Goutam KonapalapersName, Ashok K. MishrapersName, Yoshihide WadapersName, Michael E. MannpersName (2020date). “Climate change will affect global water availability through compounding changes in seasonal precipitation and evaporation.” Nature Communicationstitle·j.
Object of study — joint projected changes in mean and seasonal precipitationterm·key and evaporationterm·key across nine global land regimes
Research question — how concomitant changes in annual means and seasonal variation of precipitationterm·key and evaporationterm·key affect water availabilityterm·dom
Contribution — a regime framework showing seasonally variable regimes becoming more variable under climate changeterm·key
phenomenon · claim-type · cross-domain-transfer · mechanism · claim-relation · replication-status · limitation — grounded in the full text| phenomenon | seasonally variable precipitation regimes becoming more variable — “Increases in projected seasonal precipitation variation in already highly variable precipitation regimes gives rise to a pattern of “seasonally variable regimes becoming more variable”.” | |
| claim-type | existence — “Increases in projected seasonal precipitation variation in already highly variable precipitation regimes gives rise to a pattern of “seasonally variable regimes becoming more variable”.” | |
| method — evidence | — “Here, we employ a non-parametric analysis framework to analyze seasonal hydroclimatic regimes by classifying global land regions into nine regimes using late 20th century precipitation means and seasonality.” | |
| object of study — evidence | — “Here, we employ a non-parametric analysis framework to analyze seasonal hydroclimatic regimes by classifying global land regions into nine regimes using late 20th century precipitation means and seasonality.” | |
| research question — evidence | — “Though the combined monthly distribution of precipitation and evaporation have widespread implications for regional hydrology 15 , 16 , crop yield 17 , 18 , and ecology 19 , 20 , few studies have examined the concomitant changes in both annual mean and seasonal variation in these variables.” | |
| contribution — evidence | — “Increases in projected seasonal precipitation variation in already highly variable precipitation regimes gives rise to a pattern of “seasonally variable regimes becoming more variable”.” | |
| mechanism | moisture advection The wetter-wet-season pattern may owe to regional changes in moisture advection and evaporation. — “Therefore, we anticipate that the observed pattern may be owing to regional changes in moisture advection and evaporation.” | mechanisms[0] |
| sense | seasonal variation - read inversely from apportionment entropy, lower AE meaning more seasonal — “In our case, higher AE values imply lower seasonal variation, and lower AE imply higher seasonal variation” | senses[0] |
| relation | WA —same-as→ water availability — ( ) — “may result in changes in water availability (WA) that have implications for both humans and the biosphere” — “changes in water availability (WA)” | concept_edges[0] |
| weight | WA —same-as→ water availability — 0.95 — “may result in changes in water availability (WA) that have implications for both humans and the biosphere” | concept_edges[0] |
| relation | GPCC —same-as→ Global Precipitation Climatology Centre — ( ) — “using observed monthly gridded precipitation data from the Global Precipitation Climatology Centre (GPCC)” — “Global Precipitation Climatology Centre (GPCC)” | concept_edges[1] |
| weight | GPCC —same-as→ Global Precipitation Climatology Centre — 0.95 — “using observed monthly gridded precipitation data from the Global Precipitation Climatology Centre (GPCC)” | concept_edges[1] |
| relation | AE —same-as→ apportionment entropy — ( ) — “For quantifying seasonality, we used apportionment entropy (AE), which provides a descriptive non-parametric measure determining the seasonal variation for data, such as precipitation, that are not Gaussian distributed” — “we used apportionment entropy (AE)” | concept_edges[2] |
| weight | AE —same-as→ apportionment entropy — 0.95 — “For quantifying seasonality, we used apportionment entropy (AE), which provides a descriptive non-parametric measure determining the seasonal variation for data, such as precipitation, that are not Gaussian distributed” | concept_edges[2] |
| relation | seasonality —operationalised-as→ apportionment entropy — for quantifying — “For quantifying seasonality, we used apportionment entropy (AE), which provides a descriptive non-parametric measure determining the seasonal variation for data, such as precipitation, that are not Gaussian distributed” — “For quantifying seasonality, we used apportionment entropy” | concept_edges[3] |
| weight | seasonality —operationalised-as→ apportionment entropy — 0.9 — “For quantifying seasonality, we used apportionment entropy (AE), which provides a descriptive non-parametric measure determining the seasonal variation for data, such as precipitation, that are not Gaussian distributed” | concept_edges[3] |
| relation | WA —operationalised-as→ precipitation minus evaporation — as measured by — “We also examined the projected monthly distribution of WA as measured by precipitation minus evaporation (P–E) across the classified nine precipitation regimes.” — “WA as measured by precipitation minus evaporation” | concept_edges[4] |
| weight | WA —operationalised-as→ precipitation minus evaporation — 0.9 — “We also examined the projected monthly distribution of WA as measured by precipitation minus evaporation (P–E) across the classified nine precipitation regimes.” | concept_edges[4] |
| effect direction | increase in temperature —causes→ annual mean evaporation — increases — “Increases in annual mean evaporation over the land surface can be attributed to the increase in temperature in the CMIP5 future projections 42 and also to increase in precipitation.” — “Increases in annual mean evaporation over the land surface can be attributed to the increase in temperature” | concept_edges[5] |
| note | increase in temperature —causes→ annual mean evaporation — Hedged ('can be attributed'), citing ref. 42; an interpretation of model projections. — “Increases in annual mean evaporation over the land surface can be attributed to the increase in temperature in the CMIP5 future projections 42 and also to increase in precipitation.” | concept_edges[5] |
| relation | increase in temperature —causes→ annual mean evaporation — attributed to — “Increases in annual mean evaporation over the land surface can be attributed to the increase in temperature in the CMIP5 future projections 42 and also to increase in precipitation.” — “Increases in annual mean evaporation over the land surface can be attributed to the increase in temperature” | concept_edges[5] |
