CAMELS dataset in NetCDF/Feather formats

Motivation

The CAMELS datasets are not provided in an ideal format and takes a bit of data processing to convert them to useful and convenient forms for geospatial analyses. So, I decided to use the beloved netcdf and feather formats to make the dataset more accessible while taking care of some small annoyances! The generated files are hosted on Hydroshare.

Usage

First make sure that h5netcdf, geopandas, xarray, and requests Python packages are installed, then load the files directly like so:

import io

import geopandas as gpd
import requests
import xarray as xr

r = requests.get("https://www.hydroshare.org/resource/658c359b8c83494aac0f58145b1b04e6/data/contents/camels_attributes_v2.0.feather")
attrs = gpd.read_feather(io.BytesIO(r.content))

r = requests.get("https://www.hydroshare.org/resource/658c359b8c83494aac0f58145b1b04e6/data/contents/camels_attrs_v2_streamflow_v1p2.nc")
qobs = xr.open_dataset(io.BytesIO(r.content), engine="h5netcdf")

Methodology

This repo contains the code that I used to generate the datasets. Three data sources are available from the CAMELS dataset:

• Observed Flow: Streamflow observations for all 671 stations.
• Basin Geometries: Polygons representing basins' boundaries for all 671 stations.
• Basin Attributes: 60 Basin-level attributes for all 671 stations.

The camel_netcdf.py generates two files:

• camels_attributes_v2.0.feather: Includes basin geometries and 60 basin-level attributes that are available in CAMELS.
• camels_attrs_v2_streamflow_v1p2.nc: Includes observed flows for all 671 stations, as well as the 60 basin-level attributes. It has two dimensions (station_id and time) and 60 data variables.

Additionally, the script takes care of some small annoyances:

• Station names didn't have a consistent format and there were some missing commas and extra periods! Now, the names have a consistent format (title) and there is comma before the states.
• Station IDs and HUC 02 are strings with leading zeros if needed.

You can recreate the generated files locally using mambaforge (or conda) like so:

git clone https://github.com/cheginit/camels_netcdf
cd camels_netcdf
mamba env create -f environment.yml
conda activate camels
chmod +x ./camels_netcdf.py
./camels_netcdf.py

Example Plots

Snow fraction using camels_attributes_v2.0.feather:

The dataset camels_attrs_v2_streamflow_v1p2.nc:

Streamflow observations for USGS-01013500:

Contributions

Contributions are welcome! Please feel free to open an issue/PR if you have any suggestions that can improve the database.