This case requires that we determine the adequacy of the environmental analysis performed by the Bureau of Land Management of the Department of the Interior for its herbicide spraying program in Oregon forests. The district court found that considerable scientific uncertainty existed as to the safe level of exposure to the herbicides used. It enjoined the BLM from further spraying until it performs a “worst case analysis” under 40 C.F.R. § 1502.22.
The question is whether 40 C.F.R. § 1502.22 requires an agency to perform such an analysis when significant scientific uncertainty exists about the safety of a program and the uncertainty cannot be eliminated by further study. We conclude that it does.
I. FACTS
Southern Oregon Citizens Against Toxic Sprays, Inc. (SOCATS) is a non-profit corporation whose members live near or use forests designated for herbicide spraying by the BLM. The latter annually sprays forest lands near Medford to control non-commercial vegetation and to promote timber production.
The BLM filed a programmatic Environmental Impact Statement in 1978 to cover its spraying program for the following ten years. This program contemplated the use of Silvex, 2,4-D, and 12 other herbicides. The EIS addressed only the human health effects of Silvex. It noted that no adverse effects for the other herbicides were known.
Subsequently, the use of Silvex was suspended by the Environmental Protection Agency. The BLM has continued to spray with the other herbicides and has filed annual Environmental Assessments (EAs) to update the 1878 programmatic EIS.
In its 1979 suit to enjoin further spraying, SOCATS complained that the environmental documents prepared by the BLM were inadequate. Both parties moved for summary judgment.
The district court reviewed supporting affidavits and concluded that there was uncertainty regarding the safety of 2,4-D in small dosages. It noted particularly statements by one of the BLM’s experts, Dr. Dost, who admitted to uncertainty among the scientific community as to the carcinogenicity of 2,4-D. The court held that the scientific uncertainty, coupled with the potential danger to human health, required a worst case analysis.
It granted summary judgment to SO-CATS and enjoined the spraying from which the BLM has appealed.
SOCATS sought attorney fees under the Equal Access to Justice Act. 28 U.S.C. § 2412(d)(1)(A) (Supp.1983). The court held that SOCATS was the prevailing party but denied fees because the government’s position was “substantially justified.” Southern Oregon Citizens Against Toxic Sprays v. Watt, 556 F.Supp. 155 (D.Or. 1983). SOCATS has cross-appealed.
*1478II. ANALYSIS
A. The Need for a Worst Case Analysis
The “worst case analysis” regulation, 40C.F.R. § 1502.22, was promulgated in 1979. It is part of the Council on Environmental Quality’s (CEQ) comprehensive interpretation of the National Environmental Policy Act, 42 U.S.C. §§ 4321 et seq. (NEPA). The CEQ’s regulations are binding on administrative agencies and are entitled to substantial deference in the courts. Andrus v. Sierra Club, 442 U.S. 347, 358, 99 S.Ct. 2335, 2341, 60 L.Ed.2d 943 (1979).
The worst case analysis regulation provides:
Incomplete or unavailable information.
When an agency is evaluating significant adverse effects on the human environment in an environmental impact statement and there are gaps in relevant information or scientific uncertainty, the agency shall always make clear that such information is lacking or that uncertainty exists.
(a) If the information relevant to adverse impacts is essential to a reasoned choice among alternatives and is not known and the overall costs of obtaining it are not exorbitant, the agency shall include the information in the environmental impact statement.
(b) If (1) the information relevant to adverse impacts is essential to a reasoned choice among alternatives and is not known and the overall costs of obtaining it are exorbitant or (2) the information relevant to adverse impacts is important to the decision and the means to obtain it are not known (e.g., the means for obtaining it are beyond the state of the art) the agency shall weigh the need for the action against the risk and severity of possible adverse impacts were the action to proceed in the face of uncertainty. If the agency proceeds, it shall include a worst case analysis and an indication of the probability or improbability of its occurrence.
40 C.F.R. § 1502.22.
The worst case analysis regulation codifies prior NEPA case law. Sierra Club v. Sigler, 695 F.2d 957, 971 (5th Cir.1983). It requires disclosure and analysis of the “cost[s] of uncertainty — i.e., the costs of proceeding without more and better information.” Id. at 970; State of Alaska v. Andrus, 580 F.2d 465, 473 (D.C.Cir.1978), vacated in non-pertinent part sub nom. Western Oil & Gas Ass’n v. Alaska, 439 U.S. 922, 99 S.Ct. 303, 58 L.Ed.2d 315 (1978).
