Serious underestimation of reduced carbon uptake due to vegetation compound droughts

Anthropogenic climate change increases the frequency, intensity, and severity of climate extremes such as droughts and heatwaves, threatening ecological and societal sustainability. Drought directly limits terrestrial water availability and carbon uptake through low soil moisture (SM) and/or high vapor pressure deficit (VPD), reducing gross primary productivity (GPP) via stomatal, biochemical, and phenological responses, and can trigger agricultural losses and vegetation mortality. Compound droughts (CDs) characterized by concurrent low SM and high VPD reduce GPP more than either driver alone due to land–atmosphere feedbacks that couple SM deficits with elevated VPD. While CDs are projected to become more frequent and extreme, many assessments use statistical, quantile-based definitions of extremes that do not ensure adverse environmental impacts, leading to inconsistent global estimates and uncertainty about impacts on terrestrial productivity. From an impact-centric perspective, co-occurring drivers need not be statistically extreme to cause large impacts. Quantile-based definitions can misclassify or underestimate CDs and their impacts—for example, at high latitudes where GPP can increase with temperature/VPD, or in drylands where productivity is highly water-limited and moderate anomalies can strongly reduce GPP. This study proposes an impact-based framework defining CDs by how GPP responds to SM and VPD: identifying thresholds where low SM and high VPD strongly limit GPP, and defining vegetation compound droughts (VCDs) when both conditions co-occur in the warm season. The framework is applied to observations (GLEAM SM v3.5a, MERRA-2 VPD, FLUXCOM GPP, 1981–2017) and CMIP6 historical (1930–2014) and future simulations (SSP1-2.6 and SSP5-8.5, 2016–2100) to assess past and projected frequency, intensity, severity, and impacts on terrestrial carbon uptake.

Study design and datasets: The analysis uses CMIP6 historical (1930–2014) and future (2016–2100) simulations under SSP1-2.6 (low emissions) and SSP5-8.5 (high emissions) from 14 Earth system models with monthly soil moisture (SM), near-surface air temperature (T), relative humidity (RH) or specific humidity and pressure (for NorESM2-LM), and gross primary productivity (GPP). Vapor pressure deficit (VPD) is computed from T and RH (or specific humidity and pressure). Observational/reanalysis products include root-zone SM from GLEAM v3.5a (0.25°), T and dew-point temperature from MERRA-2 (1°) to compute VPD, and GPP from FLUXCOM (0.5°) upscaled from eddy covariance sites. All datasets are bilinearly interpolated to 1.5° × 1.5° for consistency. Land grid cells with land fraction >30% are analyzed. Warm season definition and preprocessing: For each grid cell and period (historical and each future scenario), the warm season is defined as the consecutive 3 months with highest average temperature. Monthly GPP anomalies are computed by removing 30-year centered running mean seasonal cycles (to reduce effects of radiation, temperature trends, and CO₂). Impact-based identification of drought drivers and thresholds: The response of GPP anomaly (Y, gC m⁻² day⁻¹) to SM or VPD (X) is modeled per grid cell using a two-segment linear regression with a change point c: Y = α0 + β0 X + ε for X < c, and Y = α1 + β1 X + ε for X ≥ c, enforcing continuity (α0 + β0 c = α1 + β1 c). Change-point c is estimated via grid search with bootstrap; significance of slope change (β0 ≠ β1) is tested by bootstrapping. If no significant change is detected, a univariate linear model Y = α2 + β2 X + ε is used. Drought-limited regime criteria require that in the drought-limited phase GPP increases with SM (β > 0) and decreases with VPD (β < 0). Threshold selection for impacts: For SM, the threshold is the lower of the change-point c and the x-intercept where Y = 0 (zero GPP anomaly); for VPD, the threshold is the higher of c and the x-intercept. In cases where (i) the drought-limited phase does not produce negative GPP anomalies (too weak limitations), soil drought or atmospheric aridity are not identified; and (ii) in permanently drought-limited regimes (very dry), thresholds equal the x-intercept regardless of c. Segmented regression is implemented in R (chngpt), and x-intercepts with rootSolve. Definition of compound events: Vegetation compound droughts (VCDs) are months in the warm season when SM < SM-threshold and VPD > VPD-threshold simultaneously. Statistical compound droughts (SCDs) are defined for comparison as months with SM below its 10th percentile and VPD above its 90th percentile (per month). Metrics: For CDs, frequency is the proportion of CD months in warm-season months (historical/future: 255 months over 85 years; observations: 111 months over 37 years). Duration is the mean length (months) of consecutive CD months. Intensity is the quadratic mean of normalized departures of SM and VPD from their thresholds during CD months, using historical-period standard deviations for normalization. Severity is the cumulative intensity over consecutive CD months. GPP impacts are quantified as average GPP anomaly during CD months and total GPP anomaly (sum over CD months). Dryland delineation: Drylands (hyper-arid, arid, semi-arid, dry sub-humid) are defined by UNCCD using aridity index < 0.65 (precipitation/potential evapotranspiration). Future scenarios: Two approaches are used for future VCDs: (1) apply historical thresholds to future climate to isolate climate-driven changes in SM/VPD; (2) re-estimate thresholds within each SSP to account for CO₂-induced physiological changes in GPP responses, then reassess frequency, intensity, and GPP anomalies. Sensitivity analyses: Analyses are repeated for alternative season definitions (top-3 growing months by GPP, and all growing months with GPP ≥ 70% of annual maximum) to test robustness.

