Where the Shadow Fell

What it is

The globe carries 11,898 points, one for every solar eclipse in NASA’s Five Millennium Canon of Solar Eclipses. Each sits at that eclipse’s greatest: the place where the Moon’s shadow axis passed closest to the centre of the Earth. The shape of a point is the type of eclipse, how full it is tracks the magnitude, and its colour and brightness say when it happened: recent ones warm and bright, ancient ones cool and faint.

Hover a point and the piece draws the path that eclipse swept across the world. For every total, annular and hybrid eclipse in the five millennia, this is the real track of the shadow, not a sketch of one.

Then there is the personal layer. Name a city and the field narrows to the eclipses that actually arrive there: how much of the Sun each one takes, when totality last stood over that place, when it will next, and the whole set laid out along a timeline. Step through into City Sky and you are standing in that city, looking up, with each eclipse hung where its Sun really was, at that altitude, on that bearing, over that horizon.

Questions

The same eclipse, seen from somewhere

A catalogue row describes an eclipse at its greatest: the instant the shadow axis passed closest to the centre of the Earth. For 12 August 2026 that point lies off the west coast of Iceland, the row reads total, and the track where it is true is about 294 km wide. Everywhere else on the daylit side of the planet, that same eclipse is something less.

So the piece solves each eclipse again at the coordinates you give it, from the Canon’s own Besselian elements. What it reports is obscuration: the fraction of the Sun’s disc area the Moon covers, which is what the phrase “how much of the Sun” describes. Eclipse magnitude, the number more often quoted, is a fraction of the diameter, and on a shallow partial it runs close to twice as high. New York on this date is 9.45 % obscured at magnitude 0.19. The piece always says which of the two it is showing.

One row, then, on the afternoon of 12 August 2026. Reykjavík: total. Madrid 99.95 %. Paris 92.07 %. Saint Petersburg 79.12 %. New York 9.45 %. Moscow 3.16 %. In Tokyo and Cairo the Sun is already below the horizon when it happens, so the piece reports nothing for them rather than zero.

One line in the catalogue, and no two cities get the same afternoon.

Standing in the city

Naming a city also opens the view from the ground. City Sky puts you in that city looking up, with each eclipse placed where its Sun actually stood: at that altitude, on that bearing, at that hour. The Sun that goes out is in the part of the sky it really occupied, which is the reason the mode exists.

The skyline is the city’s own. Where the piece has a built horizon profile it uses it, so the Sun sits behind the buildings and the ground that are actually there — the profile is measured in 5-degree steps around the full circle, from an eye height of 30 metres. Where a city has no profile the horizon falls back to flat, and the piece draws no skyline rather than inventing one.

The four kinds of shadow

Every eclipse in the record is one of four kinds, and what separates them is a few thousand kilometres: whether the tip of the Moon’s shadow reaches the ground, stops short of it, or misses the Earth entirely.

That there is anything to separate is a coincidence. The Sun is about 400 times wider than the Moon and about 400 times farther away, so the two discs come out very nearly the same size in our sky. A few per cent either way — and the Moon’s distance swings by more than that every month — is the whole difference between a black Sun and a ring of fire.

Each kind is set out here in two steps: the geometry that makes it, and the Sun as it looks from inside the mark that geometry leaves. Between them stood a third figure once — the band, blown up on its own patch of ground — and it went on 3 August 2026, because it said nothing the corridor does not say better in place, on the globe, at the size it really is. The numbers it stood for are all still here, in the words.

On scale. Nothing in the geometry column is to scale, and no diagram of this kind ever is. Against the real distances the gaps are about 25 times wider than drawn: the Sun stands 107.5 solar diameters from the Earth and the Moon 110.7 lunar diameters, against 4.5 and 4.8 here, and the bodies are out of proportion with them. What that column does keep is the topology — where each cone ends, and what that means for the ground — and that is the only claim it makes.

Solid, ceilinged, screened, or travelling between the first two: the pattern says which kind of shadow arrived and the weight says how much it mattered. The umbra and antumbra carry the obscuration itself, as depth; the penumbra is one screen at one weight, held deliberately below them. And within every one of them darkness still means time — older eclipses fainter, recent ones darker.

The rings at the poles

Two rings stand out on the globe, one around the Arctic and one around Antarctica, where the points crowd close together. That is real, and it is one kind of eclipse doing it. All 4,200 partial eclipses in the record sit at high latitude. In a partial eclipse the Moon’s shadow misses the Earth: its dark axis passes above or below the globe, and the moment of greatest eclipse is marked where that axis comes nearest the surface, which is always near a pole.

The total, annular and hybrid eclipses, whose shadows do reach the ground, fall everywhere else, across the tropics and the middle latitudes. The rings are the eclipses that only grazed the Earth, collected at its two ends.

