Cloud Native GIS Tools¶
Archive complete HEC projects and turn their geometry and results into open, analysis-ready cloud-native formats. Choose the library that matches your modeling system.
-
Export complete HEC-RAS projects — geometry, terrain, plans, and simulation results — to GeoParquet, PMTiles, DuckDB, PostGIS, and web-map-ready datasets.
-
Export complete HEC-HMS projects — basin geometry, meteorology, control data, and simulation results — to GeoParquet, PMTiles, DuckDB, and PostGIS.
Both libraries preserve the source project while producing portable datasets that work well in Python, desktop GIS, databases, object storage, and browser maps.
ras2cng 0.13: Rust engines¶
ras2cng 0.13 runs its heaviest steps (mapping, archiving, tiling, Cloud Optimized GeoTIFF writing, and Manning's n rasters) in Rust engines. Every Rust path was checked against RAS Mapper or ras2cng's earlier Python output, and ras2cng falls back to RAS Mapper or Python automatically wherever Rust has not been proven.
| Task | Before | ras2cng 0.13 |
|---|---|---|
| Web-map vector tiles, Bald Eagle Creek with results | 1,745 s (tippecanoe) | 36 s, and about 10% smaller |
| Inundation boundary (Davis) | 85 s (RAS Mapper) | 3.5 s |
| Manning's n raster ("Final n Values") | 45–49 s, approximate | about 2 s, identical to RAS Mapper |
| Terrain preparation for mapping, unmodified terrains | 160–240 s HEC-RAS terrain export | none, or about 7 s |
| Max depth × velocity, depth × velocity², Froude | 206 s (RAS Mapper) | 141 s |
| Arrival time and percent time inundated | 82 s (RAS Mapper) | 31 s |
| Full project archive with results | 29.1 s, 4.0 GB peak memory | 12.5 s, 0.76 GB |
| Reading mesh geometry from HEC-RAS HDF files | baseline | 4–19× faster |
| Cloud Optimized GeoTIFF of a large terrain | 70.7 s (GDAL) | 6–11 s |
| Raster web tiles: terrain / depth | 13.4 s / 4.7 s | 6.8 s / 1.6 s |
| Peak memory, terrain tiles to zoom 14 | 13.4 GB | 1.0 GB |
Larger models gain more. From 2,716 to 1,063,548 mesh cells, the project archive goes from 1.3× to 5.2× faster, and on the largest model 0.12 ran out of memory (about 34 GB) while 0.13 finished in 37 s with 4.4 GB. Reading the mesh goes from 1.5× to 33× faster. Web-map tiles with results are 7–65× faster; their time depends on how many result layers a model publishes as well as on its mesh size.
All values (median of 3 runs; peak memory of the whole process tree)
| Model | Source | Cells | Task | 0.12.0 | 0.13.0 | Speedup |
|---|---|---|---|---|---|---|
| Davis p02 | HEC example | 2,716 | Project archive with results | 3.99 s, 302 MiB | 2.98 s, 250 MiB | 1.3× |
| Davis p02 | HEC example | 2,716 | Web-map tiles with results | 27.6 s, 565 MiB | 3.96 s, 262 MiB | 7.0× |
| Davis p02 | HEC example | 2,716 | Web-map tiles, geometry only | 3.68 s, 235 MiB | 2.58 s, 201 MiB | 1.4× |
| Davis p02 | HEC example | 2,716 | Reading mesh polygons | 0.30 s, 124 MiB | 0.20 s, 127 MiB | 1.5× |
| Bald Eagle bridges p05 | HEC example | 24,718 | Project archive with results | 12.6 s, 1,243 MiB | 3.55 s, 342 MiB | 3.5× |
| Bald Eagle bridges p05 | HEC example | 24,718 | Web-map tiles with results | 561 s, 5,124 MiB | 12.7 s, 653 MiB | 44× |
| Bald Eagle bridges p05 | HEC example | 24,718 | Web-map tiles, geometry only | 11.7 s, 452 MiB | 4.29 s, 258 MiB | 2.7× |
| Bald Eagle bridges p05 | HEC example | 24,718 | Reading mesh polygons | 1.60 s, 155 MiB | 0.26 s, 144 MiB | 6.2× |
| Bald Eagle p07 | HEC example | 87,039 | Project archive with results | 31.5 s, 3,685 MiB | 5.40 s, 652 MiB | 5.8× |
| Bald Eagle p07 | HEC example | 87,039 | Web-map tiles with results | did not finish in 30 min | 27.9 s, 1,498 MiB | over 65× |
