Cartographic Projection
every map is wrong, and by exactly how much
Theme

The thread: Purpose before property — page 6

Asking which projection is best is asking an incomplete question. Every one of them minimises something, and the only useful comparison is between a projection and the job it was chosen for. Essays 121 to 144 of 153.
Two conformal maps of one square: the worst case held, and the mean square held. Lines of equal scale inside a spherical square of 25° circumradius, every quarter of a per cent about each map's own mean; one ink above the mean and another below. Left, Chebyshev's map: the boundary is itself a line of equal scale, 1.44% above the mean, and the scale varies by 2.98% in all. Right, the map whose scale is closest to constant in the mean square over the area: the lines cross the boundary, the corners rise to 2.05% above its mean, and the scale varies by 3.47% — but its root-mean-square departure is 0.804% against Chebyshev's 0.871%. What each projection optimises

Chebyshev's map is the best at its worst, and not on average

Chebyshev's criterion makes a conformal map's scale constant on the boundary of a region, and no conformal map has a smaller range of scale inside it. The same series, fitted instead to make the scale as nearly constant as possible over the area, is a different map on every region but a cap. On a spherical square Chebyshev's map is 8.3 per cent worse in root-mean-square, the other map's range is 16.5 per cent wider, and every map between them trades one for the other.

 What a seven-parameter fit's residual is made of. A published transformation accuracy is the root-mean-square residual at the common points, and here it is 1.51 metres. Almost all of it — 1.51 — is the network's own distortion, which a rigid motion and a scale cannot follow and which is present at every point of the country whether it was used in the fit or not. The transformation's own error, measured as the disagreement between the fitted parameters and the true ones over a clean grid, is 0.091 metres: 6 per cent of the quoted figure. The last bar is the control — the same fit with the distortion switched off, at 2.9e-4 metres. What the numbers refer to

What another common point buys

A seven-parameter datum fit leaves a pattern rather than noise. Six per cent of the residual it reports is the transformation's own error and the other ninety-four is distortion no seven parameters can follow — so adding common points improves a term that was already small and cannot touch the one that is quoted.

 Where the summits are, and where the water divides. A stated range 100 km across and 80 km long, its summits 3,000 m high and about 35 km from the western edge, with a plateau falling gently to the eastern edge. Over 40 km of the range the plateau carries a broad rise up to 18 km east of the summits, and a gorge cuts through the range in the middle of that stretch. The dashed line runs through the highest summits; the solid line is the watershed, computed from the land. Lightly shaded ground drains west. Where they part, the darker ground lies east of the summits and drains west: 331 km², reaching 18.1 km from the summits. The terrain is stated, not surveyed. Grids, and what a survey does

The highest summits stay where they are, and the watershed does not

A treaty that puts a boundary on the highest crests which divide the waters names one line where the water divides on the summits and two where a river has cut back through the range. On a stated range the two part by up to eighteen kilometres. They also differ in kind: an error of ten metres in every height moves the summits three metres and the watershed on the plateau more than a kilometre.

 Three criteria, three maps, and the tolerance band each one encloses. A spherical square of 25° circumradius under three conformal maps, with the two contours at ±0.5% of true scale drawn heavy and the region between them the area each map keeps inside that tolerance. Chebyshev's map holds 37.0 per cent of the area there, the mean-square map 45.4, and the map fitted to the tolerance itself 64.1. The faint lines are further multiples of the tolerance, and they say where each map spends what it is not counting: the third map's outermost contours are much further out than the other two's. What each projection optimises

The map that keeps the most ground inside a tolerance

A surveyor does not want the smallest worst case or the smallest mean square. They want as much of the territory as possible within a stated tolerance of true scale and are indifferent to how far outside the rest goes. That is a third criterion, it is answered by a third map, and on a 25° square at half a per cent it keeps 64.1 per cent of the area where the mean-square map keeps 45.4 and Chebyshev's 37.0 — bought by sending a tail out to 5.6 per cent where Chebyshev's worst is 1.5.

 A bank that dries at one tide adds a bulge to the territorial sea. A stated coast 116 km long with five bays 5 km deep, and its twelve-nautical-mile limit measured from the low-water line at two tidal datums 1 m apart. A bank 10 km offshore has its crest 0.5 m below the higher datum, so it is under water at that datum and dries at the lower one, over a disc 1.0 km in radius. Dried and inside twelve miles of the coast, its low-water line becomes part of the baseline, and the limit bulges round it: 339 km² of sea. Everywhere else the two limits are drawn on top of one another, 11.1 km² apart over the whole coast. The coast is stated, not surveyed. Grids, and what a survey does

The sea is measured from a tide, and the headlands decide how much

A territorial sea is measured from the low-water line, and the low-water line is where a chart's tidal datum meets the shore — so a datum a metre lower moves it twenty metres at a headland and a kilometre across a tidal flat. On a stated coast the twelve-mile limit barely notices the flats: fifty-six square kilometres of shore are uncovered and eleven of sea are gained. A bank that dries at the lower datum adds three hundred and thirty-nine in one step.

