C-DRONE GUIDE · 30 AUGUST 2026

Which Coordinate System for a Drone Survey: RGF93, Lambert-93, Conic Zones and NGF-IGN 1969 Heights

The point cloud has arrived, it looks clean, the report claims three centimetres. Except that once loaded into the GIS it sits forty-eight metres above the reference survey drawing — or shifted by several tens of centimetres across the length of a building. This is not a measurement failure : it is a coordinate system failure. In our experience of writing survey specifications, it is by far the leading cause of an unusable photogrammetric deliverable — and the easiest to avoid, because it is settled in three lines written before the flight. Here is what the terms RGF93, Lambert-93, conic conformal zone and NGF-IGN 1969 actually cover, why a drone's GNSS does not spontaneously produce NGF heights, what a client should require, and what to check on delivery.

Published on 30 August 2026, reviewed on 30 August 2026 — regulations in force as of August 2026.

Three different things all called the system: geodetic, projection, vertical

When a project manager writes « deliverable in Lambert-93 » into a tender, they are in fact naming only one of the three building blocks needed for a point to have an unambiguous position. Confusing them is what produces most unusable deliverables.

• The geodetic system defines the shape and placement of the reference ellipsoid on which latitudes and longitudes are computed. In mainland France this is RGF93, the French realisation of the European ETRS89 system. The WGS 84 used by consumer receivers is very close to it but not identical : treating the two as interchangeable is harmless on a basemap, but not on a survey advertised as centimetre-accurate.
• The map projection turns those angular coordinates into plane coordinates in metres, the ones a drawing or a GIS actually handles. Lambert-93 is the national projection ; the nine conic conformal zones, written CC42 to CC50 after the latitude of their origin parallel, are nine local projections designed to keep distortion small.
• The vertical datum defines the zero of elevations. In continental France that is NGF-IGN 1969 ; in Corsica, NGF-IGN 1978. A vertical datum is entirely independent of the geodetic system and the projection : a dataset can be perfectly correct in Lambert-93 horizontally and completely wrong in elevation.

This is not merely trade convention. Decree n° 2019-165 of 5 March 2019 on the national coordinate reference system, together with an order of the same day, sets the national system to which survey work carried out by the State, local authorities and their public bodies — or on their behalf — must be tied. It succeeded decrees n° 2000-1276 and n° 2006-272 without changing anything for mainland France : RGF93, Lambert-93 and the conic zones horizontally, NGF-IGN 1969 and NGF-IGN 1978 vertically. In practice, if your survey feeds a file destined for a local authority, a permitting department or a public utility operator, tying it to the legal system is not a matter of taste : it is the rule.

Why a drone does not measure NGF elevations: ellipsoidal heights and the RAF20 grid

This is the point almost no client anticipates, and the one that produces the spectacular offsets. A drone's GNSS receiver, RTK-equipped or not, knows only satellite geometry : it returns an ellipsoidal height, that is, a distance above a mathematical ellipsoid of revolution. NGF elevations, by contrast, are physical heights, tied to a level surface that follows the irregularities of the Earth's gravity field — the quasi-geoid. The two surfaces coincide nowhere.

The gap between them is anything but marginal. In mainland France, the values to subtract from RGF93 ellipsoidal heights in order to obtain NGF-IGN 1969 elevations range, according to the IGN, from roughly 43 to 56 metres — highest over the mountainous areas of the Alps, the Pyrenees and the Massif Central, lowest off the Channel coast. A DTM delivered without conversion therefore sits some fifty metres too high, uniformly. The good news : because the offset is constant at the scale of a site, it is diagnosed at a glance and corrected without reflying.

The conversion tool is the IGN's vertical conversion grid. For continental France the surface in force is RAF20 (Références des Altitudes Françaises 2020), derived from the QGF16 gravimetric quasi-geoid and succeeding the RAF18 and RAF18b versions ; for Corsica it is RAC23, updated in autumn 2023, which gives access to NGF-IGN 1978. A serious photogrammetric workflow loads that grid and applies it ; a workflow left on default settings does not always do so, particularly when the project was set up on a generic system such as WGS 84.

