Horizontal vs Vertical Datum

A full coordinate triple has three components: latitude, longitude, and height. But the three don't share a single reference frame. Latitude and longitude reference a horizontal datum; height references a vertical datum. The two are usually tracked and updated independently because they're measured by different techniques and serve different applications.

The /learn/what-is-a-geodetic-datum pillar introduces both. This article focuses on the split itself — why it exists, how the two pair, and how to handle 3D coordinates that need both.

Why the split exists

The horizontal and vertical positions of a point are measured by different observations:

• Horizontal position comes from geometric observations: GNSS satellite tracking, photogrammetry, or (historically) triangulation. The reference is an ellipsoid — a smooth mathematical surface that approximates Earth's shape.
• Vertical position comes from gravimetric observations: spirit levelling, tide-gauge records, satellite gravity missions, or terrestrial gravity surveys. The reference is the geoid — the equipotential gravity surface that approximates mean sea level.

These two reference surfaces are fundamentally different. The ellipsoid is mathematical and smooth. The geoid is physical and irregular, differing from the ellipsoid by up to ±106 m globally (see /learn/the-geoid-explained).

A unified 3D datum would force both observations into a single reference frame, but that creates more problems than it solves:

• The horizontal observation network (GNSS reference stations) and the vertical observation network (tide gauges and levelling benchmarks) are physically different sets of monuments.
• Horizontal accuracy and vertical accuracy progress at different rates as measurement techniques evolve.
• Most use cases need either horizontal (mapping, navigation) or vertical (engineering, hydrology, flood mapping), not both simultaneously.

Separating the two lets each datum be updated independently when better observations are available.

Horizontal datums recap

A horizontal datum has the four components covered in the geodetic datum pillar:

• A reference ellipsoid (e.g., GRS80, WGS 84 ellipsoid).
• An origin (centre of mass for modern global datums; surface point for legacy regional datums).
• An orientation (axes aligned to the IERS Reference Pole and Reference Meridian).
• An associated gravity model (for compatibility with vertical systems).

Common horizontal datums:

| Datum | Region | Ellipsoid | | ------------- | --------------- | --------------- | | WGS 84 | Global | WGS 84 | | ITRF | Global | GRS80 | | NAD 83 | North America | GRS80 | | ETRS89 | Europe | GRS80 | | GDA2020 | Australia | GRS80 | | OSGB36 | Britain | Airy 1830 |

Vertical datums

A vertical datum defines what “height = 0” means. Common types:

• Tide-gauge vertical datum — anchored to long-term mean sea level at one or more tide gauges. NAVD 88 (US) is referenced to Father Point, Québec. NGVD 29 (US, legacy) was referenced to 26 US tide gauges. Newlyn (UK) is referenced to a tide gauge at Newlyn, Cornwall.
• Geoid-model vertical datum — heights measured above a globally-modelled geoid. EGM2008 and EGM96 are the canonical global examples. Used internationally where no national tide- gauge reference is available.
• Ellipsoidal vertical datum — heights measured above the reference ellipsoid. Not a true gravimetric datum; what GPS produces natively. Useful for satellite-tracking and for some cartographic applications.

The EPSG registry catalogues hundreds of vertical CRSs, each tied to a specific reference and (often) a specific geoid model.

Common vertical datums:

| Vertical datum | Region | Reference | | -------------- | ------------ | ------------------------------------------- | | NAVD 88 | US, Canada | Father Point, Québec tide gauge | | NGVD 29 | US (legacy) | 26 US tide gauges, 1929 averaging | | ODN | Britain | Ordnance Datum Newlyn tide gauge | | AHD | Australia | Australian Height Datum (30 tide gauges) | | EGM2008 | Global | Global geoid model | | EGM96 | Global (older) | Earlier global geoid | | TPDN | Japan | Tokyo Peil — Tokyo Bay mean sea level |

Typical horizontal-vertical pairings

In practice, certain horizontal-and-vertical pairings are conventional:

• United States: NAD 83 (horizontal) + NAVD 88 (vertical) + GEOID18 (conversion grid between ellipsoidal and orthometric heights). The US conventional pairing for surveying, engineering, and federal mapping.
• Europe: ETRS89 (horizontal) + EVRF2019 (European Vertical Reference Frame, vertical). The INSPIRE-mandated pairing for EU geospatial data.
• Australia: GDA2020 (horizontal) + AHD (vertical).
• Britain: OSGB36 (horizontal) + ODN (vertical).
• Global / GPS: WGS 84 horizontal + EGM2008 vertical. The default pairing in any global multi-region context.

The pairing is conventional, not enforced. A workflow can mix arbitrarily — “NAD 83 horizontal + EGM2008 vertical” is unusual but legal — but doing so requires explicit attention to the height-system conversion.

A worked 3D coordinate

A surveyed monument in CONUS might have these coordinates:

Horizontal datum:    NAD 83 (2011)
   Latitude:         40.7484°N
   Longitude:        −73.9857°W

Vertical datum:      NAVD 88
   Orthometric height (H): 35.2 m

Derived:
   Ellipsoidal height (h): 35.2 + (-33.4) = 1.8 m   (using GEOID18)

The orthometric height (35.2 m above NAVD 88 sea level) is what appears on the elevation map. The ellipsoidal height (1.8 m above the WGS 84 / GRS80 ellipsoid) is what raw GPS would produce. The 35.2 m → 1.8 m gap is the geoid undulation N = −33.4 m for that location, taken from GEOID18.

