Hydrological Intensification Will Increase the Complexity of Water Resource Management

Global warming intensifies the hydrological cycle by altering the rate of water fluxes to and from the terrestrial surface, resulting in an increase in extreme precipitation events and longer dry spells. Prior hydrological intensification work has largely focused on precipitation without joint consideration of evaporative demand changes and how plants respond to these changes. Informed by state‐of‐the‐art climate models, we examine projected changes in hydrological intensification and its role in complicating water resources management using a framework that accounts for precipitation surplus and evaporative demand. Using a metric that combines the difference between daily precipitation and daily evaporative demand (surplus events) and consecutive days when evaporative demand exceeds precipitation (deficit time), we show that, globally, surplus events will become larger (+11.5% and +18.5% for moderate and high emission scenarios, respectively) and the duration between them longer (+5.1%; +9.6%) by the end of the century, with the largest changes in the northern latitudes. The intra‐annual occurrence of these extremes will stress existing water management infrastructure in major river basins, where over one third of years during 2070–2100 under a moderate emissions scenario will be hydrologically intense (large intra‐annual increases in surplus intensity and deficit time), tripling that of the historical baseline. Larger increases in hydrologically intense years are found in basins with large reservoir capacity (e.g., Amazon, Congo, and Danube River Basins), which have significant populations, irrigate considerable farmland, and support threatened and endangered aquatic species. Incorporating flexibility into water resource infrastructure and management will be paramount with continued hydrological intensification.

However, most efforts to quantify hydrological intensification have focused solely on precipitation changes (Huntington et al., 2018), which do not capture the role of rising atmospheric demand. Increasing evaporative demand (which is often described in terms of a reference, well-watered evapotranspiration rate, or ETo) is an important signal of hydrological intensification, as greater ETo dries out soil more quickly after precipitation events (Ficklin et al., 2019;Huntington et al., 2018). Actual evapotranspiration is also likely to increase as ETo rises, at least initially in places not water-limited. Increases in actual and reference/potential evapotranspiration rates have been observed (Greve et al., 2014;Greve & Seneviratne, 2015;Huntington et al., 2018;Milly & Dunne, 2016) and are projected to continue with climate change (Ficklin et al., 2019;Lavers et al., 2015;McEvoy et al., 2020). However, eventually, plants close their stomata directly in response to the increases in vapor pressure deficit (VPD) that accompany increasing ETo (Grossiord et al., 2020), which decouple actual evapotranspiration from its reference rate (Novick et al., 2016). The increases in atmospheric CO 2 can also independently reduce stomatal conductance (Ainsworth et al., 2007;Swann et al., 2016) and further decouple actual and reference evapotranspiration. Therefore, a more integrated and robust signal of hydrological intensification should consider changes in both precipitation and changes in evaporative demand driven by rising temperature but mediated by dynamic plant responses to elevated VPD and CO 2 . Prior work has characterized historical hydrological intensification based on changes in precipitation and actual evapotranspiration (Huntington et al., 2018). Future hydrological intensification has also been characterized based on changes in precipitation and reference evapotranspiration (Ficklin et al., 2019). Here, we build on the latter by describing and interpreting a novel metric to predict future hydrological intensification that treats plants as dynamic participants in the water cycle by accounting for plant stomatal closure as VPD and CO 2 increase.