| weight | increase in temperature —causes→ annual mean evaporation — 0.7 — “Increases in annual mean evaporation over the land surface can be attributed to the increase in temperature in the CMIP5 future projections 42 and also to increase in precipitation.” | concept_edges[5] |
| replication | original | |
| limitation | Implications for hydrologic extremes are unclear because climate models miss key drivers of persistent extremes. — “it is unclear what implications our findings might hold for hydrologic extremes, owing to limitations in the ability of current generation coupled climate models to capture the key drivers of persistent weather extremes” | limitations[0] |
finding · quantity · variable · definition · key-term — grounded in the full text| finding | Regimes with low precipitation seasonality get more wet-season precipitation. — “regimes become more variable in the future. Regimes with low seasonality in precipitation, instead, experience increased wet season precipitation.” | content_tags.findings[0] |
| finding | Regions with already inconsistent supply may see even more inconsistent water availability. — “This would imply that regions that already have inconsistent water supply owing to high seasonal variation in precipitation might experience even more inconsistent WA.” | content_tags.findings[1] |
| quantity | 4 — — “Out of 9 regimes, 4 show increased precipitation variation, while 5 show decreased evaporation variation coupled with increasing mean precipitation and evaporation.” | content_tags.quantities[0] |
| variable | apportionment entropy — “For quantifying seasonality, we used apportionment entropy (AE), which provides a descriptive non-parametric measure determining the seasonal variation for data, such as precipitation, that are not Gaussian distributed” | content_tags.variables[0] |
| definition | apportionment entropy — “For quantifying seasonality, we used apportionment entropy (AE), which provides a descriptive non-parametric measure determining the seasonal variation for data, such as precipitation, that are not Gaussian distributed” | content_tags.definitions[0] |
| key-terms | water availability, precipitation, evaporation, seasonality, apportionment entropy, CMIP5, hydroclimatic regimes, Bayesian model averaging, climate change, wet season |
| concept | WA — water availability — “may result in changes in water availability (WA) that have implications for both humans and the biosphere” | concepts[0] |
| concept | water availability — not defined in this paper's own words; the water left for use once evaporation is subtracted from precipitation — “may result in changes in water availability (WA) that have implications for both humans and the biosphere” | concepts[1] |
| concept | GPCC — Global Precipitation Climatology Centre — “using observed monthly gridded precipitation data from the Global Precipitation Climatology Centre (GPCC)” | concepts[2] |
| concept | Global Precipitation Climatology Centre — not defined in this paper's own words; the source of the observed gridded precipitation data — “using observed monthly gridded precipitation data from the Global Precipitation Climatology Centre (GPCC)” | concepts[3] |
| concept | AE — apportionment entropy — “For quantifying seasonality, we used apportionment entropy (AE), which provides a descriptive non-parametric measure determining the seasonal variation for data, such as precipitation, that are not Gaussian distributed” | concepts[4] |
| concept | apportionment entropy — a descriptive non-parametric measure determining the seasonal variation for data, such as precipitation, that are not Gaussian distributed — “For quantifying seasonality, we used apportionment entropy (AE), which provides a descriptive non-parametric measure determining the seasonal variation for data, such as precipitation, that are not Gaussian distributed” | concepts[5] |
| concept | seasonality — not defined in this paper's own words; how unevenly precipitation is spread over the year — “For quantifying seasonality, we used apportionment entropy (AE), which provides a descriptive non-parametric measure determining the seasonal variation for data, such as precipitation, that are not Gaussian distributed” | concepts[6] |
| concept | precipitation minus evaporation — not defined in this paper's own words; the net water balance P-E — “We also examined the projected monthly distribution of WA as measured by precipitation minus evaporation (P–E) across the classified nine precipitation regimes.” | concepts[7] |
| concept | increase in temperature — not defined in this paper's own words; projected warming — “Increases in annual mean evaporation over the land surface can be attributed to the increase in temperature in the CMIP5 future projections 42 and also to increase in precipitation.” | concepts[8] |
| concept | annual mean evaporation — not defined in this paper's own words; yearly average evaporation over land — “Increases in annual mean evaporation over the land surface can be attributed to the increase in temperature in the CMIP5 future projections 42 and also to increase in precipitation.” | concepts[9] |
Full text — tagged
Climate changeterm·key will affect global water availabilityterm·dom through compounding changes in seasonal precipitationterm·key and evaporationterm·key Abstract Both seasonal and annual mean precipitationterm·key and evaporationterm·key influence patterns of water availabilityterm·dom impacting society and ecosystems. Existing global climate studies rarely consider such patterns from non-parametric statistical standpoint. Here, we employ a non-parametric analysis framework to analyze seasonal hydroclimatic regimesterm·key by classifying global land regions into nine regimes using late 20th century precipitationterm·key means and seasonalityterm·dom. These regimes are used to assess implications for water availabilityterm·dom due to concomitant changes in mean and seasonal precipitationterm·key and evaporationterm·key changes using CMIP5term·key model future climate projections. Out of 9 regimes, 4 show increased precipitationterm·key variation, while 5 show decreased evaporationterm·key variation coupled with increasing mean precipitationterm·key and evaporationterm·key. Increases in projected seasonal precipitationterm·key variation in already highly variable precipitationterm·key regimes gives rise to a pattern of “seasonally variable regimes becoming more variable”. Regimes with low seasonalityterm·dom in precipitationterm·key, instead, experience increased wet seasonterm·key precipitationterm·key.
