KOZINSKI, Circuit Judge:
Farmers and food processors in the Pacific Northwest brought this lawsuit to enjoin the Environmental Protection Agency from suspending registrations of the pesticide dinoseb (2-sec-butyl-4, 6-dinitrophe-nol). Plaintiffs use products containing dinoseb or its salts in the cultivation of green peas, snap beans, cucurbits and caneber-[1350]*1350ries.1 As counsel for the State of Oregon dramatically proclaimed at oral argument, this case “essentially is ... about whether we’ll be able to have raspberries next year.”
Facts
The Federal Insecticide, Fungicide, and Rodenticide Act (FIFRA), 7 U.S.C.A. §§ 136-136y (1980 & Supp.1987), establishes an elaborate framework for the regulation of pesticide use in the United States. No pesticide may be sold or distributed unless it is registered with the EPA. FI-FRA §§ 3(a), 12(a)(1)(A), 7 U.S.C.A. §§ 136a(a), 136j(a)(l)(A). In order to register a pesticide, an applicant, who may be a manufacturer or user of the product, must demonstrate with sufficient scientific evidence that, “when used in accordance with widespread and commonly recognized practiced the pesticide] will not generally cause unreasonable adverse effects on the environment.” FIFRA § 3(c)(5)(D), 7 U.S. C.A. § 136a(c)(5)(D). After a pesticide has been registered, the EPA Administrator must issue a notice of his intent to cancel its registration or change its classification “ ‘whenever there is a substantial question about the safety of a registered pesticide.’” Environmental Defense Fund, Inc. v. EPA, 510 F.2d 1292, 1296 n. 4 (D.C.Cir.1975) (quoting Environmental Defense Fund, Inc. v. Ruckelshaus, 439 F.2d 584, 594 (D.C.Cir.1971); see FIFRA § 6(b), 7 U.S.C.A. § 136d(b).
Because cancellation or reclassification proceedings may take one or two years to complete,2 FIFRA authorizes the Administrator to suspend a pesticide’s registration pending the outcome of the proceedings if he determines that suspension “is necessary to prevent an imminent hazard.” FIFRA § 6(c)(1), 7 U.S.C.A. § 136d(c)(l).3 Absent an emergency, the Administrator may not issue a suspension order until he has done two things: (1) notified registrants of the pesticide that he intends to cancel the registration and that he will issue a suspension order based upon “findings pertaining to the question of ‘imminent hazard,’ ” which he must include in the notice; and (2) given registrants an opportunity for an “expedited hearing” on “whether an imminent hazard exists.” FI-FRA § 6(c)(1), 7 U.S.C.A. § 136d(c)(l).
Dinoseb is a pesticide registered for use as an herbicide, insecticide, fungicide and desiccant, and has been used in the United States for nearly forty years. It is applied primarily as a contact herbicide to control broadleaf weeds and as a desiccant on ca-neberries to suppress growth that would obstruct harvesting. Decision and Emergency Order Suspending the Registrations of All Pesticide Products Containing Dino-seb, 51 Fed.Reg. 36634, 36635 (EPA Oct. 14, 1986) [hereinafter Emergency Order]; Intent to Cancel and Deny All Registrations for Pesticide Products Containing Dinoseb, 51 Fed.Reg. 36650, 36657-58 (EPA Oct. 14, 1986) [hereinafter Notice of Intent], In October 1986 there were nearly three hundred federal registrations for pesticides containing dinoseb or its salts.
In the spring of 1986, the EPA developed doubts about the safety of dinoseb. Preliminary studies showed that dinoseb may cause serious health risks to persons exposed to it, including sterility in men and birth defects in the unborn children of pregnant women. Emergency Order, 51 Fed.Reg. at 36636-38. In October 1986, the EPA began proceedings to cancel all [1351]*1351dinoseb registrations. See Notice of Intent, 51 Fed.Reg. at 36650. On October 7, the Administrator issued an emergency suspension order under section 6(c)(3) of FIFRA, 7 U.S.C.A. § 136d(c)(3), prohibiting the sale, distribution and use of dinoseb pending the completion of the cancellation proceedings. Emergency Order, 51 Fed. Reg. at 36634, 36648.4 The order, along with a notice of intent to cancel all registrations, was mailed to the registrants of dino-seb, and was subsequently published in the Federal Register. Id. at 36634. Four registrants requested a hearing on the emergency suspension order.5 The hearing convened on October 20 before an administrative law judge but, for reasons not apparent from the record, the registrants jointly withdrew their requests ten days later.
Plaintiffs, as nonregistrant users of dino-seb, are not permitted by FIFRA to initiate an expedited administrative hearing on the suspension order.6 However, under EPA regulations they were permitted to and did petition on behalf of growers in Washington, Oregon and Idaho for a so-called “sub-part D” reconsideration of the suspension. See 40 C.F.R. §§ 164.130-133 (1986).7 The EPA denied the petitions of plaintiff Northwest Food Processors Association and others, supported by applications from the three states, with respect to green peas, snap beans and lima beans on April 1,1987. See Denial of Hearing Concerning Application to Modify the Final Suspension Order for Pesticide Products Containing Dinoseb, 52 Fed.Reg. 11119, 11121 (EPA April 7, 1987). The EPA had not acted with regard to caneberries and cucurbits at the time the district court enjoined enforcement of the suspension order. See id. at 11120.
