Earth Island Institute and the Center for Biological Diversity appeal the district court denial of their request for a preliminary injunction against implementation of a United States Forest Service restoration project involving two timber sales in the Sierra Nevada mountains. We reverse and remand.
I
In late August 2001, a large wildfire began in the Sierra Nevada mountains and quickly swept through two adjacent national forests as well as some private lands. By the time it was extinguished in September, the “Star Fire” had consumed thousands of acres in both the Eldorado National Forest and the Tahoe National Forest. Over the course of the next year, Forest Service personnel responsible for each forest developed and implemented management plans responding to the blaze. This appeal focuses on the Star Fire Restoration Project, which covered the Eldorado forest.
In the immediate aftermath of the fire, the Forest Service prepared an initial Burned Area Emergency Rehabilitation (BAER) report, which estimated that some 11% of the burned area in the Eldorado forest had experienced high fire intensity, 57% moderate intensity, and 32% low intensity. Subsequently, a more intensive survey prepared in connection with the Environmental Impact Statement (EIS) estimated that 35% of the area had experienced high intensity burns, 45% moderate intensity, and 18% low intensity, leaving 2% unburned.1
In March 2002, the Forest Service released a draft EIS that proposed logging 1,714 acres of the Eldorado Forest using tractor, skyline, and helicopter methods. Following public comment, the Forest Service issued a Final Environmental Impact Statement (FEIS) in June that again recommended this option. The plan aimed to prevent another “stand replacing” fire by removing most of the dead trees and preventing the development of excessive woody deb'ris. A “stand” is “an easily defined area of the forest that is relatively uniform in species composition or age and can be managed as a single unit.” The plan also sought to prevent soil erosion by promoting appropriate ground cover, to preserve some dead trees (often referred to as “snags”) and down logs for the use of dependent animal species, and to maximize the monetary value of dead trees by acting quickly to allow logging. Revenue from the timber sale would finance other elements of the plan.
According to the FEIS, 71% of the project area had experienced high severity burn. The report relied upon the mortality guidelines developed by Sherri Smith, a Forest Service entomologist. Based on cambium sampling2 of partially green trees and her review of both the scientific literature and some Forest Service data, Smith determined that trees with 35% green canopy or less were effectively dead; the remaining green leaves or needles would eventually turn brown and fall off. Thus, *1296trees demonstrating either less than 35% live crown or three dead cambium samples would be considered dead and marked for removal.3 The preferred alternative in the FEIS called for the retention of approximately four to eight large dead trees per acre and the removal of the remainder in any general forest areas with greater than 50% tree mortality and in Old Forest Emphasis Areas (OFEAs) with greater than 75% mortality. The guidelines placed no limit on the diameter of tree to be removed.
As one of the eleven national forests in the Sierra Nevada range, the Eldorado forest remains subject to the Sierra Nevada Framework, a comprehensive forest plan published in January 2001. The Framework established a comprehensive conservation strategy for all national forests in the area, including special limitations on logging in OFEAs and specific protection zones designed to preserve the habitat of the California spotted owl.4 According to the Framework, the Forest Service must establish 300-acre Protected Activity Centers (PACs) around all known or suspected spotted owl nesting sites. In these areas, logging is severely restricted, generally to the reduction of surface and ladder fuels.5 The Framework provides that the PACs must be maintained “regardless of California spotted owl occupancy status, unless habitat is rendered unsuitable by a catastrophic stand-replacing event and surveys conducted to protocol confirm non-occupancy.”
In addition, the Framework requires 1000-acre Home Range Core Areas (HRCAs) around each PAC in the Eldorado forest. These areas are supposed to encompass “the best available spotted owl habitat in the closest proximity to the owl PACs where the most concentrated owl foraging activity is likely to occur.” Within these areas, which are managed with the same restrictions as OFEAs, trees of more than 12" in diameter generally may not be removed, though in some circumstances slightly larger trees may be thinned. Following a “stand-replacing event,” the Framework calls for the retention of “all snags 15 inches or greater ... except to address imminent hazards to human safety.” However, in these circumstances the Framework also permits the removal of dead trees “to the extent that project analysis recommends removal to benefit landscape conditions for old forest structure and function.” If fire renders the PAC unsuitable, the Forest Service must attempt to relocate the PAC within the HRCA.
A Forest Service team surveyed the two Eldorado PACs within the Star Fire area, PAC055 and PAC075.6 Using Smith’s *1297mortality estimates, the report concluded that approximately 4% of PAC055 and approximately 13% of PAC075 remained at less than 75% tree mortality per acre. In addition, the report estimated that within the Star Fire area, only 5% of the PAC055 HRCA and 18% of the PAC075 HCRA remained at less than 75% mortality. As a result, the FEIS concluded that adjustment of PAC boundaries was impossible. The report recommended dropping both PACs from the forest plan.
In August 2002, Forest Supervisor John Berry adopted the recommendation of the FEIS but modified the preferred alternative so that no trees with green canopy would be removed from partially burned stands within the former PACs. He further determined that cambium sampling should not be used because of its relative inefficiency (though he also noted that this decision would satisfy any public concern over potential damage to trees from the sampling). Plaintiffs appealed this decision administratively, and the agency affirmed the decision in September. Drawing upon an earlier recommendation by the Regional Forester, the final administrative appeal extended the restriction on removing trees with any green canopy to all trees within the OFEAs. The Forest Service then divided the project area into two timber salvage sales and awarded the contracts to Sierra Pacific Industries.
