hydrodatasource

• Free software: BSD license
• Documentation: https://iHeadWater.github.io/hydrodatasource

Overview

While libraries like hydrodataset exist for accessing standardized, public hydrological datasets (e.g., CAMELS), a common challenge is working with data that isn't in a ready-to-use format. This includes non-public industry data, data from local authorities, or custom datasets compiled for specific research projects.

hydrodatasource is designed to solve this problem. It provides a flexible framework to read, process, and clean these custom datasets, preparing them for hydrological modeling and analysis.

hydrodatasource uses a unified URI-only design: every reader is constructed by pointing it at a directory or an s3:// URI, and a one-line open_dataset() factory resolves a dataset id to a ready-to-use reader. The design is compatible with hydrodataset.

Quick Start

If you have a registered dataset and a ~/hydro_setting.yml (see Configuration), resolve and open it in one line:

from hydrodatasource.configs.data_resolver import open_dataset

# "songliao_event" is a hydrodatasource-registered dataset (flood-event reader)
ds = open_dataset("songliao_event")

To open a custom dataset by path, construct the reader directly:

from hydrodatasource.reader.data_source import SelfMadeHydroDataset

reader = SelfMadeHydroDataset(uri="/path/to/my_dataset", time_unit=["1D"])

Reading Custom Datasets

This is the primary use case for hydrodatasource. If you have your own basin-level time series and attribute data, you can use SelfMadeHydroDataset to load it seamlessly.

1. Prepare Your Data Directory

First, organize your data into the following folder structure:

/path/to/my_dataset/
├── attributes/
│   └── attributes.csv
├── shapes/
│   └── basins.shp
└── timeseries/
    ├── 1D/                     # Sub-folder for each time resolution (e.g., daily)
    │   ├── basin_01.csv
    │   ├── basin_02.csv
    │   └── ...
    └── 1D_units_info.json      # JSON file with unit information

• attributes/attributes.csv: A CSV file containing static basin attributes (e.g., area, mean elevation). Must include a basin_id column that matches the filenames in the timeseries folder.
• shapes/basins.shp: A shapefile with the polygon geometry for each basin.
• timeseries/1D/: A folder for each time resolution (1h, 3h, 1D, 8D, 1M). Inside, each CSV file should contain the time series data for a single basin and be named after its basin_id.
• timeseries/1D_units_info.json: A JSON file defining the units for each variable in your time series CSVs (e.g., {"precipitation": "mm/d", "streamflow": "m3/s", "temperature": "degC"}). Every variable you read must be listed here.

Extended datasets may add optional directories:

• intermediate/ — interval-basin (区间流域) data with topological relationships (attributes/, timeseries/, shapes/), used by the TG basin reader.
• stations/ — gauging-station data and adjacency matrices, used by the station reader.
• forecasts/ — forecast time series, used by the forecast reader.

2. Read the Data in Python

Once your data is organized, point a URI-only reader at it:

from hydrodatasource.reader.data_source import SelfMadeHydroDataset

# Pass the absolute path (or s3:// URI) of your dataset directory as `uri`
reader = SelfMadeHydroDataset(uri="/path/to/my_dataset", time_unit=["1D"])

# Get a list of all available basin IDs
basin_ids = reader.read_object_ids()

# Define the time range and variables you want to load
t_range = ["2000-01-01", "2010-12-31"]
variables_to_read = ["precipitation", "streamflow", "temperature"]

# Read the time series data (a dict of xarray.Datasets keyed by time unit)
timeseries_data = reader.read_ts_xrdataset(
    gage_id_lst=basin_ids,
    t_range=t_range,
    var_lst=variables_to_read,
    time_units=["1D"],
)

daily_data = timeseries_data["1D"]

print("Successfully loaded data:")
print(daily_data)

# Static attributes are equally easy to read
attributes_data = reader.read_attr_xrdataset(gage_id_lst=basin_ids, var_lst=["area", "mean_elevation"])
print("\nAttributes:")
print(attributes_data)

Note on the old API. Earlier versions accepted data_path= / dataset_name= constructor arguments. These were removed in favor of the unified uri= interface and now raise ValueError.

Reader Aliases

All hydrodatasource readers are registered as aliases in READER_ALIASES, which downstream projects (e.g. hydromodel) consume alongside hydrodataset's aliases:

Alias	Class	Directory convention
selfmade	SelfMadeHydroDataset	standard dataset (attributes/, timeseries/, shapes/)
longterm	LongTermDataset	self-made dataset with long-term support
forecast	SelfMadeForecastDataset	standard + forecasts/
station	StationHydroDataset	standard + stations/
tghydro	TgHydroDatasource	standard + intermediate/ + LSTM predictions
floodevent	FloodEventDatasource	flood-event data with per-basin event markers
gages	Gages	GAGES-II public dataset
grdc	Grdc	GRDC public dataset
rainfall	RainfallReader	cleaned station rainfall
crd	Crd	China reservoir database
rsvrinflow	RsvrInflowReader	reservoir inflow data

hydrodataset's public datasets (e.g. camels_us) are also resolvable through the same open_dataset() / resolve_data_path() interface.

Configuration

hydrodatasource reads a shared ~/hydro_setting.yml (same file used by hydrodataset and hydromodel) using the unified storage.* format:

storage:
  default_source: local      # 'local' or 'cloud' — fallback when source is not given
  local:
    root: 'D:\data\hydrodatasource'   # main data root
  cache: data\cache
  s3:                        # optional — cloud (MinIO/S3) access
    endpoint_url: 'http://minio:9000'
    key: 'access_key'
    secret: 'secret_key'
    bucket: hydro-data
    prefix: hydromodel

• When default_source: local, resolve_data_path() / open_dataset() resolve against storage.local.root.
• When default_source: cloud, they resolve against storage.s3.
• If ~/hydro_setting.yml is missing, a default root ~/hydrodatasource_data is used (with a warning).

Other Features

Beyond reading data, hydrodatasource also includes modules for:

• processor: Perform advanced calculations like identifying rainfall-runoff events (dmca_esr.py) and calculating basin-wide mean rainfall from station data (basin_mean_rainfall.py).
• cleaner: Clean raw time series data. This includes tools for smoothing noisy streamflow data, correcting anomalies in rainfall and water level records, and back-calculating reservoir inflow.

The usage of these modules is described in the API Reference. We will add more examples in the future.

Installation

For standard use, install the package from PyPI:

pip install hydrodatasource

Development Setup

For developers, it is recommended to use uv to manage the environment, as this project has local dependencies (e.g., hydroutils>=0.2.0, hydrodataset>=0.3.0).

1. Clone the repository:

   git clone https://github.com/iHeadWater/hydrodatasource.git
   cd hydrodatasource
   

2. Sync the environment with uv (installs all extras, including dev tooling):

   uv sync --all-extras