Pu et al. (2025) MCI GPP: ensembling a global model- and climate-independent gross primary productivity for 2001–2023

Identification

• Journal: Scientific Data
• Year: 2025
• Date: 2025-12-11
• Authors: Jiabin Pu, Yang Chang, Si Gao, Shanning Bao, Kai Yan, Xian Sun, Nuno Carvalhais, Ranga B. Myneni
• DOI: 10.1038/s41597-025-06218-8

Research Groups

• Department of Earth and Environment, Boston University, Boston, MA 02215, USA
• Faculty of Geographical Science, Beijing Normal University, Beijing, 100875, China
• National Space Science Center, Chinese Academy of Sciences, Beijing, 100190, China
• Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China
• Department for Biogeochemical Integration, Max-Planck-Institute for Biogeochemistry, 07745, Jena, Germany

Short Summary

This study develops a novel model- and climate-independent (MCI) global gross primary productivity (GPP) product for 2001–2023 by ensembling 12 diverse GPP datasets using random forest regression and spatio-temporal tensor models. The MCI GPP product demonstrates superior accuracy against independent observations and reveals a significant global GPP increase of 5.7 Pg C yr⁻¹ per decade.

Objective

• To integrate the CO2 fertilization effect (CFE) and canopy structure stepwise into the MOD17 algorithmic framework and quantitatively assess their effects on GPP estimation.
• To generate 12 distinct GPP datasets based on Production Efficiency Model (PEM) and Two-Leaf Model (TLM) frameworks with CFE, combining MODIS GPP and other Light Use Efficiency (LUE) models, forced by multi-source climate data.
• To utilize FluxNet GPPs as a baseline for training a random forest (RF) algorithm to establish nonlinear relationships between multi-model outputs and measured GPP, thereby generating a model- and climate-independent (MCI) GPP product.
• To filter high uncertainty pixels and employ an advanced gap-filling algorithm for missing value interpolation to ensure data spatiotemporal continuity.
• To conduct a systematic accuracy verification of the generated GPP products based on FluxNet-independent eddy covariance (EC) GPP and satellite-derived GPPs across multiple scales.

Study Configuration

• Spatial Scale: Global, 0.05° spatial resolution.
• Temporal Scale: 2001–2023 (daily, aggregated to monthly and annual).

Methodology and Data

• Models used:
  • Production Efficiency Model (PEM)
  • Two-Leaf Model (TLM)
  • MOD17 GPP algorithm (with and without CO2 Fertilization Effect - CFE)
  • EC-LUE model
  • MPI-Jena model
  • Random Forest (RF) regression
  • Spatio-Temporal Tensor (ST-Tensor) completion model
• Data sources:
  • Climate Reanalysis: GMAO MERRA-2, ECMWF ERA-5 (air temperature, dew point temperature, incoming shortwave radiation), GLDAS NOAH-2.1 (net radiation, evapotranspiration, precipitation).
  • Vegetation Parameters: Sensor-independent (SI) LAI/FPAR dataset, MODIS MCD12Q1 (Plant Functional Type - PFT).
  • Atmospheric CO2: NOAA daily records.
  • In-situ Observations: FluxNet2015 (for model optimization and RF training), AmeriFlux (for independent validation).
  • Peer GPP Products for Evaluation: MODIS GPP (MOD17), GOSIF GPP, X-Base Fluxcom GPP.

Main Results

• The CO2 fertilization effect (CFE) significantly improved GPP estimation accuracy (R² increased from 0.57 to 0.62, RMSE decreased by 10% from 2.67 to 2.41 g C m⁻² d⁻¹) and corrected a persistent underestimation bias in long-term trends.
• A two-leaf model (TLM) with CFE achieved comparable accuracy to the CFE-enhanced Production Efficiency Model (PEM) (R² = 0.63, RMSE = 2.41 g C m⁻² d⁻¹), but tended to produce systematically higher GPP estimates.
• The novel model- and climate-independent (MCI) GPP product estimates an average global GPP of 141.9 ± 4.0 Pg C yr⁻¹ from 2001 to 2023.
• MCI GPP shows a significant global increase of 5.7 Pg C yr⁻¹ per decade (approximately 5.02% per decade).
• Validation against independent AmeriFlux data demonstrates that MCI GPP outperforms other global products (MOD17, GOSIF, X-Base Fluxcom), achieving an R² of 0.72 and RMSE of 1.86 g C m⁻² d⁻¹.
• High-productivity regions (>3,000 g C m⁻² yr⁻¹) are predominantly concentrated in tropical ecosystems. Global GPP is dominated by evergreen broadleaf forests (37.2 Pg C yr⁻¹), savannas (25.9 Pg C yr⁻¹), and grasslands (22.9 Pg C yr⁻¹).
• Statistically significant positive GPP trends were detected across 47.6% of global vegetated areas, with savannas, grasslands, croplands, and shrublands accounting for over 75% of the total observed GPP increase. Declining trends were observed in 4.3% of lands, primarily in the Brazilian Amazon and Central Asia.
• Temporal consistency assessment showed robust agreement between MCI GPP and MOD17 (Overall Agreement (OA) = 71.3%) and X-Base Fluxcom (OA = 72.7%), while consistency with GOSIF was comparatively lower (OA = 63.3%).

Contributions

• Systematically evaluates the quantitative impact of the CO2 fertilization effect and canopy structural traits on global GPP estimation, highlighting their importance for accuracy and long-term trend representation.
• Develops and provides a novel, robust, and high-resolution (0.05° monthly) global GPP product (MCI GPP) for 2001–2023, which is model- and climate-independent.
• Demonstrates that the MCI GPP product outperforms existing global GPP datasets in accuracy when validated against independent in-situ observations.
• Offers a valuable, long-term data asset for investigating vegetation dynamics, re-evaluating regional greening/browning trends, constraining global biogeochemical models, and assessing ecosystem responses to environmental changes and extreme events.
• Mitigates inherent uncertainties from single-model GPP estimations and climate forcing biases through an adaptive ensemble and spatio-temporal gap-filling approach.

Funding

• NASA Earth Science Division through MODIS (80NSSC21K1925)
• NASA Earth Science Division through VIIRS (80NSSC21K1960) projects at Boston University

Citation

@article{Pu2025MCI,
  author = {Pu, Jiabin and Chang, Yang and Gao, Si and Bao, Shanning and Yan, Kai and Sun, Xian and Carvalhais, Nuno and Myneni, Ranga B.},
  title = {MCI GPP: ensembling a global model- and climate-independent gross primary productivity for 2001–2023},
  journal = {Scientific Data},
  year = {2025},
  doi = {10.1038/s41597-025-06218-8},
  url = {https://doi.org/10.1038/s41597-025-06218-8}
}