The Secretary of Commerce (the Secretary) appeals a district court order that invalidated the Department of Commerce's commercial catch quota for summer flounder for 1994 and that imposed another quota in its place. Because the district court misapplied the statutory requirement that the Department set the quota in accord with the "best scientific information available," and consequently erred in invalidating the agency-set quota, we reverse.

* This case involves a challenge by a coalition of commercial fishers (the Coalition) to the 1994 commercial catch quota for summer flounder promulgated by the Department of Commerce (the Department). The Department sets the quota every year through a complex process mandated by the Magnuson Fishery Conservation and Management Act, 16 U.S.C. § 1801 et seq., and by federal regulations, 50 C.F.R. Part 625. Within the Department of Commerce are eight regional fishery management councils, established by the Act, whose job it is to develop plans for the conservation and management of the fish in each council's respective section of the American coastal waters. Among these bodies is the Mid-Atlantic Fishery Management Council ("the Council"), which has among its responsibilities the management of the summer flounder fishery, including the annual setting of the commercial catch quota.

Before it could set annual quotas, the Council was required to develop a Fishery Management Plan for Summer Flounder, which it did in 1992. Amendments to the plan were promulgated in 1993 and 1994 as well. As part of this Plan, the Council had to set for each of several years a "target fishing mortality rate." The fishing mortality rate is a statistic called F that expresses the depletion of the stock of fish attributable to fishers, whether by capture or by discard of fatally wounded fish or otherwise, in a given year. For 1994, the Fishery Management Plan set a target F of 0.53 and thus required that the commercial catch quota for 1994 be set at a level that would ensure that the actual F would not exceed 0.53. See 50 C.F.R. § 625.20(b)/(c).

The process of setting the quota began, for present purposes, with the holding of a Stock Assessment Workshop, a gathering of marine scientists whose mission it was to estimate the size, age structure, and any other relevant characteristics of current populations of various species of fish off the Atlantic coast. The numbers produced by the Workshop for summer flounder were then presented to the Summer Flounder Monitoring Committee (the Monitoring Committee), a sub-unit of the Council, on September 1, 1993 with recommendations from Council staff. The Monitoring Committee is a group of scientists constituted by the Department who are required to recommend, among other things, a commercial catch quota. The quota for 1994 was to be based on the scientific information from the Workshop and was to be designed to ensure that the "fishing mortality rate," F, did not exceed the 0.53 level previously announced as the 1994 target by the Fishery Management Plan. The Monitoring Committee recommended a quota of 16,005,560 pounds. That recommendation then went to the Demersal Species Committee of the Council and to the Atlantic States Marine Fisheries Commission, an interstate organization with which the Council cooperates. Those two bodies, meeting jointly on September 23, approved the Monitoring Committee's recommendation and passed it on to the Council which, in turn, recommended it to the Regional Director for approval. After the requisite notice and comment, during which no objections pertinent to this appeal seem to have been filed, the quota became final in early 1994.

At the heart of the case is the way the Monitoring Committee used the Workshop's data regarding the "recruitment" of summer flounder--that is, the number of new flounder expected to appear in the population in 1994. At the Committee's annual quota-setting meeting, staff member Wendy Gabriel had presented the Committee with the geometric mean of the previous five years' estimated recruitments and the values one standard deviation above and below that mean. Then, on recommendation of Council staff, and specifically following the lead of Gabriel's presentation of the recruitment statistics to the Committee, the Committee chose to recommend that the quota be set on the basis of a conservative estimate of recruitment. In particular, the staff had recommended, and the Committee then chose to use, an estimate of recruitment equal to the figure one standard deviation below the geometric mean rather than a recruitment estimate equal to that geometric mean. As a direct consequence, the Committee recommended that the previous year's quota be increased for 1994 only to about 16 million pounds (a 28% increase) whereas incorporation of the geometric-mean estimate would have boosted the quota to about 19 million pounds.

