The needle read 989 twice, and the dial never said STORMY

Dark navy screen laid out as a household wall display, masthead “Hall display · barometer · a cottage on Nantucket · Thu 4 Jan 2018 · 08:53 EST” on the left and “Source: KACK METAR · hourly” on the right, headlined “Barometer: RAIN.” A standfirst says this is one reading printed exactly as the station reports it — 989 hPa, sea-level pressure, from the automated weather station at the airport; that nothing is smoothed, averaged or rounded away; that the dial is not truncated but drawn across its full 950 to 1050 hPa in even 10-point steps with no break and no cap; that the words are the ones barometer faces have carried since long before the aneroid, STORMY, RAIN, CHANGE, FAIR, VERY DRY; and that the needle sits in RAIN, a little over a third of the way round. The read-out for the household is rain later, bring the cushions in, and take a coat to the 11:15 ferry. The main panel, headed “Sea-level pressure · now” and subtitled “Full dial, 950 to 1050 hPa · 270° sweep · five zones · nothing hidden, nothing capped”, is dominated by a large circular barometer face. Its outer band is divided into five zones, shaded from dark at the bottom of the scale to pale at the top and labelled STORMY, RAIN, CHANGE, FAIR and VERY DRY around the rim; ticks run every 5 hPa with numbers every 10 from 950 at the lower left round to 1050 at the lower right; and a single gold needle points just below 990, into the RAIN zone, with the word RAIN printed beside it. Under the dial the reading is set out as “989.1 hPa · 29.21 inHg” with a note that this is 39.1% of the way round the dial, 105.6° of the 270° sweep. Five tiles sit beside the dial. “The five words and where they start” tabulates each zone with its range in hectopascals and its lower edge in inches of mercury: STORMY up to 965.1, RAIN 965.1 to 999.0 from 28.50 inHg, CHANGE 999.0 to 1,015.9 from 29.50, FAIR 1,015.9 to 1,032.8 from 30.00, VERY DRY above 1,032.8 from 30.50, with the RAIN row picked out in gold. “Dial settings” gives minimum 950, maximum 1050, sweep 270°, major ticks every 10 hPa and minor every 5, zone edges at 28.5, 29.5, 30.0 and 30.5 inHg, set once on installation. “How to read the dial” says to find the needle and read the word behind it, that further round means higher pressure and better weather while further back means lower and worse, and that the figure under the dial is the pressure itself so nothing has to be estimated by eye. “Why the dial reads 989” says because that is what the station says. “Where the words come from” notes that the five-word face is older than the aneroid — it was on mercury wheel and banjo barometers long before Vidi’s 1844 movement inherited it — that it encodes a real piece of weather lore, low pressure going with unsettled air and high with settled, and that the zone edges are round numbers in inches of mercury because that is the unit the faces were engraved in. “No judgement required” notes the reading is reported to a tenth of a hectopascal and that at 989.1 the needle sits 9.9 hPa clear of the CHANGE edge above it and 24.0 hPa clear of the STORMY edge below — nowhere near a boundary. Eight cards run down the right. “The reading, in one figure” prints 989 hPa in large type. “The forecast, in one word” prints RAIN in large gold type, noting that no interpretation is required of the reader because the words are printed on the face. “Read-out for the household” repeats the cushions and the ferry and adds that the gutter can wait for the weekend. “The plan for today” lists the 11:15 ferry to Hyannis, cushions in off the deck, gutter at the weekend, and the Bakers arriving at four. “What the dial leaves out” says nothing, as far as the pressure goes — the whole scale is on the face, every zone is labelled, the needle is where the number puts it, no axis has been cut, no scale broken, no maximum capped and no colour ramp bent. “Why we like the dial” says you can read it from the stairs, and that it has been in the hall for eleven years without once being wrong about a Sunday. “And if the weather turns” says we shall see it out of the window. A “Basis of measurement” card credits sea-level pressure from station KACK, Nantucket Memorial Airport, from the hourly METAR of 4 January 2018 at 13:53 UTC, that is 08:53 EST — the group SLP891, or 989.1 hPa. A method line runs along the foot: one instantaneous reading, full 950–1050 hPa dial, 270° sweep, no truncation, no break, no cap, five zones at the traditional edges, reading printed to a tenth of a hectopascal. A footnote credits the Iowa Environmental Mesonet ASOS archive and says the display is a demonstration drawn by us in the manner of a household weather panel — the cottage, the household and the read-out are invented, and the figure behind them is real. — Showing the misleading chart

