Significantly wetter or drier future conditions for one to two thirds of the world's population

The study addresses the long-standing uncertainty and lack of inter-model coherence in projections of future precipitation under global warming. While temperature projections tend to be consistent, precipitation is influenced by diverse model physics, radiative forcing sensitivity, aerosols, sea surface temperature patterns, and internal climate variability (e.g., ENSO, IOD, PDO, annular modes). These factors lead to heterogeneous regional outcomes and weak agreement across climate models, limiting the effectiveness of adaptation planning. The authors propose to detect robust wetting or drying signals by analyzing continuous 120-year time series from a large ensemble of CMIP5 and CMIP6 simulations without interannual or multi-model averaging. The aim is to determine where models consistently project statistically significant, substantial changes in precipitation totals, identify seasonal dominance driving annual changes, compare CMIP generations, and quantify implications for populations at country and state scales under intermediate (RCP4.5/SSP2-4.5) and very high (RCP8.5/SSP5-8.5) emissions scenarios.

The paper situates its work within literature highlighting robust hydrological-cycle responses to warming and the greater uncertainty in regional precipitation projections compared to temperature (e.g., Held & Soden; Trenberth). It notes model-dependent precipitation responses arising from differences in model physics, climate sensitivity (including higher ECS in some CMIP6 models), aerosol forcing, and sea surface temperature biases. Internal modes (ENSO, IOD, PDO, annular modes, AMO) contribute to variability and are not phase-aligned across models, compounding projection heterogeneity. Prior multi-model assessments have identified tendencies for mid-latitude drying and high-latitude/tropical wetting, changing extremes, and strengthening contrast between wet and dry regions, but commonly rely on long-term mean differences and smaller ensembles. Issues such as ITCZ biases and coupling errors are known to affect regional precipitation patterns. Recent advances propose improved metrics for model agreement and emphasize careful treatment of independence and internal variability. The present study builds upon these by using trend-based, non-parametric methods over full time series and by integrating all available CMIP5/6 simulations to assess agreement and population exposure.

Data and scenarios: 146 GCM runs from CMIP5 and CMIP6 were analyzed: 67 runs for intermediate emissions (25 CMIP5 RCP4.5; 44 CMIP6 SSP2-4.5) and 79 runs for very high emissions (35 CMIP5 RCP8.5; 44 CMIP6 SSP5-8.5). Time period: 1980–2099 (120 years). Precipitation totals were examined annually and for calendar seasons (DJF, MAM, JJA, SON). Trend detection: For each model and grid cell, non-parametric Mann–Kendall tests assessed the presence of a monotonic trend (p<0.05), and Theil–Sen slopes quantified magnitude. The cumulative 120-year trend magnitude was compared to the local mean to define substantial change. Definition of wetting/drying signal per model and grid cell required three criteria: (i) statistically significant trend (p<0.05); (ii) direction (positive = wetting; negative = drying); (iii) cumulative change over 120 years of at least 10% of the local mean precipitation regime. Multi-model agreement metric: For each grid cell (annual and seasonal), the percentage of models satisfying the above criteria for drying and for wetting was computed, yielding separate agreement fractions. Two thresholds were considered: majority (≥50%) and 66% (two-thirds). No ensemble averaging was performed; each ensemble member was treated individually and then summarized as agreement. CMIP5 and CMIP6 ensembles were also assessed separately to compare generations; hotspot overlap was substantial. Seasonal dominance: Within regions exhibiting annual-scale agreement (≥50%), the season contributing the largest trend magnitude to the annual change was identified per model. The median across ensembles gave the dominant season (DJF/MAM/JJA/SON) for drying and wetting in each region. Population exposure: Affected populations were estimated by overlaying the wetting and drying agreement masks with 1 km gridded current population data (WorldPop) and projected future populations (to 2100 under SSPs). Grid cells within wetting or drying masks were summed globally and by country/state. Agreement thresholds were applied to ensure non-overlapping masks; analysis of distributions showed that ≥50% agreement yields distinct wet/dry areas with negligible overlap. Administrative boundaries were from GADM. Code (R) and gridded outputs are publicly available (Zenodo, Figshare).

