Human-induced intensification of terrestrial water cycle in dry regions of the globe

The terrestrial water cycle (TWC) is crucial for water resources, agriculture, and ecosystems. Although global warming has strengthened the TWC and increased global land annual total precipitation (PRCPTOT), trends are spatially heterogeneous across climatic regions. Contrary to the paradigm of “dry gets drier, wet gets wetter,” recent evidence shows PRCPTOT increasing faster in dry than in wet regions. This increase may be offset by evapotranspiration in dry areas, complicating implications for water availability. The study asks whether anthropogenic climate change (ACC) drives differing PRCPTOT changes in dry versus wet regions, whether warming and precipitation sensitivity differ between these regions, and how distinct external forcings—greenhouse gases (GHG) versus aerosols (AER)—shape these differences historically and into the future. The work investigates whether the global ACC signal in PRCPTOT remains detectable without dry regions and quantifies the contributions of external forcings to observed PRCPTOT changes in each region. Understanding these divergent responses is vital for planning, water management, and sustainable development.

Prior studies have attributed global increases in terrestrial PRCPTOT and intensification of the TWC to ACC, via thermodynamic (increased atmospheric humidity) and dynamic (circulation changes) processes. However, precipitation trends vary regionally and by climate regime, with evidence of faster PRCPTOT increases in dry regions than wet regions. External forcings can have opposing thermal effects—GHG warming versus AER cooling—leading to differing regional responses and potentially explaining divergent PRCPTOT changes. Detection and attribution studies have found ACC signals in regional and hemispheric precipitation changes, yet the detectability may vary by climate regime, especially given monsoon dynamics and aerosol impacts. The literature also highlights increased short-duration rainfall extremes, shifts in circulation (Walker and Hadley cells, jet streams and storm tracks), and land–atmosphere feedbacks (soil moisture effects) as potential mechanisms modulating regional precipitation responses to warming.

Data and region definitions: The study uses three gauge-based precipitation datasets (HadEX3, CRU, GPCC) normalized by each product’s 1981–2010 climatological mean. Data gaps from HadEX3 guide grid selection; 1432 land grid cells at 2.5°×2.5° (excluding Antarctica) with <10% missing data over 1961–2018 are retained. Dry (wet) regions are defined as grid cells within the bottom (top) 30% of climatological PRCPTOT among these 1432 cells; sensitivity tests use 20% and 40% thresholds. A PME-based (precipitation minus evapotranspiration) regionalization using GLEAM ET (1981–2020) is also applied to validate the PRCPTOT-based division. Climate model simulations: CMIP6 historical experiments include NAT (natural), GHG, AER (anthropogenic aerosols), and ALL (historical total forcing). Anthropogenic forcing (ANT) is computed as ALL−NAT. Future projections use SSP126/245/370/585. ALL is extended with SSP585 after 2014 to form H85 (1901–2100). PiControl simulations provide internal variability estimates. In total, 61 models and 309 realizations are analyzed; all data are bilinearly interpolated to 2.5°×2.5°. Diagnostics and analyses: - Trends and sensitivities: Area-averaged, normalized PRCPTOT anomalies for dry and wet regions are analyzed for 1961–2018. Linear trends use Sen’s slope; significance uses Mann–Kendall tests. PRCPTOT–temperature sensitivity (% per K) is evaluated across forcings and scenarios. A large single-model ensemble (CanESM5, 50 members) is used to examine PRCPTOT–temperature relationships by comparing 1961–2018 anomalies relative to 1861–1900 means. - Moisture metrics: Vertically integrated water vapor (VIWV) and integrated water vapor transport (IWVT) are computed and normalized: VIWV = ∫_{pt}^{ps} q dp; IWVT = ∫_{pt}^{ps} sqrt(u^2+v^2) dp, where q is specific humidity and u,v are winds. Changes in VIWV and IWVT versus global warming are assessed using model differences between 2070–2099 (SSP245/585) and 1961–1990 (ALL). - Pattern-based detection and attribution: Rotated EOF (REOF) analysis on H85 PRCPTOT (1901–2100) extracts leading spatial modes (EOF1) and PCs for dry, wet, and combined (dry+wet) regions; EOF1 serves as the anthropogenic fingerprint, explaining ~84.6–92.7% variance and closely tracking global-mean temperature. Observed and simulated PRCPTOT are projected onto EOF1 to form time series P(t); the ACC signal S(L) is the L-year trend (L=58 years for 1961–2018). Noise N(L) is the standard deviation of all L-year trends in PiControl projections. Signal-to-noise ratio SNR = S(L)/N(L) determines detectability at 90% (1.64) and 99% (2.57) confidence. Time of emergence is estimated from SNR evolution in ALL and SSP585. - Optimal fingerprinting: Observed PRCPTOT is regressed on simulated signals in one-signal (Y=(X−α)β+ε) and two-signal (Y=(XNAT−αNAT)βNAT + (Xother−αother)βother + ε) frameworks, solved via OLS and TLS. PRCPTOT time series are 3-year averaged to reduce interannual variability. Detection is established if the 90% CI of β exceeds zero. Attributable trends are computed by multiplying β by the simulated trends under each forcing. - Robustness tests: Results are tested across different dry/wet thresholds (20–40%) and using PME-based regionalization. Projections under multiple SSPs assess future evolution of dry–wet contrasts.