On April 3, 1987, with the growing season upon them, plaintiffs rushed into district court seeking relief from the EPA’s suspension order. The growers’ argument was quite straightforward: They simply could not grow their crops without dinoseb. Unlike farmers in other parts of the country, farmers in the Northwest have no substitutes for dinoseb. Climatic conditions and the prevalence of certain pests, black nightshade in particular, make dinoseb the only effective pesticide available on the market. For example, the farmers argued, without dinoseb there would be no caneber-ry crop in the Pacific Northwest, where 95 [1352]*1352percent of the nation’s commercial caneber-ry crop is grown. Potential crops losses would amount to $39.2 million this year. See Love v. Thomas, 668 F.Supp. 1443, 1449-50 (D.Or.1987) [hereinafter Dist.Ct. op.]; 2 Reporter’s Transcript (RT) at 257.
On April 15, 1987, after a two-day hearing, the district court asserted jurisdiction on the basis of section 6(c)(4) of FIFRA, 7 U.S.C.A. § 136d(c)(4). See Dist.Ct. op. at 1446, 1447. It then preliminarily enjoined enforcement of the suspension order pending completion of the EPA’s cancellation proceedings, and allowed use of dinoseb subject to twelve conditions, see pp. 1363— 1364 infra, patterning the injunction after the EPA’s modification of its suspension order as to certain other crops, see p. 1351 n. 7 supra. The court permitted limited sales of dinoseb to growers of certain crops; prohibited uncertified applicators from using the pesticide; barred “[wjomen of child-bearing age, i.e., under the age of 45,” from “any aspect of dinoseb application”; restricted the manner and extent of application of dinoseb to crops; and set standards for applicator clothing and exposure. See Love v. Thomas, No. 87-343-RE, at 2-4 (D.Or. April 15, 1987) [hereinafter Preliminary Injunction], quoted at pp. 1362-1363 infra.8 The court entered final judgment on May 1, 1987, and the EPA timely appealed.
The Parties and Their Contentions
The EPA’s appeal is supported by certain intervenors, including various labor unions that represent agricultural workers (“the unions”). Plaintiffs defend the district court’s judgment with the support of the State of Oregon, which intervened below.
The EPA argues that the district court lacked jurisdiction to review the suspension order. On the merits, the EPA argues that the court erred in invalidating the suspension order and enjoining enforcement. The unions support the EPA and in addition argue that the district court’s order denied farmworkers equal protection because it treated men and women disparately. Plaintiffs and the State of Oregon vigorously assert that the district court had jurisdiction and urge us to uphold its judgment.
Discussion
I.
The first question we must address is that of jurisdiction.9 That turns out to be a difficult issue, requiring close analysis of a rather convoluted statutory provision. As noted earlier, when the EPA commences proceedings to cancel the registration of a pesticide, it may, at the same time, issue a notice that it intends to suspend the registration pending the outcome of the cancellation proceedings. Such suspension normally does not become effective immediately. Rather, upon receipt of the notice the registrant has five days within which to request an expedited hearing on the issue of imminent hazard. FIFRA § 6(c)(2), 7 U.S.C.A. § 136d(c)(2). If a hearing is requested, the suspension is automatically stayed until the matter has been heard by an administrative law judge and the Administrator, having reviewed his findings, issues a final order. Id. But “[i]f no request for a hearing is submitted to the Agency within five days of the registrant’s receipt of the notification ..., the suspension order may be issued and shall take effect and shall not be reviewable by a courtf FIFRA § 6(c)(2), 7 U.S.C.A. § 136d(c)(2) (emphasis added).
[1353]*1353Because the expedited hearing may take a month or more,10 during which time the pesticide could continue to be used, FIFRA provides an exception to this procedure. Where the Administrator determines that an emergency exists, he may issue the suspension order “in advance of notification to the registrant.” FIFRA § 6(c)(3), 7 U.S.C.A. § 136d(c)(3). If the EPA issues such an emergency order, many of the same procedures apply as with ordinary orders,11 but the suspension goes into effect immediately and remains in effect pending the result of the expedited hearing.
The statute provides for review in the court of appeals of a suspension order issued following a hearing, notwithstanding the fact that the cancellation proceedings may be continuing before the agency. FI-FRA § 6(c)(4); 7 U.S.C.A. § 136d(c)(4). As to “[a]ny order of suspension entered prior to a hearing,” the statute provides for “immediate review in an action by the registrant or other interested person with the concurrence of the registrant in an appropriate district court.” Id.12 The scope of district court review is limited, however, to determining whether the order was “arbitrary, capricious or an abuse of discretion, or whether the order was issued in accordance with the procedures established by law.” FIFRA § 6(c)(4), 7 U.S.C.A. § 136d(c)(4). The district court action, moreover, “may be maintained simultaneously with any administrative review proceeding under [section 6(c)].” Id.
The EPA argues that judicial review is barred by section 6(c)(2), which provides, in what the EPA contends are the plainest possible terms, that if the registrants do not request a hearing, the suspension order becomes final and “shall not be reviewable by a court.” The district court rejected this contention, noting that this is not a case where the registrants failed to request a hearing: Four registrants in fact asked for an expedited hearing but eventually decided not to go forward with the process. Giving the statute a literal reading, the court held that judicial review is not barred. See Dist.Ct. op., 668 F.Supp. 1447.
Plaintiffs do not rely on the district court’s rationale. Instead, they point to the language of section 6(c)(4), which deals expressly with judicial review.13 In that section Congress provides for judicial review of two types of suspension orders, those entered after and those entered before a hearing. The latter category, plaintiffs argue, defines emergency suspension orders which, by their terms, are issued and become effective before a hearing can be held. Plaintiffs argue that if subsection (c)(2) bars judicial review, the second sentence of subsection (c)(4) becomes surplus-age.
