Plaintiffs-Appellants Natural Resources Defense Council, Inc., Environmental Defense Fund, Inc., and Alan G. Hevesi (collectively "NRDC") appeal from a May 30, 2000, judgment of the United States District Court for the Southern District of New York (Peter K. Leisure, Judge), entered upon a May 2, 2000, Opinion and Order. NRDC challenges only the district court's dismissal of its claim brought under the APA seeking judicial review of the Environmental Protection Agency's ("EPA") decision to approve total maximum daily loads ("TMDLs") for phosphorous for eight New York reservoirs. The TMDLs were submitted to EPA by the State of New York pursuant to the CWA, 33 U.S.C. § 1313. We affirm the district court's holding that the CWA does not require TMDLs to be expressed in terms of daily loads, but remand to the district court for remand to EPA for further explanation of why annual loads are appropriate in the case of New York's phosphorus TMDLs.

We presume familiarity with the district court's thorough decisions leading up to this appeal. See Natural Resources Defense Council, Inc. v. Fox, 909 F. Supp. 153 (S.D.N.Y. 1995) ("Fox I"); Natural Resources Defense Council, Inc. v. Fox, 30 F. Supp. 2d 369 (S.D.N.Y. 1998) ("Fox II"); and Natural Resources Defense Council, Inc. v. Fox, 93 F. Supp. 2d 531 (S.D.N.Y. 2000) ("Fox III").

In general, the Clean Water Act protects waterbodies by limiting discharges of pollutants into them through technology based controls. See 33 U.S.C. § 1311(b)(2)(A). As the district court noted, this "effluent limitation approach focuses on regulating, through the issuance of permits and required technology-based abatement methods, the amount of pollutants discharged by a pollution source." Fox I, 909 F. Supp. at 156. The Act also requires states to adopt water quality standards for their waterbodies based upon the uses of the waterbodies. See 33 U.S.C. § 1313(a)-(c). Where effluent limitations are not "stringent enough to implement any water quality standard applicable" to a waterbody, the CWA requires that the "State shall establish a priority ranking for such waters." 33 U.S.C. § 1313(d)(1)(A). For such waterbodies, the state is required to establish a "total maximum daily load, for those pollutants which the [EPA] Administrator identifies." Id. at § 1313(d)(1)(C). Each total maximum daily load ("TMDL") "shall be established at a level necessary to implement the applicable water quality standards with seasonal variations and a margin of safety which takes into account any lack of knowledge concerning the relationship between effluent limitations and water quality." Id. In effect, a TMDL posts a limit on the total amount of a pollutant a waterbody may receive over a period of time. See 40 C.F.R. § 130.2(i). The TMDL encompasses discharges into the water from specific sites (like factories located along a river) known as point sources, as well as from nonpoint sources, which can consist of, for example, runoff due to the agricultural use of land adjoining a river, as well as, finally, the natural occurrence of the pollutant in the waterbody (i.e., "natural background"). See id.; 33 U.S.C. § 1362(14) (defining "point source"); Trustees for Alaska v. EPA, 749 F.2d 549, 558 (9th Cir. 1984) (defining "nonpoint source"). The TMDLs the state establishes to limit the loading (or release) of pollutants into a waterbody must in turn be approved by the EPA. See 33 U.S.C. § 1313(2).

In recent years, the nineteen upstate water reservoirs which supply New York City its drinking water have suffered increasing phosphorus pollution, due both to discharges of sewage into them and runoff from nonpoint sources. The increasing quantity of phosphorus in these reservoirs has the propensity to make the reservoirs eutrophic: a state which, as one expert explained in the administrative record before us, arises from "the accumulation of phosphorus in... reservoirs [which] has caused excessive growth or nuisance `blooms' of algae and aquatic macrophytes which are harmful to a waterbody." Eutrophication "can have adverse effects on drinking water quality, ranging from aesthetic changes to potential public health risks such as the increased production of organic material which, after disinfection, leads to by- products."