One family, one saros

Eclipses don’t repeat at random. Every one in the record carries a Saros number: the family it belongs to. Members of a family recur roughly every 18 years, 11⅓ days, each time landing about a third of the way further west than the last. A family begins near one pole as a string of grazing partials, then matures into a long run of totals or annulars that can last well over a thousand years. It ends the same way, fading back into grazing partials as it reaches the other pole.

The eclipse of 12 August 2026 belongs to series 126. Seen all at once, across the whole five millennia rather than at its own moment, a family stops looking like scattered noise in the field and resolves into one drifting diagonal chain: the same alignment, returning, generation after generation. Every third return comes back near the same meridian, a step further along.

Any family can be lit on its own. The grid screen has a field that takes a series number, and clicking a number on the saros rail does the same thing. Either way the address bar picks up that series, so the link in it goes straight back to the family. It works on a phone too, where the grid itself doesn’t fit.

The catalogue as a grid

A saros is one family. Set every family side by side and the catalogue stops being a list. Each eclipse carries a saros number and a lunation number, and from those two its inex interval follows exactly, which gives all 11,898 records a row and a column. The globe unfolds into that grid, 204 columns wide, and folds back again.

What the grid is for is that it is regular enough to read forward. Step along a row and you move 18 years and 11 days at a time through one family, on into eclipses that have not happened yet. Reading it forward is how an eclipse is known centuries before anyone now alive will see it.

Every record is checked against the grid as the piece loads; a single one that was not an exact lattice point would switch the mode off. It needs a wide screen, though. Below about 900 pixels a column is thinner than four of them and there is nothing left to read, so the grid stays closed.

The day is the subject

An eclipse happens in daylight. That is the whole strangeness of it: the light goes out in the middle of an ordinary afternoon, and then comes back. So the piece rests in day: a warm, bright, starless world where the eclipses read as small dark marks, shadows on a lit ground.

Darkness only arrives when an eclipse does.

Hover a total and the daylight drains away, the real stars come out, and the black disc with its corona stands exactly where that eclipse peaked. Let go and the day returns. Day, twilight, day: it is the interaction and the argument at once.

The card you can take away

Whatever a city’s reading comes to, you can keep it. The piece builds a 1080 × 1080 card for that place — its own globe, how many eclipses reached it between 1700 and 2100, when totality last stood over it and when it is next due, and 12 August 2026 worked out for those coordinates — and hands it over as a PNG.

It is made where you are standing. The card is drawn in your browser out of data the page has already loaded, and rasterised there. Nothing is uploaded, nothing is rendered on a server, and there is no server here to render it: the piece is a set of static files and everything above happens on your machine.

What is computed, and what is not

The catalogue is NASA’s, by Fred Espenak and Jean Meeus, covering −1999 to +3000. Its 11,898 records (date, position of greatest eclipse, type, magnitude, duration, Saros series) are read verbatim; nothing is smoothed or filled in.

The shadow paths are not traced by hand. They are forward-modelled from the Canon’s own Besselian elements, the shadow axis intersected with the Earth ellipsoid, by the method in Meeus’s Astronomical Algorithms, chapter 54.

The width of a band is measured too, and separately at every point along it. A corridor is not one width: the shadow meets the ground at a different angle at every moment, so the same eclipse can be 187 km across where it is deepest and 245 km where it meets the horizon. Both limits you see are offset from the centre by the width solved for that spot, which lands on NASA’s own published limits to about a kilometre. Nothing is widened to make it easier to see and nothing is narrowed for shape, so a town on the edge of the line is on the side the geometry puts it.

The thin pale lines fanning out from a band are contours of the covered fraction of the Sun, and they are measured the same way everything else here is: the piece solves the eclipse at a point, and each line is the set of points where the answer comes out at 90, 80, 70 per cent and so on down to 10. They are solved along the whole track and separately on each side of it, because the two sides genuinely differ — on 12 August 2026 by as much as thirty points on one line. The quantity is the area of the Sun covered at the deepest moment of the eclipse there, which is the same quantity every percentage on this site quotes. Espenak’s classic maps contour magnitude instead, the fraction of the Sun’s diameter, which is always the larger number; NASA’s own recent maps contour the area, as this does, and say so on the page. Sampled along the drawn lines and asked again, they carry their own label to within a tenth of a percentage point on average and 1.2 points at worst. Where NASA has published the curves themselves — the 2017 map, whose contour geometry is downloadable — ours sit 1.7 to 3.5 km from theirs.