On September 27, 2002, two environmental organizations challenged the timber sales in federal district court and requested a preliminary injunction against their implementation, arguing that the Forest Service had violated the National Environmental Policy Act (NEPA), 42 U.S.C. § 4321, et seq., and the National Forest Management Act (NFMA), 16 U.S.C. § 1600 et seq. Because logging was scheduled to begin immediately, Plaintiffs also requested a temporary restraining order on October 1. On October 3, the district court issued a temporary injunction against logging trees with any green canopy remaining. The court also permitted Sierra Pacific to intervene on behalf of the Forest Service.
On October 11, the district court denied Plaintiffs’ request for a preliminary injunction. The court determined that the Plaintiffs were unlikely to succeed on their challenges to the Forest Service methodology and data because agencies are entitled to rely upon their own methodology and experts. The court found that the Forest Service had taken a “hard look” at the environmental issues raised by the restoration project and had not abused its discretion in preparing a separate EIS for each national forest. Finally, the court concluded that the Plaintiffs had failed to establish that the timber sales would result in irreparable harm to the California spotted owl or that the “balance of hardships” tipped in their favor. This appeal followed.
II.
Earth Island seeks a preliminary injunction barring implementation of the Star Fire timber sales until their NEPA and NFMA claims are adjudicated. The Ninth Circuit has described two sets of criteria for preliminary injunctive relief. Under the “traditional” criteria, a plaintiff must show “(1) a strong likelihood of success on the merits, (2) the possibility of irreparable injury to plaintiff if preliminary relief is not granted, (3) a balance of hardships favoring the plaintiff, and (4) advancement of the public interest (in certain cases).” Johnson v. Cal. State Bd. of *1298Accountancy, 72 F.3d 1427, 1430 (9th Cir.1995). Alternatively, a court may grant the injunction if the plaintiff “demonstrates either a combination of probable success on the merits and the possibility of irreparable injury or that serious questions are raised and the balance of hardships tips sharply in his favor.” Id. (citations omitted) (emphasis in original). “These two alternatives represent ‘extremes of a single continuum,’ rather than two separate tests.” Clear Channel Outdoor, Inc. v. City of Los Angeles, 340 F.3d 810 (9th Cir.2003). “Thus, the greater the relative hardship to [the party seeking the preliminary injunction,] the less probability of success must be shown.” Id. (citation omitted) (alteration in original).
In general, we review the denial of a preliminary injunction for abuse of discretion. Bay Area Addiction Research & Treatment, Inc. v. City of Antioch, 179 F.3d 725, 730 (9th Cir.1999). The district court, however, “necessarily abuses its discretion when it bases its decision on an erroneous legal standard or on clearly erroneous findings of fact.” Rucker v. Davis, 237 F.3d 1113, 1118 (9th Cir.2001) (en banc), rev’d on other grounds, Dep’t of Hous. & Urban Dev. v. Rucker, 535 U.S. 125, 122 S.Ct. 1230, 152 L.Ed.2d 258 (2002). When the district court is alleged to have relied on an erroneous legal premise, we review the underlying issues of law de novo. Does 1-5 v. Chandler, 83 F.3d 1150, 1152 (9th Cir.2002).
The scope of our review is generally limited to whether the district court employed the proper preliminary injunction standard and whether the court correctly apprehended the underlying legal issues in the case. Rucker, 237 F.3d at 1118. “We typically will not reach the merits of a case when reviewing a preliminary injunction.... By this we mean we will not second guess whether the court correctly applied the law to the facts of the case, which may be largely undeveloped at the early stages of litigation. As long as the district court got the law right, it will not be reversed simply because the appellate court would have arrived at a different result if it had applied the law to the facts of the case.” Id. (citations and internal quotation marks omitted).
In this case, the district court applied an improper legal standard when assessing the “possibility of irreparable injury.” Although the court began with an accurate recitation of the legal standard, it ultimately found that the Plaintiffs “failed to establish that the project will result in actual harm to the California spotted owl as opposed to speculation that some such harm could possibly occur.” The court further noted that the Plaintiffs “failed to show that measures already in place ... will not afford sufficient protection” and that the Plaintiffs “failed to identify any concrete probability of irreparable harm in any other respect.”
Each of these statements places a higher burden of proof on the plaintiff than is warranted. A preliminary injunction only requires plaintiffs to show probable success on the merits and the possibility of irreparable harm. As discussed below, the probability of success in this case is not so remote as to render the irreparable injury component irrelevant. Cf. Arcamuzi v. Continental Air Lines, Inc., 819 F.2d 935, 937 (9th Cir.1987) (“These two formulations represent two points on a sliding scale in which the required degree of irreparable harm increases as the probability of success decreases. If the plaintiff shows no chance of success on the merits, however, the injunction should not issue.”) (internal citations and quotation marks omitted).