In 1994, NRDC filed an environmental citizen suit in the United States District Court for the Southern District of New York, claiming, in pertinent part, that the State of New York had a non-discretionary duty under the CWA to promulgate TMDLs for the nineteen reservoirs providing New York City drinking water, and that its failure to do so amounted to a constructive submission of no TMDLs, leaving EPA with the non-discretionary duty to promulgate the TMDLs itself. See Fox I, 909 F. Supp. at 155. The district court refused NRDC summary judgment on this claim, holding there was a genuine issue of fact concerning whether New York had submitted TMDLs for the reservoirs. See id. at 158.

In January 1995, New York placed the nineteen reservoirs on a list submitted to EPA naming waterbodies for which technology-based pollution controls were "not stringent enough to attain or maintain compliance with applicable state water quality standards," which, for the reservoirs, is "water supply." The waterbodies on the list were given priority in the development of TMDLs for pollutants impairing water quality. In 1996, New York published a report explaining the methodology for calculating the phosphorus TMDLs for the reservoirs. The report explained that developing TMDLs "for a large watershed can be data intensive and a phased approach is often used to protect the water body while additional information is collected," and noted that initial phosphorus TMDLs (the "Phase I TMDLs") would be "based on the best available data and utilize[] simplified models." The Phase I TMDLs were to be supplemented in 1998 by a second set of TMDLS ("Phase II TMDLs") taking into account "improved data."^[1] The Phase I TMDLs were established, and after a public comment period during which NRDC offered criticisms, the TMDLs were submitted to EPA for eighteen of the nineteen reservoirs on January 31, 1997. On April 2, 1997, EPA approved the submission for eight of the reservoirs. It also declined to approve the submitted TMDLs for the remaining ten reservoirs because it concluded that pollution levels in those reservoirs did not exceed the level that required resort to TMDLs under the CWA.

In response, NRDC amended its complaint, claiming that the TMDLs submitted by New York were facially inadequate under the CWA and therefore EPA violated a non-discretionary duty under the CWA in approving TMDLs for the eight reservoirs, as well as violating the Administrative Procedure Act, 5 U.S.C. § 706 ("APA"). EPA moved, in pertinent part, for summary judgment based on New York's submission, EPA's approval of the Phase I TMDLs, and an agreement between the State and the agency establishing a schedule for submission and approval of additional TMDLs. See Fox II, 30 F. Supp. 2d at 379. In an Opinion and Order, the district court rejected EPA's motion with regard to the APA claim because "there [were] genuine issues of material fact as to whether EPA should have approved some of the 1997 TMDL submissions." Id. Nevertheless, because EPA's approval of TMDLs under the CWA is discretionary, the district court granted EPA summary judgment against NRDC on the CWA claim. See id. at 380.

On May 2, 2000, the district court issued a final ruling, finding, in pertinent part, that EPA's approval of the eight TMDLs was rationally supported by the administrative record and did not constitute a violation of the APA. See Fox III, 93 F. Supp. 2d at 557. NRDC noted that the CWA requires states to establish a "total maximum daily load" for each waterbody needing additional protection. 33 U.S.C. § 1313(d)(1)(C). Because New York's TMDLs were expressed in terms of annual-not daily-loads, NRDC argued that EPA had no authority to approve them. See id. at 554. The district court noted that while an agency's interpretation of a statute may not contradict the plain language of the statute, an agency does have authority to define "a term in a way that is reasonable in light of the legislature's revealed design." Id. at 555 (internal quotations omitted). The district court decided that "the administrative record adequately supports EPA's conclusion that, for some combinations of pollutants and waterbodies, different expressions of the total maximum load of a pollutant are optimal," and thus "EPA reasonably concluded that the best expression of the phosphorus TMDL for these reservoirs was in terms of mass per time." Id.

The CWA also states that each TMDL "shall be established at a level necessary to implement the applicable water quality standards." 33 U.S.C. § 1313(d)(1)(C). Because the TMDLs for the reservoirs relied upon a phosphorus guidance value that was developed to protect water for recreational uses, NRDC argued that EPA improperly approved the TMDLs for the reservoirs, which are used to supply drinking water. See Fox III, 93 F. Supp. 2d at 551. The district court, finding EPA's approval proper, stated that "If [the guidance value] is an appropriate phosphorus guidance value for drinking water as well as for recreational use, then EPA acted reasonably in approving TMDLs that incorporated the [guidance] value." Id.