Subject terms: Climate sciences, Climate-change impacts, Projection and prediction, Hydrology, Water resources Introduction Accessibility of water resources for human consumption and ecosystems largely depends on the spatio-temporal distribution of both precipitationterm·key and evaporationterm·key 1 – 3 . As a result, changes in characteristics of precipitationterm·key and evaporationterm·key due to human-caused climate changeterm·key in the 21st century may result in changes in water availabilityterm·dom (WAterm·dom) that have implications for both humans and the biosphere 4 , 5 . Previous studies have elucidated trends in precipitationterm·key in terms of both annual mean 6 , 7 , seasonal variation 8 , 9 , and the distribution of extreme events 10 , 11 . Studies have also examined the corresponding changes in evaporationterm·key characteristics 12 – 14 . Though the combined monthly distribution of precipitationterm·key and evaporationterm·key have widespread implications for regional hydrology 15 , 16 , crop yield 17 , 18 , and ecology 19 , 20 , few studies have examined the concomitant changes in both annual mean and seasonal variation in these variables. Moreover, the existing global climate classifications 21 – 23 that form the basis for WAterm·dom studies rarely consider seasonal variation characteristics from a non-parametric standpoint, even though they vary in a complex manner across global land regions 24 , 25 .
An analysis of the collective changes in both hydrological annual means and seasonal variations can better inform assessments of societal and ecological vulnerability with respect to potential future WAterm·dom. For instance, an increase in seasonal variability of precipitationterm·key might possibly disrupt the continuous atmospheric water supply, leading to extended dry periods in regions of unimodal precipitationterm·key distribution 26 , 27 . In regimes with high precipitationterm·key, this redistribution may result in more water concentrated over relatively short periods of time, leading to floods and operational difficulties in reservoir water management. Similarly, an increase in seasonal variation in evaporationterm·key might lead to changes in the monthly terrestrial water budget depending on the water supply regime 28 , 29 . It is thus important to understand the combined role of projected future changes in both mean precipitationterm·key and evaporationterm·key and their seasonal variations in assessing impacts on any particular region’s water supply regime.
Here, we examine spatially aggregated future projections for nine distinct regimes, characterizing joint changes in annual mean and seasonal precipitationterm·key and evaporationterm·key. Each regime displays concomitant increases in annual mean precipitationterm·key and evaporationterm·key. However, only four of the nine regimes display increased seasonal variation in precipitationterm·key. We observe a tendency for increased seasonal variation for regimes that already exhibit high seasonal variability, establishing a pattern wherein seasonally variable regimes become more variable in the future. Regimes with low seasonalityterm·dom in precipitationterm·key, instead, experience increased wet seasonterm·key precipitationterm·key.
Results Global classification of precipitationterm·key regimes In this study, we first classify the global land regions into distinct hydroclimatic regimesterm·key based on annual means and seasonal variations using observed monthly gridded precipitationterm·key data from the Global Precipitation Climatology Centreterm·dom (GPCCterm·dom) (See “Methods”) 30 . For quantifying seasonalityterm·dom, we used apportionment entropyterm·mth (AEterm·mth), which provides a descriptive non-parametric measure determining the seasonal variation for data, such as precipitationterm·key, that are not Gaussian distributed, as it captures higher-order statistics unlike parametric methods that characterize the data in terms of, e.g., coefficient of variation and standard deviation 3 . In our case, higher AEterm·mth values imply lower seasonal variation, and lower AEterm·mth imply higher seasonal variation (see “Methods” for further details).
To understand the regime classification framework qualitatively, we plotted various characteristics of the resulting regimes in Fig. 1 . The spatial distribution of regimes based on precipitationterm·key means and seasonal variations is shown in Fig. 1a . The percentage of land occupied by each classified regime is shown in Fig. 1b . The plot region in the legend is divided into nine zones, each of which is delineated with two intersecting dividing lines that pass through the limits of the respective thresholds of the two variables. Global land regions are classified into nine regimes based on percentiles thresholds (i.e., <30th—Low; 30–70th—Moderate; >70th—High) of seasonal variation (as defined by AEterm·mth) and annual mean precipitationterm·key during the 1971–2000 reference period (See “Methods”). We chose four critical zones (H p H AEterm·mth , L P H AEterm·mth , L P L AEterm·mth , H P L AEterm·mth ) to illustrate extreme scenarios based on combinations of either low (<Pr 30 ) and high (>Pr 70 ) variation and mean value of precipitationterm·key. Here, H AEterm·mth represents higher AEterm·mth values, which implies lower seasonal variation. The spatially aggregate precipitationterm·key climatology of the selected four regimes is detailed in Fig. 1c . For each grid, the month with lowest rainfall is plotted as starting month for rainfall. Regime H p H AEterm·mth witnesses a higher precipitationterm·key rate and lower intra-annual variability (i.e., high AEterm·mth). As a result, the precipitationterm·key is uniformly distributed, indicating perennial water supply for both human and ecological needs 3 , 31 . On the contrary, regime H P L AEterm·mth presents a scenario (high precipitationterm·key and high variability), where most of the precipitationterm·key is concentrated in a limited number of months. As a result, excess water needs to be stored to prevent floods as well as to enhance water supply for stakeholders in dry months. In low-precipitation regimes (L P H AEterm·mth and L P L AEterm·mth ), virtual water transfers and drought-tolerant crops are prevalent 32 . However, regions with less precipitationterm·key and higher AEterm·mth (L P H AEterm·mth ) are perennial in nature. Regions characterized by a combination of lower precipitationterm·key and AEterm·mth (L P L AEterm·mth ) are arid in nature, indicating water resources availability is extremely low. Therefore, based on these principles, the remaining regimes have precipitationterm·key conditions between these extreme limiting cases.
Classification of the precipitationterm·key regimes. a Spatial distribution of precipitationterm·key regimes based on the percentile (Pr i ) thresholds concept using mean apportionment entropyterm·mth (AEterm·mth) and annual precipitationterm·key P during the 1971–2000 reference period. b Percentage of land occupied by each regime. c Precipitationterm·key climatology of the spatially aggregated monthly rainfall climatology for regimes H P L AEterm·mth , H P H AEterm·mth , L P H AEterm·mth , L P L AEterm·mth . These regimes are selected as they represent boundary case scenarios. H, M, and L represent high moderate and low values, respectively.