The lower quota was ratified at every step of review and published in the Federal Register accompanied by four justifications that closely reflected those articulated in the original staff recommendation. As articulated in the Federal Register, the four reasons for using the lower estimate were the following: First, the summer flounder population was composed mainly of fish aged 2 and under; so an overestimate in recruitment would have great power to cause an overestimate in overall stock size and thus "would result in quotas that would exceed the target fishery mortality rate (F level)." Second, "the probability of achieving the target F level is higher at the lower harvest level" with staff estimating an 80% probability that the proposed quota would keep actual F under target F. Third, three risky assumptions--that the previous year's quota would prove to have been adhered to, that all landings get reported, and that discard rates would not increase--underlay the estimate of the stock size and suggested that the stock-size curve might be overly optimistic. Fourth, since the target F was scheduled to decrease dramatically for 1996, it was better to err on the safe side now so as to minimize the chances of having to reduce the 1996 quota even more than was already anticipated. J.A. 350. In short, the Council believed that the uncertainty in the recruitment estimates was so great and the long-term flounder population so fragile a resource for the fishers, especially in light of a coming reduction in the target fishing mortality rate for 1996, that a low estimate of recruitment was the prudent estimate.

In challenging the quota, the Coalition argued that use of the lower estimate rather than the geometric mean in calculating the quota constituted a failure to use the best scientific information available as required by 16 U.S.C. § 1851(a)(2). After holding a three-day hearing to have the administrative record explained, the district court agreed and held that the quota, therefore, represented an arbitrary and capricious decision on the part of the Department. The court concluded that only the geometric-mean estimate could constitute the best scientific information available in this case. On this basis, the court held that "the 1994 commercial catch quota is invalidated to the extent that it deviates downward from the figure reached using the best scientific information available, which was 19.05 million pounds for 1994," Fishermen's Dock Cooperative v. Brown, 867 F. Supp. 385, 386 (E.D.Va.1994), and ordered that the quota be reset at that figure. The key parts of the district court's opinion read as follows:

The Court finds that the use of figures one standard deviation below the mean was arbitrary and capricious. The use of a figure one standard deviation below the mean was chosen not because it was the best scientific information available, but solely because it increased the percentages of reaching not a balanced result but a result which protected the summer flounder stock to the detriment of the fishermen.

....

... The Council chose to implement a quota one standard deviation below the geometric mean, resulting in a quota of 16.005 million pounds rather than a quota of 19.05 million pounds, a sixteen percent difference. Defendant's designees explained that using the lower figure increased the probability of reaching 0.53, the target fishing mortality for 1994, from 59% at the geometric mean to 81% with the lower figure.

This Court finds that the Council's decision to implement a quota one standard deviation below the geometric mean failed to utilize the best scientific information available, and therefore was arbitrary and capricious.

....

... The commercial catch quota ... should be reset using the figure derived from the best scientific information available--19.05 million pounds, the geometric mean, replacing the quota set at one standard deviation below the mean, or 16.005 million pounds.

... Accordingly, this Court ... ORDERS that the 1994 summer flounder commercial catch quota be invalidated to the extent that it is less than 19.05 million pounds.

Fishermen's Dock, 867 F. Supp. at 396-97.

This appeal by the Secretary followed. A number of environmental organizations and recreational fishers' organizations have filed an amicus brief in support of the Secretary's position.

In view of the nature of the district court's proceedings in this case, it is important that we define at the outset the standard under which we review its decision. We start with the standard under which the district court reviewed the agency action. It is prescribed in the Administrative Procedure Act (APA) and incorporated by the Magnuson Act at 16 U.S.C. § 1855(b)(1)(B):

The reviewing court shall--

....

(2) hold unlawful and set aside agency action, findings, and conclusions found to be--

(A) arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law;

(B) contrary to constitutional right, power, privilege, or immunity;

(C) in excess of statutory jurisdiction, authority, or limitations, or short of statutory right;

(D) without observance of procedure required by law;....

5 U.S.C. § 706(2)(A)-(D).

Our review of district court decisions made in review of agency action is generally de novo. Thus, we generally review the agency's action from the same position as that of the district court and seek to determine, as did the district court, whether on the administrative record the agency action was arbitrary and capricious or otherwise illegal under the statutory standards quoted above. Virginia Agricultural Growers Ass'n v. Donovan, 774 F.2d 89, 93 (4th Cir.1985) ("The trial court was obligated to look to the administrative record and its view of the record is not entitled to deference."); Natural Resources Defense Council v. United States Environmental Protection Agency, 16 F.3d 1395, 1400-01 (4th Cir.1993) (court of appeals review looks to the agency 's action "to determine whether the record reveals that a rational basis exists for its decision").