Future changes to precipitation surplus and dry spell length that characterize hydrological intensification have important ramifications for a number of important societal problems. For example, not only periods of drought, but also periods of excess precipitation, can reduce agroecosystem productivity or alter the relationship between rainfall and plant growth (Felton et al., 2021;Post & King, 2020;Yin et al., 2020). Shifts in the distribution of rainfall events and dry spells have also been shown to influence landslide risk (Tichavský et al., 2019) and water quality (Loecke et al., 2017). Likely implications of increasing precipitation extremes and dry spells on water resource management are additional knowledge gaps that are directly confronted in this paper. Understanding which regions and river basins are anticipated to have larger precipitation events and longer dry spells is critical for designing and managing infrastructure for people, agriculture, and ecosystems. More frequent and severe floods and droughts are anticipated with climate change in some regions-including events that occur within the same year (Swain et al., 2018). These hydrologically intense years lead to increasing conflict among urban, agricultural, and ecological water uses, with intractable decisions for water resources managers about releasing water to alleviate flooding or storing water to meet demands later in the year (Raymond et al., 2020). Reservoir storage, conveyance, and flood control provide infrastructure to manage water resources with hydrological intensification, but paradoxically may also increase water demand because water is readily available, leaving communities that depend heavily on water infrastructure more vulnerable to hydrological intensification (Di Baldassarre et al., 2018).
Our overall objectives are twofold: (a) Understand projected changes in hydrological intensification that explicitly consider changes in precipitation, atmospheric demand, and dynamic plant feedbacks to rising atmospheric demand and elevated CO 2 , and (b) relate patterns of future hydrological intensification to conflicting objectives for water resources management in major river basins throughout the world. Our approach relies on projections from an ensemble of climate models from the latest coupled model intercomparison project, CMIP6 , and two Shared Socioeconomic Pathways (SSPs) SSP2-4.5 and SSP4-8.5 (O'Neill et al., 2016). While our previous work explored hydrological intensification for the continental United States using CMIP5 models (finding that including atmospheric demand rather than solely using precipitation resulted in increased intensification; Ficklin et al., 2019), global hydrological intensification has yet to be explored in detail using CMIP6 models. Additionally, we use descriptive information on human activities and important aquatic species occurring within river basins with considerable water storage to assess the influence hydrological intensification may have on water resource management. Quantifying hydrological intensification will improve understanding of the changing dynamics of water fluxes to and from the terrestrial surface for individual major river basins around the globe, therefore providing additional information on hydrological extremes and future water resource reliability.

Defining Hydrological Intensification
Here, hydrological intensification is defined as the annual average of daily precipitation surplus intensity (precipitation -ETo; in millimeters) and the annual average deficit time (specifically, the consecutive number of days when daily ETo > daily precipitation ;Ficklin et al., 2019). By using both supply and atmospheric demand, our definition of hydrological intensification assesses how fast the water cycle is accelerating, considering both changes in extreme precipitation and the frequency of dry spells (Ficklin et al., 2019). Thus, our framework jointly considers how the variability in precipitation and atmospheric demand determines the patterns of hydrological intensification. Additionally, previous work has not included the direct consideration of the ways that plants respond to rising VPD and/or elevated CO 2 (Ficklin & Novick, 2017;Grossiord et al., 2020;Novick et al., 2016;Swann et al., 2016;Yuan et al., 2019). In other words, like many metrics to describe drought status and/or the pace of hydrological intensification, most prior work treats plants as static participants in the hydrologic cycle. Thus, a major novelty of the present study is its explicit consideration of dynamic plant response to VPD and CO 2 .
The ETo was estimated using the American Society of Civil Engineers Penman-Monteith method for a reference grass surface, but with a formulation for the reference surface conductance that accounts for the influence of climate feedbacks on stomatal conductance, including rising CO 2 and VPD ; see Supporting Information for details). These mechanisms reduce the magnitude of ETo, and failure to account for their influence on stomatal conductance can lead to overestimation of future ETo. Because we used an improved ETo model that includes the influence of CO 2 and VPD on stomatal conductance, our ETo is not directly comparable with ETo models that do not include this advancement. At the same time, this formulation does not explicitly consider soil moisture limitations to stomatal function, and thus the ETo should not be interpreted as actual evapotranspiration; rather, it should be interpreted as the ETo rate for conditions of non-limiting soil moisture, but with the potential for VPD and CO 2 to reduce stomatal conductance. While declining soil moisture may cause additional reductions to the actual ET, VPD and soil moisture are coupled in time and space (Zhou et al., 2019), and VPD is the predominant factor limiting transpiration in many parts of the world (Flo et al., 2021;Novick et al., 2016).