The spatial distribution of regimes across global land is found to be as follows: regime H P L AEterm·mth and M P L AEterm·mth can be found mostly in the Indian subcontinent, Northeast Asia, Northern Australia, and much of north-central and south-central Africa, covering land area of ~15%. These regions are influenced mostly by monsoons 33 . Regime H P M AEterm·mth and H P H AEterm·mth mostly occupy eastern North America, Northern South America, central Africa, Western Europe, and South Eastern Asia, occupying a combined land area of 24%. These regions are mostly moist forest areas. Regime M P M AEterm·mth spans across most of central North America and is scattered in western parts of South America and Northern and central Asia, Southern part of Africa and Southern Australia amounting to ~15% of land area. Regime M P H AEterm·mth occupies most of European Continent and further extends to western Russia occupying 16% of the total land area. Regime L P L AEterm·mth occupies Northern Africa, the Middle East, and further extends to central Asia. In addition, central Australia, the interior western United States and central western South America are classified as belonging to L P L AEterm·mth, occupying a land area of 15%. L P M AEterm·mth is mostly confined to the Northern parts of North America and Siberia region occupying ~13% of land area. Finally, regime L P H AEterm·mth is confined to a small region in the central Russian region.
Trends in annual changes of precipitationterm·key and evaporationterm·key Precipitationterm·key and evaporationterm·key projections from Coupled Model Intercomparison Project Phase 5 (CMIP5term·key) models are aggregated over these nine precipitationterm·key regimes. We evaluate linear trends over the 21st century (2005–2100) using Bayesian model averagingterm·key (BMA) (31 – 32, see “Methods”) applied to both annual means and seasonal variation in precipitationterm·key and evaporationterm·key for three future scenarios (representative concentration pathway (RCP) 2.6, 4.5, and 8.5).
The changes in BMA-weighted future annual precipitationterm·key and evaporationterm·key totals and seasonal variability for the RCP scenarios 2.6, 4.5, and 8.6 are shown in Fig. 2a, b , respectively (note that the horizontal axis scales are different for both panels in Fig. 2a, b ). In Fig. 2a , the change in annual precipitationterm·key total (TOT P ) indicates an increase in precipitationterm·key in all regimes, with the least increase in regime L P L AEterm·mth , In addition, a proportional relationship wasterm·dom observed in terms of an increase in the precipitationterm·key magnitude with an increase in the emission forcing. In addition, in precipitationterm·key regimes L P H AEterm·mth , M P H AEterm·mth , and H P H AEterm·mth that are characterized by a relatively consistent water supply, the TOT P exhibit a higher magnitude of precipitationterm·key increase compared with those regimes characterized by a moderate and high seasonal variability. Conversely, regions L P L AEterm·mth , M P L AEterm·mth , and H P L AEterm·mth with inconsistent water supplies exhibit a lower magnitude of precipitationterm·key increase. The highest magnitude increase is evident in regime H P H AEterm·mth , which is characterized by high volumes of consistent water supply in which RCP 8.5 exhibits an increase of 1.3 mm/year, followed by RCP 4.5 exhibiting 0.7 mm/year and RCP 2.6 exhibiting 0.25 mm/year. In addition, a larger uncertainty in the case of GCM model use is evident in precipitationterm·key regime H P H AEterm·mth . Similar to TOT P , an increase in the annual evaporationterm·key total (TOT E ) is observed in all regimes with the lowest increase observed in regime L P L AEterm·mth . In addition, a direct proportional relationship is also evident here; the higher emission scenarios exhibit a higher magnitude of increased evaporationterm·key. As in the case of TOT P , TOT E in the precipitationterm·key regimes of L P H AEterm·mth , M P H AEterm·mth , and H P H AEterm·mth that are characterized by relatively consistent water supply exhibit a higher magnitude of evaporationterm·key increase. The greatest change is evident in regime M P H AEterm·mth , which is characterized by moderate precipitationterm·key but low seasonal variability. In this regime, RCP 8.5 exhibits an increase of 0.68 mm/year, followed by RCP 4.5 with an increase of 0.4 mm/year and RCP 2.6 with an increase of 0.19 mm/year. We observe that the changes in magnitude of TOT E are generally less uncertain than those in precipitationterm·key.
Trends in precipitationterm·key and evaporationterm·key totals and seasonalityterm·dom. Theil-sen slopes of a total annual precipitationterm·key (TOT P ) and total annual evaporationterm·key (TOT E ) magnitudes, and b precipitationterm·key apportionment entropyterm·mth (AEterm·mth P ) and evaporationterm·key apportionment entropyterm·mth AEterm·mth E . The error bars surrounding the point estimates represent the 95% confidence interval of the CMIP5term·key multimodel ensemble. H, M, and L represent high moderate and low values, respectively.