The CWA also requires that a margin of safety account for any lack of knowledge concerning the effects of a pollutant on a waterbody. 33 U.S.C. § 1313(d)(1)(C). NRDC argued that the margin of safety included with the TMDLs was insufficient. See Fox III, 93 F. Supp. 2d at 553. The district court concluded that "[t]he record makes clear that EPA gave due consideration to the margin of safety, and exercised its professional judgment with respect thereto." Id. at 554.

The CWA also requires that TMDLs take account of "seasonal variations." 33 U.S.C. § 1313(d)(1)(C). NRDC argued that the TMDLs, because expressed in terms of annual loads, could not take account of seasonal variations. See Fox III, 93 F. Supp. 2d at 556. The district court found sufficient evidence in the record to support EPA's view that because the annual loads were calculated based upon the summer growing season when phosphorus-derived algal growth in waterbodies is at its greatest, the TMDLs adequately accounted for seasonal variations. See id. NRDC filed a notice of appeal on July 28, 2000. On appeal, NRDC renews its arguments that the EPA violated the APA by approving TMDLs that were deficient because they 1) are expressed in terms of annual, not daily, loads; 2) fail to implement the applicable water standard-in this case, a standard appropriate for the water supply use of the reservoirs; and 3) fail to incorporate an adequate margin of safety.

This court reviews a district court's review of an agency action de novo. See United States v. Int'l Bhd. of Teamsters, 170 F.3d 136, 142 (2d Cir. 1999). Under the Administrative Procedure Act, a court may "hold unlawful and set aside agency action, findings, and conclusions found to be... arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law." 5 U.S.C. § 706. Review under this provision is "narrow," limited to examining the administrative record to determine "whether the [agency] decision was based on a consideration of the relevant factors and whether there has been a clear error of judgment." City of New York v. Shalala, 34 F.3d 1161, 1167 (2d Cir. 1994) (internal quotations omitted). Thus, while the "court may not substitute its judgment for that of the agency," an agency decision may be set aside where 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." Id. (internal quotations omitted).

I. Total Maximum Daily Loads.

The CWA requires states to identify waters that cannot meet their water quality standards even after implementation of the effluent limitations required by 33 U.S.C. §§ 1311(b)(1)(A), (B). See 33 U.S.C. § 1313(d)(1)(A). For such waters, "[e]ach State shall establish... the total maximum daily load, for those pollutants." Id. at § 1313(d)(1)(C). "Such load shall be established at a level necessary to implement the applicable water quality standards with seasonal variations and a margin of safety which takes into account any lack of knowledge concerning the relationship between effluent limitations and water quality." Id. Further, each state must estimate

the total maximum daily thermal load required to assure protection and propagation of a balanced, indigenous population of shellfish, fish and wildlife. Such estimates shall take into account the normal water temperatures, flow rates, seasonal variations, existing sources of heat input, and the dissipative capacity of the identified waters or parts thereof. Such estimates shall include a calculation of the maximum heat input that can be made into each such part and shall include a margin of safety which takes into account any lack of knowledge concerning the development of thermal water quality criteria for such protection and propagation in the identified waters or parts thereof.

Id. at § 1313(d)(1)(D). EPA's implementing regulations note that TMDLs "can be expressed in terms of either mass per time, toxicity, or other appropriate measure." 40 C.F.R. § 130.2(i).

NRDC claims the TDMLs approved by EPA are in violation of the CWA because they are presented in terms of annual maximum loads of phosphorus, not daily loads. NRDC argues that an annual expression of TMDLs violates the clear language of the statute, so the matter boils down to one of simple statutory interpretation. EPA, in turn, argues that the statute is silent with regard to how TMDLs should be expressed and that its own regulations make clear that any general expression in terms of units of mass over a period of time is acceptable. The district court agreed with EPA, stating: "Congress, in one sentence, directs EPA to approve TMDLs for hundreds of different pollutants in thousands of different waterbodies, and it is excessively formalistic to suggest that EPA may not express these standards in different ways, as appropriate to each unique circumstance." Fox, 93 F. Supp. 2d at 555.