We used a metric for hydrological intensification that blends information about surplus intensity and deficit time-the Surplus Deficit Intensity Index (or SDI; Equation 1; Ficklin et al., 2019) defined as: where std is the standard deviation and z is the standardization of surplus intensity (SurINT) and deficit time (DT). Surplus intensity is the annual mean of daily surplus (in mm) intensity events for days when precipitation > ETo. In simple terms, surplus intensity can be defined as "excess water," with more water leading to potential extreme hydrologic events such as floods. In agroecosystems, periods of excessive rainfall can reduce crop productivity, sometimes by an amount comparable to severe drought (Dold et al., 2017;Yin et al., 2020). In many ecosystems, larger and more intense rain events can reduce net primary productivity (Post & Knapp, 2020) and/or decrease the ratio of rainfall to net primary production, especially when heavy rainfall events are separated by longer dry spells (Felton et al., 2021;Zhang et al., 2013). Deficit time is the annual mean dry spell length (in days) for days when precipitation < ETo. Deficit time represents the time of declining water availability, either for water resources or vegetation. For ecosystems, longer periods of water limitation are particularly concerning due to the potential for interactions between declining soil moisture and rising atmospheric VPD which can stress plants from the perspective of both water supply and atmospheric demand (Humphrey et al., 2021). Longer periods of water limitation also have increased the prevalence and duration of no flow conditions in streams, causing some perennial systems to shift to intermittent flow (Zipper et al., 2021).
To estimate SDI, annual surplus intensity and deficit time are standardized (z-score) using the 1950-1980 time period to allow for spatial and temporal comparability. The denominator in Equation 1, which is fixed temporally, allows SDI to be compared between regions with different climates (and thus different supply and demand characteristics). For historical evaluation and the projected changes for surplus intensity and deficit time, we present the results in their absolute values (mm or days) or percent change from the baseline time period to aid in interpretability.

of 16
The z-scores of surplus intensity and deficit time are combined to express hydrological intensification or SDI (Equation 1). A large, positive SDI represents intensification via extreme precipitation surplus events, extended periods of water deficit, or a combination of both, while a negative SDI value represents low-intensity surplus precipitation events, small periods of water deficit, or a combination of both. While we largely focus on assessing changes of SDI and its underlying components as well as their relationship to water resource management, increased co-occurring precipitation surplus events and dry spells have been found, among others, to increase landslide events (Tichavský et al., 2019), result in crop damage (Barron et al., 2003), decrease dryland productivity (Felton et al., 2021), affect water quality (Loecke et al., 2017), increase wildfire activity (Flannigan & Harrington, 1988), and lead to unexpected impacts in coupled human and natural systems (J. Liu et al., 2007). An example of SDI is shown in Figure 1, where the bottom panel represents a more hydrologically intense year than the top panel (negative SDI value). Detailed background information on surplus intensity, deficit time, and SDI can be found in Ficklin et al. (2019).
Like most widely used metrics to describe drought status or to predict patterns of hydrological intensification, SDI is a simplification of a complex hydrological process, representing both wet and dry extremes. For example, SDI does not consider initial soil moisture content or water storage conditions, and therefore assumes that precipitation surplus events are completely converted into surface water runoff. In reality, precipitation surplus events may not be converted to water runoff if the initial soil moisture conditions are dry. Surplus intensity, however, could be an indicator of an event that increases soil moisture that would perhaps eventually lead to surface runoff. Additionally, ETo and the actual evapotranspiration that depletes soil moisture may become decoupled as ETo increases (especially in water-limited environments; McVicar et al., 2012) due to changes in water supply (Kauwe et al., 2017;Ohta et al., 2008;Peng et al., 2019), evaporative demand (Novick et al., 2016), or phenological vegetation characteristics (Donohue et al., 2010;C. Liu et al., 2017), of which are not completely characterized using SDI.

