WU et al. (2026) Dominant drivers for geographic patterns and multi-scale variability of global land‒atmosphere coupling

Identification

• Journal: Advances in Climate Change Research
• Year: 2026
• Date: 2026-03-01
• Authors: Wen-Lu WU, Hai-Shan CHEN, Siguang Zhu, Jie M. Zhang
• DOI: 10.1016/j.accre.2026.03.002

Research Groups

• State Key Laboratory of Climate System Prediction and Risk Management (CPRM)/Key Laboratory of Meteorological Disaster, Ministry of Education (KLME)/Collaborative Innovation Center on Forecast and Evaluation of Meteorological Disasters (CIC-FEMD), Nanjing University of Information Science and Technology, Nanjing 210044, China
• School of Atmospheric Sciences, Nanjing University of Information Science and Technology, Nanjing 210044, China

Short Summary

This study systematically assesses global land-atmosphere (L-A) coupling from 1958-2022, identifying five distinct regional patterns and their multi-scale temporal variability, and determining the dominant physical drivers for each region using machine learning and process network analysis. The findings reveal that while interannual signals generally dominate L-A coupling variability, specific regions like the Hot Evaporative Region exhibit strong decadal signals, with dominant drivers varying significantly across regions and seasons.

Objective

• To systematically analyze the global geographic patterns and multi-scale temporal variability of land-atmosphere (L-A) coupling.
• To identify the dominant physical drivers and their underlying mechanisms across different climatic regions and seasons.

Study Configuration

• Spatial Scale: Global land regions, excluding Greenland and the Antarctic continent.
• Temporal Scale: 1958 to 2022 (65 years), analyzed at monthly and seasonal scales (March–May, June–August, September–November, and December–February).

Methodology and Data

• Models used:
  • L-A coupling index: Energy balance approach based on evapotranspiration and potential evapotranspiration (Pearson correlation coefficient).
  • Clustering: Self-Organizing Map (SOM) neural network combined with K-means algorithm.
  • Factor contributions: Deep Forest model combined with SHAP (SHapley Additive exPlanations) algorithm.
  • Interaction analysis: Process Network using Transfer Entropy (TE).
  • Temporal analysis: 8th-order low-pass Butterworth filter (cutoff frequency of 0.2 for decadal component).
• Data sources:
  • ERA5 reanalysis dataset (ECMWF): 0.25˚ × 0.25˚ horizontal resolution, daily and monthly variables (2 m air temperature, net longwave radiation flux, net shortwave radiation flux, evapotranspiration, potential evapotranspiration, soil moisture, soil temperature, snow water equivalent, leaf area index, surface temperature, surface pressure, 2 m dew point temperature, 10 m wind speed, sensible heat flux, latent heat flux).
  • Global Land Data Assimilation System (GLDAS) dataset (NASA/NCEP): version 2.0 (before 2000) and 2.1 (after 2000), 0.25˚ × 0.25˚ horizontal resolution, monthly variables (soil moisture, soil temperature, snow water equivalent, surface temperature, 2 m air temperature, 10 m wind speed, specific humidity, sensible heat flux, latent heat flux, net longwave radiation, net shortwave radiation). Leaf Area Index (LAI) from ERA5 was used to fill gaps in GLDAS during clustering.

Main Results

• The globe was clustered into five distinct regions based on 12 L-A coupling factors: Cold Snowy Region (CSR), Cold Wet-Soil Region (CWR), Temperate Continental Region (TCR), Hot Arid Region (HAR), and Hot Evaporative Region (HER). This regionalization showed high consistency between ERA5 and GLDAS datasets.
• Strong L-A coupling hotspots were consistently observed in southern North America, eastern South America, southern Eurasia, central and southern Africa, and central Oceania across all seasons.
• L-A coupling strength exhibits pronounced seasonal and spatial variability: generally strong in March–May, peaking in June–August (especially tropical/subtropical regions), weakening in September–November (high latitudes), and weakest in December–February (northern latitudes). Regionally, HAR and TCR show strong coupling in March–May, TCR in June–August, TCR and HAR in September–November, and HAR and HER in December–February.
• Multi-scale temporal analysis revealed that interannual signals dominate the variability of global L-A coupling in most regions. Decadal signals are generally weak, with the notable exception of the Hot Evaporative Region (HER) during June–August and September–November, where decadal variability surpasses interannual contributions.
• Dominant factors for L-A coupling vary regionally and seasonally:
  • CSR: Radiative and surface energy factors (e.g., net shortwave radiation, soil temperature, net longwave radiation) dominate, particularly in March–May, influencing snowmelt and surface energy balance.
  • CWR: Eco-hydrothermal factors (e.g., specific humidity, leaf area index, soil temperature) play a central role, regulating evapotranspiration and the local water cycle.
  • TCR: Temperature-humidity factors (e.g., 2 m air temperature, surface temperature, specific humidity) primarily control coupling by regulating heat and moisture exchange.
  • HAR: Water availability (e.g., leaf area index, soil moisture, specific humidity) is the key limiting factor, enhancing latent heat flux and moisture exchange.
  • HER: Both temperature and humidity factors (e.g., 2 m air temperature, soil temperature, specific humidity) jointly influence coupling, acting synergistically due to high temperature, humidity, and dense vegetation.
• The combined importance of these dominant factors accounts for no less than 40% of the overall L-A coupling variability.

Contributions

• Provides the first systematic global assessment of dominant drivers and their interactions for land-atmosphere coupling, addressing a gap in previous regional or single-variable studies.
• Introduces a novel data-driven regionalization method using combined SOM and K-means algorithms, offering a comprehensive and dynamic global classification based on multiple L-A coupling factors.
• Employs advanced machine learning (Deep Forest, SHAP) and information theory (Process Network, Transfer Entropy) to identify and elucidate the dominant factors and their interaction pathways across different regions and seasons.
• Characterizes the multi-scale temporal variability of global L-A coupling, highlighting the prevalence of interannual signals and identifying regions with significant decadal influences.
• Advances the theoretical understanding of the physical processes underlying L-A coupling, offering new scientific insights into complex global L-A interactions and providing valuable guidance for improving land surface process modeling and climate prediction capabilities.

Funding

• National Key Research and Development Program of China (2022YFF0801603)
• National Natural Science Foundation of China (42088101)

Citation

@article{WU2026Dominant,
  author = {WU, Wen-Lu and CHEN, Hai-Shan and Zhu, Siguang and Zhang, Jie M.},
  title = {Dominant drivers for geographic patterns and multi-scale variability of global land‒atmosphere coupling},
  journal = {Advances in Climate Change Research},
  year = {2026},
  doi = {10.1016/j.accre.2026.03.002},
  url = {https://doi.org/10.1016/j.accre.2026.03.002}
}