Climate change will affect global water availability through compounding changes in seasonal precipitation and evaporation

Accessibility of water resources for human consumption and ecosystems depends on the spatio-temporal distribution of precipitation and evaporation. Human-caused climate change in the 21st century is expected to alter characteristics of both variables, affecting water availability (WA) with societal and ecological consequences. Prior studies have examined trends in precipitation annual means, seasonal variations, and extremes, as well as corresponding changes in evaporation. However, few have assessed concomitant changes in both annual means and seasonal variation of precipitation and evaporation from a non-parametric standpoint, despite their combined importance for hydrology, agriculture, and ecosystems. Traditional global climate classifications seldom consider seasonal variation using non-parametric metrics. The study aims to analyze collective changes in annual mean and seasonal variation of precipitation and evaporation, classify global land into hydroclimatic regimes based on these properties, and assess implications for future WA using CMIP5 projections. The hypothesis is that changes in seasonal variability—particularly increases—in precipitation and evaporation, together with mean changes, will differentially affect WA across regimes, potentially making already variable regimes more variable and altering wet and dry season WA.

The paper synthesizes prior work showing: robust but uncertain CMIP5 projections of precipitation means and changes in spatial patterns (Knutti & Sedláček; Chadwick et al.); shifts in seasonal characteristics and dry days (Polade et al.; Kumar et al.); increases in precipitation extremes (Kharin et al.; Sillmann et al.); and biases/changes in evaporation and evaporative demand (Mueller & Seneviratne; Milly & Dunne; Liu & Sun). It notes that existing climate classifications (e.g., Köppen–Geiger, Thornthwaite, Holdridge) largely emphasize precipitation totals, temperature, and evapotranspiration but often omit non-parametric measures of seasonality, despite documented global shifts in rainfall seasonality and dry spells (Feng et al.; Pascale et al.). Information-theoretic metrics have been used for rainfall seasonality and water resources zoning, justifying adoption of apportionment entropy (AE) to capture non-Gaussian seasonal distributions beyond parametric statistics. The literature indicates that wet seasons may become wetter and dry seasons drier under warming, and that both moisture advection and evaporation processes shape seasonal responses, motivating a joint analysis of precipitation and evaporation means and seasonality to infer WA impacts.

Data and preprocessing: Observed precipitation is from GPCC (monthly, 2.5°×2.5°, 1901–2005). Observed terrestrial evaporation is from GLEAM (daily, 1980–2015), which uses Priestley–Taylor potential evaporation adjusted by a satellite-derived evaporative stress factor. CMIP5 projections from 21 GCMs are used for precipitation and evaporation for RCP 2.6, 4.5, and 8.5. All observed and modeled fields are interpolated to a 2.5°×2.5° common grid. Seasonality metric (Apportionment Entropy, AE): For each year k, monthly values xi (i=1..12) are summed to X=∑xi. AE is computed as AE=∑(xi/X)·log2(xi/X). AE ranges from 0 (all mass in one month) to log2(12) (evenly distributed). Higher AE indicates lower seasonal variation. AE is computed for precipitation (AE_P) and evaporation (AE_E). Annual totals are TOT_P and TOT_E. Global hydroclimatic regime classification: Using observed precipitation (1971–2000), each land grid cell is classified into one of nine regimes defined by the cross-classification of precipitation annual mean (low <30th percentile, moderate 30–70th, high >70th) and precipitation seasonality (via AE_P thresholds using the same percentile scheme). The nine regimes span combinations L/M/H of mean precipitation and L/M/H of AE_P (seasonality), capturing conditions from arid and highly seasonal to wet and perennial. Regimes are mapped globally and their spatial extent quantified. Bayesian Model Averaging (BMA): To weight CMIP5 models by historical performance, BMA is applied separately for TOT_P, TOT_E, AE_P, AE_E over each regime, comparing model annual characteristics with observations (GPCC for precipitation; GLEAM for evaporation). The BMA mixture PDF uses gamma/normal components; weights wi are estimated via maximum likelihood with MCMC (DREAM), maximizing L(w)=∑log(∑wi g(Yij)). The BMA ensemble mean and RMSD are computed: Ȳ=∑wiYi, RMSD=[∑wi(Yi−Ȳ)²]¹ᐟ². Performance is evaluated by Pearson R² of spatial patterns (TOT_P R²≈0.94; TOT_E≈0.92; AE_P≈0.91; AE_E≈0.85). Trend estimation: For 2005–2100, regime-aggregated linear trends in annual totals (TOT_P, TOT_E) and seasonality (AE_P, AE_E) are estimated using the non-parametric Theil–Sen estimator for each model, then combined with BMA: ΔY_BMA=∑wiΔYi, uncertainty via ΔY_RMSD and 95% CI approximated by ΔY_BMA ± 2·ΔY_RMSD. Available water (WA) analysis: WA is defined monthly as P−E. Historical (1971–2000) and future (2070–2099) monthly climatologies are computed for each regime using BMA-averaged model weights. Wet and dry seasons are identified as the three-month periods with maximum and minimum WA, respectively, based on the historical baseline. Seasonal changes (future minus historical) in precipitation, evaporation, and WA are computed for each RCP with BMA weighting and 95% CI via the above uncertainty method. Spatial robustness: Grid-wise trends (RCP 8.5) are also computed to assess geographic prevalence of increases/decreases in AE_P and AE_E, corroborating regime-aggregated findings.

