Reconstructed and original hourly temperature observations dataset using GRIN method
Year: 2026
Type: dataset
Abstract: RECONSTRUCTED HOURLY TEMPERATURE OBSERVATIONS DATASET USING THE GRIN METHOD This repository contains two related datasets of hourly near-surface air temperature observations over France: 1. Original dataset: the original Météo-France hourly temperature observations available for the stations included in this release. 2. Reconstructed dataset: the same observational dataset after gap filling using the GRIN (Graph Recurrent Imputation Network) method described in Catteau et al. (in preparation). The reconstruction methodology, training strategy, validation, and assessment of the reconstructed dataset are described in detail in Catteau et al. (in preparation). TEMPORAL AND SPATIAL COVERAGE Both datasets cover the period from 1 January 1940 to 31 December 2025 at hourly resolution. The release contains 1,173 stations out of the 1,749 stations considered in the complete reconstruction described in Catteau et al. (in preparation). The remaining stations are not distributed in this release because their original observations are subject to more restrictive licensing conditions within the Météo-France database. Access to these additional observations may require contacting the authors and/or Météo-France to determine whether the relevant usage rights can be granted. FILE FORMAT AND STRUCTURE The datasets are provided in NetCDF format. Dimensions: - time: hourly timestamps from 1940-01-01 to 2025-12-31- num_poste: station identifier Station coordinates: - num_poste: Météo-France station identifier- Station_Name: usual station name- lambx, lamby: projected geographic coordinates (m)- lat, lon: latitude and longitude (degrees)- ZS: station elevation above sea level (m)- region: associated French mountain region, following the classification used in Catteau et al. (in preparation) Temperature variable: - t: hourly near-surface air temperature (°C), with dimensions (time, num_poste) The original dataset contains the available Météo-France observations as provided for this release. The reconstructed dataset contains the same observations complemented by values reconstructed using the GRIN methodology. Original observations are retained where available, while missing values are filled by the reconstruction method. DATA SOURCE AND LICENSING Original data source: Météo-FranceOriginal data license: Licence Ouverte 2.0 (Etalab) License of this Zenodo dataset: CC BY 4.0 ATTRIBUTION When using these data, please acknowledge both the original data provider and the processing performed by the authors: “Original meteorological observations provided by Météo-France; data processed and reconstructed by the authors using the GRIN method.” Users are also requested to cite the associated publication: Catteau et al. (in preparation). Please cite the corresponding Zenodo record when using the datasets distributed here. DATA LIMITATIONS Reconstructed values should not be considered equivalent to direct observations. Reconstruction performance varies depending on observational density, station location, elevation, and the period considered. In particular, reconstruction uncertainty increases when observational constraints become sparse. For a detailed assessment of reconstruction performance, preserved climatic signals, and remaining limitations, users should refer to Catteau et al. (in preparation).
Language: English
Subfield: Global and Planetary Change
Field: Environmental Science
Domain: Physical Sciences
SDG: Climate action
Open Access status: green
OpenAlex ID: W7204525517