| Bald Eagle p07 | HEC example | 87,039 | Web-map tiles, geometry only | 31.5 s, 715 MiB | 6.17 s, 381 MiB | 5.1× |
| Bald Eagle p07 | HEC example | 87,039 | Reading mesh polygons | 4.78 s, 238 MiB | 0.30 s, 199 MiB | 16× |
| Upper Guadalupe UPGU3, TX | FEMA eBFE (12100201) | 292,677 | Project archive with results | 97.4 s, 4,606 MiB | 29.1 s, 2,694 MiB | 3.3× |
| Upper Guadalupe UPGU3, TX | FEMA eBFE (12100201) | 292,677 | Web-map tiles, geometry only ¹ | 1,719 s, 2,120 MiB | 1,615 s, 1,321 MiB | 1.1× |
| Upper Guadalupe UPGU3, TX | FEMA eBFE (12100201) | 292,677 | Reading mesh polygons | 17.9 s, 532 MiB | 0.69 s, 416 MiB | 26× |
| Spring Creek, TX | FEMA eBFE (12040102) | 340,163 | Project archive with results | 98.4 s, 4,097 MiB | 19.2 s, 2,137 MiB | 5.1× |
| Spring Creek, TX | FEMA eBFE (12040102) | 340,163 | Web-map tiles with results | 1,066 s, 9,950 MiB | 157 s, 3,842 MiB | 6.8× |
| Spring Creek, TX | FEMA eBFE (12040102) | 340,163 | Web-map tiles, geometry only | 180 s, 1,478 MiB | 68.3 s, 721 MiB | 2.6× |
| Spring Creek, TX | FEMA eBFE (12040102) | 340,163 | Reading mesh polygons | 22.0 s, 651 MiB | 0.91 s, 489 MiB | 24× |
| Spring River, AR | FEMA eBFE (11010010) | 422,715 | Project archive with results | 129 s, 5,818 MiB | 24.8 s, 3,176 MiB | 5.2× |
| Spring River, AR | FEMA eBFE (11010010) | 422,715 | Web-map tiles with results | 1,718 s, 16,070 MiB | 178 s, 5,468 MiB | 9.6× |
| Spring River, AR | FEMA eBFE (11010010) | 422,715 | Web-map tiles, geometry only | 148 s, 1,542 MiB | 27.0 s, 778 MiB | 5.5× |
| Spring River, AR | FEMA eBFE (11010010) | 422,715 | Reading mesh polygons | 25.1 s, 694 MiB | 0.85 s, 472 MiB | 29× |
| Upper Guadalupe UPGU4, TX | FEMA eBFE (12100201) | 1,063,548 | Project archive with results | ran out of memory (about 34 GB) | 36.6 s, 4,381 MiB | — |
| Upper Guadalupe UPGU4, TX | FEMA eBFE (12100201) | 1,063,548 | Reading mesh polygons | 61.4 s, 1,537 MiB | 1.84 s, 924 MiB | 33× |
¹ Dominated in both versions by one Manning's n land-cover polygon of 1,062,037 vertices that is tiled at every zoom (1,534 of 1,564 s in a profiled 0.13 run), so it is not a mesh-size result and is left out of the figure. UPGU4 has a similar polygon (533,600 vertices); its tile runs were not made. UPGU3 tiles with results were not run.
Versions: 0.12.0 with ras-commander 0.104.0 and PyPI freestiler 0.3.0; 0.13.0 with ras-commander 0.113.0, fast-hdf 0.3.0, fast-map 0.12.0 and the CLB builds of terrio 0.5.0.1 and freestiler 0.3.0.1. tippecanoe was not installed, so both versions tiled with freestiler. Both versions write the same layers and row counts. Other work used about half of the workstation's CPU throughout, so absolute times are higher than on an idle machine; the speedups are the more reliable comparison.
Measured on the HEC-RAS example projects (Bald Eagle Creek and Davis) on a shared Windows workstation; times depend on hardware and model size. Max water surface and depth for the Bald Eagle bridges example now map in about 4.5 s end to end, with no HEC-RAS installation needed.
- Accuracy: water surface, depth, velocity, the derived products, arrival time, duration and percent time inundated matched HEC-RAS 7.0.1 RAS Mapper with zero pixel differences on the test models. Rust inundation boundaries are close to, but not identical with, RAS Mapper's.
- What still uses RAS Mapper: 1D models, terrains with modifications, shear stress, and (unless you opt in) storage areas.
- Optional fast builds: the tiling and archive speedups use the CLB builds of terrio and freestiler, published as GitHub Release downloads. See the ras2cng docs for install commands.
Open-source projects from CLB Engineering Corporation.