What Six Weeks of Autumn Actually Costs, Metro by Metro
Credit: Shutterstock
You have probably noticed it already. The streetlights come on while you are still finishing something. The dog walk that was comfortable in shorts last month now happens in the dark. Somewhere between the equinox and Halloween, the evening stops being a thing you have and starts being a thing you catch.
Here is what is actually happening, measured rather than described. We took the 50 most populous metropolitan areas in the country, pulled the coordinates of each one's principal city, and computed sunrise and sunset for every single day from September 22 through October 31, 2026 — 2,000 city-days in all. Then we ranked them by how many minutes of daylight disappear across those six weeks.
The spread is bigger than most people expect. Miami gives up 57 minutes. Seattle gives up 130. That is not a small regional difference; Seattle loses more than twice what Miami loses, over the exact same calendar. And the thing that decides it is almost entirely latitude — how far north you happen to be standing.
What follows is the whole table, counted down from the metro that barely notices to the one that loses the most. Before that, though, you deserve to know exactly how the numbers were made, because a figure like "130 minutes" is only worth something if you can check it. We have published the coordinates, the algorithm, the horizon convention, and what happened when we tested our own arithmetic against an independent authority.
One note on scope before we start. This is a measurement piece and nothing more. It tells you how much daylight your metro loses and when. It does not tell you what that means for you, and it does not suggest you do anything about it.
Three choices shape every number in the table: which cities, which coordinates, and which definition of daylight. All three are published below so you can reproduce any row you like.
The 50 metros were selected by population from the U.S. Census Bureau's Vintage 2025 estimates for metropolitan statistical areas, which carry a reference date of July 1, 2025. The cut line lands between Hartford at 1,171,426 residents and Buffalo at 1,155,653 — close enough that a different vintage could swap them.
Coordinates come from the 2025 Census Gazetteer Places file, using each metro's principal city rather than a metro centroid. A metro is a sprawling thing; the principal city is a place you can actually point at. The specific value used is the Census internal point, which is a coordinate guaranteed to fall inside the city's legal boundary. It is not a downtown marker and it is not city hall, and that distinction matters for a few cities in ways we spell out further down.
Every latitude and longitude we used is printed in the table. Nothing is hidden behind a rounding.
Daylight here means sunrise to sunset, and nothing else. We computed both using the solar position algorithm published by the NOAA Global Monitoring Laboratory, which derives from Jean Meeus's Astronomical Algorithms and assumes 0.833 degrees of atmospheric refraction — a solar zenith of 90.833 degrees at the moment of sunrise and sunset. In plain terms, the clock starts when the top edge of the sun appears over a flat horizon and stops when it disappears, with an allowance for the way the atmosphere bends light and lets you see the sun slightly before it has truly risen.
Civil twilight is not included. That is the single most common reason a reader's own app disagrees with a table like this one, and we will put a measured number on the gap shortly rather than leaving you to guess at it.
Day length itself is computed from the sun's hour angle, which means it does not depend on what any clock says. The sunrise and sunset times in the table do depend on the clock, so those use the time zone offset in effect on each date.
The same first-party approach sits behind other Daily Vitality datasets, including our look at how average sleep duration and bedtimes have shifted across four decades.
Read it from the bottom of the country upward. Number 50 is the metro that comes through the six weeks most intact; number 1 gives up the most.
The final column is the peak weekly rate — the largest amount of daylight any seven-day stretch in the window takes away. For every one of the 50 metros, that peak lands in the first week, September 22 to 29. The equinox is the steepest point of the whole year, and it is behind us the moment autumn starts.
Fourteen of the 50 metros lose 100 minutes or more. Seven lose 70 or less. The median across the whole set is 91.8 minutes, which is to say the typical large American metro gives up a little over an hour and a half of daylight between the equinox and Halloween.
By October 31, 45 of the 50 are already under 11 hours of daylight, and 14 are under ten and a half. Seattle finishes the window at 10 hours and 1 minute, the only metro in the set that comes within sixty seconds of breaking ten hours.
Set the extremes side by side with their latitudes and the relationship stops being abstract.
Across all 50 metros, total daylight lost rises by roughly 3.3 minutes for every additional degree of north latitude, and latitude alone explains nearly all the variation — the correlation between the two runs to 0.996. Longitude does not matter. Elevation barely matters. Your distance from the equator is the whole story.
That gives you a rule of thumb for any city not in the table. Find its latitude, find the nearest metro above and below it on the list, and you are within a couple of minutes.
Here is the finding we did not expect, and the one that probably matches your experience best.
The loss is not split evenly between the two ends of the day. In every single one of the 50 metros, sunset moves earlier faster than sunrise moves later — and the imbalance is not subtle. Between 57 and 66 percent of the lost daylight comes off the evening.
Seattle is the clearest case. Its sunrise slips 56 minutes later across the six weeks, from 6:57 a.m. to 7:53 a.m. Its sunset moves 74 minutes earlier, from 7:07 p.m. to 5:53 p.m. Miami's split is even more lopsided in proportional terms: 19 minutes off the morning, 37 off the evening.