The same readings on a cream background, kicked “The same station · the same dial · 120 hourly observations · KACK, Nantucket · 2–6 January 2018” and headlined “The needle read 989 twice — once on the way down, once on the way back up.” A standfirst says nothing on the display was cropped, truncated, smoothed or rescaled and every figure on it is correct; that what sets the picture is that a gauge shows a level and a storm is a rate; that pressure at this station fell 57.2 hPa in 24 hours, its fastest three-hour stretch worth 13.3 hPa on its own, and that the needle, having no time axis to fall along, simply sat where the current number put it; that the same 989 hPa therefore arrives twice, fifteen hours apart, once with 21.8 hPa still to fall and once with 21.9 already recovered, 0.1 hPa and 0.27° of arc between them; and that the two limits which decide where the needle points, 950 and 1050, were typed in by whoever installed the thing — the minimum shifting every reading and the width dividing it, two settings and neither of them the measurement. The upper panel, “The level · what the dial shows”, subtitled “Hourly sea-level pressure at KACK · 2–6 January 2018 · every observation plotted where it falls · the bands are the dial’s five words”, is a navy line chart on a pressure axis from 962 to 1032 hPa with the dial’s word zones drawn behind it as horizontal bands labelled at the right: STORMY below 965.1, RAIN 965.1 to 999.0, CHANGE 999.0 to 1,015.9, FAIR 1,015.9 to 1,032.8, the fifth zone lying above the top of the axis. The line runs near 1,025 through 2 and 3 January, peaks at 1,027.3, drifts down through the evening of the 3rd, then falls almost vertically through the 4th to a minimum of 967.3 before climbing back through the 5th and 6th to about 1,022. A gold bracket across the top marks the largest 24-hour fall, 1,024.7 to 967.5 hPa, −57.2. A marked point at 11:53 on the 4th carries a leader up to the note “11:53 — the face turns from CHANGE to RAIN. It will not change again for 26 hours.” Two gold points, A at 4 January 13:53 UTC and B at 5 January 04:53 UTC, sit at the same height and are joined by a dashed horizontal rule, each with a leader down to a caption: A is 989.1 hPa, falling 11.4 hPa in three hours, wind 29 kt gusting 50, with 21.8 hPa still to fall; B is 989.2 hPa, rising 7.4 hPa in three hours, wind 22 kt gusting 33, with 21.9 hPa already recovered. The minimum is labelled “967.3 — and still RAIN, by 2.2 hPa”. Directly below, on the same time axis with its ticks aligned, the panel “The rate · what it has nowhere to put” plots the three-hour pressure tendency for every hour as bars, gold for falls and blue for rises: small wobbles through the 2nd and 3rd, then a deepening run of gold through the 4th, reaching a labelled −13.3 hPa in three hours, transmitted as 58133, followed by a run of blue as the pressure recovers, and dots marking the eight hours at which the station itself coded the tendency into its own message. A lower-left panel, “One reading, three dials”, draws 989.1 hPa on three ranges that all really get used, each as a small dial with its needle: on the household 950–1050 dial it is 39.1%, 105.6° of the 270° sweep; on the narrower 980–1040 span common in weather widgets it is 15.2%, 41.0°; and on an 870–1090 dial wide enough for every sea-level pressure on record it is 54.1%, 146.2°. Each notes where the storm minimum of 967.3 would fall — 17.3% of the first dial, off the bottom of the second, 44.2% of the third. Beside it, “The two messages the two needles came from” prints both raw METARs exactly as transmitted, with the altimeter setting A2921 and the sea-level pressure groups SLP891 and SLP892 highlighted, and glosses them: A is wind 050° at 29 kt gusting 50, visibility 3 statute miles in rain and mist, +5 °C, 0.24 in of rain in the hour; B is wind 280° at 22 kt gusting 33, visibility 10 statute miles, −3 °C, dry. A third box explains the highlighted groups — SLP891 and SLP892 are sea-level pressure to a tenth of a hectopascal written without the leading 9 or 