S6: Price-per-m² Map — Versailles (DVF)

Intermediate CLI + Map

Capabilities: filter calculate classify grid_create spatial_aggregate

Use Case

A real-estate analyst wants to visualize the spatial structure of price per square meter in Versailles. We start from DVF mutations (Demandes de Valeurs Foncières, Etalab open data), filter to residential sales, compute price_per_m2 = valeur_fonciere / surface_reelle_bati, drop outliers (parking spots, garages, data-entry mistakes), then classify the remaining transactions into quintiles with a YlOrRd palette (pale yellow → dark red, ColorBrewer 5 classes).

We then lay a regular 100 m × 100 m fishnet over the sales extent (grid_create, Lambert93 EPSG:2154), aggregate each tile's mean €/m² from the DVF points it contains (spatial_aggregate, predicate contains), and classify the cells into quintiles with the same YlOrRd ramp to paint a clean heatmap — contiguous square tiles that read like a thematic continuous surface, much easier than scattered point dots.

Source: DVF Etalab

Source	Content	Features (Versailles 2022-2024)	Key attributes
geo-dvf Etalab	Geolocated real-estate transactions	~7,000 raw across 8 communes (Versailles + Le Chesnay-Rocquencourt + Viroflay + Vélizy + Jouy + Buc + Saint-Cyr + Bailly, 2022-2024) → ~5,100 residential after filter	`valeur_fonciere`, `surface_reelle_bati`, `type_local`, `nature_mutation`, `date_mutation`

CSV files are published per year + commune: https://files.data.gouv.fr/geo-dvf/latest/csv/{year}/communes/{dept}/{insee}.csv

Versailles = INSEE 78646, département 78.

bash

python examples/prepare_playground_data.py --city versailles
gispulse info examples/datasets/versailles_bdtopo.gpkg --layer dvf_ventes

The script concats 2022+2023+2024, builds Point geometry from longitude/latitude, drops rows without coordinates or prices, then writes the dvf_ventes layer into the GPKG.

Pipeline (8 steps)

dvf_ventes ──► filter (nature_mutation=='Vente' AND type_local in ['Maison','Appartement'])
                │                                                       # residential sales
                ▼
            calculate → price_per_m2 = valeur_fonciere / surface_reelle_bati
                │
                ▼
            filter (1500 <= price_per_m2 <= 25000)                      # DVF outlier trim
                │
                ▼
            classify → price_class (1..5) + price_color (YlOrRd)        # quintiles (points)
                         method: quantile, bins: 5
                │
                ▼
            grid_create → regular 100 m × 100 m fishnet                  # square tiles
              ref_layer: drop_price_outliers   (envelope of sales)
              cell_size: 100                    (metres)
              crs_meters: EPSG:2154             (Lambert93)
              clip_to_extent: true              (drop tiles outside sales)
                │
                ▼
            spatial_aggregate                                             # spatial attribute
              ref_layer: drop_price_outliers                              # tile ⊇ DVF points
              predicate: contains
              agg: mean_price_per_m2, max_price_per_m2, tx_count
                │
                ▼
            filter (tx_count > 0)                                         # keep tiles ≥1 sale
                │
                ▼
            classify → tile_class (1..5) + tile_color (YlOrRd)            # heatmap choropleth
                         field: mean_price_per_m2, method: quantile, bins: 5

Steps 1–4 (points) — ColorBrewer YlOrRd 5-class palette (#ffffb2, #fecc5c, #fd8d3c, #f03b20, #bd0026) attached per feature in price_color; each quintile holds ~20% of mutations.

Steps 5–8 (tiles) — grid_create emits a 100 m × 100 m fishnet in Lambert93 (exact metric) over the filtered-sales extent (~950 non-empty tiles); spatial_aggregate computes mean_price_per_m2, max_price_per_m2, tx_count per tile from the DVF points it contains; empty tiles are dropped; the final classify paints the choropleth as a high-resolution heatmap — fine-grained 100 m cells, easy thematic read, print-ready for QGIS export. Note: at 100 m on the wider S5 extent ~25 % of tiles carry a single transaction (vs ~35 % at the previous 50 m mesh on Versailles centre alone), so quintiles are more statistically stable while still keeping a fine-grained thematic read.

Rules

json

{
  "version": 2,
  "ref_layers": { "dvf_ventes": "dvf_ventes" },
  "steps": [
    {
      "id": "filter_residential_sales",
      "capability": "filter",
      "params": {
        "expression": "nature_mutation == 'Vente' and type_local in ['Maison', 'Appartement']"
      }
    },
    {
      "id": "compute_price_per_m2",
      "capability": "calculate",
      "params": { "expressions": { "price_per_m2": "valeur_fonciere / surface_reelle_bati" } },
      "input": "filter_residential_sales"
    },
    {
      "id": "drop_price_outliers",
      "capability": "filter",
      "params": { "expression": "price_per_m2 >= 1500 and price_per_m2 <= 25000" },
      "input": "compute_price_per_m2"
    },
    {
      "id": "classify_price_quintiles",
      "capability": "classify",
      "params": {
        "field": "price_per_m2",
        "method": "quantile",
        "bins": 5,
        "class_col": "price_class",
        "color_col": "price_color",
        "palette": ["#ffffb2", "#fecc5c", "#fd8d3c", "#f03b20", "#bd0026"]
      },
      "input": "drop_price_outliers"
    },
    {
      "id": "create_price_grid",
      "capability": "grid_create",
      "params": {
        "ref_layer": "drop_price_outliers",
        "cell_size": 100,
        "crs_meters": "EPSG:2154",
        "clip_to_extent": true
      },
      "input": "drop_price_outliers"
    },
    {
      "id": "aggregate_price_to_grid",
      "capability": "spatial_aggregate",
      "params": {
        "ref_layer": "drop_price_outliers",
        "predicate": "contains",
        "agg": {
          "mean_price_per_m2": ["price_per_m2", "mean"],
          "max_price_per_m2": ["price_per_m2", "max"],
          "tx_count": ["price_per_m2", "count"]
        }
      },
      "input": "create_price_grid"
    },
    {
      "id": "keep_cells_with_sales",
      "capability": "filter",
      "params": { "expression": "tx_count > 0" },
      "input": "aggregate_price_to_grid"
    },
    {
      "id": "classify_grid_choropleth",
      "capability": "classify",
      "params": {
        "field": "mean_price_per_m2",
        "method": "quantile",
        "bins": 5,
        "class_col": "tile_class",
        "color_col": "tile_color",
        "palette": ["#ffffb2", "#fecc5c", "#fd8d3c", "#f03b20", "#bd0026"]
      },
      "input": "keep_cells_with_sales"
    }
  ]
}