Statutory analysis begins with the plain meaning of a statute. See United States v. Dauray, 215 F.3d 257, 260 (2d Cir. 2000). The plain meaning can be extrapolated by giving words their ordinary sense. See id. If the plain meaning of a statute is susceptible to two or more reasonable meanings, i.e., if it is ambiguous, then a court may resort to the canons of statutory construction. See id. at 262. Although the canons of statutory interpretation provide a court with numerous avenues for supplementing and narrowing the possible meaning of ambiguous text, most helpful to our interpretation of the CWA in this case are two rules. First, when determining which reasonable meaning should prevail, the text should be placed in the context of the entire statutory structure. See id. ("[A] statute is to be considered in all its parts when construing any one of them.") (internal quotations omitted) (alteration in original). Second, "absurd results are to be avoided and internal inconsistencies in the statute must be dealt with." United States v. Turkette, 452 U.S. 576, 580 (1981); see also Dauray, 215 F.3d at 264 ("A statute should be interpreted in a way that avoids absurd results."). Finally, if the canons of statutory interpretation and resort to other interpretive aids (like legislative history) do not resolve the issue, we will give deference to the view of the agency tasked with administering the statute, particularly insofar as those views are expressed in rules and regulations that implement the statute. "[C]onsiderable weight should be accorded to an executive department's construction of a statutory scheme it is entrusted to administer." United States v. Mead Corp., 121 S. Ct. 2164, 2171 (2001) (internal quotations omitted); see also EPA v. Nat'l Crushed Stone Ass'n, 449 U.S. 64, 83 (1980) ("It is by now a commonplace that when faced with a problem of statutory construction, this Court shows great deference to the interpretation given the statute by the officers or agency charged with its administration.") (internal quotations omitted).

If the language of the statute is as plain as NRDC urges, NRDC's reading of the statute easily prevails. The CWA calls for establishment of a "total maximum daily load," not an hourly, weekly, monthly, or annual load. We believe, however, that the term "total maximum daily load" is susceptible to a broader range of meanings. Indeed, NRDC's overly narrow reading of the statute loses sight of the overall structure and purpose of the CWA. The CWA contemplates the establishment of TMDLs for an open-ended range of pollutants that are susceptible to effective regulation by such means. See 33 U.S.C. § 1313(d)(1)(c) (noting that states must establish TMDLs for all "pollutants which the Administrator identifies... as suitable for such calculation"). In the case of each pollutant, effective regulation requires agencies to determine how the pollutant enters, interacts with, and, at a certain level or under certain conditions, adversely impacts an affected waterbody. In the case of highly toxic pollutants that may work harmful effects upon a waterbody almost immediately when present at small levels, close regulation at a daily level may be most appropriate. In the case of other pollutants, like phosphorus, the amounts waterbodies can tolerate vary depending upon the waterbody and the season of the year, while the harmful consequences of excessive amounts may not occur immediately. In short, the CWA's effective enforcement requires agency analysis and application of information concerning a broad range of pollutants. We are not prepared to say Congress intended that such far-ranging agency expertise be narrowly confined in application to regulation of pollutant loads on a strictly daily basis. Such a reading strikes us as absurd, especially given that for some pollutants, effective regulation may best occur by some other periodic measure than a diurnal one. Accordingly, we agree with EPA that a "total maximum daily load" may be expressed by another measure of mass per time, where such an alternative measure best serves the purpose of effective regulation of pollutant levels in waterbodies.

Nevertheless, as with any agency action falling within the bounds of the APA, EPA's discretion in determining what period of time will govern the measurement of each pollutant's mass in a waterbody is subject to judicial review. EPA must "examine[ ] the relevant data and establish[ ] a rational connection between the facts found and the choice made." Cellular Phone Taskforce v. FCC, 205 F.3d 82, 89 (2d Cir. 2000) (internal quotations omitted), cert. denied, 531 U.S. 1070 (2001). EPA argues that a daily measure of phosphorus would be inappropriate given that phosphorus concentrations vary seasonally and annually. The record supports this view, indicating that phosphorus concentrations in waterbodies are affected "by the seasonal interplay of temperatures, density, and wind," resulting in the frequent occurrence of "very large short-term yearly variations which characterize the gradually increasing concentration." While the record makes clear why EPA or a State might opt not to measure loads on a daily basis, it remains unclear why an annual measurement of loads would therefore be more appropriate since phosphorus concentrations vary within a waterbody on a seasonal basis.

