watershed-retrieve 1.0.5

Pre-delineated MERIT-Hydro watershed boundaries for ~60,000 gauging stations across 16 countries

watershed-retrieve

Instant access to ~60,000 pre-delineated MERIT-Hydro watershed boundaries and river networks across 16 countries, served as GeoParquet. No data download required — basins are fetched on demand from a public CDN and cached locally.

Background

This library is a community contribution to the RivRetrieve ecosystem. Where RivRetrieve provides observed streamflow time series for gauging stations worldwide, watershed-retrieve adds the corresponding watershed boundaries and river networks, delineated on the MERIT-Hydro digital elevation model.

The watershed delineation was performed using a Rust reimplementation of the algorithm described in mheberger/delineator. This is the same methodology used by CAMELS-DE (Loritz et al., 2024) to derive consistent catchment boundaries for 1582 gauging stations across Germany from MERIT Hydro.

See the original proposal: kratzert/RivRetrieve-Python#87.

Installation

pip install watershed-retrieve

Quick Start

import watershed_retrieve as wr

# Zero-config — data is fetched from R2 CDN and cached locally
watershed = wr.get_watershed("portugal", "04K/04A")

# With river network
watershed, rivers = wr.get_watershed_with_rivers("portugal", "04K/04A")

# Bulk retrieval — all watersheds for a country
all_watersheds = wr.get_watersheds("portugal")

To use a local data directory instead of the CDN:

# Option 1: Environment variable
# export WATERSHED_RETRIEVE_DATA_DIR=/path/to/parquet/files

# Option 2: Programmatic
wr.configure("/path/to/parquet/files")

# Option 3: Explicit backend selection
from watershed_retrieve import Backend
wr.configure(backend=Backend.R2, cache_dir=Path("~/.my-cache"))

API Reference

Discovery

# List all supported countries
wr.available_countries()
# -> ['australia', 'brazil', 'canada', ..., 'usa']

# List gauge IDs for a country
wr.available_gauges("portugal")
# -> ['02G-02H', '02O-01H', ..., '16J-01H']  (~710 gauges)

Single Watershed

# Watershed boundary (GeoDataFrame, 1 row)
gdf = wr.get_watershed("portugal", "04K/04A")

# Watershed + river network (WatershedResult — unpackable NamedTuple)
result = wr.get_watershed_with_rivers("portugal", "04K/04A")
watershed, rivers = result

Bulk Retrieval

# All watersheds for a country
gdf = wr.get_watersheds("portugal")  # -> GeoDataFrame (~710 rows)

# Subset by gauge IDs
gdf = wr.get_watersheds("portugal", ["04K/04A", "05G/01A"])

# With rivers
result = wr.get_watersheds_with_rivers("portugal")
result.watershed  # GeoDataFrame
result.rivers     # GeoDataFrame

Gauge ID Normalization

Slashes are automatically normalized to dashes:

wr.get_watershed("portugal", "04K/04A")   # slash
wr.get_watershed("portugal", "04K-04A")   # dash — equivalent

Errors

from watershed_retrieve import (
    WatershedRetrieveError,     # base class
    CountryNotFoundError,       # invalid country name
    GaugeNotFoundError,         # gauge ID not in dataset
    DataNotFoundError,          # parquet file missing
    DataUnavailableError,       # region exists but data not yet extracted
    R2ConnectionError,          # CDN fetch failed
)

DataUnavailableError is raised for regions where gauging stations are registered in RivRetrieve but MERIT-Hydro basin delineation is pending (e.g., UK regions — the British Isles fall outside MERIT-Hydro coverage).

Supported Countries

Country	Gauges	Status
Australia	~6,200	Available
Brazil	~4,600	Available
Canada	~7,600	Available
Chile	~540	Available
Czech Republic	~830	Available
France	~5,300	Available
Germany	~190	Available
Japan	~820	Available
Lithuania	~100	Available
Norway	~4,500	Available
Poland	~1,300	Available
Portugal	~710	Available
Slovenia	~710	Available
South Africa	~1,290	Available
Spain	~1,500	Available
UK (EA)	—	Pending — MERIT-Hydro coverage gap
UK (NRFA)	—	Pending — MERIT-Hydro coverage gap
USA	~23,900	Available

Development

# Install
git clone https://github.com/CooperBigFoot/watershed-retrieve.git
cd watershed-retrieve
uv sync

# Unit tests (no data or network needed)
uv run pytest tests/ -v -m "not integration and not network"

# Integration tests (requires local parquet data)
WATERSHED_RETRIEVE_DATA_DIR=/path/to/data uv run pytest tests/ -v -m integration

# Lint & format
uv run ruff check --fix src/ tests/
uv run ruff format src/ tests/

See CONTRIBUTING.md for full development guidelines.

License

MIT