The decrease in AEterm·mth P indicates an increase in the seasonal variability of precipitationterm·key in the regimes L P L AEterm·mth , M P L AEterm·mth , and H P L AEterm·mth for all RCP scenarios as detailed in Fig. 2b . Although a negligible trend in RCP 2.6 is observed in case of regimes H P M AEterm·mth , H P H AEterm·mth , and M P M AEterm·mth , a significant increase in variability wasterm·dom observed in the other scenarios for these three regimes. However, although a negligible positive trend is observed in regimes L P M AEterm·mth and L P H AEterm·mth for RCP 2.6, the other scenarios exhibit a decrease in variability. Regime M P H AEterm·mth , which is characterized by moderate and consistent water supply exhibits a negligible change in all scenarios. Further, the changes in AEterm·mth E are not as substantial as those in AEterm·mth P . No significant trends are observed in AEterm·mth E for regimes H P H AEterm·mth , H P M AEterm·mth , H P L AEterm·mth , and M P L AEterm·mth . In addition, decrease in variability of evaporationterm·key is evident in regime L P H AEterm·mth , L P M AEterm·mth , M P H AEterm·mth , and M P M AEterm·mth . The higher emission scenarios display greater change in magnitude for AEterm·mth E . Unlike with precipitationterm·key variability, there is a pattern of decreased variability exhibited by AEterm·mth E . These contrasting future projection changes in precipitationterm·key and evaporationterm·key may result in spatially variable monthly WAterm·dom (i.e., P–E) 6 , 31 , 34 . To further assess the robustness of our results, we also computed the grid-wise trends as shown in Supplementary Figure 1 for RCP 8.5 scenario. Based on this spatially explicit analysis, we see that precipitationterm·key variability is increasing over a substantially greater region (~35.6% of the land surface) than it is decreasing (~4% of the land surface). In addition, the spatial analysis indicates that this increase in variability in more prominent in regions with high variability. Evaporationterm·key variability, conversely, is decreasing over a substantially greater region (~36% of the land surface) than it is increasing over (~6% of the land surface). Therefore, both these analyses suggested an overall trend toward increasing precipitationterm·key variability along with decreasing evaporationterm·key variability, reinforcing our main conclusions. As these metrics aggregate the monthly distributions of precipitationterm·key and evaporationterm·key at the annual scale obscuring seasonal behavior, it is worthwhile to additionally investigate, as described below, how these changes are reflected in the monthly distribution of available water (determined as the difference between precipitationterm·key and evaporationterm·key—See “Methods”).
Precipitationterm·key and evaporationterm·key role in available water change We determine which seasonal components have contributed to altering the monthly distribution of WAterm·dom (precipitationterm·key–evaporationterm·key) for each of these scenarios (See “Methods”). The BMA-weighted historical (1971–2000) monthly WAterm·dom distribution and future projections (2070–2099) for each regime are shown in Fig. 3 for each of the scenarios. The projected future monthly distribution of WAterm·dom for the wet seasonsterm·key (8th–10th month) demonstrates that the wet seasonterm·key becomes wetter, a pattern that is increasingly evident in regimes L P H AEterm·mth , M P H AEterm·mth , and H P H AEterm·mth with a consistent water supply albeit less evident in regimes L P L AEterm·mth and M P L AEterm·mth with an inconsistent water supply (regime H P L AEterm·mth exhibits a minor increase in available water). So, while seasonally variable regimes becoming more variable in terms of precipitationterm·key, that is not the case for WAterm·dom, perhaps owing to competing effects between evaporationterm·key and precipitationterm·key. The low precipitationterm·key of regime L P L AEterm·mth exhibits the least noticeable change in monthly available water, unlike the high precipitationterm·key of regime H P L AEterm·mth , wherein a more noticeable increase in available water is evident during the wet seasonterm·key. This pattern is also sensitive to the change in radiative forcing, with greater radiative forcing corresponding to greater changes. Therefore, in regions where water supply is constrained by low-precipitation amounts and a more uneven distribution (low AEterm·mth), changes in precipitationterm·key and evaporationterm·key do not affect the available water as they are more water-limited in nature. In regions with a more even distribution, however, there is a greater effect. Overall, these results indicate that the changes in monthly WAterm·dom distribution are dependent on both the particular regime and the radiative forcing.
Future monthly water availabilityterm·dom. a BMA ensemble monthly projected (RCP 2.6, RCP 4.5, and RCP 8.5) of available water (2070–2099) and historical (1971–2000) scenarios. Where, H, M, and L abbreviate high moderate and low values, respectively. Whereas, apportionment entropyterm·mth is abbreviated as AEterm·mth and annual precipitationterm·key as P.
The corresponding changes in precipitationterm·key and evaporationterm·key (relative to the historical period) during the wet and dry seasons for three future emission scenarios (RCP 2.6, 4.5, 8.5) are presented in Fig. 4 , illustrating the competition between water and energy balance at the seasonal scale. The role played by both wet and dry seasons in changing WAterm·dom are examined in Fig. 4 . In regions with high AEterm·mth values (i.e., L P H AEterm·mth , M P H AEterm·mth , and H P H AEterm·mth ), higher increases in evaporationterm·key can be observed in comparison with the moderate and low AEterm·mth regions, especially in the RCP 8.5 scenario. These changes are significant in the wet seasonterm·key but not the dry season. As noted earlier, higher emission scenarios show more pronounced evaporationterm·key increases. In the case of precipitationterm·key, regimes with high AEterm·mth values in both dry and wet seasonsterm·key display a pattern of increase similar to that in evaporationterm·key. But these changes are again only statistically significant in the wet seasonterm·key. Although greater changes in precipitationterm·key are evident in the dry season, the greater spread among models in that case hinders any confident conclusions.
Projected seasonal changes in precipitationterm·key, evaporationterm·key, and available water. Relative changes in future precipitationterm·key and evaporationterm·key during the wet (three-month average with maximum precipitationterm·key in a year) and dry seasons (three-month average with minimum precipitationterm·key in a year) derived based on the CMIP5term·key scenario relative to the historical period (1971–2000). The error bars surrounding the point estimates represent the 95% confidence interval of the CMIP5term·key multimodel ensemble. H, M, and L abbreviate high moderate and low values, respectively. Whereas, apportionment entropyterm·mth is abbreviated as AEterm·mth and annual precipitationterm·key as P.
