The timing of unprecedented hydrological drought under climate change

The study addresses when unprecedented hydrological drought conditions—defined as river discharge falling below historical variability—will first emerge under climate change. With the intensification of the hydrological cycle due to altered precipitation, snow regimes, and evapotranspiration, understanding the timing of such droughts is critical for adaptation and water resource planning, especially as historical stationarity assumptions no longer hold. While prior work has explored Time of Emergence (ToE) for various climate signals, the timing of unprecedented drought has rarely been assessed, particularly considering low-emission pathways consistent with the Paris Agreement. The authors argue that focusing on the first emergence of multi-year unprecedented drought (TFE) offers actionable urgency for adaptation, differing from strict ToE definitions of permanent exceedance. They also emphasize assessing river discharge (including low-flow propagation through river networks) rather than only local runoff to better capture water, food, and energy security risks.

Most large-scale drought projections have used time-slice or temperature-slice approaches to assess future changes, often under high-emission scenarios, and have emphasized soil moisture or meteorological indices. Few studies have examined ToE for drought; Touma et al. (2015) assessed drought ToE globally using uncorrected GCM outputs under RCP8.5 and found limited robust emergence for runoff and precipitation drought. Many ToE studies constrain emergence to permanent exceedance and largely focus on high emissions to obtain robust signals. The literature indicates regionally significant drought intensification even at 1.5 °C warming, underscoring the need to assess timing under low emissions (RCP2.6). Previous work also highlights the importance of river discharge and human influences (reservoirs, withdrawals), and the role of internal variability in masking or delaying emergence, suggesting the need for robust statistical treatment.

Design: Global multimodel, multi-scenario hydrological analysis at 0.5° resolution using ISIMIP2b simulations for 1861–2099. Historical period 1861–2005; future 2006–2099 under RCP2.6 and RCP8.5. Five global water models (GWMs: CWatM, H08, WaterGAP, MATSIRO, LPJmL) forced by four bias-corrected CMIP5 GCMs (HadGEM2-ES, IPSL-CM5A-LR, GFDL-ESM2M, MIROC5). A consistent river routing map (DDM30) was used. Human influences (withdrawals, reservoir operations, land-use changes) included historically; post-2005 kept constant at 2005 levels (2005soc) to isolate climate effects. Irrigation water demand responds to climate. Drought detection: Hydrological drought defined via a daily moving-threshold method using Qx percentiles (X=0.1) of daily discharge over the 145-year baseline (1861–2005) using 31-day windows around each calendar day (4495 samples per day). Pooling rules applied: τx allows short nondrought interruptions within a drought spell; Lenx filters out short drought spells shorter than a threshold; focus on longer, severe events. Frequency of drought days (FDD) computed annually and for seasons. Seasons and regions: Low-flow and high-flow seasons defined per grid cell as the three months with, respectively, lowest and highest historical discharge. Analyses aggregated to 59 subcontinental regions (HydroBASINS level 2) with area weighting and excluding glacial grid cells. Regional average FDD time series smoothed by 5-year windows (e.g., 1861–1865 assigned to 1865) to reduce interannual noise while retaining multidecadal variability. TFE definition: The Time of First Emergence (TFE) is the first year when the regional average FDD exceeds the historical maximum (1865–2005) and remains above for a specified consecutive duration. Main focus is TFE5: at least 5 consecutive years. To avoid censoring at the simulation end, exceedances starting 2095–2099 are excluded for TFE5. Additional analyses include TFE10 and seasonal/annual TFEs. Uncertainty and robustness: Internal variability and model structure uncertainty quantified using block-wise bootstrap resampling. For each of 20 GCM–GWM members and region, 1985–2099 FDD series was decomposed into a quadratic trend plus anomalies; five-year nonoverlapping blocks of anomalies were randomly resampled 100,000 times and added back to the trend, yielding 2 million resampled series per region. Cumulative distribution functions (CDFs) of TFE over time computed at 5-year steps (2010–2095). Uncertainty bands derived by bootstrapping CDFs from 2000 random subsets. Robustness criteria: very likely emergence if >95% of large samples show TFE by 2100; robust median TFE if >95% of bootstrapped medians occur in 21st century; unlikely if ≤33% probability by a given time (IPCC usage). Additional analyses: Total number of unprecedented years (2010–2099) counted for exceedances lasting ≥5 years; scenario differences tested via bootstrap of medians. Conversion of TFE to global mean temperature rise (AGMT) via 31-year centered GMT climatologies per GCM relative to 1850–1900. Sensitivity/experiments: Compared default variable CO2 vs fixed-2005 CO2 physiological effects (LPJmL, MATSIRO) and assessed socioeconomic evolution (rcp26soc) vs fixed 2005soc where available; noted potential influence on TFE timing.