A. We find the district court’s asserted ground for jurisdiction unpersuasive. While it is true that some of the [1354]*1354registrants sought a hearing, they subsequently withdrew their requests. It is quite obvious that the provision barring judicial review where no hearing is requested applies with equal force where a hearing is first requested and the request is then withdrawn. In either case, the registrants —those persons most directly affected by the suspension order — have waived their right to administrative remedies in the form of review by an administrative law judge. There is no less a waiver when the registrant first asks for a hearing and then changes his mind than when he asks for no hearing in the first place. See Nagel v. Thomas, 666 F.Supp. 1002, 1006-07 (W.D.Mich.1987) (dinoseb registrants' request for hearing and later withdrawal thereof is the functional equivalent of no request).
B. This leaves the question whether judicial review of an emergency suspension is barred when no registrant timely requests a hearing. As is often the case where Congress drafts complex statutes, the various provisions do not fit with surgical precision. While the statute’s meaning may be clear as to most of the situations it covers, cases may arise where two provisions in the same statute, or in different statutes, apparently conflict. It is then up to the courts to make sense of these apparent inconsistencies.
This is such a case. On the one hand, Congress provided for immediate judicial review of emergency suspension orders. FIFRA § 6(c)(4), 7 U.S.C.A. § 136d(c)(4). On the other, it provided in the most categorical terms that judicial review of suspension orders is precluded where the registrant waives its rights to an administrative hearing. FIFRA § 6(c)(2), 7 U.S.C.A. § 136d(c)(2). The conflict is exacerbated somewhat by the fact that judicial review of an emergency suspension order may be sought by any “interested person” with the concurrence of a registrant, while administrative review may be requested only by the registrant itself. If possible, we must give these apparently conflicting provisions a sensible reading that avoids redundancy or surplusage.
We find such a construction possible. FIFRA § 6(c)(4) authorizes the district courts to review EPA orders entered “prior to a hearing.” This phrase potentially embraces two categories of suspension orders: ordinary orders issued after registrants waive a hearing, and emergency orders issued before an opportunity therefor. Since ordinary orders entered after waiver of a hearing are made unreviewable under subsection (c)(2), subsection (c)(4) must authorize district courts to review emergency suspension orders, if it is to have any meaning at all.
The EPA does not seriously dispute this construction. It argues, however, that the district court may review emergency suspension orders under subsection (c)(4) only so long as they are in fact emergency suspension orders, that is, pending an opportunity for an agency hearing. If a hearing is held, the EPA’s argument continues, the final suspension order is reviewable in the courts of appeals.14 If, on the other hand, a hearing is waived (as was the case here), the final suspension order is entered by operation of law and, under subsection (c)(2), is not reviewable.
FIFRA, however, provides for the issuance of a final suspension order only in those cases where there is an administrative hearing. There is no provision for entering a separate final suspension order if no hearing is held after issuance of an emergency order, and none in fact was entered in this case.15 The EPA invites us [1355]*1355to create a procedural fiction to the effect that a final suspension order follows an emergency order whenever a hearing is waived. We see no reason for doing so. As we read the statute, when no hearing is held, the emergency suspension order simply stays in effect until the cancellation or classification proceedings are completed, or, as in this case, until stayed by the district court.
The EPA also argues that subsection (c)(3), which provides for the issuance of emergency suspension orders, by its terms incorporates subsection (c)(2), including the provision prohibiting judicial review where no registrant requests a hearing. However, the language of subsection (c)(2) indicates that this prohibition against judicial review applies only to those situations where the agency has given notice of its intent to issue a non-emergency suspension order: “If no request for a hearing is submitted to the Agency within five days of the registrant’s receipt of the notification provided for by paragraph (1), the suspension order may be issued and shall take effect and shall not be reviewable by a court.” FIFRA § 6(c)(2), 7 U.S.C.A. § 136d(c)(2) (emphasis added). In the emergency case, the suspension order has already issued and taken effect before any notice to the registrants and opportunity for a hearing. This first sentence of subsection (c)(2), unlike the other provisions of that subsection which concern hearing procedures, simply does not make sense in the emergency context and must therefore be limited to ordinary suspension orders issued after notice. Contra Nagel, 666 F.Supp. at 1007-08 (finding no jurisdiction to hear Michigan gladiolus growers’ challenge to dinoseb suspension order).
The EPA’s argument that the district courts’ jurisdiction expires with waiver of a hearing also does not fully comport with the language or drafting history of the provision. Subsection (c)(4) authorizes the district court to enter a stay of the order “pending the Administrator’s final decision with respect to cancellation or change in classification.” FIFRA § 6(c)(4), 7 U.S. C.A. § 136d(c)(4). Had Congress meant for the district court to act only until administrative review was waived or concluded, the statute would have so provided. In fact, the Senate Agriculture and Forestry Committee reported out a bill providing precisely that: “The effect of any order of the court will be only to stay the effectiveness of the suspension order, pending hearing before the Administrator.” S.Rep. No. 838, 92d Cong., 2d Sess., reprinted in 1972 U.S.Code Cong. & Admin.News 3993, 4068. That provision was amended to the current form by the conference committee. See Conf.Rep. No. 1540, reprinted in 1972 U.S.Code Cong. & Admin.News 4130, 4132-33. That Congress gave the district court authority to stay suspension orders long after any administrative hearing on suspension, up to the time when the EPA has ruled on the cancellation or reclassification, strongly suggests that the district court’s jurisdiction is not limited in the fashion suggested by the EPA.
We see one final anomaly in the EPA’s position, albeit a minor one. Subsection (c)(4) provides that the district court action “may be maintained simultaneously with any administrative review proceeding under this section” (emphasis added). As the EPA would have us interpret the statute, judicial review could only be maintained simultaneously with administrative review because if no administrative review is sought, under the EPA’s view, the right to judicial review would be cut off. We decline to change the statutory “may” to a “must.”