The Coalition points out, however, that where a district court has legitimately expanded the administrative record by holding a supplementary hearing, its decision might then be entitled to some unspecified degree of deference from our court in review. It relies on then-Judge Breyer's exploration of the intricacies of judicial review under the APA in Sierra Club v. Marsh, 769 F.2d 868, 871-72 (1st Cir.1985):

We should be more willing, or be less willing, to differ with a district court about the 'reasonableness' or 'arbitrariness' of an agency decision, depending upon the particular features of the particular case that seem to make a more independent, or a less independent, appellate court scrutiny of the administrative record appropriate. Where, for example, the district court's judgment turns on matters of fact that it has determined, or upon evidence presented by witnesses in court, or even upon lengthy district court proceedings in which knowledgeable counsel explain the agency's decision-making process in detail, we will show appropriate hesitation to overturn that judgment. [Citations omitted.] But, where the district court simply reviews a set of agency documents and, applying the same legal standard as we apply here, reaches a particular legal conclusion about the 'reasonableness' of an agency's action, we have greater legal freedom to differ with the district court's ultimate characterization of agency behavior.

Countering the suggestion that any deference is owed here, the Secretary challenges the propriety of the district court's expansion of the record, contending that its three-day hearing and ensuing expansion of the record was not warranted.

Whatever the propriety of conducting such a hearing, we do not think it need have any effect on our review process. We have carefully reviewed the record of that hearing and are satisfied that none of the testimony received actually had the effect of substantively expanding the factual record before the agency. The testimony received essentially served only to educate the district court in the complexities of geometric means, standard deviations and probabilities--an education equally available to us. Because we are therefore satisfied that the district court reached its decision wholly on the basis of the administrative record itself and its own ad hoc self-education in statistics, we owe its decision no special deference deriving from the extended hearing it conducted. Consequently, we will proceed to conduct our own independent review of the administrative record as being effectively the unexpanded record reviewed by the district court.III

The district court held that, at least in the circumstances of this case, the statutory requirement that the agency use the "best scientific information available" translated to a rule of law that the agency use the geometric mean as its estimate of recruitment in setting the commercial catch quota for 1994. The Secretary argues on appeal that the agency's action in this case was in no way arbitrary, capricious, or otherwise illegal and that, even if it was, the district court overreached its authority in imposing a new quota rather than remanding to the agency.

We agree with the essentials of the Secretary's primary argument. In explaining why, it is important at the outset to point out the limited scope of the Coalition's challenge to the agency action. It does not challenge the Fishery Management Plan or its target fishing mortality rate for 1994. It does not challenge the procedures used in producing the quota or the scientific validity of the data underlying the quota. It challenges only the Department's choice to rely on one estimate of recruitment rather than another, and contends that the district court was correct to hold that there was only one estimate consistent with use of the best scientific information available as required by 16 U.S.C. § 1851(a)(2). We disagree on that basic point. The district court's holding that the available scientific information on recruitment dictated one and only one possible quota reveals a critical misunderstanding of the nature of the information available.

The Act mandates that the agency conform to several "national standards," including the following:

Any fishery management plan prepared, and any regulation promulgated to implement any such plan, pursuant to this subchapter shall be consistent with the following national standards for fishery conservation and management:

(1) Conservation and management measures shall prevent overfishing while achieving, on a continuing basis, the optimum yield from each fishery for the United States fishing industry.

(2) Conservation and management measures shall be based upon the best scientific information available.

....

(6) Conservation and management measures shall take into account and allow for variations among, and contingencies in, fisheries, fishery resources, and catches.

16 U.S.C. § 1851(a).

No one contends that the Council's Fishery Management Plan is inconsistent with the above statutory standards. That Plan is implemented by 50 C.F.R. Part 625, which provides, in pertinent part, as follows:

§ 625.20 Catch Quotas and Other Restrictions

(a) Annual Review. The Summer Flounder Monitoring Committee will review [certain data including "recent estimates of recruitment"] ... to determine the allowable levels of fishing and other restrictions necessary to result in a fishing mortality rate of 0.53 in 1993 through 1995, and a fishing mortality rate of 0.23 in 1996 and thereafter....

