Aridity threshold of ecological restoration mitigated atmospheric drought via land-atmosphere coupling in drylands

Drylands cover about 41% of global land and are experiencing noticeable climate change alongside extensive ecological restoration. Land-atmosphere coupling strongly links land surface changes with atmospheric states and can intensify hotter, drier conditions and compound extremes. Although restoration can modify local thermal and hydrological conditions, it remains unclear whether and how ecological restoration alters land-atmosphere coupling in drylands, and whether restoration can mitigate atmospheric drought. This study targets atmospheric drought via vapor pressure deficit (VPD), asking whether ecological restoration mitigates VPD through distinct vegetation and soil moisture pathways of land-atmosphere coupling, and how these pathways vary along aridity gradients. Focusing on China’s drylands (May–September) with widespread restoration since ~2000, the study uses 1‑month lag relationships and structural equation modeling (SEM) to separate vegetation- and soil-moisture-mediated couplings and quantify their contributions to VPD variability, complemented by machine learning to attribute mechanisms. The aim is to guide dryland restoration by identifying aridity thresholds and mechanisms where restoration mitigates atmospheric drought.

Prior work shows intense land-atmosphere coupling in ecological transition zones and its role in amplifying hotter, drier climates and linking heatwaves with droughts. Restoration activities can affect local climate via biogeophysical feedbacks, including changes in evapotranspiration, soil moisture, and energy balance. Vegetation greening can reduce temperature and VPD through enhanced transpiration, whereas soil drying can increase temperature and VPD via soil moisture–atmosphere coupling. However, vegetation and soil processes interact and are not independent, leaving uncertainty about their separate roles in atmospheric drought mitigation following restoration. Large restoration programs (e.g., in China’s Loess Plateau) have documented greening and hydrological shifts, but the net effect on atmospheric aridity and extremes across aridity gradients has not been fully quantified.

Study region and periods: China’s drylands across aridity index (AI) classes from hyper-arid to humid, emphasizing dryland subareas: hyper-arid (0<AI≤0.05), arid (0.05<AI≤0.2), semiarid (0.2<AI≤0.3, 0.3<AI≤0.4, 0.4<AI≤0.5), and dry-subhumid (0.5<AI≤0.65). Two periods represent before and after major restoration: 1982–1999 and 2000–2018, based on accelerated LAI increases after 2000. Data: Monthly precipitation (CRU), temperature and VPD (MERRA‑2, ERA5‑Land), latent and sensible heat fluxes (LH, SH; MERRA‑2, ERA5‑Land), root-zone soil moisture (GLEAM), and LAI (AVHRR 1982–2000; MODIS 2000–2018). Dataset selection minimized inconsistencies via correlation and trend screening. Four SEM configurations combine reanalyses: MMG (MERRA‑2 T, VPD, LH, SH), MEG (MERRA‑2 T, VPD; ERA5‑Land LH, SH), EEG (ERA5‑Land T, VPD, LH, SH), EMG (ERA5‑Land T, VPD; MERRA‑2 LH, SH). Lag coupling and SEM: A 1‑month lag framework quantifies land-to-atmosphere impacts while accounting for VPD autocorrelation. Partial least squares regression and SEM are used to quantify pathway strengths. Path coefficients from LAI to next‑month VPD (LAI→VPD+1) represent vegetation–VPD coupling; coefficients from soil moisture to next‑month VPD (SM→VPD+1) represent soil moisture–VPD coupling. Seasonal cycles and long-term trends are removed; LAI, soil moisture, LH, and SH in SEM are averaged over current and next month to reflect lagged effects. SEMs include contemporaneous meteorological variables (temperature, precipitation, VPD) influencing their next‑month states and must satisfy fit criteria (NFI>0.90, SRMR<0.08). Only statistically significant (p<0.05) paths are analyzed; median coefficients across the four SEMs are used. Contribution decomposition: The contributions of vegetation and soil pathways to VPD variations are computed by decomposing the effects of current-month climate, vegetation (LAI), and soil moisture on next‑month VPD within the SEM framework. Mechanistic attribution: Gradient boosting machine (LightGBM) models with Bayesian optimization and SHAP analysis attribute variability in coupling strength to energy and moisture drivers: LH, SH, net radiation (Rn), relative humidity (RH), convective available potential energy (CAPE), and vertically integrated moisture divergence (VIMD). Model performance R^2 ranges from 0.71 to 0.89. Focus season: Analyses focus on May–September due to active land–atmosphere exchanges in China’s drylands.