The district court found that scientific uncertainty exists as to the safety of the herbicides and held that § 1502.22 applies because the spraying program could have an adverse iinpact on human health.
The BLWÍ does not appeal the court’s factual findings. It contends that: (1) the district court erred in requiring a worst case analysis, without also finding that the worst case is probable or reasonably likely to occur; (2) a worst case analysis is not required because the herbicides are registered by the EPA under the Federal Insecticide, Fungicide and Rodenticide Act, 7 U.S.C. §§ 136 et seq. (Supp.1983) (FIFRA); and (3) the court erred in holding that the BLM must perform a worst case analysis in an Environmental Assessment. We reject each contention.
1. The Worst Case Analysis Regulation Applies to the Herbicide Spraying Program
The district court held that scientific uncertainty about the safety of the herbicides mandates a worst case analysis, “since herbicide spraying may have a direct impact on human health.” The BLM contends that it should not have to consider impacts that are, in its judgment, neither likely nor probable. It argues that the language “significant adverse effects on the human environment” in § 1502.22, limits that regulation to situations in which the effect is reasonably probable.
Two other courts have considered the need for a worst case analysis under § 1502.22. Sierra Club v. Sigler, 695 F.2d 957 (5th Cir.1983); Village of False Pass v. Watt, 565 F.Supp. 1123 (D.Ak.1983). Both decisions were rendered after the district court’s decision here. Both support its interpretation of § 1502.22.
In Sigler, the Fifth Circuit held that the Army Corps of Engineers must analyze the *1479“worst case” of a catastrophic supertanker oil spill when evaluating the environmental consequences of a proposed oil port. The court read § 1502.22 to require separate consideration of (1) the worst case and (2) “the probability or improbability of its occurrence.” Id. at 974. It said: “that the possibility of a total cargo loss by a supertanker is remote does not obviate the requirement of a worst case analysis .... ” Id.
In Sigler, the “worst case” was an event of low probability but catastrophic effects and the scientific uncertainty concerned those effects. In contrast, this case involves a lack of information about the probability of any adverse effect. A more closely analogous situation is found in False Pass.
False Pass involved the adequacy of an EIS for lease sales of off-shore oil deposits. The court concluded that lack of information about the effect of seismic testing on endangered whale populations triggered § 1502.22. 565 F.Supp. at 1149-58. The court gave the government two alternatives: (1) obtain the missing information or, if that proved impossible, (2) prepare a worst case analysis. Id. at 1153. In either case, the government was required to reconsider the wisdom of its program.
The district court holding here accords with those later cases and with a common sense interpretation of § 1502.22. The government did not challenge the court’s finding that “there is uncertainty about what is the safe level of dosage — or if there is one.” The possibility that the safe level of dosage for herbicides is low or nonexistent creates a possibility of “significant adverse effects on the human environment.” 40 C.F.R. § 1502.22. This potential calls for a worst case analysis.
The BLM’s contention that it need not analyze a “worst case” unless it is “probable” contradicts the clear language of § 1502.22. This section requires that the agency prepare a worst case analysis “and indicate] to the decisionmaker ‘the probability or improbability of its occurrence.’ ” Sigler, 695 F.2d at 974 (emphasis in original) (quoting 40 C.F.R. § 1502.22). The agency may not omit the analysis only because it believes that the worst case is unlikely.
The cases cited by the BLM to support its refusal to consider effects that it considers improbable are easily distinguished. Trout Unlimited v. Morton, 509 F.2d 1276 (9th Cir.1974) was decided prior to the 1979 CEQ regulations. It involved consequences of dam construction that were distantly connected to the agency action. 509 F.2d at 1284. There was no information gap as to the improbability of the consequences. Here, the potential effect flows directly from the proposed action and there is scientific uncertainty regarding its likelihood.
Warm Springs Dam Task Force v. Gribble, 621 F.2d 1017 (9th Cir.1980), involved an information gap about the effect of a newly discovered fault system on a proposed dam. 621 F.2d at 1020-21. We noted that the original failure to discuss this danger violated NEPA, but held that the agency cured the defect by commissioning an extensive study that supplied the missing information. Id. at 1025-26. By eliminating the uncertainty, the agency essentially complied with 40 C.F.R. § 1502.22(a), which requires an agency to obtain information relevant to adverse impacts that “is essential to a reasoned choice among alternatives” when the cost of obtaining it is not exorbitant. It had no need to conduct a worst case analysis under § 1502.22(b).
