Yu et al. (2025) Widespread but Divergent Drought Legacy Effects on Gross Primary Productivity Across Biomes

Identification

• Journal: Global Change Biology
• Year: 2025
• Date: 2025-10-01
• Authors: Xin Yu, René Orth, Markus Reichstein, Christian Reimers, Ulisse Gomarasca, Mirco Migliavacca, Dario Papale, Michael Bahn, Ana Bastos
• DOI: 10.1111/gcb.70541

Research Groups

This study involves a large collaborative effort drawing data from multiple international eddy covariance networks and their contributing institutions, including: - FLUXNET (and its associated universities and research centers globally) - OzFlux (Australian and New Zealand Flux Research and Monitoring) - ICOS Ecosystem Thematic Centre (Integrated Carbon Observation System) - AmeriFlux (and its associated universities and research centers in the Americas) Numerous universities and national research institutes across Australia, Europe, and North America are involved as data providers for individual eddy covariance sites.

Short Summary

This study investigates the drivers of drought legacy effects on Gross Primary Productivity (GPP) across a diverse range of global ecosystems, identifying key climatic, ecosystem, and plant hydraulic traits that modulate the post-drought recovery of carbon uptake.

Objective

• To identify the potential drivers explaining the spatial variability of drought legacy effects on Gross Primary Productivity (GPP) across various plant functional types.
• To quantify the shift in climatic dependence of GPP anomalies between non-legacy and legacy periods following drought events.

Study Configuration

• Spatial Scale: Global, encompassing 70 eddy covariance sites across various continents, including Australia, Europe, North America, and the Middle East. Sites represent diverse plant functional types (e.g., forests, grasslands, savannas, shrublands).
• Temporal Scale: Multi-year periods for individual sites, generally ranging from the early 2000s to the early 2020s, with some records extending back to 1992 (e.g., US-Ha1) or 1996 (e.g., FI-Hyy). Drought legacy durations were observed up to 3 years.

Methodology and Data

• Models used:
  • Random Forest models were trained to predict GPP anomalies (GPPanom) for both non-legacy and legacy periods.
  • Shapley Additive Explanations (SHAP) were used to evaluate the contribution of individual predictors to GPPanom predictions.
  • LOESS (locally estimated scatterplot smoothing) was applied to fit response curves of SHAP values.
  • Kernel-based Conditional Independence Test (KCIT) and Hilbert-Schmidt Independence Criterion (HSIC gamma, HSIC perm) were employed for unconditional independence tests to identify significant drivers of legacy effects.
• Data sources:
  • Eddy covariance flux data from FLUXNET (FLUXNET2015), OzFlux, ICOS Warm Winter 2020, and AmeriFlux (FLXUNET, FLUXNET-1F, AMP) networks.
  • Plant functional types (PFTs) included deciduous broad-leaf forests (DBF), evergreen broad-leaf forests (EBF), evergreen needle-leaf forests (ENF), grasslands (GRA), mixed forests (MF), savannas (SAV), woody savannas (WSA), closed shrublands (CSH), and open shrublands (OSH).
  • Climatic variables: Incoming short-wave radiation (SW_IN, in watts per square meter), air temperature (TA, in degrees Celsius), vapor pressure deficit (VPD, in kilopascals), and water availability index (WAI, unitless).
  • Drought characteristics: Intensity (based on WAI anomalies), duration (relative to growing season length), and timing (relative to growing season onset).
  • Pre-drought and post-drought water limitation (based on evapotranspiration anomalies, EFanom).
  • Background climate conditions: Aridity (ratio of mean annual precipitation over mean annual net radiation), mean annual temperature, and seasonality of precipitation and temperature.
  • Forest-specific site metrics: Forest mean age (in years), maximum canopy height (in meters), species richness (count), and rooting depth (in meters).
  • Abundance-weighted plant hydraulic traits: Mean and standard deviation of wood density (in kilograms per cubic meter), water potential at 50% loss of hydraulic conductivity (P50, in megapascals), and hydraulic safety margin (HSM50, in megapascals).

Main Results

• Significant shifts in the climatic dependence of GPP anomalies were observed between non-legacy and legacy periods following drought events, with overlap percentages for climatic predictor response curves often below 50% (e.g., TA and VPD in AU-Cpr savanna showed 25.4% and 32.4% overlap, respectively).
• The classification of "Forest vs non-forest" was identified as a significant unconditional driver of drought legacy effects across all studied sites (p < 0.05 for KCIT, HSIC gamma, and HSIC perm tests).
• Within forest ecosystems, the "Ecosystem-weighted mean P50" (water potential at 50% loss of hydraulic conductivity) was a significant unconditional driver of legacy effects (p < 0.05 for HSIC gamma and HSIC perm tests).
• Other potential drivers, including concurrent drought impact, drought characteristics (intensity, duration, timing), pre-drought and post-drought water limitation, background climate conditions (aridity, mean temperature, seasonality), and other forest-specific metrics or hydraulic traits, did not consistently show significant unconditional dependence on legacy effects.

Contributions

• Provides a comprehensive, multi-site analysis of drought legacy effects on GPP across a globally diverse set of ecosystems using an extensive eddy covariance dataset.
• Quantifies the dynamic shift in ecosystem GPP sensitivity to key climatic drivers during post-drought legacy periods, highlighting altered ecosystem functioning.
• Identifies ecosystem structural characteristics (forest vs. non-forest) and specific plant hydraulic traits (mean P50) as critical modulators of drought legacy effects, offering insights into underlying physiological mechanisms.
• Employs robust machine learning (Random Forest, SHAP) and statistical independence testing methods to rigorously assess the contributions of various potential drivers.

Funding

Funding information is not explicitly provided in the supplementary material. However, the research relies on data from major international networks (FLUXNET, OzFlux, ICOS, AmeriFlux), which are supported by various national and international funding agencies and research programs.

Citation

@article{Yu2025Widespread,
  author = {Yu, Xin and Orth, René and Reichstein, Markus and Reimers, Christian and Gomarasca, Ulisse and Migliavacca, Mirco and Papale, Dario and Bahn, Michael and Bastos, Ana},
  title = {Widespread but Divergent Drought Legacy Effects on Gross Primary Productivity Across Biomes},
  journal = {Global Change Biology},
  year = {2025},
  doi = {10.1111/gcb.70541},
  url = {https://doi.org/10.1111/gcb.70541}
}