She et al. (2026) Copula–information gain-based identification of GPP response thresholds under multiscale agricultural drought

Identification

• Journal: Ecological Informatics
• Year: 2026
• Date: 2026-02-10
• Authors: Tianlong She, Chen Xu, Quanwang Chen, Yanan Wang, Yuanyuan Hong, Yechao Sun, Qiang Wang
• DOI: 10.1016/j.ecoinf.2026.103654

Research Groups

College of Resources and Environment, Anhui Agricultural University, Hefei 230031, China

Short Summary

This study develops a Copula–Information Gain (Copula–IG) framework to objectively identify Gross Primary Productivity (GPP) drought response thresholds across multiple temporal scales in China. It reveals pronounced spatiotemporal heterogeneity and scale dependence in GPP responses, with model performance improving and thresholds becoming more stable at longer drought durations.

Objective

• To characterize the spatiotemporal patterns of agricultural drought across China at multiple time scales.
• To quantitatively identify GPP drought-response thresholds across time scales using the Copula–IG model.
• To determine the driving factors underlying spatial differences in GPP drought thresholds based on eXtreme Gradient Boosting (XGBoost)–SHapley Additive ExPlanations (SHAP) analysis.

Study Configuration

• Spatial Scale: National scale for China, covering nine major agricultural regions, with data resampled to 0.5° spatial resolution.
• Temporal Scale: Monthly resolution from 2000 to 2024, analyzing short-term (SPEI-1), seasonal (SPEI-3), and long-term (SPEI-6) drought durations.

Methodology and Data

• Models used:
  • Copula–Information Gain (Copula–IG) framework
  • eXtreme Gradient Boosting (XGBoost)–SHapley Additive ExPlanations (SHAP) analysis
  • Sen–MK method (for trend analysis)
  • Three-dimensional clustering method (for drought event identification)
• Data sources:
  • Drought Index (SPEI): Global SPEI database (SPEIbase v2.11), 0.5° spatial resolution, monthly temporal resolution.
  • Gross Primary Productivity (GPP): GOSIF GPP dataset, 0.05° spatial resolution, monthly temporal resolution.
  • Land Surface Temperature (LST): MOD11C3 dataset (MODIS product series), 0.05° spatial resolution, monthly temporal resolution.
  • ERA5-Land Reanalysis Data: 0.1° spatial resolution, monthly temporal resolution, including 2 m dewpoint temperature (d2m), 2 m air temperature (t2m), total evaporation (et), evaporation from vegetation transpiration (evavt), total precipitation (pre), and soil moisture levels (swvl1: 0–7 cm, swvl2: 7–28 cm, swvl3: 28–100 cm).
  • All datasets were resampled to a 0.5° spatial resolution and monthly temporal resolution.

Main Results

• Agricultural drought in China exhibits pronounced spatiotemporal heterogeneity: short-term droughts are frequent but less persistent, seasonal droughts are fewer but more intense, and long-term droughts are the least frequent but longest in duration and highest in cumulative intensity. Drought centers are concentrated along the transitional zone between the Qinghai–Tibet Plateau and northern arid/semiarid regions.
• GPP responses to agricultural drought show significant temporal scale dependence. Event-scale SPEI–GPP correlation coefficients increase steadily to approximately |0.8| as drought duration extends from SPEI-1 to SPEI-6.
• The discriminative performance of the Copula–IG model consistently improves with increasing drought duration, with AUC values ranging from 0.60–0.75 for SPEI-1 and stabilizing within 0.65–0.80 for SPEI-6.
• GPP drought thresholds evolve from spatially scattered moderate negative anomalies at short time scales toward a more consistent negative range with reduced dispersion at longer time scales. Information Gain (IG) values are primarily concentrated between 0.05 and 0.10, indicating enhanced separation between drought-affected and non-affected GPP states.
• The mechanisms driving GPP drought thresholds transition from rapid, evapotranspiration-dominated short-term responses to structurally constrained suppression governed by cumulative soil moisture deficits under prolonged drought conditions. Shallow soil moisture, vapor pressure deficit (VPD), and temperature are dominant at short scales, while evapotranspiration factors and multi-layer soil moisture become increasingly important at seasonal and long-term scales.

Contributions

• Introduces a novel Copula–Information Gain (Copula–IG) framework for objective, information-driven identification of GPP drought thresholds across multiple temporal scales.
• Provides a probabilistic, quantitative, and spatially refined methodology for agricultural drought–ecosystem response threshold assessment.
• Enhances the understanding of nonlinear coupling mechanisms between multiscale drought processes and ecosystem responses.
• Offers an interpretable and transferable framework for high-resolution agricultural climate monitoring and early-warning systems, overcoming limitations of empirical or probability-based methods.

Funding

• National Key Research and Development Program of China [2023YFD1702105]

Citation

@article{She2026Copulainformation,
  author = {She, Tianlong and Xu, Chen and Chen, Quanwang and Wang, Yanan and Hong, Yuanyuan and Sun, Yechao and Wang, Qiang},
  title = {Copula–information gain-based identification of GPP response thresholds under multiscale agricultural drought},
  journal = {Ecological Informatics},
  year = {2026},
  doi = {10.1016/j.ecoinf.2026.103654},
  url = {https://doi.org/10.1016/j.ecoinf.2026.103654}
}