AMX-26: Cadastral Data to PMTiles

Tagline: Openness lowers the cost of trust. PMTiles lowers the cost of distribution.

A collaborative initiative to convert open INSPIRE cadastral data from European countries into PMTiles format, enabling efficient and fast geospatial information access.

Vision

This project transforms open cadastral data within the European INSPIRE framework into PMTiles, improving accessibility and performance for geospatial applications. Similar to Mapterhorn's approach for elevation data, we focus on cadastral data, with country-level PMTiles files served efficiently via rsync.

Project Status

Phase: Prototype Implementation

• Priority Countries: Czechia (first), France (next)
• Approach: Single monorepo with per-country data pipelines
• Deployment: rsync-based file hosting to pod@pod.local

Repository Structure

amx-26/
├── countries/
│   ├── cz/
│   │   ├── README.md
│   │   ├── data/
│   │   │   ├── source/      # Raw INSPIRE data (local fallback)
│   │   │   └── output/      # (deprecated - use data/output)
│   │   └── scripts/
│   │       └── convert.py   # Conversion pipeline
│   └── fr/
│       ├── README.md
│       ├── data/
│       │   ├── source/
│       │   └── output/
│       └── scripts/
│           └── convert.py
├── data/
│   ├── sources/
│   │   ├── cz/              # Czechia cadastral source data
│   │   │   └── geojsonseq/  # Per-resource GeoJSONSeq outputs
│   │   └── fr/              # France cadastral source data
│   └── output/              # Generated PMTiles files
│       ├── cz.pmtiles
│       └── fr.pmtiles
├── common/
│   ├── scripts/             # Shared conversion utilities
│   └── lib/                 # Reusable Python modules
├── docs/
│   ├── architecture.md
│   └── data-format.md
└── Justfile

Quick Start

Prerequisites

• Python 3.10+
• tippecanoe (for vector tile generation)
• just (task runner, install here)
• yq (YAML/XML query tool, brew install yq) - for parsing ATOM feeds
• xq (XML query tool, brew install xq) - alternative/companion to yq
• parallel (GNU parallel, brew install parallel) - for download concurrency

Workflow

# List all available commands
just --list

# Step 1: Fetch ATOM feed (stores JSON locally)
just fetch-feed cz

# Step 2: Show feed information
just feed-info cz

# Step 3: Download GML files from dataset feeds
just download cz

# Optional: control download concurrency
# SAMPLE_SIZE=-1 processes all feeds (0 also works), PARALLEL_JOBS controls parallelism
SAMPLE_SIZE=-1 PARALLEL_JOBS=2 just download cz

# Step 4: Convert to PMTiles
just convert cz

# Step 5: Upload to remote host
just upload cz

# Or: Full pipeline (all steps)
just process cz

Country Pipelines

Czechia (cz)

• Status: INSPIRE data source identified
• Format: Shapefile (distributed via ATOM feed)
• Source: ČÚZK ATOM Feed
• License: CC-BY
• Target: cz.pmtiles

GeoJSONSeq Outputs:

• Per-resource files in data/sources/cz/geojsonseq/*.geojsonseq
• Files are written atomically (temporary name, then rename)
• Existing files are skipped on subsequent runs for faster rebuilds
• Layer-based tippecanoe.minzoom/maxzoom metadata is added during download
• Conversion streams per-resource GeoJSONSeq directly to tippecanoe (no combined temp file)

Data Acquisition Strategy:

# ATOM feed parsing with yq/xq
yq -p xml '.feed.entry[] | select(.id == "...") | .link.@href' https://atom.cuzk.cz/CP/CP.xml

# Automated download via download.sh script
just download cz

Zoom Level Strategy:

• CadastralZoning: minzoom=5, maxzoom=13 (broad coverage, lightweight data)
• CadastralParcel: minzoom=14, maxzoom=18 (transitions at z14 from CadastralZoning)
• CadastralBoundary: minzoom=16, maxzoom=18 (high precision detail)
• tippecanoe maxzoom: 18

This ensures continuous coverage across all zoom levels without gaps where features disappear.

Coordinate Precision:

• Source GML: 1cm precision (0.01m in S-JTSK coordinate system EPSG:5514)
• Output GeoJSONSeq: 10 decimal places in WGS84 (0.01mm precision at 50°N latitude)
• Setting: ogr2ogr -lco COORDINATE_PRECISION=10
• Rationale: Future-proof for millimeter-precision data while preserving original centimeter precision

Verification: Zoom level distribution verified using tippecanoe-decode on actual tiles:

• z5: CadastralZoning only (broad coverage)
• z10: CadastralZoning only (continues until maxzoom=13)
• z14: CadastralParcel appears (transitions from CadastralZoning)
• z16: CadastralBoundary added (high-detail boundaries)
• z18: CadastralParcel + CadastralBoundary (maximum detail)

Note: PMTiles metadata shows global minzoom/maxzoom (0-18) for all layers, but actual tiles contain only the appropriate layers per zoom level as configured in GeoJSONSeq features.

