README.md

metadata

license: apache-2.0
task_categories:
  - time-series-forecasting
  - feature-extraction
  - tabular-classification
language:
  - fr
tags:
  - groundwater
  - piezometry
  - time-series
  - embedding
  - hydrology
  - ERA5
  - BRGM
size_categories:
  - 1M<n<10M

Piezometric Embedding Benchmark Dataset

Daily groundwater level time series from ~4200 French monitoring stations, with ERA5 climate covariates and hydrogeological labels.

Notebooks

Notebook	Description
01_data_exploration.ipynb	Dataset overview, label distributions, geographic maps, time series examples
02_benchmark_analysis.ipynb	Encoder comparison, whitening effect, uni vs multi, ranking

Dataset Description

This dataset supports the comparative evaluation of time series embedding methods for piezometric groundwater stations. It contains:

• Station metadata (4210 stations): coordinates, hydrogeological labels (milieu_eh), department, altitude, and derived statistics
• Univariate daily series (2000 stations, ~10.6M rows): groundwater level (niveau_nappe_eau)
• Multivariate daily series (2000 stations, ~10.6M rows): groundwater level + 3 ERA5 covariates (temperature_2m, total_precipitation, potential_evaporation)

Files

File	Rows	Size	Description
data/station_metadata.parquet	4,210	258 KB	Station coordinates, labels, properties
data/piezo_daily_uni.parquet	10.6M	38 MB	Univariate daily groundwater level
data/piezo_daily_multi.parquet	10.6M	119 MB	Multivariate (level + 3 ERA5 covariates)

Source

• Groundwater data: BRGM HubEau API
• Climate data: ERA5 reanalysis (Hersbach et al., 2020)
• Labels: BDLISA hydrogeological environments

Usage

Associated Repository

Full benchmark code, trained models, and analysis notebooks: https://scm.univ-tours.fr/ringuet/aida_embedding_benchmark

Citation