Climate Model Ensemble
To estimate projected changes in terrestrial hydrological intensity, daily CMIP6 output from 13 global climate models (GCMs) was extracted from the CMIP6 data clearinghouse at https://esgf-node.llnl.gov/search/cmip6/. The CMIP6 ensemble has increased representations of physical processes and their horizontal and vertical resolution . Models from CMIP6 span a wider range of warming and precipitation responses compared to CMIP5 models (Almazroui et al., 2020;Cook et al., 2020;Smith & Forster, 2021;Tebaldi et al., 2021;Tokarska et al., 2020).
We specifically extracted data for the historical (1850-2014) time period simulations that include climate forcings from natural and anthropogenic sources . For the projected time period (2015-2100), we used GCM output (one ensemble member from each GCM) from two SSPs (O'Neill et al., 2016;SSP2-4.5 and SSP5-8.5). The SSPs differ based on socioeconomic assumptions and each SSP represents different radiative pathways resulting in a different radiative increase at the end of 2100, where SSP2-4.5 = 4.5 W/m 2 (moderate emissions) and SSP5-8.5 = 8.5 W/m 2 (high emissions). These represent approximate end-of-the 21st century average annual temperature increases of approximately 3°C and 5°C, respectively, compared to pre-industrial conditions (Cook et al., 2020;Tebaldi et al., 2021). We focus primarily on the results from the SSP2-4.5 forcing, but provide context for changes obtained with the SSP5-8.5 forcing.
For the historical and projected time periods, we used ensembles of opportunity based on specific models and ensemble members (found in Table 1) that provided the climate variables needed to calculate surplus intensity, deficit time, and SDI. Not all variables were available for the full 1850-2100 time period (some started in 1950), and thus the baseline for this work is defined as 1950-1980. All GCM output was bi-linearly interpolated to a common 1.5° grid. To ensure the reliability of the GCM projections, we evaluate the accuracy of the GCM ensemble against reanalysis data sets for 1979-2014. We find the GCM ensemble can adequately represent historical precipitation, ETo, surplus intensity, and deficit time. A detailed analysis is shown in Tables S1 and S2 and Figure S1 in Supporting Information S1.

Projected Changes in Terrestrial Hydrological Intensification
We compare GCM ensemble mean changes for surplus intensity, deficit time, SDI, precipitation event intensity (amount of precipitation per event [defined as >1 mm]), and daily ETo for one future time period (2070-2100) and SSPs 2-4.5 and 5-8.5 compared to the baseline period. We also examine the robustness of the change in intensification amongst GCMs for each grid cell, where a robust finding has agreement on the sign of the change by at least 75% of the models (Scheff et al., 2021). To understand the relative influence of changes in precipitation and ETo on hydrological intensification, each climate variable was linearly detrended from 1950 to 2100 under SSP5-8.5, and the hydrological intensification metrics were re-assessed using one detrended climate variable and one raw (or unmodified) climate variable. For this, SSP5-8.5 was used because it exhibits the most extreme changes in temperature and precipitation, and thus the detrended signal is likely to be more apparent.

River Basin Data
While it has long been understood that the hydrologic cycle is intensifying, leading to grave implications for water management (Milly et al., 2008), understanding the co-occurrence of water surplus and deficits within years and across river basins is needed to adapt water management to climate change. Surplus intensity, deficit time, and SDI provide simple metrics to integrate precipitation surplus events and dry spells within the same year. These metrics quantify changes to evaluate the magnitude of hydrological intensification and whether changes are dominated by water surpluses or deficits. We identify potential future water management challenges by summarizing results for major global river basins that have significant water resources infrastructure. While SDI is a simplification of complex hydrological processes, quantifying the relationship between hydrological intensification (and corresponding metrics) is a crucial first step toward improving understanding of hydrological intensification effects on future water resources management decisions.