The scientific literature confirms that the conversion, correctly performed, does not degrade the result. A study by S. Erol, E. Özögel, R. A. Kuçak and B. Erol, published in 2020 in the ISPRS International Journal of Geo-Information, compared digital terrain models from airborne LiDAR and drone photogrammetry across different vertical datums, and measured an accuracy of about five to six centimetres for physical heights derived from the drone point clouds once a good local geoid model is applied (see the study on Google Scholar). In other words : the grid is not the weak link in the chain — its absence is.

Lambert-93 or a nine-zone conic projection: what projection distortion changes

Every projection distorts : flattening a portion of an ellipsoid onto a plane without altering distances is mathematically impossible. What is measured is linear distortion, the relative difference between a distance measured on the ground and the same distance read on the projected drawing, expressed in centimetres or metres per kilometre.

Lambert-93 covers the whole national territory with a single secant conic projection, which is both its great strength — one coordinate set from Dunkirk to Bonifacio, indispensable for assembling cadastral data, national layers and cross-county datasets — and its unavoidable trade-off : at the extremes of the territory, distortion is counted in metres per kilometre, with reference values of about 2.3 m/km at Dunkirk and close to 3 m/km at Bonifacio. On an 800-metre linear site in northern France, that already amounts to nearly two metres between the true length and the length read on the drawing. For a basemap it does not matter ; for a quantity take-off, a billable area calculation or a setting-out, it does.

The nine conic conformal zones, created by decree n° 2006-272 and carried over by the 2019 text, answer exactly that problem. Nine projections follow one another from south to north, each centred on a whole-degree origin parallel — hence the names CC42 to CC50 — with a 50 % overlap between consecutive zones, so that no site ends up stranded on a boundary. Within those zones the IGN gives a linear distortion between -9 cm/km and +7 cm/km : two orders of magnitude better than Lambert-93 at the edges of the territory. That is why chartered surveyors and the cadastral service have widely adopted them for local work.

The practical rule fits in one sentence : Lambert-93 for national consistency and data exchange, the local conic zone for faithful distances on a local site. And one precaution : a single dataset must never mix the two. The EPSG codes to spell out in a specification are EPSG:2154 for RGF93 / Lambert-93, EPSG:3942 to EPSG:3950 for zones CC42 to CC50, EPSG:5720 for the NGF-IGN 1969 vertical datum and EPSG:5698 for the compound Lambert-93 + NGF-IGN 1969 system. EPSG:4326 denotes WGS 84 in geographic coordinates, in degrees : useful for a web basemap, never for a metric deliverable.

Deliverable by deliverable: where the coordinate system hides

The coordinate system is not declared in the same place from one format to the next, and that is precisely why it goes missing so easily along the way. Our guide to drone photogrammetry deliverables covers each format ; here is what each one demands on the georeferencing side.

• GeoTIFF orthophoto : georeferencing is normally embedded in the file's tags, but it is common to receive a TIFF accompanied by a .tfw world file and a .prj file describing the projection. A GeoTIFF delivered with neither a .prj nor internal projection tags will load anywhere at all : ask for the .prj explicitly.
• LAS or LAZ point cloud : the file header carries a coordinate reference record. Many exports leave it empty or generic. This is the most critical format, because a cloud with no declared system ends up being realigned by hand, which destroys metrological traceability.
• DTM and DSM : these take the full brunt of vertical confusion. Always specify, separately, the projection and the vertical datum — writing « Lambert-93 » says nothing about elevations.
• DXF : the format carries no projection information at all. Coordinates in it are raw. The expected system must therefore be written into the specification and repeated in the drawing's title block, failing which nobody will be able to reconstruct it six months later.
• IFC and BIM models : georeferencing goes through the project base point and the survey point, whose coordinates and orientation must be filled in explicitly. A model whose base point is left at zero is perfectly usable internally and completely unusable for overlaying a survey, a utility network or a planning document — a topic we develop in our guides to photogrammetry and BIM on construction sites and to the building digital twin.