In any record, both datums plus the height-system conversion model must be documented. A bare “height: 35 m” is ambiguous without knowing whether it's orthometric and which vertical datum / geoid model produced it.

When the split matters

For most coordinate workflows, the horizontal-vertical split is transparent: the data's metadata names both datums, software converts as needed, the user sees a height in whichever system they expect. The split matters explicitly when:

• Integrating across datums — a federal NAVD 88 dataset and a smartphone GPS log in WGS 84 ellipsoidal heights need a geoid conversion before they can be joined.
• Crossing the NSRS modernization transition — NAVD 88 is being replaced by NAPGD2022 in the 2024–2026 US rollout. Data from before and after the transition must be transformed consistently.
• Working in a region with no national vertical datum — most parts of the world fall back to EGM2008 or EGM96 as the de facto global vertical datum.
• Sub-decimetre engineering work — the specific geoid model used (GEOID18 vs EGM2008 vs an older model) can produce decimetre-scale differences. The choice must be recorded as metadata, not assumed.

A historical note on vertical datums

Vertical datums have evolved more visibly than horizontal datums because tide-gauge networks change over time. A short US-focused timeline:

• 1929 — NGVD 29 adopted, referenced to 26 tide gauges averaging mean sea level for 1929.
• 1980s — Discovery that NGVD 29 has accumulated systematic errors (tilts and shifts) as continental relief was re-surveyed.
• 1991 — NAVD 88 adopted, referenced to a single tide gauge at Father Point, Québec. Re-levelled the entire US national network.
• 1996 — EGM96 published as the first global geoid model; becomes the de facto global vertical datum.
• 2008 — EGM2008 published with higher resolution. Replaces EGM96 for sub-metre global work.
• 2012 — GEOID12B published, the first US geoid model paired with NAVD 88 at sub-decimetre accuracy.
• 2018 — GEOID18 published, the current US conversion model.
• 2022 — GEOID2022 / NAPGD2022 announced as the upcoming vertical replacement, paired with NATRF2022 horizontal.

Each transition required (1) re-observing the national levelling network, (2) publishing transformation grids (VERTCON for horizontal-to-vertical migration), (3) updating all federal mapping product references. The /learn/nsrs-modernization support article covers the current 2024–2026 rollout.

Dynamic vs static vertical datums

A subtler distinction in modern vertical-datum work: dynamic vs static reference surfaces.

• Static vertical datum — height = 0 is fixed at one moment in time. NAVD 88 is static at the 1991 epoch. The local height-vs-mean-sea-level can drift over decades as the geoid changes (very slowly) and as the land subsides or uplifts. The static datum doesn't track these changes; updates are applied periodically as new realizations.
• Dynamic vertical datum — height = 0 follows the contemporary geoid as it evolves. NAPGD2022 (planned 2026) is a dynamic / time-aware vertical reference. Coordinates can be reported with explicit epoch metadata; height differences between epochs are first-class data.

For everyday work, the static datum is sufficient; the geoid changes by millimetres per decade, below most accuracy budgets. For climate-change-sensitive infrastructure (coastal flood defences, levees, subsidence monitoring), the dynamic model matters.

Common misconceptions

“A datum is a single 3D thing.” ISO 19111 treats horizontal and vertical CRSs as separable. A 3D coordinate combines them, but the two halves are referenced independently. EPSG even uses different codes: EPSG:4326 is 2D WGS 84 horizontal; EPSG:4979 is 3D WGS 84 with ellipsoidal height; EPSG:5703 is NAVD 88 vertical alone.

“Vertical datums are obsolete in the GPS era.” They matter more than ever. GPS gives ellipsoidal heights; what every engineering, hydrology, and surveying workflow needs is orthometric. The vertical datum is the conversion target.

“You can ignore the height datum if your accuracy budget is metres.” Within a single country, often yes — most national pairings (NAD 83 + NAVD 88, ETRS89 + EVRF) are designed to be internally consistent. Across countries or between modern and legacy datums, the offset is metre-scale and easily exceeds the budget.

“All vertical datums are the same.” They differ at the metre scale because they reference different tide gauges or different geoid models. NAVD 88 in CONUS and a Mexico vertical datum referenced to a Pacific Coast tide gauge differ by ~50 cm at the border. Cross-border infrastructure work has to handle this.

“Horizontal and vertical move together.” They don't. Plate motion changes horizontal coordinates over time (~2–3 cm/year in CONUS); vertical motion (post-glacial rebound, subsidence) is a separate effect that varies by region (millimetres to centimetres per year). For long-term records, both need temporal handling.

“If my GPS reports a height, it's already orthometric.” Sometimes — consumer devices typically apply a global geoid model internally and report orthometric. But the model used isn't always disclosed, isn't always the right one for your region, and varies between manufacturers. For survey-grade work, specify the geoid model explicitly. For consumer GPS at ±5 m horizontal accuracy, the implicit conversion is usually fine.

“A vertical datum is just a number you subtract from GPS.” It's a published model that varies across the landscape. The geoid undulation N changes by tens of metres across continental scales (and up to ~85 m globally), so the “subtraction” isn't constant — it's a grid evaluation at the input coordinate. Software libraries like PROJ or GeographicLib handle this; ad-hoc “subtract 20 m for US” rules of thumb produce errors when applied outside their calibrated region.