(b) Recommended Measures. Based on this review, the Summer Flounder Monitoring Committee will recommend ... the following measures to assure that the fishing mortality rate specified in paragraph (a) of this section is not exceeded:

(1) The commercial quota will be set from a range of 0 to the maximum allowed to achieve the fishing mortality rate specified in paragraph (a) of this section; ....

Next, in subsection (c), the regulations call for the Demersal Species Committee, the Council, and the Regional Director (both before and after public comment), each in turn, to consider what quota is "necessary to assure that the applicable fishing mortality rate specified in paragraph (a) of this section is not exceeded." (Emphasis added.) This quoted language appears four separate times in that subsection.

The sum of these provisions is that the Monitoring Committee is required to seek a fairly high level of confidence that the quota it recommends will not result in an F greater than 0.53, even as it must be equally concerned to provide the fishing industry with an "optimum yield" both in the current year and over the long term. Although the regulatory language repeatedly calls for the Council to "assure" that the target F is "not exceeded," it cannot be taken to require 100% assurance. If taken literally, that language could mean that the Monitoring Committee would have to recommend something close to closing of the fishery to "assure" an acceptable F. But, of course, the statute does not contemplate regulatory overkill. It anticipates an "optimum yield" for the fishing industry that is consistent with reasonable assurance that the actual F will be less than or equal to the target F.

Also, while the language assumes that such a quota can be established on the basis of the "best scientific information available"--and so requires that the quota be set on that basis--it appears to say nothing about how that term is to be defined. It only mandates the constitution of committees of scientists to provide and evaluate such information.

In this case, the Monitoring Committee had before it, among other things, three estimates or predictions of recruitment for 1994 that were derived from research that, at least for purposes of this appeal, everyone agrees constituted the best scientific information available. One of the estimates was the geometric mean of the estimated recruitments for the years 1988 through 1992; that is, the midpoint of the probability curve of all possible recruitment estimates for 1994 as derived from the estimated actual recruitments of the previous five years. The other two estimates for 1994 were produced by traveling one standard deviation above and below the geometric mean along the curve of 1994 recruitment estimates. As is always true of the single-standard-deviation points along a normal curve, these high and low estimates represented the upper and lower limits of a range of estimates within which there was a 68% probability that the true recruitment would eventually lie. The geometric mean was just the middle value in that range, and the single-standard-deviation values were simply standard statistical markers for indicating the degree of uncertainty in the data.

Thus, to illustrate, if the geometric mean were about 30 million fish with a standard deviation of about 10 million, all the Monitoring Committee would know from the three estimates presented to it was that the best science available predicted a 68% chance that the actual recruitment for 1994 would be somewhere between 20 million and 40 million fish. No single point estimate on the curve had any assigned probability of being the ultimately correct one. But any range of estimates on the curve had a particular and easily derived probability of containing the ultimately correct number.

One of the properties of the curve, then, was that it could provide the scientists and the administrators with the available research's best conclusions as to the ranges within which the actual 1994 recruitment was likely to lie. But another even more important property was that that curve apparently implied another curve; and this second curve provided, for any chosen point estimate of recruitment, the probability that the quota derived from that point estimate would ultimately produce an actual F for 1994 of 0.53 or less. Crucially for this case, that latter curve in fact indicated that the 19-million-pound quota, derived from the estimate at the geometric mean, carried a 59% chance of eventually producing a 1994 F of 0.53 or less, whereas the 16-million-pound quota, derived from the estimate one standard deviation below the mean, carried an 81% chance of eventually producing a 1994 F of 0.53 or less. And no one disputes the validity of these probabilities; no one suggests that they rested on anything but the best scientific information available.