- Observed historical changes (1961–2018): PRCPTOT increased significantly faster in dry regions (1.1% per decade; p<0.05) than in wet regions (0.2% per decade; p=0.18). Across thresholds, observed increases total 5.63%–7.39% (dry) versus 2.44%–2.80% (wet). - CMIP6 historical forcings: • ALL: Similar pattern with increases of ~1.2%/decade (dry) and ~0.2%/decade (wet). • NAT: No significant changes in either region. • ANT: Significant increases in both, but stronger contrast—~2.0%/decade (dry) vs ~0.2%/decade (wet). • GHG: Significant increases in both regions; AER: significant increase in dry regions but significant decrease in wet (monsoon) regions. - Gradient with aridity: The drier the selected grids within the dry category, the larger the PRCPTOT increase; conversely, selecting the very wettest grids shows weaker modeled PRCPTOT increases. - Future projections: Under SSP126, increases are weak in both regions. Under SSP245/370/585, PRCPTOT increases in both regions with a consistently faster rise in dry regions, and the dry–wet contrast grows with emissions/warming. - Warming and sensitivity: Historical regional warming is faster in dry regions (0.33 K/decade) than wet (0.23 K/decade) under ALL; this contrast vanishes under NAT. PRCPTOT sensitivity to regional warming is much higher in dry regions (3.07% K⁻¹) than wet (0.98% K⁻¹) under ALL; under NAT, sensitivity is weakly positive (dry, 0.71% K⁻¹) and negative (wet, −1.30% K⁻¹). CanESM5 large ensemble corroborates these relationships. - Moisture metrics: With global warming, VIWV increases similarly in dry and wet regions (~6.2/8.1% per K in dry vs ~6.1/7.3% per K in wet under SSP245/585), but IWVT increases more strongly in dry regions (7.7/11.1% per K) than in wet (3.6/3.8% per K), indicating stronger moisture transport to dry regions. - Detection and attribution (pattern-based): Fingerprint projections show detectable ACC signals at 90% confidence in dry+wet (SNR=1.73) and dry regions (SNR=1.68), but not in wet regions alone (SNR=1.31). Time of emergence under SSP585: detectable in dry regions by ~2012 (90%) and ~2021 (99%); in wet regions by ~2021 and ~2032. - Optimal fingerprinting: In dry regions, scaling factors β are significantly >0 for ALL, ANT, and GHG (one-signal OLS/TLS), confirming detection; not significant for NAT or AER, and no robust detection in wet regions. - Attributions (1961–2018): • Dry regions (observed 5.63%–7.39%): OLS-attributed increases—ALL: 5.22%–5.82%; ANT: 5.02%–6.17%; GHG: 5.68%–8.20%. TLS-attributed—ALL: 6.25%–7.04%; ANT: 5.89%–7.07%; GHG: 6.96%–9.93%. • Wet regions (observed 2.44%–2.80%): Attributed increases are small and uncertain—ALL: 0.06%–1.67% (OLS), 0.94%–2.08% (TLS); ANT: −0.46%–0.21% (OLS), −0.56%–0.91% (TLS); GHG: 0.19%–1.18% (OLS), 1.18%–2.14% (TLS). - Overall: The global ACC signal in increasing land PRCPTOT largely arises from dry regions; excluding dry regions substantially diminishes detectability. Faster ACC-induced intensification of the TWC in dry regions is evident historically and strengthens with higher-emission futures.