The increasing wet and dry season precipitationterm·key and evaporationterm·key provide an explanation for the changes in available water as defined by the difference between precipitationterm·key and evaporationterm·key (precipitationterm·key–evaporationterm·key). In regimes where the relative magnitude of the increase in evaporationterm·key is less than that in precipitationterm·key, a significant increase in WAterm·dom is observed. Such is the case with high AEterm·mth regions in wet seasonterm·key. This increase can thus be attributed to a larger increase in precipitationterm·key than in evaporationterm·key. In addition, regions with higher AEterm·mth exhibit a greater increase in WAterm·dom, which imply the potential for increased flood risk in regions such as Western Europe 35 , North America 36 , and South Eastern Asia that show less-seasonal variation 16 . In the case of the dry season, an increase in WAterm·dom is found in RCP 8.5 scenarios for regimes L P M AEterm·mth , M P M AEterm·mth , and H P H AEterm·mth but it is not statistically significant. The results in these cases are inconclusive owing to the large spread among models. This finding further highlights the interdependent role of evaporationterm·key and precipitationterm·key in changing seasonal WAterm·dom. Previous studies based on spatially explicit analyses have shown that the dry seasons are becoming drier in some locations around the globe 9 . Our spatial analysis in case of RCP 8.5 scenario also indicates a decrease in P–E in Europe and northern part of North America (Supplementary Fig. 2 ) in agreement with Kumar et al. 10 , 37 .
Discussion Coupled changes in seasonal variation (as measured by AEterm·mth) and annual mean precipitationterm·key and evaporationterm·key significantly impact the spatio-temporal distribution of WAterm·dom. By using appropriate thresholds applied to seasonal variations and annual means of historical precipitationterm·key data, nine land regions were identified that are characterized by different water supply regimes.
Spatially aggregated future trends over these regimes as applied to the CMIP5term·key future projections indicate an increase in precipitationterm·key annual means in all land regimes across the globe, with higher increases in RCP 8.5 as reported in previous studies 6 – 8 , 38 . Most strikingly, annual mean evaporationterm·dom is found to be increasing in all regimes, indicating a pattern of intensified response to increased water demand across all regions. Furthermore, an increase in seasonal precipitationterm·key variation is observed, especially in regions that already exhibit greater seasonal variation in precipitationterm·key. This indicates a pattern of seasonally variable regimes becoming more variable regarding precipitationterm·key. This pattern is consistent with previously reported observations of the wet seasonterm·key becoming more wet and the dry season becoming more dry 39 , 40 . In the case of wet seasonsterm·key becoming wet, vertical moisture advection and evaporationterm·key play a key role 40 , 41 . Therefore, we anticipate that the observed pattern may be owing to regional changes in moisture advection and evaporationterm·key. However, we do not observe such a pattern in the seasonal variability of evaporationterm·key. We found a decrease in seasonal evaporationterm·key variation in various regimes of the global land area. This would imply that regions that already have inconsistent water supply owing to high seasonal variation in precipitationterm·key might experience even more inconsistent WAterm·dom.
Similar to what is observed with precipitationterm·key, an increase in the annual mean evaporationterm·dom (TOT E ) is observed in all regimes with the smallest increase observed in regime L P L AEterm·mth , as found in previous studies 41 . Increases in annual mean evaporationterm·dom over the land surface can be attributed to the increase in temperatureterm·dom in the CMIP5term·key future projections 42 and also to increase in precipitationterm·key. The seasonal changes in evaporationterm·key, however, are tied to corresponding seasonal changes in surface relative humidity contrast 42 , 43 . Also, the increase in TOT E is smaller than in TOT P , a finding that is consistent with previous studies 31 , 42 . We also examined the projected monthly distribution of WAterm·dom as measured by precipitation minus evaporationterm·mth (P–E) across the classified nine precipitationterm·key regimes. Our results highlight a clear signal of increased WAterm·dom in the wet seasonterm·key especially in regimes of less-seasonal variation in precipitationterm·key.
We assessed the role of precipitationterm·key and evaporationterm·key characteristics in the changing monthly distributions of WAterm·dom. Our results indicate that the increases in WAterm·dom in wet seasonsterm·key is controlled by changes in both precipitationterm·key and evaporationterm·key. The combination of changes in precipitationterm·key and evaporationterm·key might result in an overall increase in WAterm·dom, which is more pronounced during the wet seasonterm·key and is expected to yield spatially variable annual WAterm·dom consistent with previous studies 37 , 44 , 45 . However, we show that these changes may also be dependent upon the specific regime as determined by both seasonal variation and mean precipitationterm·key changes. Even though our results indicate changes in long-term WAterm·dom, it is unclear what implications our findings might hold for hydrologic extremes, owing to limitations in the ability of current generation coupled climate models to capture the key drivers of persistent weather extremes 46 , 47 .
Overall, these changes in precipitationterm·key characteristics impact not only annual WAterm·dom but also its spatio-temporal distribution. Concomitant changes in the mean and seasonal variation in precipitationterm·key may imply significant and varied shifts in phenology 48 , 49 , reservoir management 3 , and ecosystem function depending upon the water supply regimes. The framework provided by our study complements traditional approaches used to study seasonal variability in different hydroclimatic regimesterm·key 50 , 51 . Future efforts will use this framework to quantitatively assess the implications of projected changes in seasonal and annual mean precipitationterm·key on streamflow regimes, providing further relevance to issues involving societal water use and demand.
Data Observed precipitationterm·key gridded data from GPCCterm·dom 52 at a resolution of 2.5 × 2.5 ◦ for the period of 1901–2005 at monthly scale is used. Conversely, the observed terrestrial evaporationterm·key measurements were obtained from Global Land Evaporationterm·key Amsterdam Model (GLEAM) data set available for the period of 1980–2015 53 at a daily scale. This data set incorporates the Priestley and Taylor equation to calculate the potential evaporationterm·key based on observations of surface net radiation and near surface air temperature. A multiplicative evaporative stress factor estimated from satellite estimates is then used to convert the calculated potential evaporationterm·key values to actual evaporationterm·key. A more-detailed calculation procedure is available in the work of Martens et al. 53 .
