Cloud-Native GIS Stack¶

The Problem with Traditional HEC-RAS Workflows¶

Traditional HEC-RAS result delivery relies on:

Shapefiles — row-oriented, split into multiple files, no compression, 10-character column names
File Geodatabases — proprietary format, requires Esri software to read
WMS/WFS servers — require always-on server infrastructure
Desktop GIS — spatial analysis locked to licensed desktop software

The Cloud-Native Solution¶

ras2cng converts HEC-RAS outputs to open, cloud-optimized formats:

GeoParquet¶

Format: Apache Parquet + GeoArrow geometry encoding
Benefits: Columnar storage (read only the columns you need), snappy compression (5–10x smaller than Shapefile), Arrow-native (zero-copy into pandas/DuckDB)
Access: Any language with Parquet support — Python, R, JavaScript, SQL

DuckDB¶

Format: In-process analytical SQL engine
Benefits: Reads GeoParquet directly without loading into memory, spatial extension for geometry operations, runs in serverless/embedded contexts
Access: ras2cng query, Python DuckSession, or any DuckDB client

PMTiles¶

Format: Single-file archive of map tiles (Protomaps)
Benefits: Serverless HTTP range requests — serve from S3, R2, or GitHub Pages with no tile server
Access: MapLibre GL JS, Leaflet with pmtiles plugin, any HTTP range-capable client

PostGIS¶

Format: PostgreSQL + spatial extensions
Benefits: Enterprise-grade, multi-user, GIST spatial index, full SQL with geometry operations
Access: QGIS, ArcGIS, any PostgreSQL client, SQLAlchemy

Data Flow¶

HEC-RAS project
├── *.g01.hdf  (HDF geometry)
├── *.g01      (text geometry)
└── *.p01.hdf  (plan results)
        │
        ▼
ras2cng geometry / results
        │
        ▼
GeoParquet (.parquet)          ← universal intermediate
    ├── DuckDB queries          ← serverless analytics
    ├── PMTiles generation      ← serverless map tiles
    └── PostGIS sync            ← enterprise database

Why This Matters for Flood Modeling¶

Flood model results are large (millions of mesh cells), spatial (polygon geometry), and frequently queried (depth > threshold, area statistics). The cloud-native stack handles all three:

Large: GeoParquet reads only requested columns; DuckDB streams without loading everything to RAM
Spatial: PMTiles serves flood extents to web maps; PostGIS enables spatial joins with other datasets
Queryable: DuckDB SQL filters mesh cells by depth threshold in milliseconds without a GIS application