Skip to content

Week 11 Hands-On — Georeferencing in QGIS, and the Web Mapping with AI Kickoff

Hands-On Practice · Day 21 · Thursday of Week 11 — Geoplanning and Georeferencing

Topics

  • Georeference a scanned historic map: ground control points, transformation types, residuals
  • What a site plan for the Domes for Mozambique project needs from georeferencing
  • Web Mapping with AI Experience kickoff: from a QGIS layer to a public web map with an AI assistant

Materials

Feeds Lab 10 and the Web Mapping with AI Experience.

At a glance
Goal Students georeference an image that has no coordinates of its own, understand ground control points, transformation types, and residuals, and see what a good registration looks like. Then they get the Web Mapping with AI assignment launched with a live example.
Why this week Tuesday Dr. Ames presented georeferencing and students georeferenced their own pencil sketch of their neighborhood. Today upgrades that to a real plan sheet or historic map with quality control, which is what a site plan for Lab 10's Domes for Mozambique project needs. The Web Mapping experience is due Wednesday of Week 14, so three weeks of lead time start now.
Students bring Laptop with QGIS 3.44. Their georeferenced sketch from Tuesday, if they finished it.
Graded item None on Thursday. Tuesday's Georeference Your Neighborhood Sketch (5 points) is Dr. Ames's item; students who did not finish can upload it after today's session.
Feeds Lab 10: Geoplanning, Domes for Mozambique. Due Saturday. Web Mapping with AI Experience, due Wednesday of Week 14.

Practice run before class

Tip

Thirty minutes, and do it twice if you can. Georeferencing is the least forgiving session of the semester because a bad set of control points is not obvious until the output loads.

What you need. QGIS 3.44 LTR, internet for the basemap, and a scanned map with no coordinates in it. Download a historic USGS quad of Provo from USGS topoView as a JPEG, not a GeoTIFF, so it genuinely has no georeferencing. A campus plan sheet exported from PDF to PNG works too.

Check first that Plugins > Manage and Install > Installed > Georeferencer GDAL is ticked.

Do this, in order.

  1. Add the Google satellite basemap and set the project CRS to EPSG:26912.
  2. Layer > Georeferencer..., then Open Raster and choose your image.
  3. With the Add Point tool, click a hard feature you can also find on the basemap, such as a road intersection or a bridge end. Click From Map Canvas and click the same feature on the main map. Repeat for four points spread to the corners, never along a line.
  4. Settings > Transformation Settings: Polynomial 1, resampling Cubic, target CRS EPSG:26912, an output path next to the image, and Load in project when done ticked.
  5. Read the dX, dY, residual columns and the Mean error (RMS) at the bottom. On a 1:24,000 scan a 30 m residual is a misplaced click and 3 m is fine.
  6. Untick one point and watch the RMS move. That is the quality-control moment of the session.
  7. Start Georeferencing. Set the output to 50 percent opacity and toggle it. Check whether the roads line up across the whole image or only near your control points.
  8. Add two more points in the worst area, switch to Polynomial 2, and rerun.
  9. For the kickoff half: right-click any layer > Export > Save Features As..., GeoJSON, CRS EPSG:4326, then ask an AI assistant for a single Leaflet HTML file that loads it, and serve the folder with python3 -m http.server.

You are ready when you have a georeferenced image whose roads sit under the basemap roads, and an RMS number you can defend.

The one that bites. Opening the Leaflet page straight from disk makes the GeoJSON fetch fail silently. Serve the folder over http://localhost:8000 instead, or test it on GitHub Pages.

Before class

  • [ ] An image to georeference, saved locally: a scanned historic USGS quad of Provo from USGS topoView (download the JPEG, not the GeoTIFF, so it truly has no coordinates), or a campus plan sheet PDF exported to PNG. Test it once.
  • [ ] QGIS open with the Google satellite XYZ basemap and the project CRS at EPSG:26912. The Georeferencer is under Layer > Georeferencer... in 3.44 (Raster > Georeferencer in older versions).
  • [ ] Harrison ready with a finished example web map open in a browser and the assignment page in a tab.
  • [ ] A GeoJSON export of any layer ready (steps below) for the kickoff demo.

Plan (50 minutes)

Time Segment
0:00 Mini-devotional
0:03 Why georeference: plan sheets, old maps, drone photos, your sketch
0:06 Georeferencer: open image, place four GCPs from the map canvas
0:14 Transformation settings: linear, Helmert, polynomial, thin plate spline; residuals and RMS
0:20 Run, overlay with transparency, judge it; add two more GCPs and rerun
0:26 What Lab 10's site plan needs from this
0:30 Web Mapping with AI kickoff: the deliverable, a finished example, QGIS to GeoJSON to a web page in ten minutes
0:46 Questions; where to get help

Walkthrough

1. Why

A drawing, a scan, a photo: pixels with no idea where they are. Georeferencing attaches real-world coordinates to pixel positions so the image can sit under vector data. Every renovation project starts with an old plan sheet; every site visit produces a sketch. Tuesday's sketch was the gentle version; today's image is the working version.