The reason is that solar noon itself drifts later through October. The sun's apparent position runs slightly ahead of clock time in the fall, which pushes the midpoint of the day later and hands a disproportionate share of the shortening to the back end. It is the same mechanism that produces the earliest sunset of the year in early December rather than at the solstice.
So if the autumn shift registers as an evening phenomenon rather than a morning one, the table agrees with you.
Nearly everyone assumes the equinox delivers exactly twelve hours of light and twelve of dark. It does not, and our own numbers show why the gap is worth understanding.
On September 22, 2026, our 50 metros range from 12 hours 7.9 minutes of daylight to 12 hours 10.6 minutes. Every one of them clears twelve hours, and none of them clears it by much. The National Weather Service explains the mechanism: refraction bends sunlight so the sun appears above the horizon while its actual position is still below it, which is why equinox day length runs from roughly 12 hours 6.5 minutes at the equator to about 12 hours 16 minutes at 60 degrees latitude.
Our figures land inside that published band for the latitudes we cover, which is a useful independent check on the refraction convention we chose.
There is something else in those numbers worth pausing on. On September 22, the daylight difference between Miami and Seattle — separated by nearly 22 degrees of latitude and the entire width of the country — is 2.7 minutes. By October 31 that same gap has grown to 70.4 minutes.
The equinox, in other words, is the one day of the year when the country agrees about how long the day is. Six weeks later it does not agree at all. The countdown above is really a picture of that disagreement opening up.
Day length and clock time are two different things, and the table keeps them separate on purpose.
The measurement window was chosen partly so no clock change falls inside it. Daylight saving time ends in the United States on November 1, 2026 — the day after the October 31 endpoint. Every metro that observes daylight time is on it for both endpoint dates, which means nothing in the "minutes lost" column is an artifact of an hour that moved. That column is pure astronomy.
There is a related wrinkle worth knowing about the start date. The U.S. Naval Observatory places the September 2026 equinox at 00:05 UTC on September 23, which sounds like the wrong day until you convert it: 8:05 p.m. Eastern, 7:05 p.m. Central, 6:05 p.m. Mountain, and 5:05 p.m. Pacific — all on September 22. The equinox falls on the 22nd in every U.S. time zone, which is why the window starts there.
Two entries in the table show how much wall-clock times can mislead.
Phoenix ranks 40th in daylight lost, giving up a modest 78 minutes. Yet on October 31 it records the earliest sunset of all 50 metros — 5:37 p.m., ahead of even Boston. Arizona does not observe daylight saving time, so while its neighbors spend October on daylight time, Phoenix has effectively been an hour "earlier" on the clock all along. Its days are long. Its evenings just end sooner by the clock on the wall.
Grand Rapids runs the opposite trick. It records the latest sunrise in the set on October 31 at 8:16 a.m., followed by Indianapolis at 8:12, Louisville at 8:08 and Detroit at 8:05. All four sit near the western edge of the Eastern time zone, where the clock runs ahead of the sun. None of that changes how much daylight they get. It only changes when the daylight shows up.
Daylight loss is often described as a steady subtraction, as though autumn removes the same couple of minutes every day until the solstice. It does not. The rate is fastest at the equinox and eases from there.
Because we computed all 40 days rather than just the two endpoints, the curve is measured rather than assumed. Taking differences across rolling seven-day windows:
The flattening is real but gentle, which is the honest version of the story. You are not imagining that the second half of October feels less abrupt than the first week of autumn — but the difference is a minute or so per week, not a cliff. The genuine deceleration arrives closer to the December solstice, which the Naval Observatory dates to December 21 at 20:50 UTC.
A computed table is only as trustworthy as its verification, so here is ours in full.
We cross-checked eight cities — Miami, Houston, Phoenix, Denver, New York, Chicago, San Francisco and Seattle — against the U.S. Naval Observatory's sun and moon data service, on both endpoint dates, using exactly the coordinates in our table. That is 16 independent checks spanning 25.8 to 47.6 degrees north and five time zones.
The maximum discrepancy we observed was one minute on sunrise, one minute on sunset, and one minute on total day length — and the day-length disagreement occurred in only a single case, San Francisco on October 31. Every other day-length figure matched exactly. That sits inside NOAA's own stated accuracy for its algorithm, which it puts at within a minute for locations between plus and minus 72 degrees latitude.
Worth knowing about our own source: NOAA now carries a notice that its Solar Calculator is no longer actively supported or maintained, and states the tool is for research and recreational use only. The published Meeus-derived equations are what we implemented, and they remain sound; the web tool is simply no longer being updated. For anything requiring certified sunrise or sunset data, NOAA itself points to the Naval Observatory.
If you check a figure here against your phone, you may find a longer day. The usual reason is twilight.