- Nine global precipitation regimes based on annual mean precipitation and AE-derived seasonality capture distinct hydroclimatic conditions, from arid/highly seasonal (L_P L_AE) to wet/perennial (H_P H_AE). Monsoon-influenced regions tend to fall into high-precipitation but highly seasonal regimes (H_P L_AE and M_P L_AE), while moist forests and temperate regions occupy higher AE (less seasonal) regimes. - Annual means: BMA-weighted trends indicate increases in annual precipitation (TOT_P) and evaporation (TOT_E) in all nine regimes across RCP 2.6, 4.5, and 8.5. Magnitudes scale with radiative forcing (RCP 8.5 > 4.5 > 2.6). Regimes with higher AE (more even precipitation distribution) generally exhibit larger increases in both TOT_P and TOT_E than highly seasonal regimes. • Example: In H_P H_AE, TOT_P increases by about 1.3 mm/year (RCP 8.5), 0.7 mm/year (RCP 4.5), and 0.25 mm/year (RCP 2.6), with larger model uncertainty than other regimes. • Example: In M_P H_AE, TOT_E increases by about 0.68 mm/year (RCP 8.5), 0.40 (RCP 4.5), and 0.19 (RCP 2.6), with generally smaller uncertainty than precipitation. - Seasonality (AE): Precipitation seasonality increases (AE_P decreases) in several regimes—particularly those already characterized by high seasonal variability (L_P L_AE, M_P L_AE, H_P L_AE). In contrast, regimes with high AE (LP_HAE, MP_HAE, HP_HAE) show negligible AE_P trends under RCP 2.6 but increased variability under higher RCPs. Overall pattern: “seasonally variable regimes become more variable” for precipitation. - Evaporation seasonality (AE_E): Changes are weaker and often opposite in sign to precipitation. Many regimes show decreased evaporation variability (AE_E increases), with no significant AE_E trend in some high-precipitation regimes (HP_HAE, HP_MAE, HP_LAE, MP_LAE). Higher-forcing scenarios produce larger AE_E changes where present. - Spatial robustness (RCP 8.5 grid-wise): Precipitation variability (AE_P decrease) increases over ~35.6% of land versus decreases over ~4%. Evaporation variability decreases over ~36% of land versus increases over ~6%. - Monthly WA (P−E): Wet seasons generally become wetter, especially in high-AE regimes (LP_HAE, MP_HAE, HP_HAE), with statistically significant increases in WA during wet-season months under higher RCPs. Low-AE regimes with inconsistent water supply (e.g., L_P L_AE) show smaller or negligible changes in monthly WA. In some regions (e.g., parts of Europe and northern North America), dry-season P−E decreases under RCP 8.5, consistent with prior studies. - Risk implications: Increased wet-season WA in less-seasonal regimes implies heightened flood risk potential in regions such as Western Europe, North America, and Southeast Asia. Conversely, increased precipitation seasonality in already variable regimes may exacerbate water management challenges (flood storage during short wet periods and shortages during extended dry periods).

The integrated analysis of means and seasonality reveals that climate change drives a widespread intensification of the hydrologic cycle, with both precipitation and evaporation annual totals increasing across all regimes. Crucially, precipitation seasonality is projected to intensify most where it is already high, reinforcing intra-annual variability in monsoon and strongly seasonal climates. This supports the broader narrative of a widening seasonal contrast—wetter wet seasons and, in many places, drier dry seasons—mediated by changes in vertical moisture advection and surface evaporation. In contrast, evaporation seasonality generally weakens, suggesting a smoothing of evaporative demand distribution through the year in many regions. These combined responses shape water availability: high-AE (less-seasonal) regimes experience clear increases in wet-season WA because precipitation increases outpace evaporation increases during the wet season. Low-AE regimes show muted WA changes due to strong water limitations and competing P and E responses. The spatial and seasonal heterogeneity underscores that the impacts on flood risk, reservoir operations, agriculture, and ecosystems will be regime-dependent. The framework links projected hydroclimatic changes to potential increases in flood risk in regions with historically consistent water supply, while highlighting persistent challenges in highly seasonal regimes where variability intensifies.

This study introduces a non-parametric, regime-based framework using apportionment entropy and precipitation means to classify global land into nine hydroclimatic regimes, then applies BMA-weighted CMIP5 projections to assess future changes in precipitation and evaporation means and seasonality. Key contributions include: (1) evidence that both annual precipitation and evaporation increase in all regimes, scaling with radiative forcing; (2) identification of a robust pattern of increased precipitation seasonality primarily in regimes already exhibiting high variability (“variable regimes become more variable”); (3) contrasting, generally reduced variability in evaporation; and (4) a clear increase in wet-season water availability in high-AE regimes, implying elevated flood potential. Future research should extend this framework to streamflow regimes and hydrologic extremes, integrate higher-resolution and next-generation climate models, assess adaptation implications for reservoir and agricultural management by regime, and explore regional process drivers (e.g., moisture transport, land–atmosphere coupling) underlying regime-specific responses.

- Model uncertainties: CMIP5 models exhibit substantial spread, particularly for dry-season precipitation and WA changes; this limits confidence in some seasonal conclusions. BMA weighting mitigates but does not eliminate these uncertainties. - Extremes representation: Current coupled climate models have limitations in simulating key drivers of persistent weather extremes and quasi-resonant amplification, constraining inference about hydrologic extremes. - Metric aggregation: AE and annual/seasonal aggregations may obscure sub-seasonal variability and event-scale extremes. Regime aggregation may mask important intra-regime spatial heterogeneity. - Evaporation data constraints: Observational evaporation (GLEAM) involves model assumptions (e.g., Priestley–Taylor, satellite-derived stress factors) and limited temporal coverage (1980–2015) relative to precipitation. - Definition choices: Regime thresholds (30th/70th percentiles) and the 3-month wet/dry season definitions, while justified, are somewhat arbitrary and may affect classification and inferred changes.