Where does Spain actually live?

Population density per H3 cell, cross-checked against night-lights.

NoteIn progress — ships May 26, 2026

This is the build page for P-01, the first project in Prospectra’s demographic trilogy. It’s published as the work happens. The code and figure below show the approach; the finished analysis and charts land on May 26.

The question

Spain’s population register, the INE Padrón Continuo, is published per municipality — and municipalities are wildly uneven in size. Madrid is one polygon; so is a 12-person hamlet in Soria. Ask “where do people actually live?” with that data and the map lies: it tells you about administrative borders, not density.

The fix is to throw away the borders and bin everyone into a regular grid — Uber’s H3 hexagons — then sanity-check the result against an independent signal: how brightly each cell glows at night.

The data

Layer	Source	Resolution
Population	INE Padrón Continuo	Municipal polygon
Spatial index	Uber H3	res-9 (~0.1 km²)
Validation	NASA Black Marble (VIIRS)	~500 m raster

Approach

1. Bin municipal population into H3 res-9 cells (area-weighted).
2. Aggregate night-time radiance per cell from the Black Marble raster.
3. Compare the two surfaces — where they disagree is where the register and reality part ways.

1 · Binning population to H3

On Databricks, with Apache Sedona for the spatial join and the built-in h3_* SQL functions for indexing — no GIS server, just the lakehouse:

# Databricks · PySpark + Apache Sedona
from sedona.spark import SedonaContext

sedona = SedonaContext.create(spark)

# municipal polygons (INE) → H3 res-9 cells covering each one
cells = (
    spark.table("ine.padron_municipal")
         .selectExpr(
             "muni_code",
             "poblacion",
             "explode(h3_polyfill(geom, 9)) AS h3")        # one row per covered cell
)

# split each municipality's population across its cells, weighted by area
pop_per_cell = (
    cells.withColumn("cell_pop", col("poblacion") * col("area_share"))
         .groupBy("h3")
         .agg(sum("cell_pop").alias("population"))
)

2 · Cross-checking against night-lights

Zonal statistics straight in Databricks SQL — average VIIRS radiance per H3 cell:

SELECT  h3_cell,
        AVG(radiance) AS mean_radiance
FROM    black_marble.viirs_2024
GROUP BY h3_cell

The grid, over Iberia

Illustrative H3 coverage over the Iberian peninsula. The final population choropleth and the night-lights agreement map ship May 26.

What ships May 26

• The full notebook (population binning + raster zonal stats), reproducible on Databricks Free Edition.
• Two maps: the H3 population surface, and where it agrees / disagrees with night-lights.
• A written piece, plus an X thread walking through the result.