*1299Moreover, we have often held that a Forest Service logging plan may, in some circumstances, fulfill the irreparable injury criterion because of the long term environmental consequences. See, e.g., Idaho Sporting Cong., Inc. v. Alexander, 222 F.3d 562, 569 (9th Cir.2000) (noting that the imminent and continuing logging activities presented “evidence of environmental harm ... sufficient to tip the balance in favor of injunctive relief’); Neighbors of Cuddy Mountain v. U.S. Forest Serv., 137 F.3d 1372, 1382 (9th Cir.1998) (“The old growth forests plaintiffs seek to protect would, if cut, take hundreds of years to reproduce.”) (citation omitted). As we recognized in Idaho Sporting Congress, 222 F.3d at 569, the Supreme Court has held that insufficient evaluation of environmental impact under NEPA does not create a presumption of irreparable injury. However, the Supreme Court has also recognized that “[ejnvironmental injury, by its nature, can seldom be adequately remedied by money damages and is often permanent or at least of long duration, i.e., irreparable. If such injury is sufficiently likely, therefore, the balance of harms will usually favor the issuance of an injunction to protect the environment.” Amoco Production Co. v. Village of Gambell, AK, 480 U.S. 531, 545, 107 S.Ct. 1396, 94 L.Ed.2d 542 (1987). The irreparable injury question in this case, therefore, turns on the likelihood of that injury occurring.
In dismissing the likelihood of injury to the spotted owl, the district court focused on the Forest Service decision to prohibit logging of trees with any green canopy within the former PACs and the OFEAs. Plaintiffs, however, directed much of their attention upon trees with green canopy outside the PACs that are scheduled to be logged even under the modified plan. At the preliminary hearing, their attorney argued that:
The project area is 1400 acres. The only place where they are not removing green trees is in the 600 acres of PACs. That’s it. Everywhere else they are taking green trees. That is why we’re here because we feel at the very minimum that the Home Range Core Areas need to be protected as well. Because that’s the only thing that gives the owl a fighting chance. They use their PAC for 50 percent of their foraging, and they use the Home Range Core Area for the other 50 percent. Even if you have this restriction in the PACs for not cutting any green trees, owls that are going to come back there ... that owl isn’t going to survive.
Hearing Transcript at 45-46. There is no indication that the district court considered this possibility in its assessment of irreparable harm. Nor does the record indicate the exact amount of forest with green canopy that exists within the HRCA but outside of the OFEA or the PAC.7 Under these circumstances, the court’s conclusion that Plaintiffs failed to show “actual harm” or a “concrete probability of irreparable harm” constitutes the application of an “erroneous legal standard” and thus “necessarily” an abuse of discretion. Rucker, 237 F.3d at 1118.
Ill
Even if the district court erred in its assessment of irreparable harm, re*1300mand might not be necessary if the Plaintiffs showed no possibility of success on the merits. We must therefore examine the district court’s determination that the Plaintiffs have not shown any likelihood of success on their NEPA and NFMA claims. Moreover, the degree of irreparable harm required for a preliminary injunction increases as the probability of success on the merits decreases, and vice versa. See Idaho Sporting Cong., 222 F.3d at 565; see also Kootenai Tribe of Idaho v. Veneman, 313 F.3d 1094, 1124 (9th Cir.2002) (“If plaintiffs had demonstrated a strong likelihood of success on the merits, then plaintiffs would have needed only to make a minimal showing of harm to justify the preliminary injunction.”). As a result, our assessment of the likelihood of success will also play a role in the district court’s reevaluation of irreparable harm.
Earth Island raises a number of challenges to the Star Fire plan based upon both NEPA and the NFMA. NEPA requires that a federal agency “consider every significant aspect of the environmental impact of a proposed action ... [and] inform the public that it has indeed considered environmental concerns in its decisionmaking process.” Kern v. U.S. Bureau of Land Mgmt., 284 F.3d 1062, 1066 (9th Cir.2002) (quoting Baltimore Gas & Elec. Co. v. Natural Res. Def. Council, 462 U.S. 87, 97, 103 S.Ct. 2246, 76 L.Ed.2d 437 (1983)). In order to accomplish this, NEPA imposes procedural requirements designed to force agencies to take a “hard look” at environmental consequences. Id. The NFMA requires the Forest Service to develop comprehensive forest management plans for each unit in the National Forest System. 16 U.S.C. § 1604(a). Subsequent “plans, permits, contracts, and other instruments for the use and occupancy” of the forests must be consistent with the governing forest plan, in this case the Sierra Nevada Framework. 16 U.S.C. § 1604(1); see also Wilderness Soc’y v. Thomas, 188 F.3d 1130, 1132 (9th Cir.1999). Judicial review of agency decisions under both statutes is governed by the Administrative Procedure Act, which specifies that an agency action may only be overturned when it is “arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law.” 5 U.S.C. § 706(2)(a); see also Native Ecosystems Council v. Dombeck, 304 F.3d 886, 891-92 (9th Cir.2002).
A
Plaintiffs first challenge the Forest Service conclusion that logging large trees reduces the potential damage caused by a future fire. They claim that the relevant scientific studies focus solely on smaller trees, while the few studies discussing larger trees ultimately advocate leaving most of them in place. Plaintiffs also argue that the Forest Service deliberately used scientifically questionable mortality standards to overestimate the level of tree destruction. These standards allowed the agency to remove owl habitat protections and permit more logging than the Framework would otherwise allow. According to the Plaintiffs, the scientific literature overwhelmingly suggests that a large percentage of trees categorized as dead under Smith’s criteria would survive. They also suggest that Smith’s own data contradict her conclusions and that the Forest Service ignored contradictory data contained in more recent studies, especially a 2001 study by Drs. Stephens and Finney. In addition, Plaintiffs offer expert evidence of many healthy trees on the ground more than one year after the fire, contradicting USFS predictions that trees with 65% or more crown scorch will not survive.