The future projected data from 21 different general climatic models under the CMIP5term·key version for three different RCP scenarios are used. Each of these scenarios are distinguished by their radiative forcing increases by the end of the current century (RCP 2.6, 4.5, and 8.5) relative to pre-industrial values (Supplementary Table 1 ). These scenarios correspond roughly to a 2 °C stabilization, and 3.5 °C and 5 °C global temperature increase, respectively, by the end of the century. These CMIP5term·key models are selected based on the common availability of both precipitationterm·key and evaporationterm·key variables for historical and projected scenarios.
Methods Annual precipitationterm·key and evaporationterm·key characteristics After acquisition of the data, both the observed and modeled data are interpolated to a common 2.5 × 2.5° grid from their respective original grids. We assessed the seasonal variation of monthly precipitationterm·key and evaporationterm·key using an information theory metric called as AEterm·mth. Unlike the parametric coefficient of both variation and standard deviation, this metric is non-parametric and may even encompass high-order moments 3 . Moreover, information theory metrics have been widely used as a measure of rainfall seasonalityterm·dom in both hydrologic and climatological contexts 3 , 24 , 26 , 27 . Therefore, to estimate AEterm·mth for either precipitationterm·key or evaporationterm·key over a year k , the aggregated annual quantities during the 12 months indexed are computed by summing the monthly values ( x ) over all the months in a year as where X is the aggregated value of either the precipitationterm·key or evaporationterm·key. Subsequently, the AEterm·mth is calculated as By definition, both Eqs. ( 1 ) and ( 2 ) state that when either the amount of annual precipitationterm·key or evaporationterm·key is quite evenly apportioned to each of the 12 months with the probability of 1/12, Eq. ( 2 ) assumes the maximum value of log 2 12. Conversely, the minimum value of AEterm·mth = 0 occurs when the apportionment is extended to only 1 month of the 12 month cycle with a probability of 1, thus indicating the assumption of a new value by AEterm·mth within a finite range of 0 and log 2 12. Therefore, in a year, the less variable the monthly precipitationterm·key or evaporationterm·key, the higher the AEterm·mth value. Using this definition, the seasonal variability of precipitationterm·key (AEterm·mth P ) and evaporationterm·key (AEterm·mth E ) is estimated for the study period. The aggregation of both the precipitationterm·key (TOT P ) and evaporationterm·key (TOT E ) is then used to calculate the annual precipitationterm·key and evaporationterm·key during these 12 months.
Definition of global classification of precipitationterm·key regimes The existing atmospheric climate classifications are based on broad precipitationterm·key, temperature, evapotranspiration, and biosphere characteristics 21 – 23 . However, even though the seasonal variability significantly changes across the global land regions 26 , 27 , it is often not considered for classification from a non-parametric standpoint. For instance, the occurrence of substantial precipitationterm·key with an even monthly distribution will indicate a consistent and adequate supply of atmospheric water throughout the year. Similarly, an occurrence of substantial precipitationterm·key with an uneven monthly distribution will indicate surplus water supply during a particular period of a given year and deficit water supply during another period in a given year. Therefore, a failure to consider the seasonal variability of precipitationterm·key magnitude may well lead to a misrepresentation of the actual water supply conditions. Therefore, in this study, in addition to the precipitationterm·key magnitude, the seasonal variability wasterm·dom also included for purposes of classifying the global land regions into distinct water supply regimes. Here, the global land wasterm·dom divided into nine regimes of varying magnitudes of observed annual precipitationterm·key and seasonal variability based on the period between the years 1971 and 2000. These regimes are derived based on a combination of seasonal variabilities (i.e., AEterm·mth) and magnitude using the threshold concept: with high (>Pr 70 ), moderate (between Pr 30 and Pr 70 ) and low (<Pr 30 ), in which Pr i represents the ith percentile of either the annual magnitude or the seasonal variability. The use of the 30th and 70th percentiles is based upon the wet and dry region definitions adopted by Allan et al. 54 and subsequently implemented in Liu and Allan 55 . A similar threshold based classification of various dry and wet global regions has been used elsewhere with success 56 – 58 . In using this approach, we have coupled the characteristics of both annual magnitude and seasonal variability into distinct regimes based on global percentile thresholds. As a result, this definition can capture the precipitationterm·key characteristics that might have been previously omitted in those definitions based on absolute thresholds, separate precipitationterm·key magnitude and seasonal variability considerations and regional definitions 59 . The classification of these regimes also do not explicitly consider the natural landscape (e.g., rainforest, deserts) or the human-experienced climate events (e.g., monsoons) as suggested by Trewartha 54 . As we base our analysis on both precipitationterm·key and evaporationterm·key, this scheme provides a necessary and rather simple configuration for the relative assessment of spatially aggregated changes in both precipitationterm·key and evaporationterm·key characteristics for this study. All the derived precipitationterm·key and evaporationterm·key characteristics for CMIP5term·key as well as the observed data over these regimes are spatially aggregated.
