Recipes¶

Task-shaped snippets. All of them assume you have a dataset URL from the Catalog.

Filter to one region¶

Pythonmapshaperjq

import geopandas as gpd

communes = gpd.read_file("comunas.geojson")

# codregion 13 is the Santiago Metropolitan Region.
santiago = communes[communes["codregion"] == 13]
santiago.to_file("santiago.geojson", driver="GeoJSON")

npx mapshaper comunas.geojson \
  -filter 'codregion === 13' \
  -o santiago.geojson

jq '{type, features: [.features[] | select(.properties.codregion == 13)]}' \
  comunas.geojson > santiago.geojson

Simplify for the web¶

The single most useful thing you can do with this data. Three reductions compound:

npx mapshaper comunas.geojson \
  -simplify percentage=2% keep-shapes \
  -filter-fields Comuna,cod_comuna,Region \
  -o precision=0.0001 comunas.min.geojson

Flag	Effect
`-simplify percentage=2%`	Visvalingam simplification, keeping 2% of vertices
`keep-shapes`	Prevents small islands from vanishing entirely
`-filter-fields`	Drops `objectid`, `st_area_sh`, `st_length_` and friends
`precision=0.0001`	Rounds coordinates to ~11 m

Expect roughly 70 MB → 300–800 KB. Always check the result visually: aggressive simplification creates slivers and can disconnect coastlines.

Why precision alone helps so much

The source files store coordinates with about 14 decimal places (-68.95020116247055) — nanometre precision for administrative boundaries. Six decimals is roughly 11 cm. Everything past that is incompressible noise occupying real bytes.

Convert to TopoJSON¶

TopoJSON stores shared borders once instead of twice, which matters a lot for administrative units that tile a country.

npx mapshaper comunas.geojson -o format=topojson comunas.topojson

Reading it back in the browser needs topojson-client:

import * as topojson from "topojson-client";

const topo = await fetch("comunas.topojson").then((r) => r.json());
const geojson = topojson.feature(topo, topo.objects.comunas);

Join your own data¶

The usual reason to want boundaries: you have statistics keyed by an official code and you want them on a map.

import geopandas as gpd
import pandas as pd

communes = gpd.read_file("comunas.geojson")
stats = pd.read_csv("population.csv")   # columns: cut, population

merged = communes.merge(stats, left_on="cod_comuna", right_on="cut")

The leading-zero trap

cod_comuna is stored as a number, so Camiña is 1402. The official INE/SUBDERE code is the five-character string 01402 — regions 1 through 9 all lose their leading zero this way.

If your CSV has zero-padded strings, the join silently matches nothing. Normalise one side first:

communes["cut"] = communes["cod_comuna"].astype(str).str.zfill(5)
stats["cut"] = stats["cut"].astype(str).str.zfill(5)

This is exactly the class of problem the standardised property schema exists to eliminate.

Compute area correctly¶

import geopandas as gpd

regions = gpd.read_file("regiones.geojson")

# WRONG — degrees are not a unit of area.
regions.area

# RIGHT — reproject to a projected CRS suited to the area of interest.
regions.to_crs(5361).area / 1e6   # km², SIRGAS-Chile / UTM

The area_km property already present in regiones.geojson is precomputed upstream and can be used directly. Note that st_area_sh in the same file is in square metres, not square kilometres — the units are inconsistent between the two properties and between the two files.

Merge levels into one file¶

npx mapshaper \
  -i regiones.geojson comunas.geojson combine-files \
  -o format=topojson chile.topojson

Produces a single TopoJSON with both layers, sharing the coastline arcs.