Soil moisture dominates dryness stress on ecosystem production globally

The study addresses the longstanding question of whether low soil moisture (SM) or high atmospheric water demand (vapor pressure deficit, VPD) is the primary driver of dryness stress limiting ecosystem productivity. While SM provides the direct water supply to plants and has been widely used to diagnose vegetation drought impacts, high VPD can induce stomatal closure and is hypothesized to constrain photosynthesis. Conflicting evidence and strong coupling between SM and VPD via land–atmosphere interactions have led to divergent interpretations and model representations of drought stress, with some terrestrial ecosystem models using SM-only, VPD-only, or combined formulations. This coupling complicates attribution of observed reductions in gross primary production (GPP). The purpose of this study is to decouple SM and VPD influences using global observations of solar-induced chlorophyll fluorescence (SIF) as a proxy for GPP, and to quantify their respective roles in limiting ecosystem production across climate and vegetation gradients. Clarifying this attribution is crucial for improving drought risk assessments and projections of terrestrial carbon uptake under climate change.

Prior work documents significant SM control on vegetation productivity and land–climate feedbacks, and highlights the dominant role of semi-arid ecosystems in interannual variability of the global carbon sink. Other studies stress increasing importance of atmospheric demand (VPD) for water and carbon fluxes and suggest strong stomatal sensitivity to VPD at the leaf scale. Earth system and terrestrial ecosystem models vary widely in their drought representations, from SM-only to VPD-only limitations, or both, with concerns about double counting due to SM–VPD correlation. Experimental manipulations of atmospheric humidity/temperature at ecosystem scale are scarce, and global observational constraints on the relative roles of SM and VPD are limited. The literature therefore motivates a global, observation-based disentanglement of SM and VPD effects on ecosystem photosynthesis.

• Indicator of ecosystem production: Used solar-induced chlorophyll fluorescence (SIF) as a mechanistically linked proxy for GPP. Main SIF dataset: spatially contiguous OCO-2 based CSIF (740 nm), 0.5° resolution, 2000–2016, generated via neural networks using MODIS reflectance. Additional independent datasets for robustness: GOME-2 SIF (9:30 am overpass, 0.5°, 2007–2015; two retrievals) and SCIAMACHY SIF (10:00 am overpass, 1.5°, 2002–2012). Instantaneous SIF converted to daily mean following published methods.
• Hydro-meteorological drivers: Soil moisture from ERA-Interim (0–1 m summed layers; main), MERRA-2 (root-zone), and ESA CCI (surface 0.5–5 cm). VPD computed from ERA-Interim (main) or MERRA-2 temperature and humidity. Precipitation and temperature from ERA-Interim or MERRA-2. Photosynthetically active radiation (PAR) from CERES SYN1deg Ed4A. All datasets aggregated to 0.5°.
• Ancillary datasets: Aridity index (CRU v4.01; PET and precipitation; 0.5°), tree cover (Global Forest Change v1.6; 2009; aggregated to 0.5°), plant functional types from MODIS IGBP (MCD12Q1; aggregated to 0.5°; shrubland and woody savanna split north/south of 45°N).
• Data selection: Focus on growing-season days most likely controlled by water limitations. For each pixel, retain days with temperature > 15°C, VPD > 0.5 kPa, and high radiation (PPFD > 500 μmol m−2 s−1), following prior studies, to reduce confounding by temperature/radiation limitations.
• Addressing SM–VPD coupling: Quantified correlations between SM and VPD across timescales (yearly to daily). To decouple, binned data by local percentiles (10 bins) of SM or VPD per pixel. Within each SM (or VPD) bin, SM and VPD correlations are approximately zero, enabling attribution.
• Normalization: For spatial comparability, normalized SIF by the average of SIF values above the 90th percentile per pixel.
• Estimating limitation effects without coupling: Within each SM bin, further sorted by VPD and computed change in SIF from the lowest to highest VPD bin to estimate VPD limitation under fixed SM (ΔSIF(VPD|SM)). Within each VPD bin, computed change in SIF from highest to lowest SM to estimate SM limitation under fixed VPD (ΔSIF(SM|VPD)). Two approaches used: (i) bin-endpoint differences summed over populated bins; (ii) linear fits within bins to estimate end-to-end changes. Primary results reported from approach (i); both yielded similar outcomes.
• Sensitivity metric: Defined standardized sensitivity of SIF to SM as change in SIF per 0.1 m³ m−3 SM change (vSM), to compare across regions independent of local SM range.
• Robustness checks: Tested alternative forcings (MERRA-2, ESA CCI SM; GOME-2 and SCIAMACHY SIF), standardized SIF by PAR, restricted temperature ranges, coarser temporal aggregation, and 20-bin discretization. Assessed latitudinal/zonal means and spatial patterns.
• Spatial analysis across gradients: Evaluated dependence of SM limitation across aridity classes, tree cover fractions, and within plant functional types (PFTs).