Bayesian model averagingterm·key As the CMIP5term·key models exhibit varying levels of accuracy regarding simulating historical hydrologic cycles, the BMA methodology 52 is utilized to assign higher weights to better performing models. For this purpose, we computed the yearly precipitationterm·key and evaporationterm·key totals and AEterm·mth, from the CMIP5term·key models and observed data sets, i.e., precipitationterm·key from GPCCterm·dom and evaporationterm·key dataset from GLEAM over each precipitationterm·key regime. Then the BMA approach is utilized to determine the optimal weights for each CMIP5term·key model on their ability to replicate the spatial patterns of TOT P , TOT E , AEterm·mth P , and AEterm·mth E . The BMA approach calculates ensemble of the considered CMIP5term·key models by assigning weights based on the performance of models in comparison with the observations 52 , 60 . In this approach, the probability density function ( g ) of our interest variable, Y ∈ P , E , A E P , A E E from the observed data, which is conditioned upon the 21 CMIP5term·key simulations is expressed as: where w i is the optimal weight for the i th CMIP5term·key model, f ( Y | Y i ) is a PDF of the gamma distribution of Y i . The general selection of the BMA mixture probability distribution is normal with the gamma distributions based on the suggestion of Vrugt et al. 60 and Raftery et al. 52 . Therefore, in this study, a separate analysis is undertaken for both of the prior distributions. The BMA ensemble weighted mean and standard deviation is thus expressed as: Estimation of BMA weights The optimal weights w i , i = 1, …, 21 are estimated with a maximum log-likelihood function, which is expressed as: where j = {1, … N } are observations of the considered variable. Regarding precipitationterm·key, we use the mean annual estimates from 1971 to 2000 for estimating the optimal weights. Regarding evaporationterm·key, we only use the annual characteristics from 1980 to 2005. A Markov Chain Monte Carlo algorithm 60 , which maximizes the log-likelihood function is used for estimating the weights. This algorithm in turn generates N different Markov Chains, placed separately as rows with each chain represented as a 22-dimensional vector θ = { w 1 , w 2 ,…, w 21 , σ 2 }. The candidate maximum likelihood point is then sampled from a prescribed distribution, depending on the precipitationterm·key regime. Next, in accordance with the Metropolis acceptance probability, the candidate maximum likelihood point is either accepted or rejected. If accepted, the chain moves to the next candidate model; otherwise the chain repeats the process until the optimal weight for the candidate model is estimated. Clearly, the BMA weights obtained through maximum likelihood using this procedure is comparatively more accurate than the weights obtained using the Expectation–Maximization (EM) algorithm recommended by Raftery et al. 52 . A thorough description of this algorithm is presented in Vrugt et al. 60 . The estimated BMA weights are listed in Supplementary Table 2 .
Performance and evaluation of BMA multimodel ensembles over precipitationterm·key regimes The performance of the BMA multimodel ensemble wasterm·dom assessed in terms in replicating the spatial pattern across the land regions. After obtaining the optimal weights (shown in Supplementary Table 2 ), we computed the multimodel ensemble using Eq. ( 4 ) and evaluated the performance by the metrics of the Pearson correlation. The results indicate that in terms of the TOT P , TOT E , an R 2 values of 0.94 and 0.92, respectively, is obtained. Whereas in case of AEterm·mth P and AEterm·mth E an R 2 of 0.91 and 0.85 are obtained. These values further indicate that ensemble mean of the 21 GCMs, which are studied could adequately replicate the spatial patterns.
Assessment of projected changes The projected changes of TOT P , TOT E , AEterm·mth P , and AEterm·mth E were determined using the non-parametric Theil-Sen estimator 61 . We first estimated the linear trends using Theil-Sen estimator ( Δ Y ) and the BMA ensemble trend with an uncertainty is expressed as thus, permitting the estimates of the upper and lower confidence limits as: If the changes are distributed as a Gaussian density function, then these bounds imply a ~95% confidence interval.
Available water The difference between monthly precipitationterm·key and evaporationterm·key variables is deemed reliable for approximating the potential available water for human and ecological consumption. Here, we estimate the average net available water in 1 month for the historical period of 1971–2000 and for the future scenario of 2070–2099, which is expressed as: where r m,y and e m,y represent the precipitationterm·key and evaporationterm·key in month m and year y . We then compare the monthly net available water of historical and future scenarios (RCP 2.6, 4.5, and 8.5) from each GCM model through a spatial averaging over respective regimes. The denominator 30 represents the total number of years taken into consideration. In case of WAterm·dom, we used the averaged BMA weights for each model and estimated the monthly distribution of WAterm·dom.
To explore the role of seasonal components in terms of altering the monthly climatology of available water seasonal variability within a hydrologic year, we first extracted wet and dry precipitationterm·key and evaporationterm·key components. We assumed a 3-month period in which the maximum (minimum) of available water occurs as wet (dry) seasons 40 using the same base period of 1971–2000. We estimated the changes in both the wet and dry seasons as the difference between the future scenarios (2070–2099) and the mean historical (1971–2000) spatially aggregation over each regime. The BMA weights were applied and the uncertainty wasterm·dom calculated using Eqs. ( 7 – 10 ).
Supplementary information Supplementary Information Acknowledgements This study wasterm·dom supported by the National Science Foundation (NSF) award # 1653841.
Author contributions A.M conceived the idea, and G.K. performed all analyses. Y.W. and M.M. contributed to the discussion and writing of the results along with G.K. and A.M.
Data availability The data sets analyzed during the current study are available at Earth System Grid Federation (ESGF) Peer-to-Peer (P2P) enterprise system [ https://esgf-node.llnl.gov/projects/esgf-llnl/ ]. The Observed monthly GPCCterm·dom precipitationterm·key is available at https://opendata.dwd.de/climate_environment/GPCCterm·dom/html/fulldata-monthly_v2018_doi_download.html . The evaporationterm·key data set is available at https://www.gleam.eu/ .
Code availability The accompanying MATLAB source code for Bayesian model averagingterm·key utilized in this study is available at ( http://faculty.sites.uci.edu/jasper/software ).
Competing interests The authors declare no competing interests.
Footnotes Supplementary information Supplementary information is available for this paper at 10.1038/s41467-020-16757-w.
References Associated Data Supplementary Materials Supplementary Information Data Availability Statement The data sets analyzed during the current study are available at Earth System Grid Federation (ESGF) Peer-to-Peer (P2P) enterprise system [ https://esgf-node.llnl.gov/projects/esgf-llnl/ ]. The Observed monthly GPCCterm·dom precipitationterm·key is available at https://opendata.dwd.de/climate_environment/GPCCterm·dom/html/fulldata-monthly_v2018_doi_download.html . The evaporationterm·key data set is available at https://www.gleam.eu/ .
The accompanying MATLAB source code for Bayesian model averagingterm·key utilized in this study is available at ( http://faculty.sites.uci.edu/jasper/software ).