NEPA requires that “the public receive the underlying environmental data *1301from which a Forest Service expert derived her opinion.” Idaho Sporting Cong. v. Thomas, 137 F.3d 1146, 1150 (9th Cir.1988). An agency must also “identify any methodologies used” and “make explicit reference by footnote to the scientific and other sources relied upon for conclusions in the [EIS].” 40 C.F.R. § 1502.24. Failure to provide this information “either vitiates a plaintiffs ability to challenge an agency action or results in the courts second guessing an agency’s scientific conclusions.” Idaho Sporting Cong., 137 F.3d at 1150. However, an agency is entitled to wide discretion in assessing the scientific evidence, so long as it takes a hard look at the issues and responds to reasonable opposing viewpoints. See 40 C.F.R. § 1502.9(a)-(b). Because analysis of scientific data requires a high level of technical expertise, courts must defer to the informed discretion of the responsible federal agencies. Marsh v. Ore. Natural Res. Council, 490 U.S. 360, 377, 109 S.Ct. 1851, 104 L.Ed.2d 377 (1989). “When specialists express conflicting views, an agency must have discretion to rely on the reasonable opinions of its own experts, even if a court may find contrary views more persuasive.” Id. at 378, 109 S.Ct. 1851. At the same time, courts must independently review the record in order to satisfy themselves that the agency has made a reasoned decision based on its evaluation of the evidence. Id.
In this case, the Forest Service explicitly drew upon the existing literature when estimating the size and potential damage of various levels of future fuel loads. The agency concluded that higher numbers of larger downed logs would increase fire intensity and make fire control much more difficult. The FEIS discussed this decision at length and provided both data from the Star Fire area and detailed projections of future snag falls. The FEIS also directly responded to comments raising questions about the removal of larger snags. Based on this record, we cannot say that the Forest Service acted arbitrarily and capriciously in its assessment.
Similarly, the FEIS discusses at length the development of Smith’s criteria for determining tree mortality, explaining that Smith and her colleagues merely updated a well-known set of tree mortality guidelines based on experience with recent fires. The FEIS also describes the relationship between Smith’s criteria and other published findings (such as the fact that she recommends linear measurement of crown damage rather than volume measurement and thus arrives at different numbers) and the reasons for distinguishing apparently conflicting studies cited in public comments. Finally, the FEIS explains that the Forest Service had requested but not received the Stephens & Finney data and that one of the authors had noted privately that the study dealt with prescribed fires rather than post-fire salvage criteria. The Forest Service thus provided a reasoned explanation for its decision not to incorporate this data, which is all that NEPA requires. Even if we found Plaintiffs’ interpretation of the overall scientific evidence more persuasive, that would not be enough to declare the alternative agency interpretation arbitrary and capricious.
However, Plaintiffs also charge that the USFS data is factually incorrect. That is, regardless of the specific methodology adopted for analyzing the data, Plaintiffs claim that the agency data simply does not match the reality on the ground. If true, this fact would do more than challenge the USFS methodology or its conclusion that partially green trees will likely die and thus may be counted as already dead. It would suggest that the USFS experts had relied upon factually inaccurate data or *1302that FEIS did not follow the methodology it claimed to follow.
Plaintiffs present evidence that Monica Bond, a spotted owl scientist, found much higher tree canopy readings (using a densiometer) than the FEIS describes, and that she discovered large numbers of green trees where the FEIS specifically states that no green foliage exists. Similarly, Plaintiffs claim that Dr. Ed Royce used the Forest Service’s own methodology and arrived at radically different results than those reported in the FEIS. Finally, the Plaintiffs argue that the sale area maps, issued after the final agency decision had been made, reclassify many “severely burned” areas as “cut tree” areas, meaning that the logging company may only cut marked trees (as opposed to leaving only marked trees). Plaintiffs claim that areas are only marked as “cut tree” when they have experienced mild or moderate burns, thus suggesting that the Forest Service had implicitly acknowledged a lower burn level.
At this stage, the record does not allow us to conclude that the Forest Service acted arbitrary and capriciously in relying on its own data and discounting the alternative evidence offered by the Plaintiffs. The agency plausibly explained that apparent differences in burn severity estimates across surveys and maps stemmed from the different levels of specificity and different document purposes. To the extent that the agency reasonably relied upon mortality estimates obtained in compliance with the Smith methodology and reasonably discounted alternative evidence as unreliable or preliminary, the Forest Service is entitled to use the data it collected.
We note, however, that if Plaintiffs are able to convince the district court that the agency unreasonably relied upon inaccurate data, they may be able to succeed on the merits of this claim. In Rybachek v. United States EPA, 904 F.2d 1276, 1297-98 (9th Cir.1990), we reviewed a claim that the agency had falsified data and ignored the results of other tests that would have undermined its analysis. We reviewed the data and the methodology and determined that the agency had used reasonable assumptions to estimate undetectable pollution levels and had reasonably disregarded an anomalous test result. Id. In Idaho Sporting Congress v. Rittenhouse, 305 F.3d 957, 972 (9th Cir.2002), by contrast, we invalidated the Forest Service’s use of habitat as a proxy for population trends in part because its methodology produced obviously inaccurate habitat numbers. Because a claim of factual accuracy differs from an attack on the methodology itself, the district court should revisit this argument on remand. See 40 C.F.R. § 1502.24 (“Agencies shall insure the professional integrity, including the scientific integrity, of the discussions and analyses in environmental impact statements.”)