Using the same Naval Observatory responses that validated our table, civil-twilight-inclusive figures run 45 to 64 minutes longer than sunrise-to-sunset across our eight validation cities, averaging 54 minutes. In New York on September 22, civil twilight spans 6:16 a.m. to 7:20 p.m. — 784 minutes — against 730 minutes of sunrise to sunset.
Both numbers are correct. They answer different questions. Ours answers "how long is the sun above the horizon," not "how long can you see without headlights."
Every measurement makes simplifying assumptions, and yours deserve to see ours.
Elevation is not modeled. We assume a flat, sea-level horizon everywhere. For a mile-high metro this understates absolute daylight: modeling Denver's elevated horizon would add roughly 12 to 13 minutes of daylight on each endpoint date. Reassuringly, it barely touches the ranking — Denver's six-week loss would shift from 98.0 minutes to 97.4, a difference of 0.6 minutes. That estimate uses a standard horizon-dip approximation and is labeled as an estimate, not a measurement.
Terrain is not modeled either. If a ridge sits west of you, your sun sets earlier than any table says. Astronomical sunrise and sunset assume an unobstructed horizon, which almost nobody actually has.
Coordinates are internal points, not downtowns. This produces a few entries that look wrong until you know why. San Francisco's Census internal point is 37.7272, -123.0322 — which is out in the Pacific, because the city's legal boundary includes the Farallon Islands about 27 miles offshore. The Gazetteer record gives it away: 46.7 square miles of land against 185.2 square miles of water.
We left it in place rather than quietly swapping in a downtown coordinate, and then measured what it costs. Using a City Hall location instead changes San Francisco's six-week loss from 91.04 minutes to 91.21 — a difference of about ten seconds. New York's internal point similarly lands in Brooklyn, and Denver's is pulled east by the airport annexation. In each case the effect on day length is far smaller than the effect on your eyebrows.
Three metros use consolidated-government records. Indianapolis, Nashville and Louisville resolve to their "(balance)" entries in the Gazetteer, which is the functioning municipal entity in each case.
For readers who like this kind of first-party counting, our ranking of which U.S. states report the highest rates of sleep deprivation uses the same show-your-work approach.
About the Author
Ryan May is a digital publishing leader with more than 25 years of experience turning complex information into clear, useful editorial work. At Daily Vitality, he focuses on evidence-led reporting about nutrition, fitness, healthy aging, and long-term wellbeing, informed by his own interests in cooking, yoga, cycling, and sustainable healthy habits.
Seattle, by a clear margin. It gives up 129.7 minutes — from 12 hours 11 minutes on September 22 to 10 hours 1 minute on October 31. Portland is second at 120.4 minutes and Minneapolis third at 118.0.
Miami, at 56.6 minutes. It is the southernmost principal city in the top 50 at 25.78 degrees north, and it ends the window still above 11 hours of daylight — the only metro in the set that does.
Longitude shifts when the sun rises and sets on your clock, but not how long it stays up. Latitude changes the angle at which the sun's path crosses the horizon, and that angle is what governs how quickly day length changes through the seasons. Across our 50 metros, latitude and minutes lost correlate at 0.996.
Not at all. Daylight saving time ends on November 1, 2026, one day after the measurement window closes, so no clock change falls inside it. Day length is also clock-independent by definition — moving the clock changes the labels on sunrise and sunset, not the interval between them.
Almost always because the app is reporting twilight-inclusive figures. Measured against the same Naval Observatory data we validated with, civil twilight adds 45 to 64 minutes depending on latitude. Our figures are strictly sunrise to sunset.
No. Every one of our 50 metros sees between 12 hours 7.9 minutes and 12 hours 10.6 minutes on September 22. Atmospheric refraction lets you see the sun slightly before it has risen and slightly after it has set, which tips the balance toward daylight. The date of genuinely equal day and night, sometimes called the equilux, falls a few days after the equinox at mid-latitudes.
Faster now. The peak weekly loss in every one of the 50 metros lands in the first week after the equinox. Seattle sheds 23.9 minutes in that first week and 22.6 minutes in the October 20–27 window; Miami sheds 10.5 and then 9.8. The rate eases gradually toward the December solstice.
Because Arizona does not observe daylight saving time. Phoenix has plenty of daylight — 10 hours 50 minutes on October 31 — but its clock has not been shifted forward the way its neighbors' clocks have, so its 5:37 p.m. sunset is the earliest in the set. It is a clock effect, not an astronomical one.
Across 16 independent checks against the U.S. Naval Observatory, our maximum discrepancy was one minute on sunrise, one minute on sunset, and one minute on total day length, the last occurring in a single case. Every other day-length figure matched exactly.
Yes. Find its latitude, then locate the metros immediately above and below it in the ranking; total loss climbs by about 3.3 minutes per degree of north latitude, so interpolating between neighbors will get you within a couple of minutes. For an exact figure, the Naval Observatory's sun and moon data service will compute any coordinate you give it.
