Zhu et al. (2025) The UFLUX ensemble of multiple-scale carbon, water, and energy fluxes

Identification

• Journal: Scientific Data
• Year: 2025
• Date: 2025-12-30
• Authors: Songyan Zhu, Jian Xu, Jingya Zeng, Shanning Bao, Yumeng Chen, Shuaiyi Shi, Z. Zheng, Wenquan Dong, Yapeng Wang, Jiancheng Shi
• DOI: 10.1038/s41597-025-06401-x

Research Groups

• School of Geography and Environmental Science, University of Southampton, UK
• School of GeoSciences and National Center for Earth Observation, University of Edinburgh, UK
• National Space Science Center, Chinese Academy of Sciences, Beijing, China
• Department of Economics, Business School, University of Exeter, UK
• Department of Meteorology, National Center for Earth Observation, University of Reading, UK
• Moganshan Geospatial Information Laboratory, Hangzhou, China
• Department of Earth and Environmental Sciences, The University of Manchester, UK
• Department of Physical Geography and Ecosystem Science, Lund University, Sweden
• National Satellite Meteorological Center, China Meteorological Administration, Beijing, China

Short Summary

This study introduces the Unified FLUXes (UFLUX) ensemble, a globally consistent, multi-scale dataset of terrestrial carbon, water, and energy fluxes (gross primary productivity, evapotranspiration, sensible heat) derived from eddy covariance data, satellite observations, and machine learning, demonstrating high accuracy and consistency across various spatial and temporal scales.

Objective

• To develop and present the Unified FLUXes (UFLUX) ensemble, a globally consistent dataset of gross primary productivity (GPP), evapotranspiration (ET), and sensible heat (H) fluxes, addressing the lack of sufficient spatial resolution and consistency in existing flux datasets for fragmented landscapes.

Study Configuration

• Spatial Scale: Global (0.25° × 0.25°), Europe (0.25° × 0.25°; 100 m × 100 m), UK (100 m × 100 m).
• Temporal Scale: Monthly (global, 2001–2020); Daily (Europe 0.25°, 2001–2020; UK 100 m, 2020–2022); Biannual (Europe 100 m, 2017–2022).

Methodology and Data

• Models used: Deep Forest (primary), eXtreme Gradient Boosting (XGBoost), Random Forest.
• Data sources:
  • Satellite observations: MODIS/Terra+Aqua (MCD43A4) for NDVI, EVI, NIRv; AVHRR for NDVI, EVI, NIRv; Sentinel-2 L2A for high-resolution VIs; Sentinel-1 GRD for radar data; GOME-2 SIF; GOSAT SIF; Contiguous SIF (CSIF) from MODIS and OCO-2.
  • Meteorological reanalysis: ERA5-Land (2 m temperature, 2 m dewpoint temperature, surface solar radiation downwards, surface thermal radiation downwards, 10 m u-/v-component of wind, total precipitation); CFSv2 (2 m temperature, 2 m specific humidity, downward shortwave radiation at surface, downward longwave radiation at surface, U-/V- wind components at 10 m, precipitation rate at surface).
  • Ambient CO2 concentration: CarbonTracker CT2022.
  • Land Cover: MODIS Land Cover Type Yearly (MCD12Q1) for Plant Functional Types (PFTs).
  • Eddy Covariance (EC) flux data (for training and validation): FLUXNET2015 Tier-1 (206 towers); ICOS Warm Winter dataset V1.0 (73 towers); ICOS release 2023-1 Level-2 V1.0 (61 towers).

Main Results

• UFLUX captures over 80% of flux variability (R² consistently ~0.8 for the baseline MODIS-NIRv-ERA5 member) with low mean absolute errors (MRAE values: GPP 0.23, Reco 0.21, NEE 0.43, H 0.31, LE 0.25).
• The baseline MODIS-NIRv-ERA5 ensemble member demonstrated the highest mean R² (~0.8) and lowest mean MRAE (~0.3) globally.
• High-resolution regional flux upscaling (e.g., UK 100 m) using Sentinel-1/2 and ICOS data achieved comparable performance (GPP R² 0.87, MRAE 0.28).
• UFLUX successfully reproduces climate responses and interannual patterns, aligning with existing literature and process-based models (TRENDY), though uncertainties in net carbon flux (NEE) remain higher.
• The ensemble effectively maintains ecosystem water-use efficiency (WUE) (R² 0.73, slope 0.92) and energy balance ratio (EBR) (R² 0.9, slope 0.99) at the tower level.
• UFLUX accurately reproduces NEE responses to photosynthetic photon flux density (PPFD) and ecosystem respiration (Reco) responses to air temperature (TA) across most plant functional types (e.g., croplands, forests, grasslands).
• Spatial patterns of fluxes (GPP, Reco, H, LE) align with climate classifications, showing high GPP/Reco in tropical regions and high H in arid/semi-arid regions.

Contributions

• Provides the first large-scale, globally consistent carbon, water, and energy flux information at a high spatial resolution of 100 m × 100 m, particularly valuable for fragmented landscapes.
• Introduces the UFLUX ensemble, an openly accessible dataset spanning two decades, which explicitly considers the interconnectedness between carbon and water fluxes.
• Utilizes a uniform machine learning framework to integrate multiple satellite and climate datasets, thereby reducing uncertainties and methodological inconsistencies across different flux types and scales.
• Offers a comprehensive validation demonstrating UFLUX's robust performance in capturing flux variability, reproducing key ecosystem-environment responses (WUE, EBR, NEE-PPFD, Reco-TA), and ensuring consistency across multi-scale satellite inputs.
• Delivers a critical dataset to support cross-scale climate policymaking, land management, and carbon sequestration efforts towards achieving carbon neutrality.

Funding

• National Natural Science Foundation of China (Grant 42375142)
• Chinese Academy of Sciences Pioneering Initiative Talents Program (Grant E1RC2WB2)

Citation

@article{Zhu2025UFLUX,
  author = {Zhu, Songyan and Xu, Jian and Zeng, Jingya and Bao, Shanning and Chen, Yumeng and Shi, Shuaiyi and Zheng, Z. and Dong, Wenquan and Wang, Yapeng and Shi, Jiancheng},
  title = {The UFLUX ensemble of multiple-scale carbon, water, and energy fluxes},
  journal = {Scientific Data},
  year = {2025},
  doi = {10.1038/s41597-025-06401-x},
  url = {https://doi.org/10.1038/s41597-025-06401-x}
}