The Federal Definition of a Design Storm Still Names an Atlas Published in 1961

Somewhere in a permit file there is a number in inches. It decides how much water a containment structure has to hold before the next overflow becomes a violation. The number is not measured at the site, and it is not chosen by the operator. It is looked up.

The lookup runs down a chain. A rain gauge reads once a day. A regional multiplier is applied to what it recorded. A statistical analysis pools that station with many others and produces a gridded estimate with an upper and a lower bound. A public server prints one value from that grid. A regulatory definition points at a document to say what the value means. Every link is published. Every link also drops something, and none of the drops is a mistake.

This is that chain, in order, with the governing text at each step, and a note at each step about what the next link no longer carries.

Each stage is published. What the next stage inherits is narrower than what the stage before it held.

What the Rule Asks For

Start at the enforceable end, because that is where the number has to be exact. Under the effluent guidelines for concentrated animal feeding operations at 40 CFR part 412, one form of the production-area limitation allows an overflow to escape being a violation only if the facility was, in the regulation's words, EPA, 40 CFR 412.31, "designed, constructed, operated and maintained to contain all manure, litter, and process wastewater including the runoff and the direct precipitation from a 25-year, 24-hour rainfall event."

That phrase is the hinge. It converts a weather statistic into a physical dimension: a berm height, a basin volume, a freeboard. So the regulation has to say precisely what a 25-year, 24-hour rainfall event is. It does, in the definitions section:

"Ten (10)-year, 24-hour rainfall event, 25-year, 24-hour rainfall event, and 100-year, 24-hour rainfall event mean precipitation events with a probable recurrence interval of once in ten years, or twenty five years, or one hundred years, respectively, as defined by the National Weather Service in Technical Paper No. 40, 'Rainfall Frequency Atlas of the United States,' May, 1961, or equivalent regional or State rainfall probability information developed from this source."

That is EPA, 40 CFR 412.2(i), read in full. Two things in it matter more than the date.

The first is the connector. The definition offers two sources joined by or, not and. One branch is the 1961 Technical Paper. The other branch is equivalent regional or State rainfall probability information developed from that source. Either branch satisfies the definition on its own. The text does not require both, and it does not say that a later publication supersedes the earlier one. It is a union, and reading it as a sequence of replacements changes what the rule says.

The second is that the same construction appears elsewhere with a different interior. In the cement manufacturing definitions at EPA, 40 CFR 411.31(b), the corresponding term means a recurrence interval of once in ten years "as defined by the National Weather Service in Technical Paper No. 40, 'Rainfall Frequency Atlas of the United States,' May 1961, and subsequent amendments, or equivalent regional or state rainfall probability information developed therefrom." Here the first branch carries three extra words, "and subsequent amendments," that the Part 412 definition does not carry. Same atlas, same year, two slightly different pointers.

The two sources are alternatives. Drawing this as a sequence, or as a gate that requires both, misstates the text.

The phrase is not isolated to animal feeding operations. A full-text search of the Code of Federal Regulations for "25-year, 24-hour" returns sections in 40 CFR parts 415, 418, 421 and 422 alongside Part 412 itself, and also the hazardous waste tank standards at 40 CFR 261.193 and 40 CFR 267.196, where secondary containment must be "sufficient to contain precipitation from a 25-year, 24-hour rainfall event."

The Chain

1. The gauge reads on a clock; the storm does not

The most recent volume of the National Weather Service atlas series covers Idaho, Montana and Wyoming. Its documentation states that "Precipitation measurements were obtained for 15,328 stations from a number of federal, state, and local agencies," and that "Where available, records extended through December 2023." Screening was by usable years rather than calendar span: record length "was characterized by the number of years for which annual maxima could be extracted (i.e., data years) rather than the entire period of record." Daily stations needed at least 30 data years; stations recording at sub-daily durations needed at least 20.

A daily gauge, though, does not record the heaviest 24 hours. It records the heaviest calendar day, bounded by whatever hour the observer resets it. A storm that straddles the reset is split between two readings and appears in the record as two moderate days. The Northeastern States volume states the size of the problem directly: "the true 24-hour (unconstrained) annual maximum could be up to 100 percent larger than the corresponding 1-day (constrained) value extracted from the daily records."