B
Earth Island also argues that three separate Forest Service decisions violate the Sierra Nevada Framework and thus the NFMA: the logging of trees over 20" in diameter, the de-listing of the PACs within the Eldorado Forest, and the failure to readjust those PAC boundaries. On the first point, the agency plausibly concluded that the 20" limitation applied to undergrowth thinning in live forest stands rather than salvage logging after a severe fire. Thus, the mere fact that the Star Fire Plan allows logging of some larger trees does not constitute a violation of the Framework. Plaintiffs respond that even if this general forest limitation does not apply to severely burned areas, it should still apply to the moderately and mildly burned areas included in the sale areas. However, nothing in the Framework sug*1303gests that the diameter limitation must be maintained in certain portions of a general forest area when it has been legitimately lifted in other portions the same area and the Forest Service has produced a comprehensive plan for the entire area.
As for the PAC claims, the Framework calls for maintenance of PACs “regardless of California spotted owl occupancy status, unless habitat is rendered unsuitable by a catastrophic stand-replacing event and surveys conducted to protocol confirm non-occupancy.” Earth Island must show that the USFS failed to meet these two criteria in order to establish probability of success on its claim that the USFS violated the Framework agreement.
Although the USFS characterization of this intense fire as a stand-replacing event appears reasonable, Plaintiffs raise considerable doubts as to whether the government can meet the second condition necessary to remove restrictions on logging. The government explains in great detail why the habitat appears unsuitable to spotted owl populations, largely relying upon the percentage of dead or dying trees in the area as determined under the Smith methodology. Unfortunately for the agency position, demographic surveys did not “confirm non-occupancy” of PAC075. In fact, the surveys actually turned up an owl pair. According to the FEIS, “[sjurveys for 2002 began in April. Survey of the area near the historic activity center for PAC075 has located the same banded male and a new sub-adult female near the Middle Fork American River.” Plaintiffs also supply their own evidence of owl occupancy, as owl expert Monica Boyd discovered a feather, whitewash, and pellets in August 2002.8 Thus, the agency appears to have failed the second Framework condition for lifting the restrictions on logging within the PAC.
In response, the Forest Service argues that because the habitat was unsuitable, surveys were unnecessary. The Record of Decision confirms that the general determination of unsuitable habitat, rather than any occupancy survey, drove the decision to remove PAC protections. The administrative appeal acknowledges that wildlife animals, including the spotted owl, “occasionally return[ ] to a previously occupied home range that has been severely degraded by a stochastic, stand-destroying event such as a wild fire.” The appeal concludes, however, that “this behavior does not warrant a change in the current regional definition of suitable habitat.” In essence, while Plaintiffs point to the actual presence of owls, the Forest Service simply offers reasons to classify the habitat as unsuitable.9
As authority for the proposition that surveys are not always necessary despite the Framework language, the Forest Service points to a comment posted on the agency website in response to a question in October 2001:
If the entire PAC (and much of the home range core area) has experienced a high intensity burn (70 to 80 percent mortality) and there are obviously no green trees or very few green trees remaining, there is little need to survey to protocol. The probability of occupancy is extremely low. However, if there are portions of the PAC that have remained green, the remaining suitable habitat within the defined PAC and ad*1304jacent home range core area should be surveyed to determine occupancy. There is a high probability that individual birds or pairs will merely shift use patterns and relocate the activity center. A lost PAC does not always lead to the abandonment of a home range core area.
Regional Forester memorandum, Question E5-6 (Oct. 11, 2001). Even under the Smith methodology, the Forest Service admits that some 13% of the original PAC experienced less than 75% mortality. This suggests that “portions of the PAC ... have remained green,” triggering the need for occupancy surveys under the October memorandum.
More importantly, even accepting the general approach outlined in the memorandum, in which post-catastrophe surveys are only required within “suitable” habitat areas, these surveys seem intended to find out whether or not owls are still in the area despite the catastrophe. If such surveys confirm that owls still reside within the area, the restrictions under the Framework should remain in place. After all, the Framework calls for maintenance of PACs “unless habitat is rendered unsuitable ... and surveys confirm non-occupancy” (emphasis added). In this case, the Forest Service appears to treat post-catastrophe surveys as suitable for informational purposes only. According to this approach, the agency may independently determine that the habitat is unsuitable and proceed with declassification of PACs and HRCAs while ignoring any contrary information about occupancy it obtains from the surveys it still conducts. This procedure does not seem to be a plausible interpretation of the quoted passage or of the Framework in general.
A reviewing court is not permitted to substitute its own preferred interpretation of a forest plan for that of the Forest Service. Forest Guardians v. U.S. Forest Serv., 329 F.3d 1089, 1098 (9th Cir.2003). Instead, the agency’s decision can be overturned only if the scheme is plainly erroneous or inconsistent with the forest plan. Id. In this case, given the unambiguous language in the Framework and the most plausible interpretation of the agency’s earlier comments, we cannot agree that the Forest Service fully complied with the forest plan when it ignored the confirmed presence of owls and de-listed PAC075.10
C
Plaintiffs also argue that under NEPA the Forest Service was required to prepare a single EIS covering the Star Fire restoration projects and timber sales in both the Eldorado and Tahoe national forests. The Supreme Court acknowledged that NEPA may require a comprehensive impact statement when several concurrent proposals have a cumulative or synergistic impact. Kleppe v. Sierra Club, 427 U.S. 390, 409, 96 S.Ct. 2718, 49 L.Ed.2d 576 (1976). As we recently reiterated, “[a] single NEPA review document is required for distinct projects when there is a single proposal governing the projects or when the projects are connected, cumulative, or similar actions under the regulations implementing NEPA.” Native Ecosystems, 304 F.3d at 893-94 (citations and internal quotation marks omitted).
