Net Change and Implied Flow Part Company at the Four Bcf Reclassification Line

The Line Released on September 17

In the Weekly Natural Gas Storage Report released on Thursday, September 17, 2026, the U.S. Energy Information Administration put working gas in underground storage in the Lower 48 states at 3,298 billion cubic feet as of Friday, September 11, 2026. A week earlier the same series read 3,254 Bcf. Every stock level and weekly change quoted below is read from that September 17 release and not from any later revision. The difference, 44 Bcf, went out across wire services as an injection.

It is not defined as one. The published column is a net change, and the agency's own notes define that term to reflect all events affecting working gas in storage, including injections, withdrawals, and reclassifications between base and working gas. Two of those three things are flows. The third is a bookkeeping decision made by a storage operator about gas that does not move at all.

This matters most in the closing weeks of September, when injection season ends and the peak inventory print becomes an input to winter supply arithmetic. The same care is needed on the demand side of that arithmetic, a point worked through in Count Liquefaction Fuel Before Turning an LNG Export Forecast Into Winter Gas Demand. Here the question is narrower: what does the storage number count?

Working Gas and Base Gas Divide One Reservoir

Underground storage holds two categories of gas that are physically identical. Base gas, also called cushion gas, is defined by EIA as the volume needed as a permanent inventory to maintain adequate reservoir pressures and deliverability rates through the withdrawal season. Working gas is defined as the portion of total capacity that contains gas available for withdrawal.

The weekly report publishes only the second category. EIA's annual capacity table for 2024 shows 9,319,291 million cubic feet of total storage capacity across 414 fields, of which 4,793,009 million cubic feet is working gas capacity. The residual, about 4,526,282 million cubic feet, is base gas. Just under half of the capacity in American storage fields sits outside the number the market trades on.

The split is not uniform, because the physics of holding pressure varies by reservoir type.

Aquifers carry about 70.7 percent of their capacity as base gas because an aquifer has to push water back to make room for gas and keep pushing to hold it there. Salt caverns carry about 33.1 percent, because a cavern is a sealed void that holds pressure on its own. Depleted fields, which are 328 of the 414, sit at about 45.9 percent.

Those ratios are engineering judgments, and engineering judgments get revisited. When an operator concludes that a field can run at lower minimum pressure than previously assumed, gas that was base gas becomes working gas. The reservoir is unchanged. The reported inventory rises.

Net Change, Implied Flow, and the Four Bcf Door

EIA publishes two adjacent concepts and the difference between them is the whole point. Net change keeps reclassifications inside it. Implied flow is defined as movements of working gas into or out of storage facilities that exclude reportable reclassifications, meaning those totaling 4 billion cubic feet or more. The methodology document states the rule from the other direction: if the reported reclassifications of one or more respondents reach 4 Bcf or more, EIA reflects the amount of gas actually flowing in the implied flow estimate.

So the four Bcf line is a door. Below it, a reclassification passes into the headline undetected. At or above it, the two series separate, and a reader who knows to look can see the gap.

A second threshold sits just below. The notes state that implied flow may contain the effects of unpublished net revisions to working gas levels of less than 4 Bcf. Revisions below the reportable size may therefore ride inside the estimate rather than being broken out of it. A third threshold sits lower still. The methodology records that respondents are instructed to submit revisions to data for previous weeks if those revisions were greater than 500 million cubic feet, which puts the floor for a correction a respondent is asked to file at half a Bcf. Between that floor and the 4 Bcf line, a correction exists in EIA's files and leaves no mark the reader of the headline can see.

The upper threshold is set in the Federal Register. A 2005 policy statement establishes that an unscheduled release of revisions is disseminated in a special release when the effect of reported changes is at least 10 billion cubic feet at either a regional or national level, on a Federal workday between 2:00 and 2:10 p.m. Eastern following public notification between 1:00 and 1:10 p.m. the same day. The same document carves out revised estimates resulting from reclassifications between working gas and base gas, which are instead folded into the next regularly scheduled report.

Read that carve-out slowly. A ten Bcf correction from a misreported meter gets its own announcement. A ten Bcf reclassification does not. It waits for Thursday and arrives inside the net change, where nothing labels it.

Friday at Nine, Filed by Monday, Published on Thursday

Form EIA-912 asks operators for working gas as of Friday 9:00 a.m. and tells them to return the form no later than 5:00 p.m. Eastern Standard Time on Monday. The form also instructs respondents to explain any change in working gas resulting from changes in the number or capacity of facilities being reported, or any reclassification of base gas and working gas relative to the prior week.

Three dates therefore attach to one number: a Friday morning reading, a Monday filing, and a Thursday publication. Anything that happens between Friday morning and the Thursday release belongs to the following week's print, no matter how much it moves prices in the interval. Release-day calendars carrying more than one reference period are a recurring trap in energy statistics, taken apart in One EIA Release Day Carried Two Reference Months.

Eighty-Eight Operators Stand In for 414 Fields

The weekly figure is a sample estimate, not a count. EIA's 2024 sample reselection documentation reports that 88 operators were selected for the new sample, against 85 in the prior one, with ten operators added and seven dropped. Those operators account for 97 percent of the total measure of size in the Lower 48. The methodology notes that roughly 80 to 85 respondents are used for estimation in practice, typically accounting for at least 90 percent of all storage volumes in a given stratum.

The sampling frame has six strata: East, Midwest, Mountain, Pacific, South Central-Salt, and South Central-Nonsalt. The published report shows five regions. The salt and nonsalt split inside South Central is a sampling design feature that surfaces as a reporting subtotal, not a presentational choice added later.