They also stop, and where they stop is not a shortcut. A contour of “how deep it got here” means nothing at a place where the deepest moment happened after the Sun had set, so the lines end at the curve where maximum eclipse falls on the horizon — the same edge Espenak’s maps draw them between. On 12 August 2026 that is most of the ladder on the evening side: those lines are not somewhere off the coast of Africa, they are not anywhere, and the field they belong to does not fade to zero there so much as run out. The two sides of that eclipse are not alike, either, and it is worth saying by how much: walking outward from greatest eclipse, a westward line runs a median 5 689 km and the covered fraction really is down to nothing when it stops, while an eastward one stops after 3 444 km with a median 83 % of the Sun still hidden. Ninety-four per cent of the eastward lines are cut while more than a fifth of the Sun is still covered. What ends along that edge is the day, not the eclipse — which is why it looks like a wall on the map and a sunset from the ground. The ladder of levels is uneven for that reason. Even steps of ten points would put seven of its twelve lines on ground that, east of this track, no eclipse ever reaches: walking outward from greatest eclipse toward Warsaw the day ends at 2 820 km with 83 % of the Sun still hidden, toward Berlin at 2 890 km at 85 %, toward Moscow at 2 665 km at 79 %. Of 360 lines walked eastward, 318 leave the map still deeper than 60 %. So the ladder is closer spaced where that half of the world actually is — 95, 90, 85 and 80 % — and the shallow end, which only exists out over the Atlantic, keeps its wider steps. Toward New York the same walk crosses twelve lines; toward Berlin it crosses three, and before this it crossed one. Two smaller things are also true and worth having in writing rather than in the source. The lines are drawn through solved points and straight between them, so between two points a line is a chord and not a curve; that is the half-point above. And where a contour sits so far out that its own construction folds back on itself — which happens on this eclipse to one line, on one side, where the Sun’s covered fraction barely changes over a thousand kilometres and the line’s position stops being pinned down — it is broken rather than drawn through the fold.

Where the shading stops, it stops on that same curve. It is not drawn outward from the centre line any more but read off the covered fraction itself, sampled across the ground and cut along the same twelve levels as the lines — so it ends where the field ends and nowhere else. On 12 August 2026 that leaves Warsaw, Vilnius and Vienna unshaded: at all three the deepest moment falls with the Sun already below the horizon, so there is no maximum to draw, and Espenak’s own maps stop at the same place. The Sun was still 75 to 85 % covered on the way down at each of them, and the percentage on a city card is solved at your own coordinates — it holds where the shading does not reach.

The same geometry does the personal work, and this is the part worth stating plainly: when you choose a city, the piece solves the local circumstances at your coordinates, rather than simply reporting what type the eclipse was somewhere. It works out how much of the Sun was covered there, at what hour, and whether the Sun had even risen. An eclipse total over the Pacific is a shallow partial over Paris, and the card says so. Eclipses whose Sun never cleared the local horizon are dropped rather than counted.

Air bends light near the horizon and this piece does not model it. The correction lifts a setting Sun by about half a degree, so where the Sun goes down during an eclipse the real event runs three or four minutes longer than the times given here, and covers more of the Sun than the percentage says. Rome on 12 August 2026 shows the size of it: the geometry gives 69 %, and a watcher in Rome will see about 76 %. Espenak leaves refraction out of his own circumstances and every figure here is checked against his, so it is left out here too.

The hour on a card is the local civil clock: what a watch in that city reads, with summer time where it applies. Standard time zones are a 19th-century invention, so for an eclipse older than its city’s zone the hour is local to the place itself.

Two things decide what a city keeps. The first is a floor and it does not move: an eclipse that covered less than 5 % of the Sun’s area at your coordinates is not drawn anywhere in the piece. Five per cent of the area is the same quantity the city percentages quote. Written as magnitude, the fraction of the Sun’s diameter, the same eclipse reads about 12 %.

The second is yours. A city opens showing the partials that covered over 50 % of the Sun, and the two words under PARTIAL on the chart are where that is set. Press any depth and every partial above the floor comes back. Paris opens with 58 of the 147 eclipses that reached it, and the other 89 are one press away. That 50 is the area again, the same quantity as the floor and as the percentages on the card. Totals and annulars stay at either setting, since no depth cut can reach them.

A card you download does not follow this control. It holds to a stricter 85 % of the Sun’s area on its own, regardless of where the screen’s setting sits when you press the button — the two are independent numbers for two different artefacts, a page you are looking at and a picture you keep.

So the count on a city’s card is the count of what is drawn, and it is shorter than the record behind it unless you have pressed any depth. A card you download says neither which setting it was made at nor what the other number would have been. This paragraph is where that is said.

Above the floor nothing is dropped for being shallow. Depth sets the ink: below three quarters of the Sun’s diameter covered, the shallower it was the paler it draws, so a grazing partial sits as texture and a deep one carries over it. The three-quarter mark is Espenak’s own. The astronomer who computed this catalogue draws that line himself, in his catalogue of “major” eclipses, and the word he uses for them is major.

