Wang et al. (2025) Unveiling the accuracy of global GPP products in data-scarce mountain ecosystems of Southwest China

Identification

• Journal: International Journal of Applied Earth Observation and Geoinformation
• Year: 2025
• Date: 2025-10-25
• Authors: Yu Wang, Xiaojun She, Chongjing Zhu, Jian Chen, Debin Kong, Weiyu Shi, Xiaobin Guan, Qiaoyun Xie, Xiaojie Gao, Yuanchao Wang, Yao Li
• DOI: 10.1016/j.jag.2025.104908

Research Groups

• Chongqing Jinfo Mountain Karst Ecosystem National Observation and Research Station, School of Geographical Sciences, Southwest University, Chongqing, China
• School of Resource and Environmental Sciences, Wuhan University, Wuhan, China
• School of Engineering, The University of Western Australia, Perth, Australia
• Centre for Water and Spatial Science, The University of Western Australia, Perth, Australia
• Harvard Forest, Harvard University, Petersham, USA
• Key Laboratory of Remote Sensing and Digital Earth, Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing, China

Short Summary

This study evaluated four global Gross Primary Production (GPP) products (BESS, GOSIF, MOD17, VPM) against eddy covariance observations from 11 flux towers in data-scarce mountain ecosystems of Southwest China. It found that GOSIF had the highest correlation but consistently overestimated GPP, while BESS showed the lowest RMSE and better captured interannual variations, highlighting the need for improved model parameterization in complex regions.

Objective

• To evaluate the performance of four global GPP products (BESS, GOSIF, MOD17, VPM) at the site level using eddy covariance (EC) observations across diverse vegetation and topographic conditions in Southwest China.
• To assess the temporal dynamics and seasonal accuracy of these GPP products, including inter-annual trends and phenology-based GPP partitioning.
• To examine spatial patterns and vegetation-type differences in GPP estimates across the Southwest China region.

Study Configuration

• Spatial Scale: Southwest China (approximately 1.14 x 10^6 km^2), encompassing Sichuan, Guizhou, Yunnan provinces, and Chongqing municipality. Validation performed at 11 flux sites representing four vegetation types (Evergreen Broadleaf Forest, Deciduous Broadleaf Forest, Grassland, Open Shrubland). GPP products at 5 km or 500 m spatial resolution.
• Temporal Scale: Flux data with a minimum of three years of records per site, aggregated to 8-day averages. GPP products cover periods from 2001–2019 or 2001–2022, with daily or 8-day temporal resolution.

Methodology and Data

• Models used:
  • Breathing Earth System Simulator (BESS)
  • GOSIF (Solar-Induced Chlorophyll Fluorescence-based)
  • MOD17 (MODIS GPP algorithm)
  • Vegetation Photosynthesis Model (VPM)
• Data sources:
  • Validation Data: Eddy covariance (EC) observations from 11 flux towers in Southwest China (ChinaFLUX and Chongqing Jinfo Mountain Karst Ecosystem National Observation and Research Station). Raw flux data processed to 30-minute intervals, filtered, gap-filled, and partitioned into GPP using REddyProc tool.
  • GPP Products: BESS (process-based, integrates radiative transfer, canopy photosynthesis, ecosystem respiration, uses reanalysis meteorological forcing), GOSIF (derived from OCO-2 SIF, MODIS, and reanalysis data), MOD17 (Light Use Efficiency (LUE) concept, MOD15 LAI/FPAR inputs), VPM (LUE concept, EVI and climate-driven scalars).
  • Ancillary Data: MODIS Land Cover Type (MCD12Q1) product for vegetation classification. Meteorological data (downwelling shortwave radiation, air temperature, relative humidity) from automatic weather stations at flux towers. Bayesian land surface phenology (BLSP) model used to determine start-of-season (SOS) and end-of-season (EOS).

Main Results

• GOSIF exhibited the strongest correlation with GPP-EC (R^2 > 0.61) but consistently overestimated GPP, especially during summer, with the highest RMSE (0.202 mol C/(m^2 day)) and bias (0.126 mol C/(m^2 day)) across all products.
• BESS achieved the lowest RMSE (0.168 mol C/(m^2 day)) and effectively captured interannual GPP variations, including drought-induced declines.
• MOD17 and VPM generally showed lower accuracy, tending to underestimate peak GPP during the growing season, particularly in evergreen broadleaf forests.
• All products performed better during the non-growing season (R^2 > 0.58 for BESS, GOSIF, MOD17) compared to the growing season (BESS highest R^2 = 0.39, RMSE = 0.203 mol C/(m^2 day)).
• Spatially, mean annual GPP estimates varied significantly among products, ranging from 92.32 mol C/m^2 (BESS) to 124.66 mol C/m^2 (GOSIF).
• GOSIF consistently overestimated seasonal carbon uptake across all vegetation types, with the largest overestimations in open shrubland (34.84 mol C/m^2) and evergreen broadleaf forest during the growing season.
• Deciduous broadleaf forests showed the highest level of agreement among models, while open shrublands presented the greatest discrepancies.
• Solar-induced chlorophyll fluorescence (SIF) consistently demonstrated the strongest correlation with GPP-EC (R^2 = 0.66) compared to FPAR derived from EVI (R^2 = 0.47) and NDVI (R^2 = 0.31).
• GOSIF showed the most widespread and significant positive GPP trends from 2001 to 2019, with the highest annual increase in deciduous broadleaf forests at 2.75 mol C/(m^2 year).

Contributions

• Provides the first comprehensive multi-product GPP validation using a dense eddy covariance flux network in the topographically complex and data-scarce mountain ecosystems of Southwest China.
• Quantifies the strengths and limitations of different GPP modeling approaches (process-based, SIF-based, and LUE-based) at site, seasonal, and regional scales, highlighting their varying sensitivities to vegetation types and environmental conditions.
• Emphasizes the critical need for model improvements, including the integration of physiological indicators like SIF, enhanced representation of vegetation structure and phenology, and better accounting for region-specific environmental stressors.
• Offers a valuable baseline for refining GPP modeling frameworks in Southwest China and provides guidance for the application and development of GPP products in similarly heterogeneous landscapes worldwide.

Funding

• National Natural Science Foundation of China (42201349 and 42001288)
• Chongqing Municipal Science and Technology Bureau (cstc2024ycjh-bgzxm0043)

Citation

@article{Wang2025Unveiling,
  author = {Wang, Yu and She, Xiaojun and Zhu, Chongjing and Chen, Jian and Kong, Debin and Shi, Weiyu and Guan, Xiaobin and Xie, Qiaoyun and Gao, Xiaojie and Wang, Yuanchao and Li, Yao},
  title = {Unveiling the accuracy of global GPP products in data-scarce mountain ecosystems of Southwest China},
  journal = {International Journal of Applied Earth Observation and Geoinformation},
  year = {2025},
  doi = {10.1016/j.jag.2025.104908},
  url = {https://doi.org/10.1016/j.jag.2025.104908}
}