Further, the BLM’s contention that the worst case is improbable is unsupported by the court’s unchallenged findings. The court found scientific uncertainty regarding the likelihood of the worst case occurring. The BLM’s belief that its herbicides are safe does not relieve it from discussing the possibility that they are not, when its own experts admit that there is substantial uncertainty. When uncertainty exists, it must be exposed.
2. Registration of a Herbicide Under FI-FRA Does Not Alter the BLM’s Duty to Prepare a Worst Case Analysis
The BLM and amicus Monsanto contend also that a worst case analysis is not required because the herbicides have been *1480registered by the EPA under the Federal Insecticide, Fungicide, and Rodenticide Act, 7 U.S.C. §§ 136, et seq. (Supp.1983) (FI-FRA).
This argument is foreclosed by Oregon Environmental Council v. Kunzman, 714 F.2d 901 (9th Cir.1983), which involved the sufficiency of an EIS for the use of pesticides that were registered under FIFRA. We said there:
One agency cannot rely on another’s examination of environmental effects under NEPA.... “Thus, the mere fact that a program involves use of substances registered under FIFRA does not exempt the program from the requirements of NEPA.”
Id. at 905 (quoting Citizens Against Toxic Sprays, Inc. v. Bergland, 428 F.Supp. 908, 927 (D.Or.1977)). The BLM must assess independently the safety of the herbicides that it uses.
3. The Worst Case Analysis Regulation Applies to an Environmental Assessment (EA)
The BLM prepared a programmatic EIS in 1978 and annual EAs for individual applications. The original EIS was prepared before the effective date of the 1979 CEQ regulations. The district court held that the EIS was adequate. See 40 C.F.R. § 1506.12. It held, however, that subsequent EAs were deficient because they did not comply with § 1502.22.
The BLM argues that the regulation applies only “[w]hen an agency is evaluating significant adverse effects on the human environment in an environmental impact statement.” 40 C.F.R. § 1502.22 (emphasis added). It notes that the regulation is codified in part 1502 of the CEQ regulations, which deals with impact statements. Therefore, the BLM contends, the regulation does not apply to EAs.
The district court rejected this argument and reasoned that the EA was an integrated part of the overall environmental analysis. This view more closely reflects the purposes of NEPA and the CEQ regulations.
“We start with the premise that a federal agency has a continuing duty to gather and evaluate new information relevant to the environmental impact of its actions.” Warm Springs Dam, 621 F.2d at 1023. This continuing duty is especially relevant where the original EIS covers a series of actions continuing over a decade. See Council on Environmental Quality, Forty Most Asked Questions Concerning CEQ’s National Policy Act Regulation, 46 Fed.Reg. 18026,18036 (1981) (CEQ, Forty Questions). In general, an EIS concerning an ongoing action more than five years old should be carefully examined to determine whether a supplement is needed. Id.
The CEQ regulations encourage agencies to “tier” their environmental impact statements “to eliminate repetitive discussions of the same issues and to focus on the actual issues ripe for decision at each level of environmental review.” 40 C.F.R. § 1502.-20. The BLM uses annual EAs to supplement its programmatic EIS. The EAs must discuss new circumstances or information “relevant to environmental concerns and bearing on the proposed action or its impact.” 40 C.F.R. § 1502.9(c)(l)(ii).
An EA need not conform to all the requirements of an EIS. Foundation for North American Wild Sheep v. United States Department of Agriculture, 681 F.2d 1172, 1178 n. 29 (9th Cir.1982). The EA must support the reasonableness of the agency’s decision not to prepare a new or supplemental EIS. Id. Together, the EA and the programmatic EIS must “provide the information ‘necessary reasonably to enable the decision-maker to consider the environmental factors and to make a reasoned decision.’ ” Oregon Environmental Council v. Kunzman, 714 F.2d at 904. The label of the document is unimportant. We review the sufficiency of the environmental analysis as a whole.
The programmatic EIS and the EA were inadequate without a worst case analysis. Including a worst case analysis in the EA allows the BLM to consider the scientific uncertainty in the least cumbersome manner, without having to prepare a new *1481or supplemental EIS. See Warm Springs Dam, 621 F.2d at 1027 (Kennedy, J., concurring) (approving agency decision to respond to a new environmental concern without preparing a formal EIS or supplement).
