This methods note grew out of an exploration of mental spatial representation. It introduces georeferencing, easily overlooked sources of error, and relevant tools. The original note drew on Brad Skopyk’s tutorial on georeferencing historical maps.
Chapter 1: Putting a map into geographic space↗
Georeferencing establishes a relationship between image locations and known geographic coordinates, allowing a scanned map to be overlaid and compared with other spatial data. It first answers “where does this belong?” It does not automatically make every feature on the map accurate.
The original note considered four uses:
| Use | What georeferencing helps accomplish | What still needs attention |
|---|---|---|
| Showing map extent | Locate the region represented by a map | Approximate placement is not precise measurement |
| Transferring features | Digitize boundaries, settlements, roads, and other features into a GIS | The original note’s main purpose; local positional error matters |
| Three-dimensional display | Drape a map over terrain or use height to display attributes | Elevation or attribute values require additional data; georeferencing does not generate them |
| Understanding cartography | Compare scale, layout, and agreement across regions | Differences can also arise from projection, scanning, or registration choices, rather than the cartographer alone |
Different purposes call for different tolerances and preserve different information. Showing where a map belongs and measuring a historical boundary should not demand the same standard of precision.
Chapter 2: What does a control point connect?↗
Ground control point pairs connect a location on the scan to the same identifiable place in reference data: pixel coordinates on one side, a position in the target coordinate system on the other. The ArcGIS Pro documentation recommends spreading points across the image rather than concentrating them in a small area. Adding points does not necessarily improve the registration.
“At least three points” is not a universal requirement. For a two-dimensional affine transformation, the mapping from pixel coordinates to target planar coordinates can be written as:
The six coefficients describe translation, rotation, scaling, and shear. Three non-collinear source points can determine them; additional points support fitting and checking. This model maps straight lines to straight lines, rather than bending them into arbitrary curves. More complex transformations have different point-count and geometric requirements; consult the relevant tool documentation.
2.1 First ask whether the landmark is reliable↗
A historical map may have been built from known base features, with the features of interest drawn in relation to them. Error can therefore enter through both the reference framework and the features added later. Distinguishing them requires knowing when, why, and by whom the map was made, and which positions were actually surveyed.
Churches and road intersections are not inherently more reliable than rivers or lakeshores. Buildings can move, roads can be rerouted, and rivers can change course; a shared name does not guarantee a shared site. What matters is identifying the same place and assessing its stability and reliability at the dates represented by both datasets.
2.2 Fit is not accuracy↗
Control-point residuals describe the difference between transformed positions and the reference positions assigned to them. Low residuals do not establish accuracy across the whole map. A flexible transformation can fit control points closely while distorting the regions between them.
To evaluate locations not used in fitting, reserve separate check points and report the units, spatial distribution of error, and quality of the reference data. Rather than deleting high-residual points until a number looks good, ask whether each reflects a misplaced click, a changed landmark, or a difference in the historical map worth preserving.
Chapter 3: Tools and a basic workflow↗
QGIS Georeferencer and ArcGIS Pro both provide control-point and transformation tools. QGIS is an open-source option; this note does not rank the two by their feature sets. The original three-step workflow remains useful, with interface locations depending on the version in use.
- Set up. Load the scan and reliable reference data, establish the target coordinate reference system (CRS), and decide whether the goal is display or measurement.
- Add control points. Identify corresponding places, record sources and uncertainty, and distribute points across the edges and interior. Reserve some locations for independent checks.
- Transform and inspect. Choose a transformation, resampling method, and output resolution appropriate to the distortion and research purpose. Inspect residuals and local deformation before exporting.
Save the control-point table, coordinate system, transformation settings, and reasons for choosing them alongside the image. An aligned map still needs a record of how it was aligned.