We recognize that this issue is far from clear-cut. The EPA’s position is not implausible and, quite frankly, neither side advances an entirely satisfactory construction of a statute that obviously was the product of some controversy and considera[1356]*1356ble compromise. The courts have long recognized, however, a presumption in favor of judicial review of administrative actions. See Block v. Community Nutrition Inst., 467 U.S. 340, 350-51, 104 S.Ct. 2450, 2456, 81 L.Ed.2d 270 (1984) (presumption only overcome when “the congressional intent to preclude judicial review is ‘fairly discernible in the statutory scheme’ ”); Ruff v. Hodel, 770 F.2d 839, 840 (9th Cir.1985) (“[t]he bar to judicial review ... requires a ‘persuasive reason to believe’ that Congress intended to preclude judicial review”) (quoting Abbott Laboratories v. Gardner, 387 U.S. 136, 140, 87 S.Ct. 1507, 1511, 18 L.Ed.2d 681 (1967)); Moapa Band of Paiute Indians v. Department of Interior, 747 F.2d 563, 565 (9th Cir.1984) (“[preclusion of judicial review is not lightly inferred, and usually will not be found absent a clear command of the statute”). Moreover, we construe prohibitions against judicial review narrowly. Cf. Wallace v. Christensen, 802 F.2d 1539 (9th Cir.1986) (en banc); Moapa Band, 747 F.2d at 565. In light of these considerations, we find plaintiffs’ interpretation the most plausible, and hold that the district court had jurisdiction to review the EPA’s suspension order.
II.
Subsection (c)(4) provides that an emergency suspension order shall be reviewable by the district court “solely to determine whether the order of suspension was arbitrary, capricious or an abuse of discretion, or whether the order was issued in accordance with procedures established by law.” We review the district court’s legal determinations de novo and its factual findings for clear error. United States v. McConney, 728 F.2d 1195, 1200-01 (9th Cir.) (en banc), cert. denied, 469 U.S. 824, 105 S.Ct. 101, 83 L.Ed.2d 46 (1984). On the question whether an injunction was properly issued, once the facts and law are established, we review for abuse of discretion. Colorado River Indian Tribes v. Town of Parker, 776 F.2d 846, 849 (9th Cir.1985).
A. As a preliminary matter, the EPA argues that the district court improperly considered evidence beyond the four corners of the administrative record. Generally, judicial review of agency action is limited to review of the record on which the administrative decision was based. Citizens to Preserve Overton Park v. Volpe, 401 U.S. 402, 420, 91 S.Ct. 814, 825, 28 L.Ed.2d 136 (1971). “[T]he focal point for judicial review should be the administrative record already in existence, not some new record made initially in the reviewing court.” Camp v. Pitts, 411 U.S. 138, 142, 93 S.Ct. 1241, 1244, 36 L.Ed.2d 106 (1973), quoted in Florida Power & Light Co. v. Lorion, 470 U.S. 729, 743, 105 S.Ct. 1598, 1606, 84 L.Ed.2d 643 (1985); Friends of the Earth v. Hintz, 800 F.2d 822, 828-29 (9th Cir.1986).
We have recognized, however, certain exceptions to this general rule. The court may find it necessary to review additional material to explain the basis of the agency’s action and the factors the agency considered. Friends of the Earth, 800 F.2d at 829; Asarco, Inc. v. EPA, 616 F.2d 1153, 1159-60 (9th Cir.1980). Moreover, the court may consider, particularly in highly technical areas, substantive evidence going to the merits of the agency’s action where such evidence is necessary as background to determine the sufficiency of the agency’s consideration. Asarco, 616 F.2d at 1160. Nonetheless, the court may not weigh the evidence to determine the correctness or wisdom of the agency’s decision. Id. at 1160-61.
The statutory scheme here strongly suggests that this is an appropriate case for consideration of matters outside of the administrative record. Subsection (c)(4) carefully apportions jurisdiction to review the EPA’s suspension order between the courts of appeals and the district courts: “final order[s] on the question of suspension following a hearing shall be reviewable in accordance with section 136n [in the courts of appeals].” By contrast, emergency suspension orders, issued without the benefit of an agency hearing, are to be reviewed by the district courts. A fair implication is that, where no administrative hearing has been held, the district court — a court of first instance equipped to consider [1357]*1357evidence — may look to matters outside the administrative record. Otherwise, it would have made little sense for Congress to give the district courts jurisdiction to review emergency orders; it might as well have left all agency review in the courts of appeals. See S.Rep. 970, 92d Cong., 2d Sess., reprinted in 1972 U.S.Code Cong. & Admin.News 4092, 4113 (bill vests jurisdiction over emergency orders in the district courts because, in the absence of an administrative hearing, there would be no factual record for an appellate court to review).
We therefore conclude that the district court was well within its authority in considering evidence outside the administrative record. The evidence it considered was precisely of the type that it could properly take into account in determining whether the agency acted arbitrarily or in a manner inconsistent with the statute. Moreover, the district judge carefully limited the scope of the inquiry, noting that, “[i]n considering the evidence presented outside the administrative record, I did not question the scientific merit of the EPA’s decision.... Instead, I was concerned that the EPA had not considered all relevant factors_” Dist.Ct. op., 668 F.Supp. at 1448.