10, and A2921 is the altimeter setting, 29.21 inches of mercury, identical in both messages. Nine cards run down the right. “What a gauge cannot show” prints 0.27° in large type — the angle between the two needles, narrower than the needle drawn over it, with fifteen hours, a 21.8 hPa fall and a 21.9 hPa recovery lying between them. “The range is typed in” says a gauge has no baseline and no zero, only a minimum and a maximum, with the needle landing at value minus minimum over maximum minus minimum — the minimum shifting the reading, the width dividing it — so the same reading is 39.1%, 15.2% or 54.1% of the way round depending on which pair was chosen, 105.2° apart at the extremes. “It is a truncated axis, bent round” notes that the face starts at 950, so by the same arithmetic that gives a bar chart its zero this is a truncation the curve merely hides; that it cannot simply be restored, because a barometer drawn from zero would be useless; and that a gauge therefore has no baseline to put back, which is why its range is a judgement and a judgement has to be declared. “The zones are a verdict, not a measurement” notes five words and four edges drawn once and before any data arrived — the exact inverse of class-break shopping — knowing nothing of the station, the season or the direction of travel, so every reading from 965.1 to 999.0 hPa gets the same answer, which is a noun. “The word barely moved” records that across the five days the face showed three of its five words, FAIR, then CHANGE at 05:53 on the 4th, then RAIN at 11:53, that last change coming with 29.5 hPa of the fall still ahead, after which RAIN stood for 26 hours and 27 observations. “It never said STORMY” notes the lowest reading of 967.3 hPa is 28.564 inHg against a STORMY edge at 28.5, so the needle stopped 2.2 hPa short of the only word that would have described the afternoon, while the station gusted to 53 kt. “The rate was in the same message” explains that every three hours the station coded the tendency into its own report — 58133 at 17:53 meaning a fall of 13.3 hPa in three hours — and that the same message carried the remark PRESFR, pressure falling rapidly. “The same needle, twice” tabulates A against B: pressure 989.1 and 989.2, dial word RAIN and RAIN, three-hour change −11.4 and +7.4, gust 50 and 33 kt, visibility 3 and 10 statute miles, temperature +5 °C and −3 °C. A “Basis of measurement” card credits 120 hourly METAR observations for KACK from the Iowa Environmental Mesonet ASOS archive of NWS/FAA reports and notes that the three-hour changes, computed by us from the sea-level series, agree to a tenth of a hectopascal with the eight tendency groups the station transmitted, which encode station-level pressure. Five notes run along the foot: “Every observation plotted”, “Hours where they fall”, “Nothing smoothed or interpolated”, “One scale throughout” and “The rate on the same axis as the level”. A closing footnote gives the window’s extremes and largest 24-hour fall; records that the strongest gust in the window is the 53 kt peak-wind group in the 19:53 report on the 4th while the hourly gust column tops out at 50 kt; names the system as the explosively deepening East Coast cyclone of 4 January 2018 and says that the bomb-cyclone criterion of Sanders and Gyakum (1980) is defined on a cyclone’s central pressure rather than on a fixed station, working out at 18.3 hPa per 24 hours when evaluated at 41° N; gives the two extremes the 870–1,090 hPa dial is drawn to cover, 870 hPa in Typhoon Tip on 12 October 1979 and, on the high side, the WMO’s two figures of 1,083.3 hPa at Agata, Russia, on 31 December 1968 below 750 m and 1,089.4 hPa at Tosontsengel, Mongolia, on 30 December 2004 above it; notes that the dial words and their edges are the traditional ones and that makers place them a little differently, which is part of the point; and states that both drawings are ours, that the cottage, the household and the read-out on the first are invented, and that every measurement on both is real. — Showing the misleading chart