The study addresses whether ACC differentially intensifies the terrestrial water cycle across dry and wet regions and whether the global PRCPTOT increase is primarily driven by changes over dry lands. The findings show that anthropogenic forcings produce faster regional warming and a higher precipitation sensitivity to warming in dry regions, generating larger PRCPTOT increases than in wet regions. Thermodynamic effects (enhanced moisture content) occur in both regions, but dynamic processes and land–atmosphere feedbacks preferentially strengthen moisture transport and vertical ascent over dry regions. Stronger land–sea thermal contrasts and altered circulation (jet streams, storm tracks) under warming enhance landward moisture transport, while soil moisture declines amplify turbulent fluxes and low-level ascent over dry areas. In contrast, in many wet (monsoon) regions, dynamic changes under GHG warming (e.g., weakened Walker/Hadley circulations and modified stratification) partially offset thermodynamic moistening, dampening PRCPTOT increases; aerosols further reduce precipitation in these regions through thermodynamic and dynamic pathways. Detection and attribution analyses confirm that the anthropogenic fingerprint is robustly detectable in dry regions but not in wet regions over 1961–2018, and the cumulative ACC signal in global land PRCPTOT is predominantly contributed by dry regions. These results have high relevance for water resource management, infrastructure planning, and risk mitigation, given potential dual outcomes in dry regions—easing water scarcity but increasing flood risk.

This work demonstrates that anthropogenic climate change has intensified the terrestrial water cycle more strongly in dry regions than in wet regions during 1961–2018, with faster warming, higher precipitation sensitivity to warming, and stronger moisture transport underpinning the contrast. Detection and attribution results show the global ACC signal in increasing land PRCPTOT mainly originates from dry regions; without them, the signal is weak or absent. Future projections indicate that under medium-to-high emission pathways, both the rate of increase and the dry–wet contrast will grow, emphasizing the need for targeted adaptation for both water scarcity and flood hazard management in dry regions that host a large share of the global population. Future research should reassess ACC signals in other TWC components (evapotranspiration, runoff, soil moisture, snow) considering strong human interventions, refine representations of aerosol effects and monsoon dynamics, and further examine land–atmosphere feedbacks and circulation changes that drive moisture transport into drylands.

- Metric choice: The study uses PRCPTOT as the primary proxy for TWC intensification. Other components (ET, runoff, soil moisture, snow) are influenced by substantial human interventions (e.g., dams, irrigation, land use), complicating clean attribution; thus, ACC signals in these components are not directly assessed here. - Regionalization: Dry and wet regions are defined by PRCPTOT quantiles (bottom/top 20–40%), which, while tested for sensitivity and validated with PME-based regions, remain a methodological choice that may affect regional categorization. - Model dependence and forcing separation: Attribution relies on CMIP6 simulations and their representations of GHG and aerosol effects; uncertainties in aerosol-cloud interactions, monsoon dynamics, and circulation responses can affect results, particularly in wet regions where detection is weaker. - Detection in wet regions: The anthropogenic signal is not robustly detectable in wet regions over 1961–2018, indicating larger internal variability and/or forcing uncertainties; attributable contributions there have wider confidence intervals. - Data coverage and normalization: Although extensive gauge-based datasets were used and normalized, spatial inhomogeneities and observational uncertainties, especially in data-sparse regions, may influence trend estimates.