pyhazus

A standalone, cross-platform Python implementation of the FEMA Hazus loss estimation methodology, reconstructed from the Hazus 7.0 / 7.2 installers and the FEMA technical manuals. It has no ArcGIS Pro, SQL Server or Windows COM dependency at runtime.

Status: early development. Flood (Hazus 7.2) is ported:

• study-region creation from FEMA state inventories;
• depth grids, scenario creation and area weighting;
• general building stock: damage, economic loss, damage counts and depreciated loss (14 result tables);
• essential facilities and utilities;
• debris, shelter, and average annualized loss.

Each step is checked against the original Hazus T-SQL, run in SQL Server as a reference oracle (see Verification).

Install

pip install -e ".[geo,dev]"

geo adds rasterio, geopandas and shapely (needed for depth grids). tools adds pymssql, needed only to export Hazus databases from SQL Server.

Use

A study region is a directory with one Parquet file per Hazus table (hzCensusBlock.parquet, flSchemeMapping.parquet, hzCareFlty.parquet, ...). Census blocks, the boundary and facilities are GeoParquet in EPSG:4326.

Create one from a FEMA state inventory (the Hazus state download, <ST>.zip):

1. Export the state once from SQL Server. This is the only step that needs SQL Server; see docs/extraction.md.

   hazus inventory export-state DC states/DC

2. Create regions without SQL Server, as Hazus's aggregation does (see docs/region.md):

   hazus region create states/DC region/ --counties 11001

An existing Hazus study-region database can also be exported directly with hazus inventory export MyRegion region/.

Run a single-event scenario:

hazus flood run region/ results/ --riverine depth.tif --units meters

Run a scenario with return periods (three or more give AAL):

hazus flood run region/ results/ --riverine d10.tif@10 --riverine d100.tif@100 --riverine d500.tif@500

Results are written as one file per Hazus result table (flFRGBSEcLossByTotal, flFRCareFlty, flFRDebris, ...), as Parquet or, with --format csv, CSV.

From Python:

from hazus.flood import DepthGrid, create_scenario, run_scenario
from hazus.io import read_region

region = read_region("region/")
rasters = create_scenario([DepthGrid("depth.tif", "riverine", units="meters")], "work/")
results = run_scenario(region, rasters)
results["flFRGBSEcLossByTotal"]

Fidelity

The port reproduces Hazus 7.2 as it behaves, including its numeric semantics and its bugs: float32 arithmetic, truncation into integer columns, T-SQL evaluation order, and the quirks listed in docs/flood-dataflow.md and docs/flood-gbs.md. Where Hazus would stop with an unhandled error, the port raises a clear exception instead, or documents its choice.

Verification

pytest runs unit tests everywhere; GitHub Actions runs them on Linux, Windows and macOS. The SQL differential tests (tests/test_sql_parity*.py) run when a SQL Server with the Hazus databases attached is reachable (HAZUS_SA_PASSWORD set; see docs/extraction.md). They load synthetic study regions, run the original stored procedures, and compare every result table with the Python output. Integer results must match exactly. tests/test_sql_parity_state.py does the same on a study region that Hazus built from a FEMA state database.

Area weighting (depth grid → flooded share of each block) runs in ArcGIS in Hazus, so there is no SQL to compare with. tests/test_hll_crosscheck.py compares it instead with Hazus 5.1 runs published in the Hazus Loss Library. Two packages are covered: the Little Blue River in Missouri (7 scenarios, 4 ft grids) and Lake Ontario at Hamlin Beach in New York (5 scenarios, 1 m grids of 35,175 × 5,915 cells, 252 of 670 blocks on the Great Lakes). In each scenario, flooded shares agree within 0.25 percentage points on average and mean depths within 0.07 ft. A block flooded by only a few cells (at most 0.13 % of the block) can appear on one side only. Hazus 5.1 area-weights on the depth grid's own projected cells rather than 7.2's 0.00001° cells, so these comparisons use tolerances.

Layout

• src/hazus/: the library (import hazus), with the Hazus 7.2 lookup tables in src/hazus/data/
• tools/: one-time tooling to unpack the FEMA installers and export their databases
• docs/: reconstructed methodology notes
• tests/: unit tests and SQL differential tests

The FEMA installers, extracted payloads and decompiled sources are not part of this repository (see .gitignore).

License

The code is released under the BSD 3-Clause License. The Hazus lookup tables in src/hazus/data/hazus72/ are FEMA data exported unchanged from the Hazus 7.2 installer. The BSD license covers this project's code, not those tables.