First, the court heard testimony from two EPA officials: Dr. James Lamb, Special Assistant to Assistant Administrator for Pesticides and Toxic Substances; and Roger Holtorf, an economist with the Economics Analysis Branch of the Benefits and Use Division of the Office of Pesticides Programs. Holtorf, the principal EPA witness, was a team leader for the dinoseb cancellation proceedings and was involved in providing information on the economic consequences of dinoseb for the suspension hearing. 2 RT at 188, 190. The district court first carefully determined that Hol-torf was intimately familiar with the evaluation process carried out by the EPA. Id. at 192-93. Holtorf then testified about the information considered by the EPA on the effects of suspension and the extent of its investigation and evaluation of these effects on plaintiffs’ crops. In the district court’s judgment, such evidence was necessary in light of plaintiffs’ contention that the EPA’s fact-finding procedures were inadequate. This testimony was properly admitted as evidence necessary to explain the basis for the agency action. See Asarco, 616 F.2d at 1159-60.
Second, the district court considered evidence developed or brought to the attention of the EPA after the issuance of the suspension order, evidence pertaining to the relative efficacy of alternative pesticides in the Pacific Northwest and potential costs of the suspension. Much of this evidence was developed during plaintiffs’ subpart D proceeding or in the course of the EPA’s reconsideration as to other crops. See p. 1351 n. 7 supra. The court reviewed the evidence not in order to judge whether the EPA properly weighed the costs and benefits, but only to determine whether it considered them at all. So limited, the court’s consideration of evidence was entirely proper. See Asarco, 616 F.2d at 1160. We turn therefore to the merits of the district court’s decision.
B. The EPA may only issue a suspension order if it determines that continued use of the pesticide during cancellation proceedings would pose an "imminent hazard.” The statute defines that term as a situation where continued use “would be likely to result in unreasonable adverse effects on the environment or will involve unreasonable hazard to the survival of a species declared endangered or threatened. ...” FIFRA § 2(i) (amended 1973), 7 U.S.C.A. § 136(i). In turn, “unreasonable adverse effects on the environment” means “any unreasonable risk to man or the environment, taking into account the economic, social, and environmental costs and benefits of the use of any pesticide.” FIFRA § 2(bb), 7 U.S.C.A. § 136(bb) (emphasis added). Before suspending all dinoseb registrations, the statute thus requires the EPA to consider the benefits as well as the risks of its use, including the economic consequences of suspension.16
[1358]*1358The EPA admitted in its October 14, 1986, Notice of Intent that “[f]or some ... sites, such as green peas, snap beans, cane-berries, and hops, the extent of economic impacts is uncertain.” 51 Fed.Reg. at 36656. Based on this admission and other evidence, the district court found that “[bjecause of the way in which the EPA gathered information, the impact of a ban on dinoseb on the northwest was treated in a cursory, unacceptable fashion.” Dist.Ct. op., 668 F.Supp. at 1449. While concluding that dinoseb’s potential dangers were properly substantiated, id. at 1448, the district court held that “the EPA did not consider all relevant factors and, hence, made a clear error of judgment resulting in [a] finding that their entry of the emergency suspension order was arbitrary and capricious.” Id. at 1448-49.
After a careful review of the administrative record and the supplemental testimony before the district court, we must agree. The EPA’s evaluation of the relevant factors under FIFRA was incomplete and rushed and, under the circumstances of this case, simply not adequate to justify the emergency suspension of plaintiffs’ use of dinoseb.
First, the EPA gave itself insufficient time to comply with the statutory requirement that it balance risks and benefits. The EPA became especially concerned about the hazards of dinoseb in late June of 1986 when it received two studies from registrants indicating environmental risks. 1 RT at 144-45, 147. Around August 1, 1986, it ordered its Benefits and Use Division to prepare an evaluation of dinoseb, including production and usage information, a biological assessment and an economic impact assessment. 2 RT at 190, 193, 214. It established a September 1 deadline, which left what EPA economist Holtorf characterized as “a very short time frame.” Id. at 190-91. The division was given only two weeks within which to evaluate nearly one hundred registered “sites,” or crops, listed on dinoseb labels. 1 RT at 126; 2 RT at 197. As Holtorf testified, “roughly a couple of weeks is about all total that we had to evaluate this; and the staff available for evaluation, they were fully employed with other projects.” 2 RT at 207.
Given this tight schedule, the EPA personnel responsible for evaluating the benefits of dinoseb never quite got around to studying plaintiffs’ crops. The agency’s standard methodology called for the staff to rank the various sites either by total acreage treated or by total pounds of dino-seb applied. With the information at hand, the EPA staff identified roughly 15 sites. 2 RT at 197. Starting with the largest crops, soybeans, peanuts and potatoes, the staff was supposed to work its way down the list until no other usages could be identified. Id. at 197, 206. However, by the sixth and seventh sites, green peas and snap beans (each of which accounted for approximately 2 percent of total dinoseb usage in the United States, id. at 197), the EPA had to stop because it ran out of staff and resources. As economist Holtorf testified, “we flat ran out of resources when we got about halfway down the list which would be roughly peas and beans.... We have a relatively small staff in relation to our work load, Your Honor. The Federal has been somewhat on a slide in terms of resources over the last few years.” 2 RT at 206-07. The EPA thus never reached caneberries and cucurbits, which ranked below green peas and snap beans.
As a result of the incomplete study process, the district court found, the EPA used “scanty” data with respect to those crops, information “inconsistent with or contradicted by data readily available to, but not considered by, the EPA.” Dist.Ct. op. at 1449. For example, a principal document in the administrative record, relied upon by the decision-makers, was a study titled “Dinoseb: Summary of Biological and Economic Impacts of a Suspension/Cancellation.” (EPA Aug. 1986) (Plaintiffs’ Exh. 1) [1359]*1359[hereinafter August Summary]; see Dist.Ct. op., 668 F.Supp. at 1449. That document, dated August 1986 but attached to a cover memorandum dated September 2, 1986, was unfortunately both flawed and incomplete.17 It was only a less detailed version of a May 1986 document that, by its own description, was “limited in scope ... and should not be considered as a benefit analysis.” Dinoseb: A Summary of Biological and Economic Effects 4 (EPA May 1986) (Plaintiffs’ Exh. 62) [hereinafter May Summary]; see Dist.Ct. op. at 1449. Although Holtorf testified that “there were telephone calls made to individuals at various Land Grant universities,” he also stated that “[f]rom the time of the emergency suspension activity began until the document was prepared, there was no opportunity to gather additional outside information_” 1 RT at 124.18 The district court found that “[t]he September document was not updated.” Dist.Ct. op. at 1450.