 The triangle and the ellipse are two pictures of the same three sights. Three bodies at azimuths of 20°, 140°, 250° with independent errors of 2 km, drawn twice: as the cocked hat their lines bound, dashed, and as the 95 per cent confidence ellipse of the least-squares point, outlined. The ellipse is 4.25 by 3.78 km and covers 50.5 km²; the average hat from the same sights covers 7.06. The triangle is seven times the smaller and holds the ship a quarter of the time; the ellipse holds it 95 per cent of the time, which is what it was constructed to do. The hat is a picture of how much the sights disagree, and the ellipse is a picture of where the ship is. Paths and directions

A confidence ellipse is honest only where the sheet keeps angles

The cocked hat covers seven square kilometres and holds the ship a quarter of the time; the confidence ellipse from the same three sights covers fifty and holds it ninety-five per cent, which is what it was built to do. On a plotting sheet whose longitude is not corrected for latitude that ellipse fails in two unrelated ways — its centre is displaced, which costs coverage and depends on the navigator's own assumed position, and its shape is misread, which costs nothing but meaning and depends only on the latitude.

 What a folded net is accurate to, beside what the map it carries is wrong by. For a cube globe of 100 mm radius: each tolerance of cutting and folding a paper net, as a displacement on the assembled sphere, and the seam corner of each of the three face maps beside it. The fold's own error is the largest at 0.955° — 106 km at Earth scale — because a degree of dihedral error moves the far edge of a face by a degree of the face's own radius, which for a cube is 54.7°. The conformal face map's corner is 0.769°, below the whole budget of 1.000°; the gnomonic's is 20.2°, twenty times above it. The families

The map is finer than the paper it is folded from

Thirteen measurements price a polyhedral map in angles, areas and lengths, and none of them is about paper. A hand-folded net is accurate to about a degree of dihedral angle, which displaces the far edge of a cube's face by 0.955° of arc — 106 km at Earth scale — and that floor does not shrink when the globe is made larger, because a degree is a degree. Against it the conformal face map's seam corner is 0.769° on a cube and 0.0040° on an icosahedron, where it is 165 times finer than the sheet can be folded to.

 Straight baselines across a fringe of islands. A stated coast 220 km long, a gentle undulation 10 km in amplitude, with 8 islands off it: six between 5 and 12 km offshore and two rocks at 30 and 26 km. The dashed line is the twelve-nautical-mile limit measured from the low-water line of the mainland and every island. The solid lines are the straight baselines enclosing the most water under one reading of the article — segments of at most 24 nautical miles departing at most 20° from the coast's general direction — 6 segments, the longest 44.4 km — and the solid curve is the limit measured from them. Tinted: 988 km² of water landward of the lines, now internal waters. Shaded: 68 km² of sea beyond the dashed limit, against a territorial sea of 8,122 km². Neither rock is reached. The coast is stated, not surveyed. Grids, and what a survey does

A straight baseline moves more water inward than sea outward

The seventh article of the law of the sea lets a coast fringed with islands measure its territorial sea from straight lines joining chosen points, and gives no length, no angle and no scale for the choice. On a stated coast the lines a common reading permits add sixty-eight square kilometres of sea at the outer edge and turn nine hundred and eighty-eight into internal waters. Drop the test of direction and the outer rocks come into reach: eighteen hundred added, forty-six hundred enclosed.

 The same 2-pixel road at three latitudes, zoom 5. The dark bar is the mark as drawn — 2 pixels, identical in all three panels, because that is what the stylesheet says. The pale band behind it is the ground that mark covers, drawn to one common ground scale: 9.78 kilometres at the equator, 6.92 at 45° and 1.70 at 80°. The reader sees the dark bar and is being told about the pale one. What a machine does with it

The road is drawn two pixels wide

Seven rungs measure what a screen map does to position. Nothing on a map is a point: every mark has a width, the width is chosen in pixels, and a two-pixel road covers 9.78 kilometres of ground at the equator and 1.70 at 80° north. That is a generalisation applied at a strength varying by a factor of six across one sheet, by a stylesheet with no latitude in it.