This piece used to cut the set at 85 % of the Sun’s diameter and show you what was left, which made a city’s count smaller than its history without saying so. That is worth admitting rather than quietly fixing. At that floor Paris’s chart carried 15 of the 131 eclipses that had reached it, and 12 August 2026, the date this piece was built for, was missing from New York’s.

Sun positions in the city sky (altitude and azimuth for one place at one instant) are computed the same way, with the gap between Terrestrial and Universal Time carried explicitly, because over five millennia it grows to hours.

Universal Time is the world’s shared clock: the time on the Greenwich meridian, with no summer time and no zone shifted onto it. When this piece names the moment an eclipse peaked — 17:46 UT on the card the globe raises — that is the instant it happened, given on one clock for the whole Earth. It is not the time on any particular clock on the ground: the same instant is late afternoon in Iceland and the middle of the night in Tokyo. A card about a city gives that city’s own wall time instead, summer time included, because there the question is what time you would have looked up.

Which is also why most eclipses here are given a date and no hour. The catalogue dates an eclipse on a uniform clock — one that ticks evenly, the clock the orbits are computed on. Universal Time is tied to the turning Earth, which does not turn evenly, and the gap between the two has to be recovered from records of eclipses people actually watched. Far from those records it is estimated, and the estimate is soft: at the start of this catalogue the gap is nearly thirteen hours and is itself uncertain by about an hour. An hour of the Earth’s turning is the 16° of longitude named below, and printing 14:21 against it would be naming the minute when we do not reliably know the hour. So the rule is this: a time given to the minute claims to be good to a minute, and the hour is printed only over the years where the uncertainty stays under one — 954 to 2115. That is 2,750 of the 11,898 eclipses here, and all of the ones inside the 1700–2100 the piece opens on. Outside it the date stands on its own, and the date is not in doubt.

One consequence of that gap looks like a mistake and is not. Because the two clocks are hours apart in the deep past, a single instant can fall on different calendar days on the two of them: an eclipse the catalogue dates to 12 June happened, in Universal Time, late on the 11th. Inside the window above the gap is minutes rather than hours, so only eleven eclipses in the whole catalogue still show it — ones that peak within a few minutes of midnight, where the card reads 23:56 UT, 17 Jan under a heading of the 18th.

That gap is also where the piece runs out of certainty, which is worth saying plainly. ΔT can be measured only against observations, and there are none ahead of us and few far behind; outside the observed record it is extrapolated, and the further from that record a date sits the softer the extrapolation gets. What it moves is the Earth’s rotation under the shadow, so it shifts where a given instant lands rather than when the eclipse occurs. For a date thousands of years out the geometry holds and the date is right; the meridian it is pinned to is good to about a degree of longitude between roughly 100 and 2300 CE, widening to some 8° at this catalogue’s far end in 3000 CE and some 16° at its start in 2000 BCE, on Espenak’s own published figures for that uncertainty.

ΔT itself is inherited rather than computed. One value per eclipse is read from the same NASA page the Besselian elements come from and reproduced to a third of a second, so the shadow this piece draws is the shadow Espenak computed. His figures follow Morrison and Stephenson’s 2004 model, and work published in 2016 and 2021 has revised it since: on the newer consensus the oldest tracks in this catalogue would sit about 2.7° of longitude from where they are drawn. That is well inside the 16° stated above. The estimate at the centre of that band has moved, and anyone checking these paths against a present-day source should know which model they are on.

The stars are real stars: every one down to naked-eye magnitude 6.5, 8,920 of them, at their true positions, with their true brightness and colour.

What the corona is, and what it is not

The corona that appears at totality is an illustration, not a photograph of the eclipse you are hovering. No eclipse here wears another eclipse’s corona.

What it obeys is real. Its radial falloff follows the classical white-light measurements of Baumbach and van de Hulst. Its overall shape follows the solar cycle: round, with streamers at every latitude near maximum; flattened into wings, with polar plumes, near minimum. That cycle is driven by the observed sunspot record, month by month back to 1749. Its fibrous micro-texture is derived from a real photograph of a real corona, flattened and stripped of its large-scale structure so that only the grain survives.

Outside the record, the piece makes no claim. For an eclipse in 1200 BCE or in 2400 CE nobody counted the spots, so every cycle-dependent feature collapses to a neutral middle: the corona is drawn, but it argues nothing about a Sun no one watched.

A note on dates

Dates before 15 October 1582 are Julian-calendar dates, as published in the NASA canon and as historians write them: the Thales eclipse is 28 May 585 BCE. From that date onward they are Gregorian.