The paucity of the information in the possession of the EPA when it summarily suspended use of dinoseb in the Northwest is disturbingly revealed by this document. The report states at the outset that, although the Benefits Division usually considers information from “all possible data sources” during cancellation proceedings,
due to the sensitive nature of suspension proceedings, outside resources were not used to estimate the current biological and economic impacts for either a suspension or a cancellation of dinoseb. While this impact analysis is the best that can be provided with current data limitations, it is possible that the impacts could change with new information from outside sources.
August Summary at 1-2.19 The summary’s findings are woefully incomplete. They note, for example, that “[w]ith other crops, such as caneberries and hops, there is uncertainty as to the extent of impacts because of substantial data gaps.” Id. at 10. They go on to state that “[t]he overall impacts from a loss of dinoseb on green peas and snap beans are not as well defined [as those for potatoes and peanuts].... Based on indications that the dinoseb alternatives may not be as effective, output losses are possible but could not be accurately assessed with available data.” Id. at 11-12 (emphasis added). The tables provided in the summary reveal starkly the EPA’s uncertainty as to the economic impact of suspension on the crops grown by plaintiffs.20
[1360]*1360In addition to limited data, the EPA conducted only a cursory evaluation of the availability of alternative pesticides and the consequent economic impact of suspension; it gave no particular attention to the Northwest at all.21 Even though consideration of the comparative performance of the challenged pesticide and its alternatives is a standard part of the EPA’s benefit analysis, 2 RT at 227, the May summary study stated that there had been no such analysis of dinoseb and its alternatives, and that as a result the authors could not estimate yield or revenue effects. May Summary at 7; 2 RT at 220.
Holtorf testified that the EPA never completed a “full efficacy evaluation” for peas and beans. 2 RT at 206. At the time of the suspension, the EPA was thus “very uncertain as to the levels of these alternatives” for dinoseb and “because of the uncertainties, [the EPA staff was] a little reluctant to estimate exact yield losses that could result.” Id. at 207; see also 1 RT at 131-32. The EPA also had no evidence of what effect dinoseb suspension would have on caneberries and made no effort to find any. 1 RT at 133. As a result, although it had studied potatoes and peanuts in greater depth, the EPA was not able to evaluate the economic effects of dinoseb suspension on growers, regionally or nationally, “for some of the lesser used sites,” including green peas and snap beans. 2 RT at 210.
In its own defense, the EPA argues that it ordered suspension on a nationwide basis, and was entitled to rely on nationwide findings as to alternatives and economic impact. For example, it found that only about a third of the nation’s green pea acreage and 20 percent of the snap bean acreage was treated with dinoseb. 2 RT at 198-99, 201. It therefore extrapolated the costs of suspension likely to be suffered by pea and bean farmers in such states as Wisconsin to all parts of the country and concluded there “would be a minor effect on a national basis.” 2 RT at 208 (green peas); id. at 209, 241 (snap beans). However, as the district court pointed out, there was no pressing need to enter an immediate nationwide ban on dinoseb: “The agency did have the power to suspend registration on a crop by crop and/or area by area basis.” Dist.Ct. op. at 1449. The suspension order was entered in October while the spraying season for the crops here at issue was not until the following March or April, some six months later. See Dinoseb: A Summary of Biological and Economic Impacts of an Emergency Suspension, table 1 (EPA Aug. 1986) (Plaintiffs’ Exh. 63). The EPA therefore had ample time to give the Northwest crops separate and careful attention.
There were good reasons for the EPA to have done so. At the time it entered the order suspending dinoseb nationally, the EPA was aware that the Pacific Northwest was subject to unusual conditions that made reliance on national figures tenuous, if not completely arbitrary. 2 RT at 209-10. For example, Northwest green pea farmers have to contend with black nightshade, a weed that produces a toxic berry about the same size as a pea. 1 RT at 43. The agency was aware that dinoseb was the only pesticide effective against nightshade. See, e.g., Notice of Intent, 51 Fed.Reg. at 36657 (“[m]ost of the alterna[1361]*1361tives provide poor control of black nightshade”); May Summary at 7-8 & table 2; 2 RT at 220-21, 223. Without dinoseb, farmers would be unable to keep nightshade out of the pea harvest, and processors would reject the crops infested with the toxic berries. See, e.g., Letter from W.H. Kosesan, Oregon Department of Agriculture, to Donald R. Stubbs, EPA, at 2 (Feb. 13, 1987) (Plaintiffs’ Exh. 36); 1 RT at 14-15, 26, 43-44, 81-82; see also May Summary, table 2 (“[mjarginal producers with serious broadleaf e.g. nightshade, infestations may be forced out in a few years”). Although the EPA knew that nightshade posed some problem, it did not follow up on this issue, nor did it specifically evaluate the impact of suspension on the Northwest region. 1 RT at 131-32, 154; 2 RT at 209, 221-22.22
By extrapolating data from other regions, the EPA also did not consider, as it should have, the particular economic circumstances of farmers in the Pacific Northwest. For example, 95 percent of the commercial raspberry crop is grown in that region, and dinoseb is the only herbicide known to be effective as a chemical pruning agent in suppressing the primocane stage of raspberry growth; this “cane burning” increases crop yield and permits mechanical harvesting. Kosesan Letter at 3-4; Schwisow Letter (Plaintiffs’ Exh. 43) at 4 (raspberry yield increases of 45 to 90 percent); 1 RT at 50-51, 116-17. The industry faces strong competition from Canadian producers across the Washington border, where dinoseb use is permitted. Without dinoseb, the entire American raspberry crop might prove unprofitable. See Schwi-sow Letter (Plaintiffs’ Exh. 43) at 4; 1 RT at 90-91.