• SM–VPD coupling: Strong negative correlation between SM and VPD at yearly scales globally; correlations decrease toward daily scale but remain substantial in many regions, confounding attribution when not accounted for.
• Decoupling via binning: Within percentile bins of SM or VPD at daily scale, SM and VPD are approximately uncorrelated, enabling separate effect estimation.
• Dominant role of SM: Across 71.3% of vegetated land with valid data, the absolute SM limitation |ΔSIF(SM|VPD)| exceeds the VPD limitation |ΔSIF(VPD|SM)|. VPD dominates in 26.7% of areas.
• Magnitude of effects: On average globally, moving from wettest to driest SM under constant VPD reduces SIF by up to 14.9%. Changing VPD from lowest to highest under constant SM has a much smaller mean effect (−3.8%). In many regions, ΔSIF(VPD|SM) is close to zero after removing SM–VPD coupling.
• Regional patterns: Strong SM limitation in mid-latitudes and semi-arid regions (southern North America, central Eurasia, southern Africa, Australia). In tropical Africa near the equator, VPD effects can exceed SM effects. In boreal and tropical regions, both SM and VPD effects on SIF are small, with radiation and temperature being more limiting.
• Ecosystem gradients: SM limitation effects are largest in semi-arid ecosystems (shrubland, grassland, savanna). Regions with lower tree cover exhibit larger SM stress responses. Within the same PFT, sensitivity to SM varies widely, indicating substantial within-PFT heterogeneity likely linked to species composition and plant hydraulic strategies.
• Robustness and caveats: Spatial patterns of dominance are robust across forcing datasets and methodological variants. Morning overpass SIF (GOME-2, SCIAMACHY) yields weaker VPD effects, implying potential underestimation of VPD impacts when using morning observations.
• Implication on prior estimates: Many previous assessments likely overestimated VPD impacts on ecosystem production by not accounting for SM–VPD coupling.

By removing the confounding SM–VPD correlation through percentile binning, the study demonstrates that low soil moisture is the principal constraint on ecosystem photosynthesis across most vegetated regions, resolving the debate over relative influences of SM and VPD at ecosystem scale. The strong SM limitation in semi-arid biomes implies that soil water availability is a key driver of interannual variability of terrestrial carbon uptake and that projected dryland expansion could amplify soil-moisture control on the global carbon cycle. The weak VPD effect after accounting for coupling suggests stomatal responses to atmospheric demand observed at leaf scale do not necessarily translate into ecosystem-level reductions in photosynthesis, likely due to compensating processes (e.g., canopy integration, water storage, hydraulic strategies, and changes in water-use efficiency). These findings call for re-evaluation of model formulations that apply strong VPD limitations and underscore the need for models to disentangle and parameterize SM versus VPD effects consistently with observations for reliable drought risk and carbon-climate projections.

The study provides global, observation-based evidence that soil moisture, rather than vapor pressure deficit, predominantly limits ecosystem production under dryness across most vegetated land. After accounting for SM–VPD coupling, VPD effects are much smaller and often negligible. SM stress peaks in semi-arid ecosystems and in areas with lower tree cover, and sensitivity to SM varies substantially even within the same PFT. These insights inform improvements to terrestrial ecosystem models by emphasizing SM-driven constraints and guiding the incorporation of plant hydraulic processes. Future work should translate SIF-derived reductions into carbon flux changes, incorporate deeper soil and groundwater contributions, refine diurnal sampling to better capture VPD effects, and leverage expanded ecosystem-scale experiments to constrain process representations.

• Observational proxies: SIF, while closely linked to GPP, carries uncertainties related to atmospheric attenuation, cloudiness, canopy structure, and the use of MODIS reflectance (including some morning observations) in CSIF.
• Temporal sampling: Morning SIF datasets (GOME-2, SCIAMACHY) tend to underestimate VPD effects; diurnal variations may not be fully captured despite conversion to daily means.
• Soil moisture representation: Main analyses rely on shallow to root-zone SM from reanalyses/satellite; deep soil moisture and groundwater contributions for deep-rooted vegetation are not explicitly included, potentially underestimating SM effects in some systems or mischaracterizing accessible water.
• Filtering and binning: Analyses restricted to warm, high-VPD, high-radiation days to isolate water limitations; results may not generalize to cooler/low-radiation conditions. Percentile binning requires sufficient data density; some pixels have fewer than 10 populated bins.
• Spatial resolution and aggregation: Forcings aggregated to 0.5°; sub-grid heterogeneity in vegetation, soils, and microclimate may not be resolved.
• Attribution scope: Focus on SM and VPD; other drivers (temperature, radiation, phenology, nutrient status) are filtered or controlled but may still contribute locally. Lack of large-scale ecosystem humidity/temperature manipulation experiments limits direct validation of decoupled effects.