Sharp-eyed readers of the HD Map Handbook pointed out something we had not made explicit enough: geographic coordinates are as much about time as they are about place.
That sounds strange at first. A coordinate tells us where something is. What could time have to do with it?
Quite a lot, as it turns out.
The Earth beneath our feet is not a perfectly static reference surface. Tectonic plates move. Reference frames evolve. Different coordinate agreements can place the same physical point a meter or two apart while both coordinates are perfectly correct inside their respective systems.
For an autonomous machine, that distinction matters.
A map is supposed to tell a machine where it is and where it can go. But if the coordinate system doesn’t also tell us relative to what and when, two perfectly respectable coordinates can describe subtly different versions of reality.
And that is exactly the kind of subtlety that can turn into a very non-subtle bug.
A coordinate is an agreement
One of the ideas we’ve added to Chapter 5 is a sharper definition of what a coordinate actually represents.
A coordinate isn’t simply a number attached to a point on Earth. It is an agreement about how that number relates to the physical world: we will measure position relative to this origin, along these axes, in these units, at this date.
That last phrase is easy to overlook.
We tend to think of a reference frame as something fixed: establish the origin, define the axes, publish the transformation, and you’re done.
But the Earth’s surface doesn’t cooperate.
Tectonic plates move continuously, so reference frames have to account for that motion, and modern geodetic systems increasingly make the time component explicit.
In other words, a coordinate carries a field we don’t always write down. It answers two questions at once: where is it, and when?
Datums have dates
Chapter 6 now says so out loud, in a new section on datums and dates.
WGS84 is not one frozen realization of the Earth’s coordinates. It is a family of frames, re-anchored as measurements improved and the planet kept moving, which is why a coordinate can carry a tag like WGS84 (G1762): the GPS week that variant was aligned to.
North America has a second answer: NAD83 (the North American Datum of 1983), re-readjusted in 2011 into a frame fixed to the North American plate. The two describe the same patch of ground a meter or two apart, and both are correct inside their own agreement.
The frames also creep. Positions in a frame fixed to the Earth as a whole move past a plate-fixed survey mark by a few centimeters a year.
That leads to one of the simplest lines we’ve added to the Handbook:
Leave the date off and the map does not fail, it ages.
That is a much better way of thinking about temporal reference than treating it as an obscure geodesy detail.
Your map doesn’t suddenly become wrong at midnight on some particular date. It gradually becomes a description of a physical world that is no longer quite the same world.
Height above what
There is another coordinate question that is easy to miss: height above what?
Latitude and longitude get most of the attention because they are familiar. But a three-dimensional coordinate needs a vertical reference as well.
Chapter 6 gives that relationship its own section and reduces it to one line:
H = h - N
H is orthometric height, the elevation a surveyor sets a benchmark to. h is ellipsoidal height, what the GNSS receiver reports. N is the geoid undulation, the gap between the two surfaces, read from a geoid model covering the site.
We worked it through with the Handbook’s Harbor Yard example, including both directions of the failure. Skip the conversion while georeferencing the map and the vehicle sits 22 m below the pavement the cloud was captured on. Skip it on the vehicle side instead and the vehicle floats 22 m above it.
The important point isn’t memorizing the equation. It’s understanding that “altitude” is not a sufficiently precise description of a physical height.
A machine can be given a perfectly valid height value and still be given the wrong answer for the question it actually needs answered. That is another reason the coordinate itself needs context.
Four questions before you trust a coordinate
The revised Chapter 6 closes its argument with a simple checklist:
Where?
Relative to what?
When?
Height above what?
There’s a nice relationship here with an earlier idea in the Handbook. Chapter 2’s four questions ask where the vehicle stands. These four ask what a coordinate means.
That distinction is useful because it turns what can feel like abstract geodesy into a practical engineering habit.
Before trusting a coordinate, ask what agreement produced it. Ask which reference frame it belongs to. Ask what date or epoch applies. And ask what physical surface its height is measured from.
The map ages even when the road doesn’t
This is perhaps the counterintuitive part.
If you drive back to the same Harbor Yard five years from now, the road probably hasn’t moved in any way that a human driver would notice.
But the geodetic coordinates assigned to that road can change relative to a different reference frame. Nothing is “wrong.” The physical place is still the physical place, and the coordinate is still a valid coordinate.
The problem only appears when we quietly combine coordinates defined under different agreements, reference frames, or dates and assume they are describing exactly the same thing.
For an autonomous system, that assumption can be dangerous. The machine doesn’t have our intuitive sense that two maps are “basically the same place.” It needs the spatial relationships to be explicit.
That is why these details matter.
What we changed in the Handbook
These updates came from readers pointing out places where the explanation could be sharper, so both chapters have been revised, Chapter 6 the deeper pass.
Chapter 5 now stamps the coordinate agreement itself with a date, gives the frame tree its survey anchor, and points forward to the one-line conversion between ellipsoidal height and elevation.
Chapter 6 took the larger pass: a new section on dated datums, the vertical equation with a figure showing the geoid undulating around the ellipsoid, a new failure mode for a frame with no date, and the Before you trust a coordinate checklist. There is a short Try this [10 min] exercise at the end.
Two corrections went in with them. The UTM section now says distortion is zero along the two standard lines and grows from them in either direction, rather than from the zone center. And the glossary gained five entries: geoid, epoch, NAD83, ellipsoidal height, and orthometric height.
We also deliberately avoided pinning the discussion to a particular announced release date for the next generation of U.S. datums. The important engineering idea is that the replacement carries an epoch on purpose, not a particular calendar date that may change while the Handbook is being written.
Keep pointing out what we missed
One of the reasons we’re writing the HD Map Handbook as a living book is precisely because we don’t expect to get everything right the first time.
The field is full of details that seem obvious once someone points them out. Sometimes those details turn out to be footnotes. Sometimes they turn out to change the way you think about the entire subject.
The distinction between where and when is one of the latter.
If you find something in the Handbook that is unclear, incomplete, or simply makes you stop and say “wait, what about…?”, send it our way.
The best parts of a living reference are often the things its readers teach it.
Read the updated chapters: Coordinates, Frames, and Reference Systems and Geographic and Projected Coordinate Systems.
If the territory is new, Why HD maps matter makes the system-level case for why a map carries this much weight in the first place.
