A map that has survived two hundred years of handling can still be ruined in an afternoon at a scanner. That is the uncomfortable truth behind most digitization projects: the failures rarely come from the ambitious parts. Nobody loses a collection because they picked the wrong georeferencing tool. They lose it because the scan was made at 150 dpi, every control point was dropped in the same corner, and the file is called scan_004_final_final.tif. What follows is a working list of those ordinary mistakes, the ones that show up again and again in map libraries, local history archives, and personal projects that started with good intentions.
Scanning: too little resolution is fatal, more is not automatically better
Resolution is the first place people economize, and it is the one place where you cannot recover later. A 50 × 70 cm sheet captured at 400 dpi optical yields roughly 7,900 × 11,000 pixels. The same sheet at 150 dpi gives you about 3,000 × 4,100 — enough to read the title, not enough to read the parish names, the hachures, or the pencil annotations in the margin. Upsampling afterwards adds pixels, not information.
For printed maps, 300 dpi at original size is a sensible floor. Dense copperplate engraving, small-scale topographic sheets, and manuscript corrections justify 400–600 dpi. Capture in 24-bit colour even when the map looks black and white: paper tone, foxing, and the difference between iron-gall ink and a later pencil note are all data, and they disappear the moment you let the software binarize the image.
Turn off the helpful features
Scanner drivers love to help. Auto-contrast, despeckle, sharpening, and descreening all alter the pixels before you have a chance to look at them. Save the master as an uncompressed or LZW-compressed TIFF, straight from the device, and do any cosmetic adjustment on a copy. If the archive needs a JPEG for the website, that JPEG is a derivative, not the record.
If you are photographing instead of scanning
Overhead capture is often the only option for oversize sheets, and it goes wrong in predictable ways. Lock the white balance and exposure instead of leaving them on auto, put a colour target in the frame on the first shot, and keep the camera parallel to the sheet. Aim for around 30% overlap between frames and shoot at a fixed aperture so the edges stay sharp. Uneven lighting and vignetting are the two things that make stitched composites obvious, especially on toned or stained paper.
Handle less, capture more
Clean, dry hands; support the sheet from below with a rigid board rather than lifting it by a corner; no tape, no weights directly on the surface. If an edge is torn, photograph the damage before any repair is attempted. One good capture under controlled conditions is worth five rushed ones.
Georeferencing: the control points decide everything
Three points will technically run an affine transformation. That does not make three points a good idea. Aim for eight to fifteen well-scattered control points, and check their distribution on a map before you accept the fit. A county survey georeferenced entirely from points along one river will look excellent near the river and rotate badly thirty kilometres out, because nothing constrained the far side.
Watch the residuals, not just the RMS. A root-mean-square error is an average, and one point that is four hundred metres off can hide behind nine that are nearly perfect. Open the point table, sort by residual, and ask why the worst points are worst. Often the answer is a misread graticule intersection or a click on the wrong side of a thick neatline.
Do not force the fit
Reaching for a high-order polynomial to make the residuals drop is the most common self-deception in the workflow. A warped sheet or a genuinely distorted original survey will not become accurate because the software was allowed more freedom. If paper deformation is the issue, a spline transformation with a dense, honest set of points is defensible — but record that you used it, because the resulting raster is no longer a simple rectification of the source.
The map's own coordinates are not ground truth
Old sheets reference different prime meridians — Paris, Ferro, Washington — and different ellipsoids, and magnetic declination has moved since the surveyor's compass was read. Where a map disagrees with modern data, it is worth asking which one is wrong about what. Coastlines on a decent Admiralty chart are usually reliable; inland rivers, boundaries, and roads may have been sketched, generalized, or drawn to support a claim rather than to describe the ground.
Metadata debt accumulates faster than you think
Everything you know about a file today will be forgotten in five years, and the person who inherits the folder will not be able to guess it. Build the metadata template before the first scan, not after the hundredth. At minimum, record:
- Institution, collection, and shelfmark or inventory number
- Title, date or date range, surveyor, engraver, and printer
- Original scale, dimensions, and stated projection
- Condition notes and any annotations added after publication
- Capture device, operator, date, optical resolution, and colour profile
- Georeferencing method, transformation type, number of control points, and RMS
- Software names and versions used at each stage
A plain spreadsheet with fixed columns is enough. The point is consistency, not sophistication.
File names and folders: the myth of "final"
Spaces in file names break scripts and links; so do accents and slashes. Adopt a pattern early and keep it, something like collection_shelfmark_recto_400dpi_v01.tif. Use ISO dates (2024-03-11, not 11/03/24) so files sort correctly. Avoid "final" entirely: the moment you name something final, someone produces a better scan of the same sheet.
Keep a simple three-folder structure — raw captures, edited masters, derivatives — and never edit a file in place. A read-only master folder has saved more projects than any backup policy.
Quality control that actually catches problems
Check every master at 100% zoom in all four corners and along the neatline. Cropped borders, cut-off marginalia, and clipped plate marks are the most frequent scan defects, and they are invisible in a thumbnail. Confirm that the georeferenced file opens correctly in a GIS application other than the one that produced it, since export bugs tend to surface on the second read.
Have a second person review a sample of about one in ten sheets against the physical originals. A fresh pair of eyes catches transposed dates, misfiled sheets, and upside-down verso scans far faster than the person who made them.
A short checklist before the scanner warms up
- Resolution and colour depth agreed, and written into the project notes
- Scanner enhancements disabled; master format set to TIFF
- Colour target captured with the first frame of the session
- Metadata spreadsheet open and populated for the sheet in hand
- Naming convention and folder structure already in place
- Control point strategy decided before any clicking begins
- Residuals reviewed and the transformation type documented
- Corner and neatline check done at full resolution
None of this is glamorous, and that is the point. Good digitization is mostly a matter of refusing to take shortcuts at the stages that feel tedious, so that the interesting work later — comparison, overlay, tracing change over time — rests on images you can trust.
Photo: Nothing Ahead / Pexels