 The same baseline, turned. The part of a 20 km height difference a degree-360 geoid model omits, against the direction the baseline runs, at four anisotropy ratios. A ratio of one is the model rung 9 used and is a flat line — the isotropic covariance cannot depend on a direction, by construction, which is the whole of the objection. At a ratio of two the same baseline omits 165 millimetres along the grain and 241 across it. What the numbers refer to

The correlation is not the same in every direction

Every number in the measurement before this one came out of Σ cₙ Pₙ(cos ψ) — a covariance that depends on the angular distance and nothing else. Ground has grain: at a modest anisotropy the same 20 km baseline omits 165 millimetres along it and 241 across, and the isotropic answer understates the worse direction by 18.4 per cent.

 Weighting the low sight turns the region the fix is in. Four bodies, three of them between 44° and 58° up on one side of the ship and one 5° up behind, with each sight's error made of a sextant part of 0.5′ and a tenth of its refraction correction: 0.94, 0.95, 0.93 km for the high bodies and 2.05 km for the low one. Solid: the 95 per cent ellipse of the fit that weights each sight by its variance, 2.30 by 1.55 km. Dashed: the ellipse a navigator draws who gives every sight the same variance, 2.63 by 2.02 km, its long axis turned 20.1° from the weighted one. Dotted: where that unweighted fix actually scatters, 2.98 by 1.57 km, turned 27.7° from the ellipse drawn for it. All three are centred on one point. Paths and directions

A low sight is worth keeping only if it is weighted

A body five degrees above the horizon is seen through so much air that its refraction correction is ten times a high body's, and a tenth of that correction uncertain doubles the sight's error. Given the same weight as the others, a sight below seven degrees makes the fix worse than throwing it away; weighted by its variance it never does. The ellipse drawn without weights is turned twenty degrees from the right one, and a navigator who must guess the weights loses far less by guessing high.

 The seam corner falls faster than the paper and crosses it at 720 faces. For an icosahedron globe of 100 mm radius with every face divided ν ways, the gnomonic face map's worst seam corner against what a hand-folded paper net is accurate to. On twenty faces the corner is 9.19°, 12.8 times the whole budget of 0.716°. The corner falls as ν to the power -1.97 and the fold's error as ν to the -0.97, so the two lines converge; the budget stops falling once the crease, a fixed quarter-millimetre radius of paper, is larger than the fold. The corner passes under it at ν = 6, 720 faces, where it is 0.315° against 0.321°. The families

The seam falls faster than the paper

Divide each face of an icosahedron ν ways and two things shrink at different rates. The fold's error falls as 1/ν, because it is the face's own radius times a degree. The gnomonic seam corner falls as 1/ν², because it needs both a tilt between faces and a distance along their edge. So the gnomonic, 12.8 times the whole assembly budget on twenty faces, goes under it at 720 faces on a 100 mm globe. No globe of any size needs more than 5,120, and what it costs is paper folded out of sight: 5.8 per cent of the sheet at that crossing.

 Twenty-four versions of one shape, and not one of them gains area. The same closed boundary rotated twenty-four times and simplified at the same tolerance. If the area error were noise the values would straddle zero and their mean would fall towards it; they do not. Every one is negative, the mean is -0.4644 per cent, and the mean is 71 standard errors from zero. A bias of that size cannot be removed by averaging over more boundaries, which is the only defence anybody has against a rounding error. What a machine does with it

A thousand features are wrong in the same direction

The area a simplification costs is unpredictable in sign for one feature. Over a population it is not: twenty-four presentations of one shape all lose area, the mean is seventy standard errors below zero, and no amount of aggregation removes it.

 Five readerships, four winners, one library. The best projection under each stated readership, with the share of the sphere carrying half that readership's weight. The criterion is Kavrayskiy's throughout, every map is in its normal aspect, and the library is the same twenty. Four different projections win. The flat weighting at the top is the one every earlier comparison used, and naming it as a readership rather than as the absence of one is the whole of what this changes: it is the assumption that everybody who will ever read the map is spread evenly over the globe. What is taught wrongly

The area weighting was a readership all along

Twelve measurements have taken a projection comparison apart and found the comparer's own freedom in every part of it. All twelve rested on one thing none of them examined: that a square kilometre of empty ocean counts exactly as much as a square kilometre of city. That is not the absence of an assumption about the reader — it is the assumption that everybody who will ever look at the map is spread evenly over the globe. Named as one readership among five, it produces a different winner from three of the other four, and against a North Atlantic readership its choice is fifth of twenty and 33 per cent worse.