We recognize that the question of whether the annual loads set by New York and approved by EPA account for seasonal variation was litigated below, resolved in favor of EPA, and not raised by NRDC on appeal. See Fox, 93 F. Supp. 2d at 555-56. But the question before the district court was whether the TMDLs accounted for seasonal variations, as the CWA requires. The district court found the annual loads accounted for seasonal variations because the record shows the loads were calculated based upon phosphorus levels during the growing season, when conditions are optimal for algal growth. See id. at 556. The question before us is not whether the maximum load was calculated to account for seasonal variations but, rather, whether the annual period over which the load is measured is appropriate for tracking and monitoring something that fluctuates on a seasonal basis. The record shows that the "flow" or the amount of phosphorus contributed to a waterbody by each source will "vary widely in response to season, storm events, and other random factors." Given this and the fact that phosphorus contributed to a waterbody during one season can impact algal growth during other seasons (when, due to seasonal factors, biological activity increases), seasonal regulation of phosphorus flows may be more appropriate. Of course, we do not suggest how best to regulate phosphorus. Instead, we remand for EPA to justify how the annual period of measurement takes seasonal variations into account.

II. Implementation of Applicable Water Standards.

The CWA requires that each TMDL "be established at a level necessary to implement the applicable water quality standards." 33 U.S.C. § 1313(d)(1)(C). The New York State Department of Environmental Conservation ("NYSDEC") has determined that the use, and hence applicable water quality standard, for the reservoirs is "water supply." New York's water quality standard for phosphorus requires the presence of "[n]one in amounts that will result in growths of algae, weeds and slimes that will impair the waters for their best usages." N.Y. Comp. Codes R. & Regs., tit. 6, § 703.2. In 1993, the NYSDEC determined that a numerical guidance value of twenty micrograms of phosphorus per liter of water ("ug/L") would meet the standard imposed by the regulation with regard to waterbodies used for recreational purposes. New York relied on the same figure when determining the TMDLs for the reservoirs, while recognizing that a "guidance value" developed for the "protection of aesthetics for primary and secondary contact recreation.... may not be stringent enough to protect the drinking water supply." New York declared it "will continue to investigate if the [guidance value developed to protect water for recreational purposes] is sufficient to maintain water quality in the reservoirs, with the future option to revise this critical concentration." NRDC argues that the TMDLs for these reservoirs are not those necessary, under the CWA, to "implement the applicable water quality standards" because a less stringent standard, designed to protect water for recreational uses, has been applied to waterbodies designated for water supply use, the highest use.

According to the NYSDEC, the value of 0.020 ug/L "total phosphorus is based upon [waterbody] user surveys conducted in New York" as well as Vermont and Minnesota in which citizens "are asked to best describe the physical condition of the lake with respect to algal levels and the recreational suitability of the lake at the time of sampling." During the comment period for New York's proposed TMDLs, New York met NRDC's objection to the use of an aesthetic criteria in determining appropriate phosphorus levels by stating: "We acknowledge that the current guidance value... is based on aesthetic conditions affecting recreational uses.... Since it is the only presently available phosphorus criteria, it was used for the Phase I TMDL analysis. This is appropriate, especially in New York, where the best use designation of a waterbody for drinking water expressly incorporates all other best uses set forth in our regulations." However, as NRDC argues, this does not get around the fact that the suitability of drinking water is being evaluated by a lesser, and therefore possibly inappropriate, water use standard. New York conceded "that a different phosphorus guidance value or standard may be necessary to reduce eutrophication and/or to protect surface waterbodies that are a source of water supply. NYC is in the process of collecting additional data for the development of a phosphorous standard specifically designed for the protection of waters as a water supply source."