In addition, specific crop effects, such as the loss of various caneberries and the contamination of green peas by black nightshade, could have tremendous secondary effects in the Northwest, none of which the EPA considered. As the district court found, “[c]rop losses of this magnitude will result in business failures not only for farmers, but for the food processing industry as well.” Dist.Ct. op. at 1451. These losses were not included in the $39.2 million estimate, which covered only the losses to growers. Id. Plaintiffs presented testimony that food processors and other buyers come to the Northwest for the pea and bean crops, and then purchase a fuller line of vegetables while there. 1 RT at 86-87. Should the caneberry, pea or bean crops fail or prove unmarketable because of the suspension, the effects might ripple throughout the Northwest agricultural sector: Buyers would establish sources of supply elsewhere; processors might find it unprofitable to maintain plants in the region without those key crops; producers of other, ancillary, crops may be unable to sell their harvests as well. See Dist.Ct. op. at 1451. The effect of crop losses growing from discontinued use of dinoseb thus could seriously harm the entire regional economy.23
It may well be that, despite these dire economic effects, the suspension would be justified by the health risks to applicators and other farmworkers.24 Without any in[1362]*1362vestigation of those economic effects, however, the EPA could not do even a rough and ready balancing.
C. The EPA claims that the health risks were equivalent for all crops, and that the need to act decisively in removing dangerous pesticides from use justifies its less than exhaustive inquiry at the emergency suspension stage. See Environmental Defense Fund, Inc. v. EPA, 465 F.2d 528, 537 (D.C.Cir.1972) (expedited hearing on imminent hazard “is to make a preliminary assessment of evidence, and probabilities, not an ultimate resolution of difficult issues”). Complete review of all uses, it claims, is the proper function of the full cancellation proceedings. See Environmental Defense Fund, Inc. v. EPA, 510 F.2d 1292, 1303 (D.C.Cir.1975) (“[a] more careful exploration of the availability of alternatives for minor uses would be contemplated for the final determination on cancellation vel non”). Because the Pacific Northwest sites represented a relatively small percentage of dinoseb use nationally, the EPA contends that it was not required to prepare a full evaluation of economic consequences.
With all due respect to the EPA and its overworked staff, such insensitivity to the local economic problems caused by its decision is unbecoming and inappropriate. Crop losses of over $39 million may look like small potatoes from Washington, D.C., but, as the district court found, such losses would cause very serious economic hardships to the people of the Northwest who would have to bear them. There may well be very minor uses that are of economically trivial importance and therefore need not be separately studied before a suspension order is entered. See, e.g., Environmental Defense Fund, 510 F.2d at 1303 (challenged pesticide “is used on less than 5 percent of the total citrus acreage, and the ALJ noted that much of that use was a kind of ‘just in case’ insurance, applied even in the absence of knowledge that the pest exists in the pertinent grove”). The effects in the Northwest documented here surely do not fall into that category. As the district court noted:
[SJeveral processing plants are the principal payroll in small Washington and Oregon communities_ If these processing plants cease to operate because of business losses, community stability will be profoundly and adversely affected. The State of Oregon anticipates and proved a potential and severe impact on its economy if these losses disrupt the agricultural industry to the extent predicted.
Dist.Ct. op., 668 F.Supp. at 1451. There is ample support in the record for the district court’s conclusion. See pp. 1360-1361 supra.
The EPA’s inattention to the Northwest’s special problems might nevertheless have been justified if there had been a genuine emergency that necessitated an immediate suspension of all dinoseb uses and foreclosed consideration of the problems faced by different regions of the country. However, the record does not document any particularly compelling circumstances requiring the EPA’s haste. In order to issue a suspension order without notice and prior opportunity for a hearing, the EPA must first find that there is an emergency that justifies immediate suspension. FIFRA § 6(c)(3), 7 U.S.C.A. § 136d(c)(3). The district court found that the principal basis for the “emergency” was the spraying of the potato crop in the early fall of 1986. See Dist.Ct. op. at 1449.25 By the time the EPA published the suspension order on October 14, 1986, however, the potato crop had already been [1363]*1363treated with dinoseb.26 Moreover, as noted, dinoseb use need not have been banned on a nationwide basis. See id. The EPA had another six months — until March or early April 1987 — to study the situation of the Pacific Northwest. In the absence of an ongoing health hazard to create an emergency, giving the Benefits Division insufficient time and resources to evaluate the various crops appears to have been entirely arbitrary.
The Administrator estimated that it would take four months to complete a suspension hearing; Holtorf testified that the economic impact evaluation and comparative performance review for the remaining crops would have been performed between September 1986 and March 1987. 2 RT at 285; see also 1 RT at 156 (Lamb testimony). The EPA, however, neither used that time to compile evidence justifying an emergency order, nor took advantage of the winter hiatus in dinoseb application to hold a suspension hearing. By suspending first and asking questions later — if at all— the EPA acted without due regard to the commands of FIFRA. We therefore agree with the district court that the emergency suspension order was arbitrary and capricious, an abuse of discretion, and was not issued in accordance with the provisions of FIFRA.