Spatial/temporal changes: By mid-21st century (2036–2065), FDD during the low-flow season increases significantly over 25% of land under RCP2.6 and 28% under RCP8.5 (KS test, p=0.05; >60% sign agreement). Pronounced increases (>twofold) affect 6% (RCP2.6) and 9% (RCP8.5) of land. Hotspots include the Mediterranean, Western/Central Europe, Middle East, Central Asia, United States, Central and southern South America, West–South Africa, and Australia. In many hotspots, FDD increases in both low- and high-flow seasons, implying year-round low-flow intensification. TFE occurrence: For TFE5 (low-flow season), robust median TFEs detected in 11 (RCP2.6) and 18 (RCP8.5) out of 59 regions. High likelihood (>66%) during the 21st century in 4 regions (RCP2.6) and 13 regions (RCP8.5). Earliest and most robust regions (RCP8.5): Southwestern South America (SWS) median TFE5 ≈ 2020; Mediterranean Europe (MED) ≈ 2035; Northern Africa (NAF) ≈ 2040. By 2050, TFE5 detected in 98% (SWS), 93% (MED), and 67% (NAF) of large samples; uncertainty in cumulative probability at 2050 is <2%, <5%, and <10% respectively. More than 95% of resamples show TFE5 by 2045 (SWS), 2050 (MED), and 2080 (NAF). Scenario benefits: Under RCP2.6, only SWS has >95% likelihood of TFE5 by 2100; MED shows relatively high likelihood, while many other regions see delayed or no median TFE5 within the century. Differences between scenarios grow in the second half of the century; cumulative probabilities under RCP8.5 increase steadily, whereas under RCP2.6 they slow or stabilize around mid-century. The total number of unprecedented years (2010–2099) is significantly lower under RCP2.6 than RCP8.5 in many regions (18 of 59 with TFE5 under either scenario). Longer consecutive durations: Under RCP8.5, TFE10 appears in 13 regions; seven regions (including SWS, MED, NAF, Southwestern and Southern North America, Madagascar, Western and Central Europe) exhibit median TFEs for consecutive exceedance lasting to century-end (>20–70 years). Under RCP2.6, TFE10 robustly appears only in SWS and MED. High-flow and annual scales: TFE5 during the high-flow season and annually is detected in similar regions; under RCP8.5, SWS, MED, NAF, and the Middle East show robust TFE5 (>95%). Temperature framing: Mapping TFEs to AGMT shows several regions likely face unprecedented drought at relatively low warming. Median TFE5 corresponds to AGMT below or near 2 °C for SWS and MED, and 2–2.5 °C for NAF and Madagascar. Likelihoods at 2 °C: SWS 76% (±10%), MED 52% (±10%), NAF 36% (±11%); at 1.5 °C: SWS 49% (±10%), MED 15% (±7%), NAF 19% (±9%). >95% likelihoods are reached around AGMT ~2.9 °C (SWS), ~3.3 °C (MED), ~4.2 °C (NAF). Achieving Paris targets reduces TFE likelihoods to unlikely (<33%) in most regions.

The findings directly answer when unprecedented multi-year hydrological drought conditions are first expected to emerge regionally, highlighting substantial near- to mid-term risks even under low emissions for some hotspots (notably SWS and MED). Emission pathway choice critically alters both timing and likelihood, with RCP2.6 substantially delaying or preventing emergence in many regions and reducing the total time spent under unprecedented drought. Nonetheless, some regions face high likelihood of emergence irrespective of scenario, necessitating early adaptation to protect water, food, energy systems, and ecosystems. The conversion to AGMT underscores that keeping warming to 1.5–2 °C considerably lowers emergence probabilities, supporting the Paris Agreement’s effectiveness for drought risk reduction. The study also stresses inter-seasonal linkages: high-flow season deficits can propagate into more severe low-flow seasons. Uncertainties related to internal variability, climate sensitivity, vegetation physiological responses to CO2, and future human water use remain, but the ensemble and bootstrap approach provides probabilistic, robust timing estimates that can inform planning horizons and preparedness.

This work provides the first global assessment of the time of first emergence (TFE) of unprecedented regional hydrological drought based on river discharge, using a multi-model, bias-corrected framework with explicit human water management. It shows significant increases in drought-day frequency across multiple hotspots and robust TFEs in 11 (RCP2.6) to 18 (RCP8.5) regions, with particularly early and robust emergence in Southwestern South America, Mediterranean Europe, and Northern Africa under high emissions. Emission mitigation markedly reduces both the likelihood and duration of unprecedented drought, with Paris-aligned warming levels generally lowering probabilities to unlikely by century’s end for most regions. However, given early emergence in some areas even under low emissions, proactive adaptation is indispensable. Future research should refine drought–impact linkages for sectoral thresholds, expand climate ensemble sizes and initial-condition ensembles, better represent vegetation CO2 physiology and dynamics, and include evolving socioeconomic water use and infrastructure changes to narrow uncertainties and improve regional relevance.

Key limitations include: reliance on four bias-corrected CMIP5 GCMs and five GWMs, leaving residual model and climate sensitivity uncertainties; internal variability treated via bootstrap rather than large initial-condition ensembles; incomplete representation of vegetation physiology and dynamics (only LPJmL and MATSIRO include stomatal CO2 responses; only LPJmL includes dynamic vegetation), with potentially opposing effects on evapotranspiration and drought; future human water use, reservoir expansion, and land-use changes held at 2005 levels (2005soc), likely underestimating drought emergence where withdrawals grow—limited tests with rcp26soc suggest earlier TFEs; drought definition parameter choices (Qx=0.1, pooling thresholds) and the focus on FDD may affect detected timing; aggregation to 59 large regions improves robustness but may mask finer-scale heterogeneity; use of a nontransient historical threshold may not capture evolving baselines; conversion to AGMT depends on GCM-specific warming trajectories, with late-century TFEs capped at 2084 for temperature mapping.