- Robust hotspots of agreement in drying include Mediterranean Europe (Greece, Spain, Portugal), North Africa (Morocco, Tunisia), Central America and the Caribbean (Cuba, Dominican Republic, Haiti, Jamaica, Puerto Rico), parts of South America (Chile, southern South America, eastern Brazil, Amazon), southern Africa (Namibia, Botswana, South Africa, Mozambique), and Western/Southwestern Australia. - Robust hotspots of agreement in wetting include high latitudes (Finland, Scandinavia, Canada, Russia, Greenland, Svalbard), much of Asia (India, Bangladesh, China, Japan, South Korea, Southeast Asia), Central and parts of East Africa, and the northwestern United States. - Magnitudes: Countries with high agreement often show large cumulative changes. Drying: potential cumulative changes up to about −21% under intermediate emissions and up to −55% under very high emissions (examples include Trinidad and Tobago, Morocco, Grenada, Gibraltar). Wetting: cumulative increases exceeding +35% under intermediate emissions and +48% under very high emissions in some locations (e.g., Greenland, Svalbard, Nauru, Kiribati). - Population exposure (≥50% agreement threshold, current population): Intermediate emissions: 38% (~3.0 billion people) affected (3.3% drying ~266.52 million; 34.7% wetting ~2.76 billion). Very high emissions: 65.6% (~5.0 billion) affected (11% drying ~875.47 million; 54.6% wetting ~4.35 billion). - Population exposure using future projections to 2100: 35.5% (~3.26 billion) under intermediate and 61.4% (~4.64 billion) under very high emissions. - Using a stricter 66% agreement threshold: current affected population ranges from 9.9% (1.2% drying, 8.7% wetting) under intermediate to 42.9% (7% drying, 35.9% wetting) under very high; future affected population ranges from 6.2% (1.0% drying, 5.2% wetting) to 39.7% (7.0% drying, 32.7% wetting). - Seasonal patterns: Strong JJA drying agreement over parts of central/northern Europe (e.g., UK, Germany, Austria, Belgium, Switzerland), with DJF wetting agreement; SON drying over southwestern Australia and northeastern Brazil; MAM–JJA gradients across the United States (MAM drying-to-wetting from south to northwest; inversion in JJA); Africa shows wetting in the north and drying in the south across seasons. Seasonal dominance: SON dominates drying over Iberian Peninsula and SW Australia; SON dominates wetting over China, India, Central Africa; MAM dominates wetting over the United States. - CMIP5 vs CMIP6: Broadly similar patterns; CMIP6 tends to show greater wetting agreement especially at high latitudes, likely linked to higher equilibrium climate sensitivity. Agreement is generally stronger under very high emissions; the increase from intermediate to very high is more pronounced for drying than for wetting. - Regions with limited annual-scale agreement include central Europe, Southwest Asia, Australia (some regions), parts of the west coast of Africa, and parts of South America, where seasonal signals can be strong but offset annually.

The analysis demonstrates that a large fraction of the global land area exhibits robust, model-agreed trends toward wetter or drier precipitation totals under continued greenhouse gas emissions. The spatial pattern aligns with known large-scale circulation changes under warming: expansions of the Hadley cell, weakening Walker circulation, shifts in the ITCZ, land–ocean warming contrasts, and poleward moisture transport. These mechanisms promote drying in many mid-latitudes (e.g., Mediterranean, southern Africa, parts of South America and Australia) and wetting in high latitudes and parts of the tropics and subtropics (e.g., much of Asia and central Africa). The study’s trend-based, time-series approach reduces sensitivity to internal variability and avoids the potential masking effects of interannual or multi-model averaging, thereby clarifying where and when different models tell consistent storylines. The projected impacts are substantial: by 2100, approximately 3–5 billion people living today (or 35–61% based on future population projections) are in regions where precipitation totals are expected to change significantly. Seasonal analyses reveal that annual-scale signals can be composed of opposing seasonal trends (e.g., European DJF wetting vs JJA drying), emphasizing the need for season-specific adaptation planning. Interpretation must consider model structural limitations (e.g., ITCZ biases) and correlated genealogy among ensemble members, though hotspot patterns remain consistent when accounting for model dependence. The findings have direct relevance for water resource planning, disaster risk reduction, and climate adaptation policies in both drying and wetting hotspots.

This study introduces a robust, trend-based framework to detect multi-model agreement on significant wetting and drying of precipitation totals, using the full time series from 146 CMIP5/CMIP6 simulations without ensemble or interannual averaging. It maps global hotspots of change, identifies seasonal drivers, and quantifies exposed populations at country and state scales under intermediate and very high emissions. Key contributions include: (i) a clear delineation of regions with substantial drying (e.g., Mediterranean, southern Africa, parts of the Americas and Australia) and wetting (e.g., high latitudes, much of Asia, central Africa); (ii) evidence that moving from intermediate to very high emissions particularly amplifies drying agreement and magnitude; and (iii) estimates that one to two thirds of the world’s population may be affected by century’s end. Implications are immediate for adaptation planning, including seasonal water management, flood risk mitigation in wetting regions, and drought preparedness in drying regions. The authors note that strong emissions reductions could lessen the extent of drying-impacted regions and populations, though wetting extents may be less sensitive. Future research could extend this framework to include other hydrological variables (e.g., evapotranspiration, soil moisture), explicitly assess extremes alongside totals, refine model-independence weighting, and explore attribution of regional trends to circulation and SST pattern changes.

- Focus on precipitation totals only: atmospheric demand, evaporation, and other hydrological components are not included, potentially underrepresenting drought risk aspects tied to evapotranspiration. - Model structural biases: Known issues such as ITCZ position biases and cloud-related errors can affect regional precipitation patterns, especially in the tropics and Southern Ocean-influenced regions. - Ensemble dependence: Although many runs are used, multiple realizations of related models reduce effective independence; however, hotspot patterns remained consistent under subset sampling accounting for genealogy. - Scenario plausibility: Very high emissions (RCP8.5/SSP5-8.5) may be less likely under current policies, but are used to bracket an upper-bound impact envelope. - Spatial heterogeneity and seasonal compensation: Some regions show limited annual-scale agreement due to opposing seasonal trends; local planning should consider seasonal detail and subnational variability. - Threshold choice: While a ≥50% agreement threshold is justified to avoid overlap and reveal impacts, using a stricter threshold (66%) substantially reduces affected area/population estimates, reflecting uncertainty in model agreement magnitude.