 Cut into Goode's lobes, the ranking of uncut sheets turns over. Mean angular deformation over the sphere for eight equal-area pseudocylindricals, each read two ways: as one uncut sheet, and cut into Goode's lobes with every point measured from its own lobe's central meridian. Uncut, Eckert IV is the best named map at 27.70° and the one-term curve fitted to the uncut sheet beats Goode's composite by 6.83°. Cut, Eckert IV falls to 21.34°, second from last, the curve fitted uncut reads 20.77° against the composite's 18.48, and the plain sinusoidal beats the composite at 17.88°. A curve refitted for the cut sheet reaches 17.01°. The families

Scored lobe by lobe, the homolosine loses to its own lower half

Read as one uncut sheet, a single smooth curve beats Goode's composite by 9.6 degrees. But the homolosine is used cut into six lobes, so here it is scored cut, with each point measured from its own lobe's meridian, and the ranking turns over. Eckert IV, the best named map uncut, falls to second from last. The curve that beat the composite is now 2.3° worse than it. The plain sinusoidal, the composite's own lower half, wins at 17.88° against 18.48. The best smooth curve for a cut map is the sinusoidal with a few per cent of correction.

 Eighteen ways to close an archipelago within a hundred miles. A stated archipelago about 580 km across: three main islands near the centre (A, B and C, 23,576 km² of land between them) and thirteen smaller islands round them, 26,970 km² of land in all. The dot on each island is its outermost point, where a state drawing the widest lawful system would put its basepoint. Solid: the lawful system enclosing the most water, 186,617 km² at a water-to-land ratio of 6.96 to one, 12 segments, the longest 86.6 nautical miles. Dashed: the lawful system enclosing the least, 159,032 km², through the main islands A, B and C. Dotted: the two segments that would take in the outlier P to the south-east, 122.3 nautical miles and 100.9 nautical miles — both over the hundred-mile limit. The archipelago is stated, not surveyed. Grids, and what a survey does

Written in numbers, the archipelago rule still leaves eighteen

The article on archipelagic baselines gives its tolerances as numbers: no segment over a hundred nautical miles except three per cent of them, which may reach 125, and water between one and nine times the land. On a stated archipelago 8,799 closed systems keep the main islands inside, the ratio passes all of them and the length test eighteen, and the eighteen still differ by 27,585 square kilometres. Thin the land and the ratio does the deciding; the lower bound decides nothing; and the three-per-cent exception can be used only by a system padded to thirty-four segments.

 How long a network has before it breaks its own tolerance. The years until the worst baseline in a 400-kilometre network exceeds five millimetres plus one part per million of its length, which is an ordinary first-order specification. A boundary zone breaks it in 16 years; a plate interior takes 1819; a rigid plate never does, at any speed, because a rigid body keeps every distance it has. Bars are clipped at five thousand years. What the numbers refer to

A grid stops fitting the ground it was laid on

A hundred-kilometre baseline across a plate boundary changes by 815 millimetres in fifty years, and the same baseline on the fastest plate in the model changes by nothing at all in the same fifty. The interval before an ordinary first-order specification is broken is six years in one place and never in the other.

 One of these three lines does not slope. Three lengths at each zoom, for a world coordinate near the antimeridian. The pixel halves with every level, as it must. The spacing between representable double-precision values is far below it and halves with it. The spacing between representable SINGLE-precision values does not move at all — 2.0 metres at every zoom, because it is a property of the size of the number and the world does not get smaller. The two cross at zoom 17, and past it a vertex snaps to a lattice coarser than the pixels it is drawn into. What a machine does with it

The renderer runs out of numbers before the zoom does

Eleven essays price the pyramid in exact arithmetic. The pipeline that draws it carries single precision, where a world coordinate near the antimeridian quantises to two metres — at every zoom, because the number does not get smaller when the pixel does. At zoom 22 that is fifty-four pixels.

 The query, the box it becomes, and the 14 per cent it loses. A query for everything within 400 kilometres of a place at 55° north, drawn in the stored coordinates of Web Mercator. The ring is the true answer's edge; the rectangle is the box the index is given, sized from the scale factor at the query point. The arcs outside the rectangle are true neighbours the index never returns — 13.6 per cent of the rim, reaching 5.0 per cent of the box's own half-width beyond it — and nothing downstream can tell they are missing. What a machine does with it

The query a fast path actually answers

Fifteen rungs price what a stored coordinate means and none asks what it is searched with. No index answers "within two hundred kilometres of here"; an index answers "inside this rectangle of stored coordinates", and the rectangle is built by somebody's arithmetic — which at 55° north silently drops a tenth of the true answer on a conformal projection and three fifths of it on an equal-area one.