*1305In this case, because there is no comprehensive plan covering both forests, Plaintiffs may only prevail by showing that the separate actions are “connected, cumulative or similar,” and that the Forest Service acted arbitrarily and capriciously in failing to prepare a single, comprehensive EIS. Id. (citing 40 C.F.R. § 1508.25).11 Although federal agencies have considerable discretion to define the scope of NEPA review, some actions must be considered together to prevent an agency from “dividing a project into multiple ‘actions,’ each of which individually has an insignificant environmental impact, but which collectively have a substantial impact.” Thomas v. Peterson, 753 F.2d 754, 758 (9th Cir.1985).
The Ninth Circuit applies an “independent utility” test to determine whether actions are “connected” for purposes of NEPA review. See Native Ecosystems, 304 F.3d at 894. (“Where each of two projects would have taken place with or without the other, each has ‘independent utility’ and the two are not considered connected actions.”). As in Native Ecosystems, the two restoration projects in this case have independent utility in that they each generate revenue and implement distinct forest conservation measures, and each plan would go forward without the other. Id. Thus, the two projects do not meet the Ninth Circuit test for connected actions.
Because the two projects may have “cumulatively significant impacts” on spotted owl habitat, they might instead fall under the definition of “cumulative actions” in 40 C.F.R. § 1508.25(a)(2). We have required the Forest Service to prepare a single EIS for multiple timber sales when they formed part of a single timber salvage project, were announced simultaneously, were reasonably foreseeable, and were located in the same watershed. Blue Mountains Biodiversity Project v. Blackwood, 161 F.3d 1208, 1214-15 (9th Cir.1998). Many of the same factors do not seem to apply here. Unlike the multiple timber sales in Blue Mountain, the boundary between sale areas in this case predated the agency decision, and the sales and analyses proceeded on separate time schedules. Moreover, nothing in the record suggests that the agency intended to segment review to minimize cumulative impact analysis. Churchill County v. Norton, 276 F.3d 1060, 1079-80 (9th Cir.2001). Rather, the FEIS explicitly discusses the cumulative impact of many elements of Tahoe’s corresponding Red Star Restoration Project. Although the regulations might support a decision to proceed with a comprehensive EIS for “cumulative actions,” we cannot say that the Forest Service acted arbitrarily under this circuit’s case law in deciding not to do so.
Plaintiffs insist that the two projects instead meet the definition of “similar actions” in 40 C.F.R. § 1508.25(a)(3): “actions which, when viewed with other reasonably foreseeable or proposed agency actions, have similarities that provide a basis for evaluating their environmental consequences together, such as common timing or geography.” We agree that the *1306many similarities in underlying cause, proposed solution, and general geography place the post-fire sales in the Tahoe and Eldorado national forests into this category.
Plaintiffs insist that “similar” actions must be considered together within a single EIS, a position echoed by language found in at least two Ninth Circuit cases. See Northwest Res. Info. Ctr. v. Nat’l Marine Fisheries Serv., 56 F.3d 1060, 1067 (9th Cir.1995) (“an agency is required to consider more than one action in a single EIS if they are ‘connected actions,’ ‘cumulative actions,’ or ‘similar actions’ ”) (citing 40 C.F.R. § 1508.25); see also Native Ecosystems, 304 F.3d at 893-94. However, the regulations actually accord the agency more deference for “similar actions” than the other two categories. Compare 40 C.F.R. § 1508.25(a)(1) (“Connected actions ... should be discussed in the same impact statement”) and § 1508.25(a)(2) (“Cumulative actions ... should therefore be discussed in the same impact statement”) with § 1508.25(a)(3) (“An agency may wish to analyze [similar] actions in the same impact statement. It should do so when the best way to assess adequately the combined impacts of similar actions or reasonable alternatives to such actions is to treat them in a single impact statement.”) (emphasis added). Neither case actually involved a claim about “similar actions,” nor did either decision analyze that particular regulatory provision. See Native Ecosystems, 304 F.3d at 894-95; Northwest Res. Info. Ctr., 56 F.3d. at 1067-68. Moreover, several other Ninth Circuit cases describe the regulations as requiring the preparation of a comprehensive EIS only for cumulative and connected actions. See Churchill County, 276 F.3d at 1076; Wetlands Action Network v. U.S. Army Corps of Eng’rs, 222 F.3d 1105, 1118 (9th Cir.2000); Thomas v. Peterson, 753 F.2d at 758-59.
After comparing the clear language of the regulations with the isolated sentence contained in two cases that do not discuss “similar” actions, we are forced to conclude that the agency was not required to prepare a single EIS for these actions, at least under the circumstances presented here: prior administrative boundaries, different patterns of ownership and destruction, disparate timetables, and separate supervisory personnel. Given these factors, the agency may have reasonably concluded that two separate review documents constituted “the best way to assess adequately the combined impacts of similar actions.” 40 C.F.R. § 1508.25(a)(3). We cannot say that the Forest Service acted arbitrarily.
