lildrip

Stochastic rainfall disaggregation with the Bartlett-Lewis process

lildrip takes coarse-resolution rainfall data (e.g. hourly) and generates finer-resolution series (e.g. 10-minute intervals) using the Bartlett-Lewis stochastic process — a well-established model in hydrology.

Quick start

pip install lildrip

from lildrip import BartlettLewisModel

model = BartlettLewisModel()
events = model.identify_events(fine_series, inter_event_gap_minutes=30)
params = model.calibrate(events)

disagg = model.disaggregate(coarse_series, fine_interval_minutes=10)

Web API

pip install "lildrip[api]"
uvicorn api.main:app --host 0.0.0.0 --port 8000

Then POST /calibrar with a high-resolution CSV or POST /desagregar with coarse data + pre-calibrated parameters. Both endpoints accept configurable column names (time_column, rainfall_column).

See the full API docs below.

Demo

Run the examples from the repository root:

# No data?  Generate sample CSV files first:
python examples/generate_demo_rain.py

# Full pipeline (calibrate → disaggregate → plot):
python examples/bartlett_lewis_demo.py

Model parameters

The Bartlett-Lewis model describes rainfall through five parameters calibrated via the Method of Moments.

Project structure

lildrip/
├── src/lildrip/
│   ├── __init__.py
│   ├── bartlett_lewis_model.py   # Core model
│   └── plotting.py               # Visualisation helpers
├── api/
│   ├── app.py                    # FastAPI application
│   └── main.py                   # Uvicorn entry point
├── examples/
│   ├── generate_demo_rain.py     # Generate synthetic data
│   └── bartlett_lewis_demo.py    # Calibrate + disaggregate demo
├── tests/
│   └── test_model.py
├── Dockerfile
├── pyproject.toml
└── README.md

License

MIT