- Coverage and effectiveness: VCDs are detected over 66% of global land (excluding Antarctica/Greenland) in observations and 91% in CMIP6, with consistent negative GPP anomalies for soil droughts, atmospheric aridity, and VCDs. High-latitude regions generally lack atmospheric aridity/VCDs due to temperature limitation and positive GPP–VPD relationships. Model–observation spatial agreement of VCD GPP anomalies: r = 0.54; area-weighted mean GPP anomaly during VCDs: CMIP6 −0.61 ± 0.42 gC m⁻² day⁻¹; observations −0.15 ± 0.10 gC m⁻² day⁻¹. - Impact-based vs quantile-based definitions: Quantile-based SCDs yield positive GPP anomalies over >14% of land (notably high latitudes) due to unaccounted positive GPP–VPD relationships. The frequency of SCDs (CMIP6: 2.6 ± 1.0%; observations: 3.5 ± 1.6%) is only about 11% of VCD frequency (CMIP6: 24.5 ± 8.8%; observations: 32.0 ± 11.7%). Higher VCD frequency arises from higher SM thresholds (median around 30th percentile or higher in many regions) and lower VPD thresholds (often ~70–80th percentile) for adverse GPP impacts versus fixed 10th/90th percentiles for SCDs. - Carbon loss underestimated by SCDs: Global total GPP anomalies due to VCDs are −1.44 PgC yr⁻¹ (CMIP6) and −0.31 PgC yr⁻¹ (observations), whereas SCDs capture <26% of these (CMIP6: −0.24 PgC yr⁻¹; observations: −0.08 PgC yr⁻¹). - Event characteristics: Compared to SCDs, VCDs have 50% higher intensity on average and 35% longer duration, leading to roughly double severity (VCD severity: CMIP6 1.30 ± 0.20 sd; observations 1.42 ± 0.37 sd; SCD severity: CMIP6 0.67 ± 0.25 sd; observations 0.72 ± 0.55 sd). VCDs tend to occur consecutively within warm seasons. - Hotspots and drivers: VCDs are most frequent and severe in drylands with low mean SM, high mean VPD, and strong SM–VPD coupling; drylands contribute >41% of global VCD-induced total GPP anomalies, underscoring their dominant role in interannual land carbon sink variability. Quantile-based methods depend strongly on SM–VPD tail dependence/correlation and under-detect VCDs where co-occurrence exists without strong correlation. - Future projections with historical thresholds: VCD frequency increases across >81% of land; global frequency rises by 10.9 ± 12.8% (SSP1-2.6) and 14.3 ± 16.5% (SSP5-8.5). Intensity increases by 0.41 ± 0.30 sd (SSP1-2.6) and 0.99 ± 0.61 sd (SSP5-8.5), driven mostly by increased VPD and, over >50% of land, also by reduced SM. Amazon Basin and mid-latitudes are projected hotspots of intensification. - Future projections with scenario-specific thresholds (accounting for CO₂ effects): Frequency changes are small (SSP1-2.6: +1.6 ± 4.3%; SSP5-8.5: −0.9 ± 5.5%), as elevated CO₂ raises VPD thresholds and lowers SM thresholds for adverse impacts. Nevertheless, intensity increases markedly (by 56% in SSP5-8.5), and total VCD-induced GPP losses increase by 0.34 PgC yr⁻¹ (SSP1-2.6) and 0.37 PgC yr⁻¹ (SSP5-8.5) relative to historical. - Comparison with SCDs in the future: While SCD intensity and negative GPP anomalies increase regionally, projected total GPP anomaly reductions are only −0.05 PgC yr⁻¹ (SSP1-2.6) and −0.01 PgC yr⁻¹ (SSP5-8.5), less than one-sixth of VCD-induced reductions, due to severe underestimation of CD frequency by quantile-based methods. - Additional insights: The added effect of low SM often dominates the co-occurrence impact over mid-latitude dry regions. Sensitivity tests show most VCD-induced GPP losses occur during the warm season; 3-month warm-season losses account for ~84% of losses across all growing months.