Be the first to star this exhibit

A hall barometer reads 989 hPa on a full, untruncated 950–1050 dial and prints the word RAIN. Every figure on it is right. But a gauge shows a level and a storm is a rate: pressure at that station fell 57.2 hPa over the following 24 hours, and the same 989 arrives again fifteen hours later with the sky clearing — a tenth of a hectopascal and a quarter of a degree of arc between the way down and the way back up. The face changed its word twice on the way down and then not again for 26 hours, and at its lowest reading it was still 2.2 hPa short of the only word on it that would have described the afternoon.

Hourly METAR observations for station KACK (Nantucket Memorial Airport, 41.253° N, 70.060° W), 2–6 January 2018 UTC, from the Iowa Environmental Mesonet ASOS archive of NWS/FAA reports — 120 hourly sea-level pressures, winds, gusts and raw message text. Maximum 1,027.3 hPa on 3 January at 14:53 UTC, minimum 967.3 hPa on 4 January at 20:53 UTC; largest 24-hour fall 1,024.7 to 967.5 hPa, 19:53 on the 3rd to 19:53 on the 4th. Three-hour changes computed by us from the sea-level series, agreeing to a tenth of a hectopascal with the eight 5appp tendency groups the station transmitted, 2018Added September 6, 2026

01The claim

One reading, printed exactly as the station reports it: 989 hPa of sea-level pressure, from the automated weather station at the airport. Nothing is smoothed, averaged or rounded away. The dial is not truncated — it is drawn across its full 950 to 1050 hPa, every hectopascal of it, in even 10-point steps, with no break in the scale and no cap at the top. The ticks are evenly spaced, the whole range is on the face, and the number is printed underneath so nothing has to be estimated by eye. The words are the ones barometer faces have carried since long before the aneroid: STORMY, RAIN, CHANGE, FAIR, VERY DRY, with their edges at 28.5, 29.5, 30.0 and 30.5 inches of mercury. The needle sits a little over a third of the way round, in RAIN — 9.9 hPa clear of the CHANGE edge above it and 24.0 clear of the STORMY edge below, nowhere near a boundary, so no judgement is required of anybody. Read-out for the household: rain later. Bring the cushions in off the deck, and take a coat to the 11:15 ferry. The gutter can wait for the weekend. And if the weather turns, we shall see it out of the window.

