Projection barely matters on a city map and decides everything on a map of the whole world

What it is

Absolute projection distortion grows with coverage: on a city map, any reasonable projection's distortion approaches zero (a city is tiny relative to Earth's curvature), while on a global map the projection choice determines the credibility of the whole structure. Projection is therefore not a globally fixed configuration—global, national, provincial, and city maps each have suitable projections and parameters; a product rendering every map in the same Web Mercator wastes the choice on city maps (harmless) and distorts global choropleths (harmful).

Why it happens

Coverage governs projection because distortion grows with distance from the projection's center or standard lines: take Albers with standard parallels at 25°N and 47°N—distortion over China is negligible, but the same parameters render North America badly; projection parameters only work when matched to the data extent. The practical strategy is tiered selection: global extent uses equal-area world projections (Mollweide, equal-area cylindrical, Robinson-type compromises); continental or large-country extents use conic or azimuthal projections customized to the range (Albers, Lambert, azimuthal equal-area); sub-national extents have negligible distortion and simplify to local systems (UTM or local coordinates). Web products face the reality that basemap services usually ship only Web Mercator tiles—compromises include reprojecting the data layer (transform to equal-area for analysis and rendering, then back to tile coordinates for overlay) or accepting Mercator for small-extent applications (distortion negligible). Parameters must be recomputed when coverage changes: switching the data from China to the US requires resetting Albers's standard parallels and central meridian; reusing China's parameters introduces visible skew.

Where it stops holding

Matching coverage to projection hits an implementation-cost boundary: a reprojection pipeline (transform on ingest or at render time) is real engineering, and teams should tier the investment by "does the data carry area/distance comparison"—maps showing only position may accept a unified projection to simplify the system. Interactive maps spanning multiple ranges (continuous zoom from globe to street) create visual jumps if the projection switches mid-zoom; mainstream web maps therefore run Mercator throughout and accept high-latitude distortion—an engineering-cartography compromise where products should compensate in analysis features (compute in equal-area coordinates, display in Mercator). The negligibility judgment belongs to the analytical purpose: negligible for showing position, not negligible for comparing areas.

Applying it

• Select projections in tiers by coverage: equal-area world projections for global, customized conic/azimuthal for large regions, simplified local systems for city-level.
• Recompute projection parameters (standard lines, central meridian) whenever data coverage changes; never reuse across extents.
• When web basemap and data-layer projections differ, reproject the data layer before overlay, or declare the limitation in the caption.
• Verification: compare the map's coverage against the projection's suitable range; a global choropleth still on Mercator is a coverage-projection mismatch instance.

Related

• Same group: U12.01.1 No planar projection can preserve area, angles, and distances simultaneously · U12.01.2 Area-type maps must use an equal-area projection · U12.01.3 The Mercator projection severely inflates areas at high latitudes · U12.01.5 The projection used should be stated in the caption
• Nearby: U12.01.3 The Mercator projection severely inflates areas at high latitudes · U12.01.1 No planar projection can preserve area, angles, and distances simultaneously
• Search terms: projection selection · albers parameters · map extent