Download

scenario-6-rules.json

Execution

bash

gispulse run examples/datasets/versailles_bdtopo.gpkg \
  --layer dvf_ventes \
  --rules playground/scenario-6-rules.json \
  -o output/versailles_price_map.gpkg

gispulse serve output/versailles_price_map.gpkg

Expected Result

Output schema — point layer (step 4)

Column	Type	Source	Description
`geometry`	Point	source	DVF parcel centroid
`date_mutation`	date	source	Sale date
`nature_mutation`	string	source	"Vente" after filter
`type_local`	string	source	"Maison" or "Appartement"
`valeur_fonciere`	float	source	Sale price (€)
`surface_reelle_bati`	float	source	Built surface (m²)
`price_per_m2`	float	step 2 (`calculate`)	Price per square meter
`price_class`	int	step 4 (`classify`)	Quintile 1..5
`price_color`	string	step 4 (`classify`)	Hex color (YlOrRd palette)

Output schema — 50 m tile choropleth (step 8)

Column	Type	Source	Description
`geometry`	Polygon	step 5 (`grid_create`)	100 m × 100 m square tile in Lambert93
`row`	int	step 5 (`grid_create`)	Fishnet row index
`col`	int	step 5 (`grid_create`)	Fishnet column index
`mean_price_per_m2`	float	step 6 (`spatial_aggregate`)	Mean €/m² of contained DVF points
`max_price_per_m2`	float	step 6 (`spatial_aggregate`)	Max €/m² observed in this tile
`tx_count`	int	step 6 (`spatial_aggregate`)	Number of mutations in tile
`tile_class`	int	step 8 (`classify`)	Quintile 1..5 of mean price
`tile_color`	string	step 8 (`classify`)	Hex color of the choropleth

Versailles 2022-2024 quintile edges (typical, after outlier trim)

Q1 (< ~€5,200/m²): pale yellow #ffffb2 — peripheral segments, atypical units
Q2 (€5,200 → €6,400/m²): light orange #fecc5c
Q3 (€6,400 → €7,300/m²): orange #fd8d3c — market median
Q4 (€7,300 → €8,500/m²): red-orange #f03b20
Q5 (> ~€8,500/m²): dark red #bd0026 — Notre-Dame, Château district

Quintile edges are recomputed dynamically: bins shift if you change the period or the spatial filter.

Full interactive playground

Live 8-step pipeline (requires demo backend).

Points (DVF) — per-mutation gradient

filter_residential_sales (orange) — keep only Maison / Appartement sales
compute_price_per_m2 (cyan) — ratio valeur_fonciere / surface_reelle_bati
drop_price_outliers (orange) — 1500 ≤ price/m² ≤ 25000 €
classify_price_quintiles (red) — quintiles + YlOrRd palette → color gradient on points

Choropleth (tiles) — 50 m heatmap

create_price_grid (teal) — 100 m × 100 m fishnet in Lambert93 clipped to the DVF extent (~950 non-empty tiles)
aggregate_price_to_grid (purple) — spatial_aggregate: per tile, mean price_per_m2 of contained DVF points (+ max, + count)
keep_cells_with_sales (orange) — drop empty tiles (tx_count > 0)
classify_grid_choropleth (red) — quintiles on mean_price_per_m2 + YlOrRd → heatmap choropleth

DVF popup: date, type_local, valeur_fonciere, surface_reelle_bati, price_per_m2, price_class. Tile popup: row, col, mean_price_per_m2, max_price_per_m2, tx_count, tile_class. Legend: each quintile ~20% (points, then tiles), same palette, continuous thematic read.

Try it live

Live Demolists available capabilities (filter, calculate, classify) used by the S6 pipeline

GET/capabilities

Live Demolists demo datasets, including versailles_bdtopo with the dvf_ventes layer loaded for this scenario

GET/datasets

Next steps

S5: Green Spaces — another Versailles workflow
Vector capabilities — classify, filter, calculate
DVF Etalab — 2014-2025, updated twice per year, all communes

S6: Price-per-m² Map — Versailles (DVF) ​

Use Case ​

Source: DVF Etalab ​

Pipeline (8 steps) ​

Rules ​

Execution ​

Expected Result ​

Full interactive playground ​

Try it live ​

Next steps ​