Cartography in Python
Last updated on 2026-10-20 | Edit this page
Overview
Questions
- How do I turn a correct map into a finished one that others can read and print?
- How should layers, labels and a basemap be ordered and styled?
- When do small multiples need a shared scale?
Objectives
- Style and order layers by their role: subject, supporting features, context and reference
- Label features legibly and by rule, with halos
- Add a basemap responsibly, a locator inset, a true scale bar and a north arrow
- Build small multiples with separate or shared classifications
- Export maps for print and screen
From a correct map to a finished map
The Fundamentals of Map Design episode covers what a thematic map shows: map elements, normalization, classification and color. This episode is about finishing: turning a correct choropleth into a map that a reader outside your project understands at a glance and that prints cleanly. The running example is the share of households without a car in Lafayette and West Lafayette, Indiana, with parks and city limits on top.
Visual hierarchy, layer by layer
Every layer has a role, and its styling should match it:
| Role | Example | Styling |
|---|---|---|
| Subject | The choropleth of car-free households | Strong color, full opacity |
| Supporting features | Parks, city limits | Distinct but secondary: one hue, thin or dashed lines, some transparency |
| Context | Basemap, neighboring areas | Quiet: gray, low contrast |
| Reference | Labels, scale bar, north arrow, source | Small, legible, never decorative |
Order follows from the roles: polygons of the subject at the bottom, supporting outlines and points above them, labels on top, and the basemap underneath everything. A layer on top of a choropleth needs either no fill, transparency, or a strong outline so it does not hide what is underneath.
Labels
Labels are reference information. They should be findable, legible and not in the way.
- Label sparingly. Label what the reader needs to orient themselves or to understand the story: the largest parks, the two city names. Labeling all 68 parks makes none of them readable.
- Use a halo (a thin outline in the background color, called a buffer in QGIS). It keeps text readable over any color and is the single most effective labeling technique.
- Use type to signal kind. Common conventions: italics for water and areas (rivers, parks, neighborhoods), upright for places and points, larger or capitalized type for more important names.
- Place labels inside or next to their feature, never across a boundary. For polygons, a point guaranteed to be inside (a “representative point” or “point on surface”) is safer than the centroid, which can fall outside an L-shaped polygon.
- Choose rules, not exceptions. “Label parks over 30 hectares” is reproducible; hand-picking labels is not.
Basemaps
A tiled basemap (OpenStreetMap, Esri, CARTO, Stamen) adds streets and place names for free, at a cost: visual noise, a dependence on an online service, and license conditions. Choose a quiet, gray basemap for thematic maps, keep its attribution text, and do not rely on it for print if you need full control or very high resolution. Basemap tiles are in Web Mercator (EPSG:3857); software reprojects them to your map’s CRS, which can blur their labels slightly.
Locator insets, scale bars and north arrows
- A locator inset (a small map of the state or country with the study area highlighted) answers “where is this?” for readers who do not know the area. It does not need a scale bar or legend of its own.
- A scale bar is only true on a map in a projected CRS with little distortion over the map (a UTM zone or a state plane for a county). In degrees or Web Mercator at large extents the scale changes across the map, and a single bar misleads.
- A north arrow is only essential when north is not up, but it is cheap and many readers expect one.
Small multiples and shared scales
To compare several variables or years, repeat the same map in a grid of small multiples. The eye compares patterns across panels far better than it decodes several variables in one crowded map.
Decide deliberately whether panels share a classification:
- Same variable or same units (one rate in different years, two rates you want to compare): use the same class breaks and colors in every panel, or the same dark color will mean different values.
- Different variables (income in dollars, children in percent): each panel needs its own classes and legend, and the reader compares patterns, not colors.
Export
| Destination | Format | Notes |
|---|---|---|
| Print, a paper, a poster | PDF (vector) | Sharp at any size; set the page or figure size to the printed size |
| Slides, a website | PNG at 200 to 300 dpi | PNG keeps sharp edges; JPEG blurs lines and text |
| Further editing (Illustrator, Inkscape) | SVG | Every element stays editable |
Check the exported file, not just the screen: font sizes, line widths and legend spacing often look different at print size.
The same map in QGIS
The Cartography in QGIS episode makes the same map with rule-based styling and labeling and a print layout with a locator inset. Notebook 01c of this module covers the basics (choropleths, classification, palettes) used here.
The toolkit in code
| Task | Python |
|---|---|
| Layer order |
zorder= on each plot() call;
alpha= for transparency |
| Zoom |
ax.set_xlim(), ax.set_ylim() from a
layer’s total_bounds
|
| Labels with halos |
ax.text(..., path_effects=[patheffects.withStroke(linewidth=3, foreground="white")])
at representative_point()
|
| Basemap | contextily.add_basemap(ax, crs=layer.crs, source=contextily.providers.Esri.WorldGrayCanvas) |
| Legend for several layers | Build Patch and Line2D handles and pass
them to ax.legend(handles=...)
|
| Locator inset |
inset = ax.inset_axes([left, bottom, width, height]),
then plot the state and the highlighted county |
| Scale bar |
ax.add_artist(ScaleBar(dx=1, units="m")) from
matplotlib-scalebar, for maps in meters |
| Shared classes |
scheme="UserDefined", classification_kwds={"bins": [...]}
in every panel |
| Export |
fig.savefig("map.pdf"),
fig.savefig("map.png", dpi=300, bbox_inches="tight")
|
Challenge 1: What is wrong with this map?
A map of poverty rates by tract uses a satellite basemap at full opacity, red and green classes, labels for every street, a scale bar on a Web Mercator map of the whole United States, and no source line. List the problems and a fix for each.
- Satellite basemap: too busy for a thematic map; use a quiet gray basemap or none, and keep the attribution.
- Red and green classes: unreadable for many readers with color vision deficiency, and a diverging scheme for a one-directional rate; use a sequential palette (see the Accessibility episode).
- Labels for every street: clutter; label only what orients the reader.
- Scale bar on a national Web Mercator map: the scale varies greatly with latitude; drop it or use an equal-area projection and a note.
- No source line: add the dataset and year.
- Shared: the same variable over time, so the same color must mean the same value in every year. Compute breaks from all four years together. (b) Separate: four variables in different units; give each panel its own classes and legend, and let readers compare patterns.
- Style each layer by its role; the subject is strong, context is quiet
- Label sparingly, by rule, with halos; italics for areas and water
- Use quiet basemaps and keep their attribution; scale bars need a projected CRS
- A locator inset shows where the map is; small multiples compare, with shared breaks only for comparable values
- Export PDF for print and PNG at 200 to 300 dpi for screens, and check the exported file
Notebooks
| Notebook | What you will do |
|---|---|
| 2a. Layers, Labels and Basemaps | Stack tracts, parks and city limits, zoom to the cities, label the largest parks with halos, add a basemap and a multi-layer legend |
| 2b. Insets, Small Multiples and Export | Add a locator inset, a scale bar and a north arrow, compare four variables in small multiples, share a scale, and export PDF and PNG |