While at first blush New York's use of an aesthetic water quality standard to protect drinking water seems a cause for concern, in the end, EPA's primary concern in determining whether to approve the TMDL is whether or not the TMDL will "implement the applicable water quality standard[ ]." 33 U.S.C. § 1313(d)(1)(C). As the district court concluded, "If 20 ug/L is an appropriate phosphorus guidance value for drinking water as well as for recreational use, then EPA acted reasonably in approving TMDLs that incorporated the 20 ug/L value." Fox III, 93 F. Supp. 2d at 551. In approving New York's TMDLs, EPA noted that the guidance was based on an aesthetic criteria, but stated that nevertheless "the phosphorus guidance value is meant to control excessive and nuisance growths of algae and other aquatic plants. This guidance value is used by the NYSDEC to reduce nuisance algal blooms to acceptable levels and therefore, it indirectly address [sic] the effects of eutrophication and cultural eutrophication." Most importantly, EPA concluded that "NYSDEC's guidance value is below EPA's recommended level of 25 ug/L, and should, therefore, be sufficiently protective to control nuisance aquatic growth and to protect against other indirect effects of eutrophication and accelerated cultural eutrophication."

EPA's 1986 manual, "Quality Control for Water," notes that a criterion to "control nuisance aquatic growths... currently is evolving," but also points out that "[m]ost relatively uncontaminated lake districts are known to have surface waters that contain from 10 to 30 ug/L total phosphorus as P" (citation omitted). The manual thus concludes that "[t]o prevent the development of biological nuisances and to control accelerated or cultural eutrophication, total phosphates as phosphorus (P) should not exceed... 25 ug/L within [a] lake or reservoir." The administrative record thus lends substantial support to EPA's approval of the TMDLs.

NRDC objects that nothing in the manual suggests that a level of 20 ug/L is sufficient to protect drinking water. However the manual recommends a maximal level of 25 ug/L to prevent "biological nuisances" and "cultural eutrophication" of reservoirs, and the problems phosphorus creates for drinking water stem from biological nuisances and cultural eutrophication. To be sure, the manual instructs that "[e]utrophication problems may occur in waters where the phosphorus concentration is less than that" recommended as the maximal level due to "natural conditions." As NRDC points out, the manual explicitly states, "[n]o national criterion is presented for phosphate phosphorus for the control of eutrophication." However, EPA is not seeking to enforce a national standard in this instance. As all the parties recognize, additional research oriented to the specific conditions of New York's reservoirs would be optimal. In the meantime, EPA's hands are not tied just because it must act based on scientific knowledge that is incomplete or disputed. "In the face of conflicting evidence at the frontiers of science, courts' deference to expert determinations should be at its greatest." Cellular Phone, 205 F.3d at 90. Therefore, EPA's determination that New York can formulate its TMDL for phosphorus using an aesthetic criterion is not arbitrary and capricious at this point in time.

III. Margin of Safety.

The CWA requires that a TMDL incorporate "a margin of safety which takes into account any lack of knowledge concerning the relationship between effluent limitations and water quality." 33 U.S.C. § 1313(d)(1)(C). The TMDLs for the pertinent reservoirs incorporate a margin of safety ("MOS") amounting to ten percent of the critical load. Thus, the critical load for a given reservoir-that is, the estimated maximal amount of phosphorus that the reservoir can receive without becoming eutrophic-is typically reduced by ten percent in order to arrive at the appropriate TMDL. New York explained: "For the NYC reservoirs, the TMDL will be set at 90% of this critical load, which utilizes a 10% MOS. The MOS component accounts for the inherent uncertainty in modeling the effects of pollutant loads on the quality of the receiving waterbody. In this particular case, the MOS accounts for yearly variations in the hydrology and the reservoir response to the phosphorus load" (internal citation omitted). In approving the TMDLs, EPA noted that in lieu of any

`standard' or guideline for choosing a specific margin of safety, best professional judgment and the available information are used in setting a margin of safety. (In many cases, a separate margin of safety is not used, but is inherent in the conservative assumptions used in the model). The close calibration between the predicted and observed phosphorus concentrations indicates the model and assumptions are justified and that a higher margin of safety is not warranted.

NRDC argues the margin of safety fails to meet "the clearly delineated Congressional specifications" because there are not adequate findings in the record that the margin of safety used addresses "any lack of knowledge concerning the relationship between effluent limitations and water quality," as the CWA requires.