Two families of projects make these requirements non-negotiable. As-built surveys of buried utility networks, where the statutory accuracy class presupposes a tie to the legal system and an explicit vertical datum ; and missions flown in support of a surveying practice, described in our guide to drone subcontracting for chartered surveyors, where the practice imposes its own system because it is the one signing the final document.

What the specification must state and what to check on delivery

On the specification side, six lines are enough to avoid nearly every unpleasant surprise :

• the geodetic system (RGF93 in mainland France) and the exact EPSG code of the expected projection — EPSG:2154 or the site's conic conformal zone ;
• the vertical datum, by name : NGF-IGN 1969 in continental France, NGF-IGN 1978 in Corsica — plus an explicit statement that ellipsoidal heights are not accepted as a final deliverable ;
• the conversion grid used (RAF20 for the mainland) and the obligation to state it in the processing report ;
• the target accuracy, in centimetres, separately for horizontal and vertical components ;
• the proof of checking : a table of residuals at independent check points measured outside the georeferencing ;
• the formats and their metadata : a mandatory .prj beside every raster, the system declared in LAS/LAZ headers, base point coordinates for an IFC model.

On the acceptance side, reading the residuals table is what separates a real check from a formality. Three signatures are almost visible to the naked eye. Random residuals of a few centimetres, positive and negative with no structure : normal behaviour for a properly georeferenced survey. A constant vertical offset, identical in sign and value at every point : the signature of a missed datum conversion — if it approaches fifty metres, the RAF20 grid was never applied ; if it is a few decimetres, it is often an antenna height or a base station elevation entered wrongly. A residual proportional to the distance between points : the signature of a projection or zone error.

This insistence on independent checks is not client over-caution. A study by M. Štroner, R. Urban, T. Reindl, J. Seidl and J. Brouček, published in 2020 in the journal Sensors, compared three georeferencing strategies across two sites — ground control points alone, onboard GNSS RTK alone, and a combination of both — and shows that the vertical component can carry relatively large systematic errors, particularly over terrain that is hard to reconstruct, even where the horizontal component remains sound (see the study on Google Scholar). Elevation is indeed what should be checked first, and at points the contractor did not use to control the block. In the same vein, a study by G. Forlani and co-authors, published in 2018 in Remote Sensing, assessed the quality of DSMs produced from drone flights georeferenced by onboard RTK and stresses that the block control configuration — ground control points only, camera stations only, or camera stations with at least one ground control point — directly determines the model's vertical consistency (see the study on Google Scholar). Choosing RTK, PPK and the number of ground control points is the contractor's method : our guide to RTK/PPK and when to require centimetre accuracy sets out what is reasonable to expect, and our guide to ground sampling distance (GSD) what it changes about the level of detail. The client specifies a checked result and a system, not a technology.

2026 prices (excl. VAT), on top of a photogrammetric survey : €400 to €900 for a tie to the legal system with placement and centimetre-grade GNSS survey of ground control points, depending on the number of points and site access ; €300 to €600 for a conformity check on delivery, with independent check points and a residuals table appended to the report ; €150 to €400 to convert an already-produced deliverable to another system, projection or vertical datum ; €100 to €300 for an additional export in a second system or a second georeferenced format. A statutory tie signed by a chartered surveyor is quoted separately and charged in addition. Set against the cost of an unusable deliverable — and against the remedies available for non-compliant deliverables, always slower than a line in a specification —, those amounts are trivial. Our drone surveying and photogrammetry page presents the full range of deliverables ; request a quote stating the expected EPSG code, the vertical datum and the software that will use the data.

Frequently asked questions

Request a free quote

Also worth reading

• Drone Services on the Canal Seine-Nord Europe Site: Construction Monitoring and Archaeological Support
• Balcony, Loggia and Railing Inspection by Drone: What the Flight Sees, What It Cannot Conclude, and the Prices
• Drone Helipad Inspection: Surface, Markings, Lighting and Approach Paths — Method and Prices