Our conclusion in this regard is based on the circumstances of this case. Our holding should not be interpreted as suggesting that where human health is imminently threatened, the EPA may not lawfully issue an emergency suspension before considering the economic circumstances of every affected region. In this case, however, at the time the EPA entered its order, the agency had already received substantial information regarding a potentially disastrous and unique economic impact of its contemplated action on one affected geographic area. It issued the emergency suspension order without evaluating that impact, and then failed to use available time after entry of the order and before advent of the spring use season to evaluate the risks and benefits in the Pacific Northwest. Under those circumstances, the EPA acted arbitrarily and capriciously.
III.
Having found the EPA’s suspension order to be arbitrary, capricious and an abuse of discretion, we turn to the district court’s grant of relief to plaintiffs. As noted above, the district court did not merely stay the suspension order. It handed down a five-page preliminary injunction permitting use of dinoseb on the crops at issue in Oregon, Washington and Idaho. The court specified the conditions for such use;
1. That dinoseb be sold (1) only to growers of green peas, snap beans, cucurbits, and caneberries in Oregon, Washington, and Idaho (2) in a quantity not to exceed that required to treat his acreage for each of those crops at the maximum application rates stated below. Dealers must maintain records of sales to growers.
2. Only certified applicators may use dinoseb; other persons, even if they are operating under the direct supervision of a certified applicator, may not use dino-seb.
3. Women of child-bearing age, i.e., under the age of 45, may not be involved in mixing, loading, or any aspect of dinoseb application.
4. A warning label must appear on the product specifying that (1) women of childbearing age may not use the product, (2) all reasonable efforts should be made to minimize indirect exposures to women of child-bearing age, (3) the product also poses risks to male reproduction, (4) is acutely toxic, and (5) the product may be applied only by certified applicators.
5. Aerial spraying is prohibited.
6. Mixing/loading of dinoseb is prohibited except from closed systems.
[1364]*13647. Ground application of dinoseb is prohibited except by the ‘barrel sucker’/ground boom/tractor system.
8. Mixer/loaders and applicators must wear chemically resistant disposable coveralls (Tyvek suits) and chemically resistant gloves when mixing or loading dino-seb. Applicators or other personnel may remove such protective clothing immediately before entering the tractor cab to avoid cab contamination, but must carry an unused set of gloves and coveralls in their cabs, to be used in the event of spraying equipment malfunction and repair during application.
9. That maximum application rates for dinoseb be set as follows: 3 pounds ai/A for green peas, 4.5 pounds ai/a for snap beans, 4.5 pounds ai/a for cucurbits, and 2.5 pounds for caneberries.
10. That dinoseb be applied to a maximum of 80 acres per day per applicator.
Preliminary Injunction at 2-4. As to the spraying of green and dry peas, chickpeas, lentils, snap beans and cucurbits, the court added two “additional conditions”:
11. Ground application is prohibited when wind conditions exceed ten miles per hour.
12. Tractor cabs must be closed and equipped with positive pressure ventilation systems.
Id. at 4. The court further specified that caneberries could only be sprayed during conditions that would “prevent any drift of the herbicide,” and that “it may be applied by open tractors equipped as stated in paragraph 7 above, if operators wear chemically resistant disposable overalls of the type specified above.” Id. Last, the court ordered Oregon to enforce these conditions, and applied the order to the two other states if they agreed to enforce the conditions. Id. at 4-5.
Such a detailed and carefully crafted order — balancing the rights and interests of the affected parties — would normally be entirely appropriate for a court sitting in equity.27 However, the court is not free to use its full powers of equity because FI-FRA strictly limits the remedy the district court may provide if it deems the EPA’s emergency suspension order improper: “The effect of any order of the court will be only to stay the effectiveness of the suspension order, pending the Administrator’s final decision with respect to cancellation or change in classification.” FIFRA § 6(c)(4), 7 U.S.C.A. § 136d(c)(4) (emphasis added). Under the plain language of the statute, the district court could either let the suspension order stand or stay its effectiveness. The court was not empowered to craft an alternative suspension order. That task was reserved for the EPA.
Because the court exercised its discretion under the mistaken impression that it could tailor an injunction so as to limit the pesticide’s use, we cannot be sure that it would have been willing to allow the unrestricted use of dinoseb. We therefore vacate the district court’s injunction and remand for a redetermination of whether to stay the EPA’s suspension order.
The EPA is not without recourse. If the district court does stay the suspension, and even if does not, the EPA may initiate an expedited suspension hearing in which plaintiffs and other interested parties may participate. Or, the EPA may itself modify its suspension order, much as it did with respect to dry peas and lentils, in response to the pending subpart D petition. The district court would, of course, take any such modification into account in deciding whether to issue the stay, or, once issued, in deciding whether to keep the stay in effect. See Fed.R.Civ.P. 60(b)(5)-(6); Nicacio v. United States Immigration & Naturalization Serv., 797 F.2d 700, 706 (9th Cir.1985) (“a court which issues an injunction retains jurisdiction to modify the terms of the injunction if a change in circumstances so requires”); 11 C. Wright & A. Miller, Federal Practice & Procedure § 2961, at 558-600 (1973). Thus, while the district court may, under FIFRA, suspend the EPA’s order until completion of eancel-[1365]*1365lation proceedings, this does not necessarily mean that dinoseb will be freely usable in the Northwest during that entire period. We leave these matters for the court and parties to sort out on remand.
Conclusion
We affirm the district court's ruling that the EPA order was arbitrary and capricious. We vacate its order staying the suspension, and remand for further proceedings consistent with this opinion.
AFFIRMED IN PART, REVERSED IN PART AND REMANDED.