This is not a case where the agency has "relied on factors which Congress has not intended it to consider." Shalala, 34 F.3d at 1167. EPA's approval of the MOS was guided in part by the close calibration between predicted and observed phosphorus concentrations in the New York reservoirs. New York relied on the Reckhow Land Use Model to calculate the amount of phosphorus entering the reservoirs. New York decided modeling results were "acceptable for adopting a Phase I TMDL when the predicted total phosphorus concentration is within n 20% of the observed phosphorus concentration." Reliance on the Reckhow Model was justified because it had "been previously applied to several of the New York City reservoirs with good results." The Reckhow Model allows calculation of the amount of phosphorus coming from nonpoint sources based on the use of land adjoining waterbodies contributing to the reservoirs. If the model is not closely calibrated, (i.e., if an agricultural use of land, for example, contributes significantly more than the expected amount of phosphorus to the adjoining water body), regulation of phosphorus concentrations in the reservoirs becomes an uncertain matter. Where the estimate of how much phosphorus all the point and nonpoint sources will contribute to a reservoir over a given period tends to correlate with the concentrations actually found in the reservoirs, regulators can be more assured that limiting source discharges to levels that will fall short of the TMDL will result in a phosphorus concentration in the reservoir that does not exceed the critical load. As New York explains, "[w]hile the critical and observed phosphorus loads are obtained by actual phosphorus measurements in the reservoirs, the proportion of the load allocated to point and nonpoint sources will be determined by the modeling."

EPA could reasonably conclude that given the close calibration of the Reckhow Model, a larger margin of safety was not warranted. The margin of safety accounts for uncertainty regarding the effects of phosphorus on water quality by ensuring that the TMDL maintains phosphorus levels below the critical load the reservoir can tolerate. The Reckhow model aids in this task since its accuracy ensures a limited range of fluctuation between projected phosphorus levels and actual levels.

NRDC takes issue with the adoption of a ten percent margin of safety, arguing that no scientific or mathematical basis prescribed this percentage as opposed to any other. As EPA explained, because "there is no `standard' or guideline for choosing a specific margin of safety, best professional judgment and the available information are used in setting [it]." While the MOS may thus be set with an uncomfortable degree of discretion, requiring that EPA show a rigorous scientific methodology dictates one course of action as opposed to another and would effectively prevent the agency from acting in situations where action is required in the face of a clear public health or environmental danger but the magnitude of that danger cannot be effectively quantified. "[A]s long as Congress delegates power to an agency to regulate on the borders of the unknown, courts cannot interfere with reasonable interpretations of equivocal evidence." Public Citizen Health Research Group v. Tyson, 796 F.2d 1479, 1505 (D.C. Cir. 1986). Were it clear, for instance, that a widely used and reliable scientific methodology was applicable to determining a margin of safety and that EPA had turned a blind eye to recommendations based on such a methodology, NRDC's challenge would have more force. But simply to reject EPA's efforts to implement the CWA because it must respond to real water quality problems without the guidance of a rigorously precise methodology would essentially nullify the exercise of agency discretion in the form of `best professional judgment.' Finally, it is worth noting that approval of the Phase I MOS was based, in part, on the limited information available. The EPA approval contemplates revision of the MOS as more information becomes available: "As additional reservoir data and loading data become available, Phase I model assumptions are being reexamined under Phase II."

It is clear that the amount of phosphorus the New York reservoirs can tolerate and still provide good quality drinking water is open to dispute and in need of additional research. However, based on information in the record, we find EPA's approval of the margin of safety incorporated in the New York TMDLs along with its finding that the TMDLs were established "at a level necessary to implement the applicable water standards" of the reservoirs was supported by substantial evidence in the record and, accordingly, not in violation of the APA. We also hold that the CWA does not require that all TMDLs be expressed strictly in terms of daily loads. Nevertheless, while EPA has discretion to approve TMDLs expressed in other periodic measurements where those differing measurements are best suited to regulating a given pollutant, we do not find adequate reasons in the record before us to explain why phosphorus is best regulated by TMDLs expressed in terms of annual loads. We therefore remand to the district court for remand to EPA for explanation of how expressing the New York TMDLs in terms of annual loads will account for seasonal fluctuations in the levels of phosphorus in waterbodies.