On the basis of this best available scientific information, then, the Monitoring Committee had to estimate what the true recruitment would be. If it estimated too high and set the quota accordingly high, then the smaller actual recruitment would be decimated by fishers operating under the high quota, with the result that F would exceed 0.53. Alternatively, the Monitoring Committee could estimate that actual recruitment would be low and thereby increase its chances of not exceeding target F. Since the Fishery Management Plan called for assurance that target F would not be exceeded but not such absolute assurance as would result in grievously undershooting target F and denying the fishing industry its "optimum yield," the Agency's decisionmakers necessarily had some discretion to decide what precise degree of assurance it would seek within the uncertainty of the data.

In exercising that discretion, these decision-makers consistently offered the four reasons recited above for recommending use of the lower of the two recruitment estimates and thus the lower of the two quotas that they had considered in light of the data presented: first, the truncated age structure of the summer flounder population, which magnified the risk to target F involved in any overestimate of the size of the recruitment class; second, the general proposition that a lower recruitment estimate provided a higher probability of assuring that actual F would turn out to be less than or equal to target F; third, the recognition that certain assumptions underlying the estimate of the flounder stock size might be overly optimistic; and, fourth, the belief that, since target F was to decrease significantly in 1996, it was better to deal with the current uncertainty in recruitment estimates by erring on the conservative side rather than risking an exacerbation of the painful quota decrease that had to come soon in any case.1 J.A. 350.

These reasons do not justify the precise choice of an 81% probability of success or the choice to plug into the formula the specific recruitment estimate located at the one-standard-deviation-below-the-geometric-mean mark. See Brief of Appellee-Coalition at 36-39. Nor do they consistently explain why the choice of a conservative recruitment estimate was the best way to modify the quota rather than using a higher recruitment estimate but then simply lowering the quota itself to take account of all the uncertainties in the data. But they do reflect the Monitoring Committee's understanding that the recruitment estimate was the main source of uncertainty along the way to making a stock-size estimate and then setting a quota; and they do justify setting a quota that, taking that uncertainty into account, offered a high probability--if not 100% assurance--of achieving the regulatory goal of not exceeding target F. Moreover, at the Monitoring Committee's meeting and later in the process, the point was made that, "happenstantial" as the lower figure might be, it had the benefit of both providing a high probability of staying under target F and still allowing a substantial (28%) increase in the quota from the year before. It thus showed commercial fishers that the Department's measures were not simply relentless attacks on the industry but beneficial over the long term for all concerned as the Act contemplated. J.A. 189, 193, 325-27. The option of setting the quota so as to provide a 90% or 95% probability of success was also considered but rejected as too heavy a short-term burden on the fishing industry. J.A. 325, 193.

In this situation, to assert, as the district court did, that the "best scientific information" required use of the geometric-mean estimate to dictate a quota that had only a 59% chance of not causing F to be exceeded, rather than use of an estimate that would provide an 81% chance of not causing F to be exceeded, is to misconstrue what the best scientific information really shows in the context of the statute and regulations. A quota "based on" that information and designed to "assure" that the target F was not exceeded while still providing the fishing industry with an "optimum yield" could not properly be determined by a court in judicial review to be, as a matter of law, only one that happened to provide a 59% chance of not exceeding F. The district court was correct that the specific choice of an 81% probability largely because it happened to correspond to one of the three recruitment estimates presented to the committee--when those three estimates were presented merely as indicators of the degree of uncertainty in the data--was, in a sense, arbitrary. But, within the terms of the statute and regulations, so would have been the choice of a 59% probability. See Brief of Amici Curiae at 27-28.

When the regulations say that the Committee's recommendation should "assure" that target F is not exceeded, they do not say what probability of success (as derived from the best scientific information available) constitutes "assurance" of success. As long as everyone agrees, as everyone does, that the regulations do not require 100% assurance, the choice of how much assurance to indulge in must be a policy choice left to the reasonable exercise of the discretion of the statutorily authorized decision-makers. And that choice inevitably contains a degree of arbitrariness.

In the event, those decisionmakers seem to have allowed themselves to gravitate to a specific number--within the general range suggested by their reasoning--largely because that number happened to have been on the table as a standard deviation. If allowing themselves to gravitate in that way constituted arbitrariness in the selection of the final number within the acceptable range of assurance, then the Monitoring Committee indulged only in the kind of arbitrariness that is inherent in the exercise of discretion amid uncertainty and not in the kind of arbitrariness that the statute condemns when it exists in tandem with capriciousness.

