly found it could reliably predict the extent of sea-ice loss in the Arctic through 2100, see, e.g., 76 Fed. Reg. at 7,641–44, 7,672, and, as the Service acknowledges in its briefing, at least implicitly held it could predict the walrus's response to sea-ice loss through the same date.3 Yet, the 2017 Decision found the effects of climate change on the Pacific walrus beyond 2060 to be “based on speculation, rather than reliable prediction.” 82 Fed. Reg. 46,643. The new decision noted only that the Service had “less certainty” on “how the species will respond” to changes in environmental conditions, without explaining why it now found 2060 (rather than some other time period) to be the appropriate window. See id.
*8 The Assessment again adds little clarity to that conclusion. The document notes that observations of the Pacific walrus's “responses to the effects of climate change from 2007 ... to the present are likely the most realistic information when evaluating the future,” and that “even these observed responses have changed as Pacific walruses have adapted their behaviors.” But, the review team then stated only that it had “less confidence in [its] ability to predict the potential behavioral and physiological adaptations of Pacific walruses ... because of the extensive time between now and 2100.”
But, of course, the amount of time until 2100 was known at the time of the 2011 Decision. Simply reiterating generic uncertainty that was known at the time of the prior finding does not meet the agency's burden to explain its change in position. This is particularly so here, because even projections with some degree of uncertainty can have value in the ESA-listing process. See Alaska Oil & Gas Ass'n v. Pritzker, 840 F.3d at 680.
The Service's brief marshals evidence suggesting potential adaptation and argues that it would be rational to conclude that predictions could accordingly not be made beyond 2060. But, the detailed explanation it offers is in neither the 2017 Decision nor the Assessment. See id. at 681–82; Zinke, 900 F.3d at 1072–73 (finding decision insufficient where Service did not explain why uncertainty of climate change impacts on species meant not listing was appropriate).
V
The Service may be able to issue a decision sufficiently explaining the reasons for its change in position. But, the 2017 Decision does not do so, and we may not ourselves come up with those reasons from the large and complex record. We therefore reverse the summary judgment below, with directions to the district court to remand to the Service to provide a sufficient explanation of its new position.
REVERSED and REMANDED.
Footnotes
* The Honorable Clifton L. Corker, United States District Judge for the Eastern District of Tennessee, sitting by designation.
1 The Service examined other potential stressors to the Pacific walrus but did not deem any a population-level threat. See 76 Fed. Reg. at 7,639–74.
2 The Service filed and later voluntarily dismissed a cross-appeal. Ctr. for Biological Diversity v. Bernhardt, No. 19-36031 (9th Cir. Aug. 17, 2020), ECF No. 27.
3 The Service “acknowledge[s] its modification of the appropriate ‘foreseeable future’ timeframe” in the 2017 Decision, but elsewhere it appears to suggest that 2100 was a timeframe relevant only to the sea-ice loss analysis.