Defining compound droughts from an impact-based perspective using GPP responses to SM and VPD resolves key deficiencies of quantile-based definitions that overlook regional ecohydrological regimes. The approach prevents misclassification in high-latitude, temperature-limited regions (where higher VPD can enhance GPP) and reveals substantial underestimation of frequency, intensity, severity, and carbon losses in low- and mid-latitude regions and especially in drylands. Strong land–atmosphere coupling in drylands amplifies co-occurring low SM and high VPD and their impacts on productivity, explaining their outsized contribution to global carbon sink variability. Future projections indicate increasing VCD frequency and intensity under both low- and high-emissions scenarios. Accounting for CO₂ fertilization and physiological adjustments suggests little change in frequency but significant intensification and larger GPP losses, particularly in the Amazon and mid-latitudes. These findings imply higher risks to the land carbon sink, vegetation growth, and food production than indicated by quantile-based assessments. The results also highlight the need to improve Earth system model representations of drought processes (including extreme event dynamics, plant hydraulics, and drought legacies) to reduce uncertainty and better capture the sensitivity of carbon uptake to compound water stress.

This study introduces an impact-based framework to identify vegetation compound droughts (VCDs) based on local GPP responses to soil moisture and vapor pressure deficit. Relative to quantile-based definitions, VCDs are far more frequent and severe and cause substantially larger reductions in terrestrial carbon uptake—especially in drylands—implying that risks to the land carbon sink have been seriously underestimated. Projections show VCDs will intensify and, depending on scenario, also become more frequent, leading to greater carbon losses despite potential CO₂-driven physiological amelioration. The work advances understanding of compound drought risks and their carbon-cycle impacts and underscores the need for adaptation and mitigation strategies, including reducing fossil-fuel emissions and implementing water management and agricultural measures (e.g., cross-basin transfers, improved irrigation efficiency, drought-tolerant crops). Future research should refine model representations of plant hydraulics, biodiversity effects, and drought legacies to better project ecosystem responses under increasing climate extremes.

- Model limitations: Earth system models may underestimate vegetation sensitivity to drought and fail to fully represent extreme events (e.g., ENSO-driven) and key plant hydraulic processes (including mortality), potentially underestimating VCD impacts. - Methodological assumptions: The impact-based method assumes concurrent-month GPP responses to SM and VPD; it does not capture drought legacy effects and asynchronous ecosystem responses, which are widespread and imperfectly represented in models. - Season definition: Analyses focus on warm seasons (and alternative growing-season definitions in sensitivity tests), which may not align perfectly with phenological growing seasons everywhere (e.g., Mediterranean climates, evergreen tropics). - Threshold estimation uncertainties: Segmented regressions and x-intercept choices introduce statistical uncertainty; thresholds vary spatially/temporally and depend on data quality and model biases in SM, VPD, and GPP. - Scenario treatment: When using historical thresholds, physiological acclimation to CO₂ is not considered; when using scenario-specific thresholds, uncertainties in modeled CO₂ responses and variability remain. - Observational uncertainties: GLEAM SM, MERRA-2 climate, and FLUXCOM GPP have inherent uncertainties from retrievals, assimilations, and upscaling that can affect threshold estimation and anomaly magnitudes.