The GIS team says the coordinates are correct. The BIM team says the model is correct. So why the hell is the road going through the building?
The two can be true. A GIS layer exists in a projected coordinate system with large numbers and a defined datum. A BIM model typically exists in a local engineering grid with its own origin (near the building), its own units, and its own idea of north. If you put them both in the same viewer, they can appear aligned. Whether they are aligned or not is down to the transformation between them, and that transformation is often missing, undocumented, and/or wrong.
In this post, we look at why two correct models can end up disagreeing, what settings lead to the mismatch, why aligning by eye is not a good solution, what the cost of small offsets is on site, and what a repeatable GIS/BIM handover actually looks like. We finish with a way of working out where a particular mismatch originated.
Why can two correct models disagree when combined?
Each model is right in its own frame
A BIM model is measured from a project origin. In Revit, the Internal Origin is the fixed zero point used for calculation, the Project Base Point is the local origin for dimensions, and the Survey Point carries the real-world position and orientation for shared coordinates. A GIS layer is measured from the false origin of a map projection. Neither is incorrect. They are just different rulers.
The transformation is the missing third model
To join the two, you need a set of parameters: where the BIM origin is in the map system, how far its axes are rotated, and what scale and units you're using. In the IFC standard, these are all in IfcMapConversion, which defines the local origin as an easting, northing, and orthogonal height, rotates the x-axis, and applies a single scale to all three axes. Its target is an IfcProjectedCRS which records the datums, the projection and the zone. The IFC4 documentation itself makes note that IfcMapConversion does not deal with the projection from the geodetic coordinate system.
In practice, the information is often very incomplete. The GeoBIM benchmark found that georeferencing has simply not been a priority for designers and software developers so it's not routinely stored or used and is not always read or exported in a consistent manner. A 2024 ISPRS paper states that the five levels of georeferencing as defined by Clemen and Görne can all be written into an IFC file at the same time and independently of each other. An IFC file can therefore contain contradictory positions, and different viewers may end up with different positions.
Where do the mismatches come from?
Datums and transformation choices
Moving between datums requires a transformation, and there may be several options. PROJ can fall back on a "ballpark" transformation if a precise one isn't available, and operations using such ballpark transformations will have unknown accuracy. Two different tools selecting different operations could end up with different coordinates for the same point, and that difference would not necessarily be flagged in any way.
Local grids, scale and north
A local engineering grid is flat. The Earth isn't. Esri's documentation makes it clear that the relationship between surveyed ground distance and grid distance is via a combined scale factor (the grid scale factor times an elevation factor). The elevation component is easily overlooked. Our figures suggest it's around 157ppm at 1,000m above the ellipsoid (some 16cm over a km). The same page cautions users that 'north' on a survey document can be geodetic, magnetic, grid, or locally defined, so rotation is another source of discrepancy.
Vertical datums
GNSS provides ellipsoidal heights whereas the design drawings and benchmarks we're used to will be in orthometric heights. The two are related by the geoid height. NGS states that geoid heights in the conterminous US are between about -8m and -53m and between about +75m and -100m globally. Esri throw in another caveat in that no vertical coordinate system is defined for BIM data, and if it's not known, then the ArcGIS clients will assume that the x, y, and z units are the same.
Units
A feet v metres mix-up is big enough to spot. The more subtle one is the US survey foot versus the international foot. NIST states that the survey foot is only 2 parts per million longer, yet a State Plane coordinate of 1,000,000 ft will be out by 2 ft if you use the survey foot instead of the international foot, and the error can easily be in the tens of feet in some coordinates. The survey foot was dropped at the end of 2022, but of course legacy data will still contain it.
Why is eyeballing it not a reliable solution?
A good match near the anchor can hide drift
Nudge the model until it's sitting on the basemap, and you've fixed one location. You haven't fixed scale or rotation. According to our maths, a rotation error of just 0.1° will move a point by around 1.7m at 1km, and the building you nudged to may well be the only one you actually looked at.