2. Ground control points

  1. Layer > Georeferencer... In the Georeferencer window, Open Raster and choose the image. It appears in its own canvas with pixel coordinates.
  2. Add Point tool. Click a feature you can also find on the basemap: a road intersection, a building corner, a bridge end. In the dialog, click From Map Canvas, then click the same feature on the QGIS map. The GCP appears in the table with source pixels and destination coordinates.
  3. Repeat for four points, spread to the corners of the image. Say the rule: spread wide, avoid a line, use hard corners not soft edges.

3. Transformation settings and residuals

  1. Settings > Transformation Settings. Walk the list: Linear (shift and scale only, needs 2 GCPs), Helmert (adds rotation, 2 GCPs), Polynomial 1 (affine, 3 GCPs, the default for a flat scan), Polynomial 2 (6 GCPs, corrects gentle warps), Thin Plate Spline (bends locally to fit every point exactly, which hides bad points). Resampling Cubic for imagery, Nearest neighbor for a categorical scan. Target CRS EPSG:26912. Output file next to the image. Tick Load in project when done.
  2. Look at the GCP table: the dX, dY and residual columns appear once there are enough points for the chosen transformation. The Mean error (RMS) is at the bottom. A residual of 30 m on a 1:24,000 scan is a misplaced click; 3 m is fine.
  3. Turn off one point (untick it) and watch the RMS change. That is quality control, and it is why TPS with four points is meaningless: it forces zero residual everywhere.

4. Run and judge

  1. Start Georeferencing. The output loads over the basemap. Set its opacity to 50 percent and toggle it. Do the roads line up across the whole image or only near the GCPs?
  2. Add two more GCPs in the worst area, switch to Polynomial 2, rerun. Better in the middle, and the corners? Discuss what more points buy and when they stop helping.
  3. Optional: Vector digitize a building footprint from the georeferenced 1950s map and compare to today's imagery. That is a change-detection analysis in two minutes.

5. Lab 10 connection

Domes for Mozambique places a site inside a project boundary and plans around it. If the site drawing comes as an image, this is how it becomes a layer. Also point out that Lab 10 starts by choosing a CRS for Mozambique (UTM 36S or 37S), the Week 8 lesson applied somewhere new.

Web Mapping with AI kickoff (16 minutes, Harrison)

The assignment is on the site: Web Mapping with AI Experience. Cover:

  1. The deliverable in one sentence. A public web page with an interactive map showing at least one layer you made in QGIS, built with an AI assistant, plus a short note on how you used the AI and what you checked.
  2. A finished example. Open one: pan, zoom, click a feature, see a popup. This is the bar; it is not high.
  3. Live, in ten minutes. Right-click any layer in QGIS > Export > Save Features As..., format GeoJSON, CRS EPSG:4326 (web maps expect degrees). Open an AI assistant and paste a prompt like: Write a single HTML file that uses Leaflet from a CDN, shows an OpenStreetMap basemap centered on Provo, Utah, loads data.geojson from the same folder, styles it, and shows each feature's attributes in a popup. Save the reply as index.html next to the GeoJSON. Open it in a browser (a local server may be needed for fetch; the assignment page has the one-line command). It works, or the AI fixes what it got wrong when you paste the error back.
  4. Publish. Create a GitHub repository, upload the two files, enable Pages in Settings. The page URL is the submission. The assignment page walks this with screenshots.
  5. Where to get help. Harrison's office hours and the Week 12 lab hour. Start this week; the last week of the semester is the final project.

Common snags

  • Georeferencer menu is missing. It is a core plugin; Plugins > Manage and Install > Installed > Georeferencer GDAL must be ticked.
  • Output loads far from the basemap. The target CRS did not match the map canvas CRS at the time the GCPs were captured. Redo with both at 26912.
  • RMS column is blank. Not enough points for the transformation; Polynomial 2 needs six.
  • The image is upside down or mirrored. A scan from a PDF sometimes carries a rotation flag. Rotate it in an image editor first, or just place the GCPs and let Helmert handle rotation.
  • fetch of the GeoJSON fails in the browser. Opening index.html from disk blocks local file loads. Run python3 -m http.server in the folder and open http://localhost:8000, or publish to Pages and test there.