Tippecanoe Settings:

• Do not drop features: --no-feature-limit
• Increase per-tile size limit: --maximum-tile-bytes 1000000
• Temporary directory: ./tmp/amx-26-cz/tippecanoe-tmp (avoids small system temp space)
• Progress display: Real-time processing progress shown during conversion
• Processing may take several hours for full dataset (44 GB, 84M+ features)
• Attribution: --attribution "© ČÚZK" (CC BY 4.0 license)

Current Status:

• Generated PMTiles: 13GB (10.7M tiles, z0-18, optimized)
• Layers: CadastralBoundary (62M features), CadastralParcel (22M features), CadastralZoning (13K features)
• Average tile size: 1.29KB
• Max tile size: 504KB

France (fr)

• Status: Planned
• Format: To be determined
• Target: fr.pmtiles

Web Viewer

Interactive web map viewer for Czech cadastral data:

• URL: https://amx-project.github.io/amx-26/
• Tech Stack: Vite + MapLibre GL JS + Protomaps basemaps
• Features:
  • Czech cadastral layers (zoning, parcels, boundaries) with red outlines
  • Labels for cadastral zones and parcels
  • 3D terrain with hillshade (dark theme optimized)
  • Layer control for visibility/opacity management
  • Parcel information on hover (label, area, national reference)
  • Loading overlay during tile streaming
  • Hash-based navigation for sharing locations

Local Development:

# Install dependencies
just site-install

# Run development server
just site-dev

# Build for production (outputs to docs/)
just site-build

# Preview production build
just site-preview

Data Sources:

• Cadastral PMTiles: pmtiles://https://tunnel.optgeo.org/cz.pmtiles
• Basemap: Protomaps (OpenStreetMap data)
• Terrain: Matterhorn DEM (terrarium encoding)

Optimization

PMTiles Optimization with vt-optimizer-rs

The project uses vt-optimizer-rs to analyze and optimize vector tiles:

# Install vt-optimizer-rs (requires Rust)
git clone https://github.com/unvt/vt-optimizer-rs.git /tmp/vt-optimizer-rs
cd /tmp/vt-optimizer-rs
cargo build --release
cp target/release/vt-optimizer ~/.cargo/bin/

# Inspect PMTiles structure
vt-optimizer inspect data/output/cz.pmtiles

# Inspect with JSON output for analysis
vt-optimizer inspect data/output/cz.pmtiles --report-format json > cz-report.json

Attribute Reduction Strategy

Based on web viewer usage analysis, the following attributes can be removed to reduce file size:

CadastralBoundary (6 attributes → 2 attributes):

• Keep: estimatedAccuracy, estimatedAccuracy_uom (precision metadata)
• Remove: beginLifespanVersion, gml_id, localId, namespace

CadastralParcel (9 attributes → 4 attributes):

• Keep: label, areaValue, areaValue_uom, nationalCadastralReference
• Remove: gml_id, localId, namespace, beginLifespanVersion, pos

CadastralZoning (13 attributes → 1 attribute):

• Keep: label
• Remove: LocalisedCharacterString, beginLifespanVersion, gml_id, localId, namespace, nationalCadastalZoningReference, originalMapScaleDenominator, pos, script, sourceOfName, text, language

Implementation: Attributes are filtered during GeoJSONSeq generation via jq in countries/cz/scripts/process_feed.sh, so tippecanoe ingests already-trimmed features.

Benefits:

• Reduced tile file size (observed 21GB → 13GB, about 38%)
• Reduced average tile size (2.07KB → 1.29KB, about 37%)
• Faster tile transmission and display
• Lower bandwidth requirements
• Improved performance in web viewer and other clients

vt-optimizer report excerpt (from tmp/cz_report.json):

{
  "overall": {
    "tile_count": 10744660,
    "total_bytes": 14158698421,
    "avg_bytes": 1317,
    "max_bytes": 516523
  },
  "vector_layers": [
    {
      "id": "CadastralBoundary",
      "fields": {
        "estimatedAccuracy": "Number",
        "estimatedAccuracy_uom": "String"
      }
    },
    {
      "id": "CadastralParcel",
      "fields": {
        "areaValue": "Number",
        "areaValue_uom": "String",
        "label": "Mixed",
        "nationalCadastralReference": "String"
      }
    },
    {
      "id": "CadastralZoning",
      "fields": {
        "label": "String"
      }
    }
  ]
}

Development

Adding a New Country

1. Create a new directory: countries/{country}/
2. Add country-specific README.md with data source details
3. Create scripts/convert.py tailored to your data format
4. Update Justfile with country-specific tasks

Shared Libraries

Place reusable conversion logic in common/lib/. For example:

• INSPIRE data parsing
• PMTiles encoding utilities
• Metadata handling

Deployment

Upload to Pod Server

Files are hosted via rsync to a remote pod server:

just upload cz
# Executes: rsync -av data/output/cz.pmtiles pod@pod.local:/home/pod/x-24b/data/

Tile Access URLs

Once deployed, Czech cadastral tiles are available at:

• PMTiles (direct): https://tunnel.optgeo.org/cz.pmtiles
• TileJSON metadata: https://tunnel.optgeo.org/martin/cz
• XYZ tiles: https://tunnel.optgeo.org/martin/cz/{z}/{x}/{y}

The martin server provides TileJSON compatibility for web mapping libraries (Leaflet, Mapbox GL, etc.)

References

• INSPIRE Directive
• PMTiles Specification
• Matterhorn Project (elevation data)
• amx-project
• x-24b

Data Sources:

• ČÚZK Cadastral Data ATOM Feed - Czech cadastral data

Tools:

• yq - YAML/XML query tool (brew install yq)
• xq - XML query tool (brew install xq)

License

See LICENSE