River basin descriptive information was extracted from HydroATLAS (Linke et al., 2019), which is a global data set of basin demographic and environmental characteristics. We focused on two variables in the HydroAT-LAS database that are directly related to water management: population (Center for International Earth Science Information Network -CIESIN -Columbia University, 2018) and percent area irrigated (Siebert et al., 2015). Additionally, the number of total endangered and threatened fish species was examined (IUCN, 2021) and then integrated with the HydroATLAS database. For this work, we relate changes in SDI to potential water resource management conflicts, as the magnitudes of precipitation and ETo are dominate factors that determine how precipitation is partitioned into runoff or evapotranspiration (Berghuijs et al., 2017;Blöschl et al., 2013;Budkyo, 1974), especially when summarized at the annual time step.

Global Hydrological Intensification
Our results suggest that precipitation surpluses are projected to increase, even after accounting for increases in evaporative demand. It is important to reiterate, though, that this work uses a version of ETo that incorporates the influence of CO 2 and VPD on stomatal conductance (as fully described in Yang et al. [2019]). As expected, and was found in previous work (Greve et al., 2019;Yang et al., 2019), not incorporating the influence of CO 2 and VPD on stomatal conductance resulted in a larger increase in ETo ( Figures S2 and S3 in Supporting Information S1). This increased globally averaged surplus intensity (1%-2% without ETo that incorporates the influence of CO 2 and VPD compared to ETo with CO 2 and VPD), deficit time (2%-5% increase), and SDI (∼0.1 increase in z-score) with larger changes found in drier regions (as found in Cui et al. [2021]). Regardless of whether the influence of CO 2 and VPD on stomatal conductance is incorporated in ETo, the changes in surplus intensity, deficit time, and SDI do not change sign and follow the same statistical significance as changes with the influence of CO 2 and VPD as is discussed in more detail below.
For the 2070-2100 time period, we find an 11.5 ± 4.6% (GCM ensemble mean and interquartile range) and 18.5 ± 4.7% increase in surplus intensity under SSP2-4.5 and SSP5-8.5, respectively, relative to the 1950-1980 baseline, with a robust agreement amongst GCMs (75.6% of the terrestrial surface for SSP2-4.5 and 80.1% for SSP5-8.5; Figure 2 for SSP2-4.5 and Figure S4 in Supporting Information S1 for SSP5-8.5). For both SSPs, the largest increases in surplus intensity occur in the northern and midlatitudes and generally follow patterns of increased precipitation event intensity (Figures 3 and S5 in Supporting Information S1; significant [p < 0.05] GCM ensemble mean Spearman correlation of 0.60 for SSP2-4.5 and 0.61 for SSP5-8.5). At the global scale, the CMIP6 ensemble generally predicts increases in precipitation event intensity for both SSPs (7.5% for SSP2-4.5 and 12.1% for SSP5-8.5) relative to baseline conditions (Figures 3 and S5 in Supporting Information S1), consistent with previous work using CMIP6 models (Cook et al., 2020;Tebaldi et al., 2021). Only India has consistent decreases in ETo, a result consistent in other studies that show projected decreases in drought for this region 7 of 16 (Aadhar & Mishra, 2020;Cook et al., 2012). Both SSPs show more widespread increases in surplus intensity than average precipitation intensity, suggesting that precipitation events that eclipse ETo are projected to become more frequent and intense (Figures 2 and S4 in Supporting Information S1). Global average projections of precipitation event intensity, daily ETo, and hydrological intensification components can be found in Figure S6 in Supporting Information S1.
Even while surplus intensity is projected to increase globally under climate change, dry spell deficit time is also projected to lengthen (Figure 2 for SSP2-4.5 and Figure S4 in Supporting Information S1 for SSP5-8.5). Overall, for SSP2-4.5 and SSP5-8.5, GCM ensemble mean deficit time increased by approximately 5.1 ± 3.1% and 9.6 ± 7.2%, with an area of robust change of 31.0% and 41.1%, respectively. Significant increases in deficit time occur in eastern North America, the Amazon region, southern Africa, and large portions of Europe and western Asia. These areas are where average daily ETo is projected to have a significant relative increase (>10%) and are coupled with minor changes or decreases in average precipitation event intensity (Figure 3). Changes in both average precipitation intensity and daily ETo are significantly (p < 0.05) correlated to changes in deficit time. Spearman correlation between late 21st century changes in precipitation event intensity and deficit time was −0.48 for SSP2-4.5 and −0.58 for SSP5-8.5, whereas the correlation between changes in daily ETo and deficit time was 0.25 for SSP2-4.5 and 0.34 for SSP5-8.5 (both significant at p < 0.05).