 The sea surface, against the surface heights are measured from. The stated mean dynamic topography — how far the sea stands above the geoid — drawn on Mollweide, with the eight tide gauges marked. It ranges over 2.09 metres, it is smooth, and it is what every national datum's zero is sitting on. Nothing here is a model's output: it is a stated closed form, chosen to have the observed sign structure and the observed size. What the numbers refer to

Every country's zero is a different surface

Ten rungs measure a height against a geoid and treat the geoid as one object. No national datum is on it: each is pinned to the mean sea level at one tide gauge, the sea surface stands up to two metres from the geoid, and the eight European zeros measured here spread over 462 millimetres — a step no levelling can remove and nobody's error.

 A ruler on a cartogram is not measuring anything. Twelve pairs of places, each measured on the ground and on the page of the triangular, x first cartogram of four cities, with the page scaled so that the median pair reads exactly right — the most generous calibration available. The line is that calibration. The worst pair, Tokyo to Quito, reads 21598 km for a ground distance of 14439 km, an error of 49.6%. The open marks are the same pairs on the equal-area map the cartogram was built from. Measuring distortion

Everything else on the page pays for the areas

A cartogram gets one quantity exactly right and every other reading a page supports is collateral. A ruler on it is out by 35 per cent after the most generous calibration available, and ten of sixty triples of places change which one is in the middle.

 Eight hundred surveys, each a poor estimate of the scale, find the patch where the published covariance is four times too optimistic. The middle one of nine agencies, by how much of the patch its map flags: a region covered by 800 local surveys at random, each tied to five control stations of a regional network whose published covariance is honest everywhere except a disc covering 15% of the region (dashed), where the control's true errors are 4 times its published variance; one survey in ten has a monument moved five centimetres. Filled dots: the 357 surveys whose own residuals put the scale above 2. Shaded: the 18 of 441 grid cells where the count of such surveys within 0.2 of the region's width exceeds what an honest region exceeds there one time in twenty, over the whole map; 18 of them lie in the disc. Grids, and what a survey does

An agency can map where its published covariance is optimistic

The local surveys tied to a regional network each estimate, badly, how far the network's published covariance understates the truth, and the agency that receives them can pool what they say. When the understatement is confined to one part of the region, pooling it into one number dilutes it: a patch covering 15 per cent of the region with errors four times their published size reads as 1.30 for the whole. Counted place by place, the same surveys find the patch — two agencies in three find it with 400 surveys and nearly every one with 800, flagging almost nothing outside it. Fed back as one number the scale removes little of the false alarms inside the patch and costs nearly a third of the test's power outside; fed back as a map it cuts the false alarms inside from 20 per cent to 3 and costs little outside.

 One smooth term beats the water's best join; the land needs three. Equal-area spacing curves with no join — the exponential of a cosine series in latitude — fitted to each surface on that surface's own lobes at its own best join, two lobes north and four south, each map at its own best proportions; the gain is over Goode's composite rejoined at the surface's best latitude (zero line). On the water to 88°, against the composite joined at 75° (12.27°): 1 term 12.03°, 2 terms 11.89°, 3 terms 11.73°, 4 terms 11.70°. On the land outside Antarctica, against the composite joined at 60° (13.83°): 1 term 14.31°, 2 terms 14.31°, 3 terms 13.71°. The families

A smooth curve beats the water's best join from its first term

On the land, a smooth equal-area curve with no join needed three terms before it beat Goode's composite rejoined at the land's best latitude, and on the land's own lobes it needs three again. On the water a single term beats the water's best composite, a second and third add more, and the best curve is 0.54 degrees of mean angular deformation better than the composite joined at seventy-five and 1.6 better than Goode's. It is nothing like the composite in shape: its parallels never close to less than seven eighths of the equator's spacing, so the map the water wants is the sinusoidal reshaped by a few per cent at every latitude, not the sinusoidal with a cap. The lobes do not move under it. And where a separate join for each hemisphere was worth three hundredths of a degree, a separate curve for each is worth a quarter of a degree.

 One rule, applied once, to one dataset. Two features 3 metres apart on the ground, and a snapping tolerance of 5 map units on Web Mercator. The tolerance's reach is the curve; the pair's separation is the line. They merge everywhere below about 52.8° and stay separate above it — the same features, the same rule, the same run. A pipeline that cleans a global dataset in one pass produces a topology that changes at a latitude nobody chose and nothing records. What a machine does with it

A tolerance in map units is not a tolerance

A snapping tolerance is a number, and the number is in whatever units the file is in. Five map units on Web Mercator is 4.97 metres of ground at the equator and 0.87 at eighty degrees — so a rule that merges two features three metres apart merges them everywhere below 52.8° north and refuses everywhere above it, in one pass, over one dataset, with nothing recording where the boundary is.

All threads