FLUXtrapolation: Hourly Ecosystem Fluxes from 207 Global Sites, 2015-2022

207 FLUXNET eddy-covariance sites across diverse climates and biomes provide hourly observations from 2015 to 2022. The dataset, created by Ulrich Weber and hosted on MPI-BGC Dataverse, includes three ecosystem flux targets and 13 input covariates from meteorological and satellite sources. It serves as a standardized benchmark for evaluating model extrapolation in flux prediction.

Use Cases

• Benchmarking machine learning models for flux prediction under distribution shifts based on the described extrapolation challenges.
• Analyzing relationships between satellite-derived indices (e.g., EVI, NDWI) and ecosystem fluxes like GPP and ET.
• Studying spatiotemporal heterogeneity in carbon and water fluxes across different biomes and geographic regions.
• Diagnosing model failures in domain generalization tasks for environmental time-series forecasting.

Strengths

• Covers 207 distinct sites, providing a wide geographic and biome representation.
• Includes hourly data over an 8-year period from 2015 to 2022.
• Features three key ecosystem targets (GPP, NEE, ET) and 13 input covariates derived from multiple sources.
• Processed through the standardized FLUXCOM-X framework for scientific grounding.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for large-scale training.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

MPI-BGC Dataverse, author Ulrich Weber.

Collection Method

Observations from FLUXNET eddy-covariance sites, processed through the FLUXCOM-X framework.

Time Range

2015 to 2022

Freshness

Last updated 2026-05 06 08:22:01; freshness should be verified.

Geography

Global, spanning a wide range of climates, biomes, and geographic regions.