The fix is a multiplier, and the multiplier is regional. In the Northeastern States volume the daily correction factors run 1.11 at one day, 1.04 at two, 1.03 at three, 1.02 at four, 1.01 at seven and 1.00 beyond seven, with hourly factors of 1.08, 1.04, 1.02, 1.01 and 1.00. In the Interior Northwest volume the same tables read 1.17, 1.06, 1.04, 1.03, 1.02 and 1.00 for daily durations, and 1.13, 1.05, 1.03, 1.02, 1.01 and 1.00 for hourly. The Alaska volume publishes a third set, beginning at 1.12 for one day.

What the next step no longer carries is the position of the storm inside the day. It is not recovered; it is replaced by a regional average for what a clock boundary typically costs. The same substitution appears whenever a continuous process is read at fixed instants, as when a streamgage reports discharge from a rating curve rather than a direct measurement.

2. Stations that fail the screen leave no trace in the grid

The screening threshold is itself a local judgment. The Northeastern States volume retained 1,218 daily stations with an average record length of 66 years. Alaska could not meet the usual bar at all. Its documentation is explicit about the departure: "Typically, in other NOAA Atlas 14 volumes, only stations with at least 30 data years were considered for frequency analysis, with allowances made for isolated stations or stations recording at very short intervals. Since there were not enough stations that satisfied that requirement in Alaska, all stations with 15 data years or more were included in the initial dataset. Several isolated and/or hourly stations with 9-10 years of data were also retained."

A reader pulling a depth off the finished grid sees a smooth surface. Whether the nearest station cleared 30 years, 15 years or 9 is in the volume documentation, not in the number.

3. The published product is an interval; the requirement is a single depth

The atlas does not publish a point value alone. The Interior Northwest volume states that "The estimates and associated upper and lower bounds of the 90% confidence interval are provided at 30-arc second resolution." The Alaska volume describes how those bounds are produced: "A Monte Carlo simulation procedure, as described in Hosking and Wallis (1997), was used to construct 90% confidence intervals (i.e., 5% and 95% confidence limits) on both AMS-based and PDS-based precipitation frequency curves."

A structure cannot be built to an interval. The regulation asks for containment of the event, and the design calculation takes one depth. The bounds exist, they are published, and they stop at the edge of the spreadsheet. This is the same narrowing that happens when a flood map carries a zone label but no elevation: the map is still correct, and the thing it cannot say is the thing the builder needs.

4. Two series produce two meanings of "25-year"

Before the Precipitation Frequency Data Server prints anything, it asks the user for a data type, precipitation depth or precipitation intensity, and for a time series type, partial duration or annual maximum. The choice is not cosmetic. The Alaska volume spells out what the label means under one of them: "When used with the AMS-based frequency analysis, ARI does not represent the 'true' average period between exceedances of a given precipitation magnitude, but the average period between years in which a given precipitation magnitude is exceeded at least once." Under the other, the volume describes an indirect conversion: "PDS-based precipitation frequency estimates were calculated indirectly from the Langbein's formula (Langbein, 1949) that transforms a PDS-based average recurrence interval (ARI) to an annual exceedance probability (AEP): AEP = 1 - exp(-1/ARI)."

An annual maximum series counts at most one event per year, so a year with three extreme storms contributes one value and the other two are discarded. A partial duration series keeps them. The depth that gets copied into a design file does not record which selection produced it.

5. The estimate carries no expiry date

The method rests on an assumption that is stated plainly. The Interior Northwest volume: "The precipitation frequency analysis methods used in NOAA Atlas 14 are based on the assumption that the annual maximum series used in the analysis are stationary." The Alaska volume puts the same assumption in climate terms, and describes the decision that followed from testing it: "Therefore, the assumption of stationary climate was accepted for this project area and no adjustment of AMS magnitudes was recommended."

The Interior Northwest volume reports what its own trend tests found before making that call: "neither test detected trends in approximately 90% of the stations at both durations," and "The null hypothesis of no trends in relative AMS data could not be rejected at 5% significance level for either region." That is a defensible finding for that region and those durations. It is also a finding with a date on it that the published depth does not carry. A number built on stationarity looks identical in year one and year forty.

