GeoJSON Simplifier

Why simplification is usually the fix for a slow map

Government boundary files are digitised at survey accuracy — a single district can carry fifty thousand vertices describing a river bank nobody will ever zoom into. Rendering that in a browser means parsing megabytes of JSON and asking the GPU to draw detail smaller than one pixel. Dropping to a few hundred vertices per shape typically cuts file size by 80–95% with no visible difference at city zoom levels, and it is almost always a bigger win than any other optimisation.

How the algorithm decides what to remove

Douglas-Peucker draws a straight line between the first and last point of a segment, finds the vertex furthest from that line, and keeps it only if the distance exceeds your tolerance — then recurses on both halves. The consequence is that it preserves corners and characteristic bends while deleting the near-collinear runs that make up most of a detailed coastline. Tolerance here is in degrees, so 0.0001 is roughly 11 metres at the equator.

• 0.00001 (~1 m) — barely visible, still worth 20–40% savings
• 0.0001 (~11 m) — safe default for city-level web maps
• 0.001 (~110 m) — country and regional overviews
• 0.01 (~1.1 km) — world maps and thumbnails only

What you give up

Simplification changes geometry, so measurements change with it: area and perimeter both drift, usually shrinking. Adjacent polygons are simplified independently, which means shared borders can pull apart and leave slivers or overlaps — this tool is not topology-aware. And a ring can never be reduced below four positions, so tiny parcels stop shrinking at a certain point. Measure area before you simplify, and keep the original file as your source of truth.

A sensible workflow

Start at the second or third level and step up until the preview visibly deviates from the original, then step back once. Check the vertex count rather than the file size — a 90% vertex reduction is the signal you want. Then keep two copies: the full-resolution file for analysis and measurement, and the simplified one for display. If the map still struggles after simplification, the problem is feature count rather than vertex count, and clustering or vector tiles is the next step.