Hackathon 2025: Streamflow Forecasting and Water Resource Regulation

I. Data

The objective of this challenge is to develop a model to predict streamflow levels across different river basins, aiding in sustainable water management strategies. The provided dataset includes time-series observations of water flow at specific stations within the selected watersheds, with a prediction horizon of 4 weeks.

Spatial dimension:

• Covers key French river basins (Adour-Garonne, Rhône-Mediterranean) and the Doce River basin in Brazil.
• Data is available for 30 stations for training and 39 stations for evaluation for France and for Brazil we have 9 stations for training, 13 stations for evaluation.

Temporal dimension:

• Training data spans from 1990 to 2003.
• Evaluation/inference data spans from 2004 to 2009. The evaluation dataset consists of multiple time segments, including a 4-week historical period for streamflow and weather (temperature, ...), 1 week of weather forecast data, and a 4-week prediction horizon for streamflow.

Participants will have access to the following feature types:

Spatiotemporal Climate Variables:

• Temperature at 2m above ground
• Total precipitation
• Evaporation rates
• Soil moisture (volumetric soil water)

Geospatial Features:

• Soil properties (bulk density, clay content, sand/silt ratio). Source: Soilgrids
• Hydrological divisions (watersheds and sub-basins). Source: BD Carthage for France and SNIRH for Brazil
• Altitude Digital Elevation Model (NASA SRTM). Source: NASA SRTM 30m
• Dams positions in France. Source: Hydrographic nodes

All features are provided in formats suitable for direct analysis and integration. You can download all the data below.

For more details about the hackathon see Here --> https://www.codabench.org/competitions/4335/