D
Finally, Plaintiffs argue that the Eldorado study failed to consider adequately the cumulative impact of foreseeable actions in the Tahoe forest. Even if a single, comprehensive EIS is not required, the agency must still adequately analyze the cumulative effects of the projects within each individual EIS. See Native Ecosystems, 304 F.3d at 896 n. 2 (differentiating cumulative impact analysis from the discretionary determination of the EIS scope). Cumulative impact involves the “incremental impact of the action when added to other past, present, and reasonably foreseeable future actions, regardless of what agency (federal or non-Federal) or person undertakes such other actions.” 40 C.F.R. § 1508.7; see also Kern, 284 F.3d at 1075 (discussing this analysis). The government points to the many sections of “cumulative impact” analysis in the FEIS. However, the reviewing court needs to explore the specific parameters of this analysis; the fact that such a section exists is *1307not enough. See Native Ecosystems, 304 F.3d at 896 (noting that the Environmental Assessment had a cumulative impact section but that it failed to discuss the road density amendments at issue).
In this case, the Plaintiffs suggest that the Eldorado Forest FEIS failed to consider adequately the cumulative impact on spotted owl PAC075 within the Tahoe National Forest. Because spotted owls depend upon the surrounding HRCA in addition to the PCA, the destruction of the Eldordo HRCA might significantly impact those owls, even if they “reside” exclusively within Tahoe.
Although the FEIS discusses cumulative impacts on several occasions, only twice does the FEIS mention the role of the Eldorado HRCA in supporting PAC075 within Tahoe. Critically, both passages assume the Tahoe PAC075 has been removed due to lack of suitable habitat. The report also acknowledges that the spotted owl exhibits “high site fidelity” and notes that a pair has returned to the area. The report states that, when partially burned stands are counted, some 304 acres of suitable habitat remain on the former HRCA within the Eldorado Forest. Because this amount was not enough to sustain a breeding pair, the report recommended that the PAC and HRCA protections be removed, though some green patches within the former PAC would be subject to a limited operating period pending further surveys.
The Tahoe Forest has now decided not to de-list PAC075 due to the presence of the owl pair. Given the difficulty of constructing an adequate HRCA within Tahoe Forest, the HRCA area within Eldorado Forest could play an important role for the pair of owls in the Tahoe PAC through the surviving patches of green cover and nearby snags. The Eldorado FEIS never assessed the potential role of the remaining suitable habitat within the former HRCA for a maintained Tahoe PAC075 despite the acknowledged presence of owls in the area. This omission amounts to an insufficient consideration of cumulative impact under NEPA.
The government first responds that the owl nest is in Tahoe, that the Tahoe PAC is being preserved, and that the Plaintiffs failed to prove how close the nest lies to Eldorado. However, the Eldorado FEIS itself reveals the close connection: “The activity center for owl territory PAC075 is actually located on the Tahoe National Forest.... Since the activity center for PAC075 is close to the Eldorado National Forest boundary, a 300-acre PAC for PAC075 was delineated on the Forest.” The close proximity suggests the remaining viable portions of the former Eldorado HRCA may play a role in supporting the Tahoe PAC, triggering the requirement for analysis of the cumulative impact.
The government also argues that Tahoe made its listing decision in November 2002, some five months after Eldorado completed its FEIS.12 However, the gov-*1308eminent admits that the Tahoe decision was based upon the same May 2002 surveys showing actual owl presence that the Eldorado FEIS recognized. The Sierra Nevada Framework only permits de-listing of an owl PAC when “habitat is rendered unsuitable by a catastrophic stand-replacing event and surveys conducted to protocol confirm non-occupancy.” Given that the Eldorado FEIS itself acknowledged the return of owls to Tahoe PAC075, see FEIS at M-16 & M-45, the agency should have reasonably foreseen that Tahoe would not — indeed, could not — de-list PAC075. Thus, the failure to consider the impact of Tahoe’s likely decision appears to be a serious omission.
In supplemental briefing, the government stressed that both forest supervisors surveyed a great deal of territory and found very limited suitable habitat. The agency also emphasized that the FEIS explicitly acknowledged the presence of the two owls, considered the available habitat, and approved the sale nonetheless. Essentially, the government argues that because it reasonably determined the entire area to be unsuitable for owl habitat, it could treat the temporary presence of owls as an aberration. From this point of view, the Tahoe decision to maintain the PAC was irrelevant, since it did not imply the continuing presence of owls. However, at the time Eldorado issued the FEIS, surveys confirmed the presence of owls in their traditional nesting area within Tahoe.13 Given that the Sierra Nevada Framework calls for the maintenance of an HRCA surrounding a PAC, that the same Framework forbids de-listing of a PAC during owl occupancy, and that the record showed actual evidence of owl occupancy, the failure to consider maintaining additional snags and green canopy throughout the former HRCA is not reasonable.
Finally, the government argues that the agency did take reasonable precautions against the Tahoe decision by modifying the preferred alternative to prevent the removal of any trees with partial green canopy from former PAC and OFEA areas. However, Plaintiffs argue primarily that the owls will be damaged by the removal of snags within the PAC and removal of partially green trees from the surrounding HRCA. As mentioned, Plaintiffs claim that a great deal of green forest still exists within the HRCA but outside of the PAC and the OFEA. Thus, they have focused this portion of their NEPA claim on trees not affected by the administrative modifications. Because the FEIS failed to consider the effect that removal of these trees and snags might have on the Tahoe PAC, even though the agency could have reasonably anticipated that the Tahoe forest would be obliged to maintain the PAC, we cannot say that the FEIS sufficiently considered the cumulative impacts of the Star Fire sale.