02The trick

Everything on that face is true, and the read-out it produces is wrong, because a gauge shows a level and the thing the household wanted to know is a rate. The reading is 08:53 on the morning of 4 January 2018, and over the next seven hours the pressure at that station fell to 967.3 hPa. Measured end to end the fall is 57.2 hPa in 24 hours, from 1,024.7 at 19:53 on the 3rd to 967.5 at 19:53 on the 4th, with its fastest three-hour stretch worth 13.3 hPa on its own. A needle has one mark and no time axis, so none of that can appear on it; the dial simply sat where each successive number put it. The clearest way to see what that costs is a pair of readings from the station’s own record. At 13:53 UTC on the 4th the pressure was 989.1 hPa, falling 11.4 points in three hours, wind 29 kt gusting 50, visibility three statute miles in rain, +5 °C, with 21.8 hPa of the fall still ahead of it. At 04:53 UTC on the 5th it was 989.2 hPa, rising 7.4 points in three hours, wind 22 kt gusting 33, visibility ten miles, −3 °C, dry, with 21.9 hPa already recovered. The two raw messages even report the same altimeter setting, A2921. On a 100-hectopascal dial with a 270° sweep, one tenth of a hectopascal is 0.27 of a degree — narrower than the needle drawn over it — so the way down and the way back up are, to the instrument, the same picture. Two smaller mechanisms ride along, and both are what make a gauge different from a bent axis. The first is that the needle’s position is not a property of the number alone. There is no baseline and no zero here: the needle lands at the value minus the minimum, over the maximum minus the minimum, so the minimum shifts every reading and the width divides it — two settings, neither of them the measurement, and both typed in once on installation. The same 989.1 hPa is 39.1% of the way round a 950–1050 dial, 15.2% of a 980–1040 one — the narrower span weather widgets often ship with — and 54.1% of a dial running 870 to 1,090, which is wide enough for every sea-level pressure on record. That is 105.2° of arc between the extremes, from one measurement, and on the 980–1040 dial the afternoon’s minimum would have been drawn pinned to the bottom peg with nothing to say how far past it the storm went. Worth naming plainly: a face that starts at 950 is a truncated axis, by exactly the arithmetic that gives a bar chart its zero. The curve is what hides it, and the standfirst’s proud “not truncated” is about the top of the scale rather than the bottom. Nor can it simply be put back, because a barometer drawn from zero would be useless — which is the honest version of the problem. A gauge has no baseline to restore, so its range is a judgement, and a judgement has to be declared. The second is the words. Five of them, four edges, drawn once and identical in every hallway the instrument is ever hung in, and they convert a continuous quantity into a verdict — the exact inverse of choosing class breaks after the data arrives, and no safer for it. Across the whole five days the face showed three of its five words. It turned from FAIR to CHANGE at 05:53 on the 4th and from CHANGE to RAIN at 11:53, with 29.5 hPa of the fall still ahead, and then it stopped: RAIN stood for the next 26 hours and 27 observations, across the rest of the drop, the minimum and the whole recovery. At the lowest reading of the event, 967.3 hPa, with the station gusting to 53 kt, the needle was still 2.2 hPa short of STORMY, the only word on the face that would have described the afternoon. And the rate was never missing from the data. Every three hours the station coded the pressure tendency into the same message the dial read its number out of: 58133 at 17:53 on the 4th means the pressure fell 13.3 hPa in three hours, and that message also carried the plain-language remark PRESFR, pressure falling rapidly. The gauge took the level out of the wire and dropped the rest of it. (Both drawings are ours; the cottage, the household and the read-out on the first are invented, and every measurement on both is real.)

03The fix

Give the number back its history. A single value almost never means anything on its own, and the smallest honest fix costs no more space than the dial did: a short line of the last day or two with the current reading printed beside it, so the level, the direction and the rate arrive together. Where the rate is the finding, plot the rate — a bar per hour of the change over the previous three hours turns this station’s 4 January into a picture nobody could misread, and it is a number the system already had, because the sensor that reported the level reported the tendency in the same breath. That is worth checking before building anything: the quantity you need is very often already in the feed, dropped by the widget rather than missing from the data. When you do draw both, put them on one time axis, ticks aligned, so the reader can carry a moment from the level down to the rate without doing arithmetic in their head. Then treat the gauge’s two limits as what they are, an axis. Print the range in the caption the way you print a unit, hold it fixed across every gauge on a page, run the minimum to zero wherever the quantity has one — and where it does not, say so, because an undeclared range is a truncation that happens to be curved. Never let a value past the maximum be drawn pinned to it: a needle against the peg is a clipped axis in a second body, and it should own up with an arrow and a printed figure. Keep the coloured zones only for decisions that genuinely are categorical, say at each edge what the threshold is and who set it, and print the number as well as the word, since a reading one unit inside a boundary and a reading at the far end of it are the same colour and not the same news. None of this makes a gauge useless. It is a genuinely good instrument for the job it was built for — glanced at, live, read for which side of a line you are on, which is why cockpits and boilers still have them — and a barometer face is honest about that job the moment you notice what a real aneroid has that a dashboard tile does not: a second, manually set pointer, parked against the needle the last time you looked, so the instrument can show you the one thing a single mark cannot. The distinction to hold on to is between a position and a change: a gauge answers where the number is, and only a series answers where it is going. And for comparing several quantities, drop the form entirely — a row of gauges is a row of separate scales, each with its own hand-typed limits, which is the free-scale small multiple in circular dress. One sorted, zero-based bar chart puts them all on the same ruler, in less room than the dials it replaces.