Hand-moving leaves no audit trail
Esri's georeferencing tools will store a manual shift in a WLD3 world file, and the Move tool can make use of coordinate values for the destination. That is a perfectly valid method if the offset is derived from known points. It's a risk if it's derived from the eye, as the next person to look at it will only see a file and a shift with no explanation as to why.
How do small positional differences actually affect the real work?
Utilities, roads and structures
Utilities are the most sensitive as clearances are tight, and the cost of a miss is very high. We've covered that in our post on mapped versus located utilities. The same applies to a road alignment which crosses a building footprint by half a metre in the model, a culvert whose invert is set a metre too high, or a retaining wall which has been staked from a layout that's been converted by eye.
Vertical offsets hide longest
A horizontal shift is visible in plan. A vertical shift can go completely undetected, even though gravity drains, road gradients and floor levels all depend on small height differences. If the height datum is not explicitly stated, then you're one wrong assumption away from having tens of metres of geoid separation. Height errors also get fed into hydraulic models, as we demonstrated in our post on DEM quality and flood map accuracy.
Distance from the origin matters too
Autodesk's Revit training material states that there is a geometry limit of 20 miles from the Project Base Point. Large survey coordinates are therefore better acquired via Acquire Coordinates rather than by moving the model a great distance from its internal origin.
What should a GIS and BIM handover checklist contain?
- Name the coordinate reference systems. State the horizontal and vertical systems with codes and versions, and the epoch that one uses. "UTM" or "local" are not definitions.
- State the unit and whether the coordinates are grid or ground, along with the combined factor and the reference point it applies to.
- Supply control points. List named points with coordinates in both systems, physically monumented where possible. On a World Bank funded wind farm project, our scope was to supply the control points in the form of concrete monuments, establish the permanent benchmarks, and pass the control network coordinates to the consultant and client and the work was subsequently reviewed and accepted against the technical specifications.
- Define the vertical reference. Give the name of the datum and the geoid model and state whether the heights are ellipsoidal or orthometric.
- Record the transformation. Note the method, parameters, software, and date, and store them in the file: IfcMapConversion and IfcProjectedCRS in IFC, shared coordinates in Revit. Be careful not to have multiple or conflicting georeferencing entries in the same file.
- Run an independent check. Compare the check points that were not used in the alignment, report the residuals, and agree on the tolerances prior to handover.
Control, documented transformations and independent checks form the very foundation of any good topographic survey or GIS product.
How do you know whether the problem originated in GIS, CAD/BIM, the transformation or the export?
Begin with a short list of control points whose coordinates you have faith in. Check them at every stage: survey, GIS, CAD or BIM, and then export. The stage at which the residual goes off is the stage at which the problem occurs. The pattern of the error usually points to the cause:
| What you see | Likely origin | Quick test |
|---|---|---|
| Model sits kilometres away, or near 0,0 | Wrong CRS or zone, or no georeference | Compare one control point's coordinates in both files |
| Constant offset of a few metres | Datum transformation or origin definition | Transform one known point with a documented operation |
| Error grows with distance from the anchor | Grid-to-ground factor or units | Compare the distance between two control points in both models |
| Error grows sideways from the anchor | Grid, true or project north | Compare a bearing between two control points |
| Constant vertical offset of metres to tens of metres | Ellipsoidal versus orthometric height, or geoid model | Check a known benchmark |
| Correct in one viewer, wrong in another | Conflicting georeferencing in the file, or viewer settings | Inspect the stored entries and re-export with shared coordinates |
If the numbers match in the source files and don't match after export then check your export settings first. In Revit for instance DWG and DXF exports have an option to include shared coordinates.
Final thoughts
Models that just look aligned on screen have only demonstrated that they look aligned. True alignment in the field comes from a documented transformation, shared control, stated units and vertical datum and a check that someone other than the author has actually run it.
Three things to take away. Firstly, treat the transformation as data (write it down, store it in the file and don't ever leave it in someone's head). Secondly, align from control points rather than by eye. Thirdly, diagnose by pattern as a shift, a scale error, a rotation, and a vertical offset each have their own distinct signature.
If your GIS and design teams are about to be exchanging models, get in touch and we can go through the control points and CRS definitions prior to anything being put out on site.