To dispute this conclusion, the Coalition cites a handful of cases for the proposition that the APA requires a tighter connection than the agency has articulated here between the specific regulation adopted and the regulation's justifications. But evaluation of agency reasoning is inevitably an ad hoc enterprise, and the Coalition's cited cases simply do not present factual situations similar enough to our case to suggest that the agency failed to justify itself adequately. In fact, these cases do little but reaffirm the requirement that the agency have engaged in reasoned decision-making within the specific regulatory context.

Thus, in Gas Appliance Mfrs. Ass'n v. Department of Energy, 998 F.2d 1041, 1047 (D.C.Cir.1993), the court found a Department of Energy cost-benefit analysis inadequate because the Department had offered no information at all on how private companies might be able to conform to the assumptions of the analysis, much less what the costs of conforming might be or how those costs might compare to the benefits to be had. In Parravano v. Babbitt, 837 F. Supp. 1034, 1046 (N.D.Cal.1993), similarly, the court found no record basis at all for the Secretary of Commerce's increase in a "salmon escapement floor." In Chemical Mfrs. Ass'n v. Environmental Protection Agency, 28 F.3d 1259, 1265 (D.C.Cir.1994), the court rejected the EPA's use of a model when the agency failed to respond meaningfully to a challenger's "specific detailed evidence of a poor fit between the agency's model and that party's reality"; but, of course, the Coalition in this appeal does not even challenge the Council's model. Finally, in Motor Vehicle Mfrs. Ass'n of the United States v. State Farm Mut. Auto. Ins. Co., 463 U.S. 29, 43, 103 S. Ct. 2856, 2866-67, 77 L. Ed. 2d 443 (1983), the Supreme Court said that usually "an agency rule would be arbitrary and capricious if the agency has relied on factors which Congress has not intended it to consider, entirely failed to consider an important aspect of the problem, offered an explanation for its decision that runs counter to the evidence before the agency, or is so implausible that it could not be ascribed to a difference in view or the product of agency expertise." The Court then went on to invalidate the agency's action on the basis of its utter failure to consider obvious alternative actions. State Farm, 463 U.S. at 46-51, 54-57, 103 S. Ct. at 2868-71, 2872-74. And it reminded those who would challenge agency action that often "the available data do not settle a regulatory issue, and the agency must then exercise its judgment in moving from the facts and probabilities on the record to a policy conclusion." State Farm, 463 U.S. at 52, 103 S.Ct. at 2871. Far from supporting the Coalition's position, State Farm offers a description of legitimate exercise of agency discretion that neatly suggests the Council's actions in this case. In sum, these cases actually require an agency to do what the Council did here--that is, reason its way to a decision without pretending that that decision reflected some degree of rational perfection beyond what the inherent uncertainties of the available data permitted. See Leather Industries of America v. Environmental Protection Agency, 40 F.3d 392, 409 (D.C.Cir.1994) ("Where the agency's line-drawing does not appear irrational and the [challenger] has not shown that the consequences of the line-drawing are in any respect dire ... we will leave that line-drawing to the agency's discretion."); see also 1 Kenneth C. Davis and Richard J. Pierce, Jr., Administrative Law Treatise, § 7.5, at 326-332, 3d ed. (1994) (citing cases and arguing the futility of courts' expecting agencies to reduce every exercise of discretion to precise argument).

Our independent review of the record satisfies us that the agency's process of setting the 1994 quota was conducted in good faith, pursued with a proper understanding of the law, based on the best scientific information available, and adequately justified by the agency. If there was an inevitable element of arbitrariness in the decision, there was not the least caprice. The district court could not properly hold that the quota was not based on the best scientific information available where, as we hold, the record demonstrates that the agency fully understood the meaning of the data before it and chose to adopt a quota that, on the basis of that information, offered a high probability of meeting its regulatory mandate while also allowing the fishing industry to increase its harvest for 1994 over that of 1993. Accordingly, we will reverse the district court's judgment, uphold the Department's 1994 commercial catch quota, and dismiss the Coalition's action.2

REVERSED.