EIA states that its established sampling and estimation methodologies have not changed, and that differences between the two samples were small and not statistically significant at the 5 percent level. That is a claim about the sample change, not a claim that sample estimation equals a census.

The Census Arrives Two Months Behind

There is a census. Form EIA-191 collects monthly data from all operators rather than a sample, and it lands roughly two months after the weeks it covers. The agency publishes the comparison itself. Its performance evaluation reports that for January 2020 through December 2022 the average absolute difference between the weekly and monthly series at the national level was 13 Bcf, or 0.5 percent of the average monthly working gas level of 2,738 Bcf for the period, with a root mean squared difference of 17 Bcf.

Among the causes EIA lists is respondents' reclassifications between base gas and working gas, or other inventory adjustments, during the two-month lag between collecting the two series. The reclassification problem is not a theoretical edge case that analysts invented. It is one of the agency's own stated explanations for why its two measures of the same quantity disagree.

Salt and Nonsalt Move on Different Clocks

The week ending September 11, 2026 shows why the regional table deserves more attention than the national line.

East added 22 Bcf, rising from 773 to 795. Midwest added 26, from 908 to 934. Mountain added 2, from 239 to 241. Pacific lost 1, from 290 to 289. South Central lost 5, from 1,044 to 1,039, and the entire regional decline sat in salt, which fell from 227 to 222 while nonsalt held at 817.

Salt caverns cycle several times a season because they fill and empty quickly, so a salt draw inside an injection month is usually a price response rather than a supply problem. Nonsalt landing exactly flat is the stranger reading, and a reclassification or a rounding convention can produce it as easily as physical balance can.

The Comparison Columns Are Built, Not Observed

The report's year-ago and five-year-average columns are not raw past readings. The methodology describes constructing a daily working gas time series, from which five-year averages are computed for each day of the year using the preceding five years of daily estimates. The benchmark against which this Friday is judged is an interpolated construct whose five-year window rolls forward every year.

The raw weekly history is available and worth checking separately. Lower 48 working gas stood at 3,423 Bcf for the week ending September 12, 2025, against 3,298 Bcf this year, a gap of 125 Bcf on comparable weekly readings.

Concrete Framework

Five questions, in order, before a storage print becomes a view.

One. Read net change and implied flow as two numbers. If they agree, no reportable reclassification occurred. If they differ, the gap is the reclassification, and it is not supply.

Two. Check the regional subtotals against the national line. A 44 Bcf build assembled from two large positives and two negatives describes a different market than one spread evenly.

Three. Separate salt from nonsalt inside South Central before concluding anything about the region. They answer different signals on different timescales.

Four. Treat the year-ago and five-year columns as constructed benchmarks. When the comparison matters, pull the raw weekly history and compute the gap directly.

Five. When the monthly EIA-191 census catches up, compare it to what the weekly series said at the time. A persistent one-directional regional gap is information about reporting, not noise.

Where This Doesn't Apply

Implied flow parts from net change only when a respondent reports reclassifications totaling 4 Bcf or more, which is why the two carry the same value in a week without one. The distinction here is a conditional risk, not a standing haircut to apply to every print. Discounting each weekly number by an assumed reclassification allowance would be wrong far more often than right.

The 13 Bcf average weekly-to-monthly gap is also small against seasonal swings. Set beside a weekly change of the size printed for September 11, 44 Bcf, a difference running 0.5 percent of the average monthly working gas level rarely flips a seasonal narrative. It matters for level calibration and end-of-season peak estimates, not for whether storage builds in September.

The base gas share is a capacity concept, drawn from an annual table, and it does not imply that roughly half of any given week's reported inventory is hidden. Working gas in storage during shoulder season runs well below working gas capacity, so the two quantities are not proportional.

None of this transfers cleanly outside the United States. European inventory reporting uses its own definitions of technical and commercial capacity, and the thresholds described here are specific to EIA forms and EIA policy.

What to Watch Next Week

The report covering the week ending September 18, 2026 was scheduled for September 24. Four things are worth checking when it lands.

First, whether net change and implied flow are printed at the same value. They usually are. The weeks when they are not are the weeks the headline is carrying something other than gas.

Second, whether nonsalt South Central stays pinned near 817 Bcf. Two consecutive flat weeks in a subregion of that size would be worth a look at the footnotes rather than at the weather.

Third, the shape of the East and Midwest builds as injection season closes. Those two regions together added 48 Bcf in the week to September 11, while the rest of the country subtracted 4, which is how a 44 Bcf national build gets assembled.

Fourth, any unscheduled release. Under the 2005 policy that signal arrives at 1:00 p.m. Eastern as notice and 2:00 p.m. as data, and it means a reported change of at least 10 Bcf at either a regional or national level that is not a reclassification. Those are the corrections the system surfaces immediately, which is what makes the reclassifications it does not surface worth tracking yourself.

Primary documentation: the U.S. Energy Information Administration, Methodology for EIA Weekly Underground Natural Gas Storage Estimates; the U.S. Energy Information Administration, Underground Natural Gas Storage Capacity annual table; the Federal Register, Policy Statement on Unscheduled Release of Revisions to the Weekly Natural Gas Storage Report; and U.S. Energy Information Administration, Form EIA-912 Weekly Underground Natural Gas Storage Report. The weekly-to-monthly comparison is documented in the U.S. Energy Information Administration, Differences Between Monthly and Weekly Working Gas in Storage, which reports an average absolute error of 12 Bcf, or 0.4 percent, over January 2010 through December 2024; the 13 Bcf and 0.5 percent figures used above come from the U.S. Energy Information Administration, Weekly Natural Gas Storage Report Performance Evaluation for 2020 through 2022.