6. Which document is current depends on which side of a state line

The National Weather Service maintains a table of NOAA Hydrometeorological Design Studies Center current precipitation frequency documents, organized by state and by storm duration. The twelve volumes of the current atlas series were published across two decades: volumes 1 and 2 in 2004, volume 3 in 2006, volumes 4 and 5 in 2009, volume 6 in 2011, volume 7 in 2012, volumes 8 and 9 in 2013, volume 10 in 2015, volume 11 in 2018, and volume 12, covering Idaho, Montana and Wyoming, in NOAA Atlas 14 Volume 12, in 2024.

Oregon and Washington are not in any of them. For durations from one hour to 24 hours, the current documents for those two states are NOAA Atlas 2 volumes 10 and 9 respectively, both from 1973. For durations longer than 24 hours, it is Technical Paper 49, from 1964. For durations of an hour or less, it is a 1986 study. Volume 12 lists what it supersedes, and the list names NOAA Atlas 2 volumes 1, 2 and 5 along with Technical Paper 49; the Oregon and Washington volumes of Atlas 2 are not on it.

So a facility in Idaho and a facility a short drive west in Oregon can both be subject to the same federal sentence about a 25-year, 24-hour rainfall event, and read their depth out of documents published 51 years apart.

Publication years from the NOAA HDSC current-documents listing; latest revision years, where different, are not used here. Ages run from 65 years for the 1961 paper named in the definition, through 62, 53 and 40 for the Oregon and Washington documents, to 22, 11, 8 and 2 for the atlas volumes shown.

What Atlas 15 Changes, and What It Does Not

A replacement series is underway. The National Water Prediction Service describes NOAA Atlas 15 as "the new authoritative, spatially continuous National Precipitation Frequency Atlas of the United States, currently under development," and describes the methodological break as "a shift from a stationary assumption (i.e., extreme precipitation events do not change significantly over time) to a nonstationary assumption (i.e., extreme precipitation events change over time)." A Montana pilot dataset was released on September 26, 2024. The published schedule puts preliminary estimates for the contiguous states out for peer review in September 2026, with Volume 1 covering the contiguous states by 2027 and areas outside them by 2028. Until then the same page states that Atlas 14 "is the existing national standard and authoritative source for precipitation frequency information today."

What a new atlas does not do by itself is edit a definition in the Code of Federal Regulations. Branch one of 40 CFR 412.2(i) still reads Technical Paper No. 40, May 1961. A full-text search of the CFR for the phrase "NOAA Atlas 14," run while preparing this piece, returned no sections at all. The current atlas moves the ground under branch two; branch one is a citation, and citations change when someone amends them.

Where This Doesn't Apply

• Where a state has its own rainfall probability information. Branch two of the definition is doing real work in much of the country. Where a state has developed equivalent information, the 1961 paper is not what a permit writer is reading, and the vintage question becomes a state question rather than a federal one.
• Where flooding is not set by point rainfall depth. Coastal surge, snowmelt-driven floods and floods governed by upstream routing are not described by a 24-hour depth at a point. The chain here explains a design storm, not a flood. Coastal elevations run through a different measurement chain, one where the reference surface itself is an average over past years.
• Where the record already resolves shorter intervals. The correction factors described above exist because observations are bounded by fixed clock times. A record that resolves the interval being analyzed does not need that adjustment, and the factors quoted here belong to the specific volumes that published them. They are not portable between regions.
• The 90% interval is not a failure probability. It describes uncertainty in fitting a frequency curve to a sample of station years. It says nothing about whether a particular basin will overflow, which depends on antecedent storage, the state of the structure and what else is running that day.
• Screening thresholds vary by volume. The 30-data-year daily bar is described in the volumes as typical, not universal, and at least one volume documents accepting stations with far shorter records because the network could not supply more.

The recurring shape here is not that any step is wrong. Each publication states its assumptions, its screening rules, its correction factors and its uncertainty bounds, and each is available to read. The narrowing happens at the handoffs. A gridded estimate with two bounds becomes one printed depth. That depth becomes a number in a design file. The number meets a definition naming a 1961 document or an equivalent developed from it, and the definition does not ask which of the two produced the figure. By the time the value is a berm height, everything needed to reconstruct it sits one step back, in a document nobody in the chain has particular reason to open.