• Land-atmosphere coupling accounts for about 30% of atmospheric drought (VPD) variability; the soil moisture pathway contributes roughly twice as much as the vegetation pathway.
• Along the aridity gradient, both LAI and soil moisture are negatively related to next‑month VPD (mitigating VPD increases). Strongest negative vegetation effects occur in 0.3<AI≤0.5: LAI→VPD+1 path coefficients are −0.10 (0.3<AI≤0.4) and −0.11 (0.4<AI≤0.5; p<0.05). Strong negative soil moisture effects occur in 0.4<AI≤0.65: SM→VPD+1 coefficients are −0.39 (0.4<AI≤0.5) and −0.31 (0.5<AI≤0.65; p<0.05).
• Pathway contributions to VPD variability: vegetation–VPD coupling contributes 10.19% (0.3<AI≤0.4) and 12.36% (0.4<AI≤0.5); soil moisture–VPD coupling contributes 22.92% (0.4<AI≤0.5) and 25.18% (0.5<AI≤0.65), nearly double the vegetation pathway.
• Restoration impacts vary by aridity: • 0.3<AI≤0.4: Greening and soil wetting strengthen both vegetation- and soil-moisture couplings, mitigating VPD increases. • 0.4<AI≤0.5: Greening with soil drying strengthens vegetation coupling and weakens soil moisture coupling, net mitigation of VPD increases. • 0.5<AI≤0.65: Greening with soil drying weakens vegetation coupling and intensifies soil moisture coupling, amplifying VPD increases.
• Changes in pathway contributions after restoration: vegetation pathway contribution rises by 8.90% (0.2<AI≤0.3), 6.09% (0.3<AI≤0.4), and 12.95% (0.4<AI≤0.5), but decreases by 6.10% (0.5<AI≤0.65). Soil moisture pathway contribution increases markedly in drier classes: +21.16% (AI≤0.05), +19.88% (0.05<AI≤0.2), +37.51% (0.2<AI≤0.3), but only +0.08%, +2.86%, and +8.83% for 0.3<AI≤0.4, 0.4<AI≤0.5, and 0.5<AI≤0.65, respectively.
• Mechanisms: In 0.3<AI≤0.4, Rn and CAPE most influence vegetation coupling; CAPE reductions strengthen negative vegetation coupling. RH primarily controls soil moisture coupling; RH decreases amplify its negative effect. In 0.4<AI≤0.5, Rn is the dominant control on vegetation coupling; increased Rn slightly weakens negative soil moisture coupling. In 0.5<AI≤0.65, CAPE and LH shape vegetation coupling (lower LH strengthens, lower CAPE weakens the negative effect); CAPE and SH shape soil moisture coupling (lower CAPE weakens, higher SH strengthens the negative effect).
• Transitional zones (semiarid to dry-subhumid) exhibit the strongest coupling, consistent with theory that evapotranspiration is limited by water in arid zones and by radiation in humid zones.
• Implications: Restoration mitigates atmospheric aridification (VPD increases) in semiarid areas (0.3<AI≤0.5), potentially reducing concurrent heatwave–drought probabilities; in dry-subhumid areas (0.5<AI≤0.65), restoration may amplify atmospheric drought.

Separating vegetation and soil moisture pathways clarifies how restoration modifies land–atmosphere feedbacks affecting atmospheric drought. In semiarid transition zones, restoration-driven greening enhances transpiration and energy partitioning, strengthening negative vegetation–VPD coupling and, depending on moisture conditions, can also reinforce negative soil moisture–VPD coupling—together mitigating VPD increases. In drier or wetter ends of the gradient, constraints on evapotranspiration (water-limited in arid, radiation-limited in humid) limit the capacity of restoration to influence VPD via coupling. The findings explain where restoration reduces vulnerability to atmospheric drought and compound extremes, emphasizing climate-background sensitivity and the need for context-specific strategies (e.g., vegetation types, planting density). They also quantify how temperature and precipitation changes propagate to VPD through the two pathways, highlighting the role of coupling in amplifying or damping climate-change impacts on extremes. Data-source-related differences in coupling strength estimates, especially for soil moisture coupling in 0.5<AI≤0.65, underscore uncertainties and the importance of multi-dataset assessments.

Ecological restoration in China’s drylands intensifies land–atmosphere coupling and, within defined aridity thresholds, mitigates atmospheric drought via enhanced vegetation and soil pathways. The study introduces a pathway-resolved SEM approach, quantifies contributions to VPD variability (with soil moisture generally contributing about twice the vegetation pathway), and identifies semiarid zones (0.3<AI≤0.5) as optimal targets for restoration to alleviate atmospheric aridification and related extremes. Mechanistic attribution using ML highlights key roles of RH, Rn, CAPE, LH, and SH in modulating coupling changes post-restoration. Future work should expand to global scales, evaluate teleconnections and atmospheric circulation responses to large-scale restoration, and refine multi-source datasets and model frameworks to reduce uncertainties.

• Results depend on reanalysis and remote-sensing datasets; disagreements among data sources (e.g., soil moisture–VPD coupling changes in 0.5<AI≤0.65 across SEM configurations) introduce uncertainty.
• Median aggregation across four SEMs may mask dataset-specific behaviors; although all SEMs meet fit criteria (NFI>0.90, SRMR<0.08), path estimates remain sensitive to inputs.
• Generalizability beyond China’s drylands is limited; outcomes depend on local climate, soil properties, water availability, species selection, and restoration design (e.g., afforestation in hyper-arid regions may exacerbate water stress).
• The analysis focuses on May–September and on 1‑month lag effects; other seasons or lag structures may reveal additional dynamics.