We hold that the BLM must prepare a worst case analysis before it may resume spraying and the annual EA is an appropriate place to include it.
B. Denial of Attorney Fees
SOCATS cross-appeals from the court’s denial of attorney fees sought under the Equal Access to Justice Act. 28 U.S.C. § 2412(d)(1)(A) (Supp.1983).
We review that decision for an abuse of discretion. United States v. 101.80 Acres of Land, More or Less, in Idaho County, Idaho, 716 F.2d 714, 728 (9th Cir. 1983); Foster v. Tourtellotte, 704 F.2d 1109 (9th Cir.1983). This standard applies even where summary judgment was granted in the underlying action. See Foster, 704 F.2d at 1110. The court’s interpretation of the Act is subject to de novo review. Id. at 1111.
The Act authorizes attorney fees and expenses for prevailing parties in civil litigation involving a government agency unless the court finds that the agency’s position was “substantially justified or that special circumstances would make an award unjust.” 28 U.S.C. §'2412(d)(1)(A). The test is one of reasonableness. To defeat an award, the government must show that “its case had a reasonable basis both in law and in fact.” Hoang Ha v. Schweiker, 707 F.2d 1104, 1106 (9th Cir.1983).
The district court determined that SO-CATS was the prevailing party but held that the government’s position was substantially justified. 556 F.Supp. at 157. It gave two reasons: (1) the BLM had prevailed on three out of four legal issues, and (2) § 1502.22 was a convoluted regulation which was difficult to interpret.
The first reason was improper. SOCATS’ success on specific issues does not affect its status as a prevailing party. See Hensley v. Eckerhart, — U.S. —, 103 S.Ct. 1933, 76 L.Ed.2d 40 (1983). It also does not affect the substantial justification of the government’s position. The four legal theories advanced by SOCATS were alternative approaches to a single desired remedy, an injunction against further spraying. SOCATS gained that remedy. The time spent by SOCATS on unsuccessful legal theories may be considered only in determining the amount of an award. Id. at 1943. See 28 U.S.C. § 2412(d)(1)(C).
To support its finding that the government’s position was substantially justified, the trial court noted also the complexity of § 1502.22 and the absence of case law.
We do not agree that § 1502.22 is difficult to interpret. It is straightforward and means what it says. Sierra Club v. Sigler, 695 F.2d at 973. Much of the BLM’s confusion about it was self-induced.
Nevertheless, we cannot say that the court abused its discretion by denying fees. The analysis required by the regulation is exceptional. F. McChesney, CEQ’s “Worst Case Analysis” Rule for EISs: “Reasonable” Speculation or Crystal Ball Inquiry?, 13 Envtl.L.Rep. (Envtl.L.Inst.) 10069 (1983).
Further, the only reported case interpreting the worst case analysis regulation before this decision supported the government’s position. See Sierra Club v. Sigler, 532 F.Supp. 1222 (S.D.Tex.1982), rev’d, 695 F.2d 957 (5th Cir.1983). With our limited scope of review, we affirm the court’s decision to deny fees.
The situation on appeal is different. After the appellate decision in Sigler, the BLM had notice that its reading of § 1502.-22 was untenable. It could also have consulted the CEQ’s own interpretation of its regulations. See CEQ, Forty Questions, 46 Fed.Reg. at 18032. The Council stated explicitly that § 1502.22 applies to low probability/catastrophic impact situations, and to those where the probability of an impact is unknown. Id. The government’s position was no longer justified on appeal.
Within 30 days of this opinion, SOCATS will file with the clerk in quadruplicate a certificate upon which this court can base a fee award. See 28 U.S.C. § 2412(d)(1)(B). *1482It will give the information required by Kerr v. Screen Extras Guild, 526 F.2d 67 (9th Cir.1975), cert. denied, 425 U.S. 951, 96 S.Ct. 1726, 48 L.Ed.2d 195 (1976). The BLM may respond within 21 days . thereafter.
CONCLUSION
We affirm the decision that the scientific uncertainty regarding the safety of the BLM’s spraying program requires it to prepare a worst case analysis.
We affirm the denial of attorney fees below. The judgment is modified to allow reasonable fees on appeal.
AFFIRMED AS MODIFIED.