IV
The Plaintiffs have demonstrated a reasonable probability of success on the merits of some of their claims, and the district court applied an improper legal standard of irreparable harm. We also note that in its discussion of the balance of hardships, the district court only compared the Plaintiffs failure to demonstrate irreparable harm with the “loss in value of deteriorating timber” to Sierra Pacific. Under Ninth Circuit case law, however, it is also appropriate to consider the broader public interest in the preservation of the forest and its resources. See Kootenai *1309Tribe, 313 F.3d at 1125 (“We have already-decided that, in a case such as this one where the purpose of the challenged action is to benefit the environment, the public’s interest in preserving precious, unreplen-ishable resources must be taken into account in balancing the hardships.”). Those resources include any living trees outside the PACs and OFEA, as well as the California spotted owl. See also Sammartano v. First Judicial District, 303 F.3d 959, 974 (9th Cir.2002) (holding that the public interest is “better seen as an element that deserves separate attention in cases where the public interest may be affected.”). We therefore reverse the decision of the district court and remand for further proceedings consistent with this opinion.
REVERSED AND REMANDED.
. The parties dispute the significance of these different estimates. The Forest Service argues that the BAER report is designed to assess watershed impacts rather than tree mortality. The second, more intensive survey discovered that low levels of soil moisture had produced a greater percentage of surface matter consumption (and related tree mortality) than had been apparent initially. Plaintiffs reject this explanation and offer expert declarations that the initial BAER estimate is a more accurate reflection of the true condition of the forest.
. Cambium is a formative layer of trees that gives rise to new cells.
. As discussed below, Plaintiffs vigorously dispute both the methodology and the specific mortality estimates produced.
. Unlike the northern spotted owl, the California spotted owl is not currently listed as threatened or endangered. At the time of the FEIS, the U.S. Fish and Wildlife Service had classified the species as "sensitive” and had ordered a review of whether it should be classified as endangered or threatened. In February 2003, the Service decided that the species did not warrant additional protection under the Endangered Species Act. See Endangered and Threatened Wildlife and Plants; 12-Month Finding for a Petition to List the California Spotted Owl, 68 Fed. Reg. 7580 (Feb. 14, 2003).
. Examples of surface and ladder fuels are pine needles, shrubs, or an understory layer of trees, which can provide fuel contributing to intense wildfires.
. Because the PAC075 owl nest is located in the Tahoe National Forest along the river boundary with the Eldorado forest, another PAC with the same number surrounds the nest on the Tahoe side of the river. Plaintiffs insist that the entire PAC075 should be considered as a single area, subject to a single *1297EIS. The Forest Service responds that the two regions were numbered identically but not managed jointly, thus permitting separate evaluation.
. Plaintiffs argue that several hundred acres of partially green trees within the HRCA but outside the PAC and OFEA would be cut even under the modified plan. The FEIS concludes that if all partially burned stands were considered suitable habitat, some 236 acres of suitable habitat would remain on the former HRCA outside of the PAC. See FEIS M-45. The record does not reveal the exact overlap between the OFEA and the other two areas, though Plaintiffs argue that a substantial portion of the 200-acre OFEA lies within the PAC.
. Both parties appear to agree that no owls were spotted in PAC055 during 2002.
. The administrative appeal acknowledged soon-to-be-published research by Monica Bond that owls may return to and breed in burned areas but declined to modify its findings.
. Naturally, the agency is not permanently trapped by the language of the Framework. In fact, the Forest Service has recently proposed significant changes to the Sierra Nevada framework. See Notice, Supplemental Environmental Impact Statement, 68 Fed. Reg. 16758 (Apr. 7, 2003). See also U.S. Dep’t of Agrie., Sierra Nevada Forest Plan Amendment, available at http:// www.fs.fed.us/r5/snfpa/draft-seis/index.htm (describing changes proposed in the draft EIS, including easing the “detailed and prescriptive standards and guidelines’’ on California spotted owl protections).
. As the Supreme Court has explained,
The determination of the region, if any, with respect to which a comprehensive statement is necessary requires the weighing of a number of relevant factors, including the extent of the interrelationship among proposed actions and practical considerations of feasibility. Resolving these issues requires a high level of technical expertise and is properly left to the informed discretion of the responsible federal agencies. Absent a showing of arbitrary action, we must assume that the agencies have exercised this discretion appropriately-
Kleppe, 427 U.S. at 412, 96 S.Ct. 2718 (citation omitted).
. The government also argues that uncertainty over the ultimate Tahoe listing decision forced the Eldorado supervisor to make a "worst case assumption that the Tahoe PAC, too, would be delisted.” According to the government, because Tahoe ultimately decided to keep the PAC in operation, the owls ended up with more protection than they would have had under the agency assumption. However, this assumption was only the "worst case” from the ex ante point of view of the owls. From the point of view of the Eldorado decision to remove protections in the surrounding Home Range Core Area, the worst case scenario would have been to assume that Tahoe would not de-list its PAC, which is precisely the assumption Plaintiffs argue should have been made. The assumption used by the agency made the preferred Eldorado decision more, rather than less, likely.
. Although the government claims that no spotted owls have been present in either PAC since mid-August 2002, the record only states that one August 2002 survey did not confirm owl presence. Plaintiffs presented alternative evidence of owl presence in August 2002.