SDI generally follows the same spatial patterns and magnitudes as surplus intensity and deficit time with increases for much of the terrestrial surface (Figure 2 for SSP2-4.5 and Figure S4 in Supporting Information S1 for SSP5-8.5). For SSP2-4.5 and SSP5-8.5, we find an increase in SDI for the global terrestrial surface with a 0.61 ± 0.21 and 0.88 ± 0.24 z-score increase, respectively, with large areas of robust agreement for both SSPs. We We re-assessed changes in hydrological intensification under SSP5-8.5 to understand the relative importance of individual trends in both precipitation and ETo in causing hydrological intensification. We find that detrending precipitation and keeping ETo at the original values results in slightly larger increases in hydrological intensification (global mean SDI z-score of 1.04) with the biggest differences were found in the midlatitudes and northern latitudes, driven by increases in surplus intensity after detrending as well as increases in deficit time in portions of Asia ( Figure S7 in Supporting Information S1). Detrending ETo and keeping precipitation at its original values results in a decrease in hydrological intensification (terrestrial global SDI z-score of 0.62). As was the case for detrended precipitation, the largest differences were found in the midlatitudes and northern latitudes; however, the reason largely stemmed from a lack of increase in surplus intensity with deficit time overall playing a smaller role ( Figure S7 in Supporting Information S1).
While detrending precipitation did result in an increase in hydrological intensification, many of the increases were smaller when compared to the large decreases for SDI with a detrended ETo. For many grid points, detrending ETo changed the direction of hydrological intensification from an increase to a decrease, as is shown in Figure 4. An increase in hydrological intensification from a detrended precipitation is counterintuitive; however, this result indicates that when precipitation is detrended many of the smaller precipitation events no longer eclipse ETo, and because we find an increase in precipitation event intensity for much of the terrestrial surface (Figures 4 and S5 in Supporting Information S1), this results in a lower number of small precipitation events that are included in the annual average of surplus intensity (4.8% reduction), thus resulting in an increase in surplus intensity magnitude. Additionally, with precipitation detrended, deficit time increased because the smaller precipitation events that broke up dry spells were no longer included (Figure 4). Similarly, when ETo is detrended (which increased for much of terrestrial surface; Figures 4 and S5 in Supporting Information S1), smaller precipitation events that eclipse ETo are included (10.4% increase), resulting in a decrease in the surplus intensity annual average. These smaller events tend to break up time periods where ETo > precipitation intensity, thus also decreasing in average deficit time. Additionally, defining hydrological intensification using precipitation and atmosphere demand led to more intensification as compared to solely using precipitation (as is found in Giorgi et al. [2011]). For the global average, incorporating atmospheric demand increased projected surplus intensity by ∼3%-4%, deficit time by 3%-4%, and SDI by 0.25 (not shown).

Implications and Recommendations for Water Resource Management
An increase in surplus intensity events suggests the possibility of more flooding (Blöschl et al., 2019;Davenport et al., 2021;Mallakpour & Villarini, 2015) with subsequent impacts to soil moisture and groundwater recharge (Cook et al., 2020;Mankin et al., 2019;Smerdon, 2017), water quality (Wheater & Evans, 2009), biodiversity, and ecosystem health (Maestre et al., 2015). Additionally, Famiglietti et al. (2021) found that extreme wet events are just as important as extreme dry events for controlling vegetation green-up anomalies. These impacts, in turn, affect other hydrologic variables such as infiltration and evapotranspiration. Increases in deficit time are a common proxy for droughts (Vicente-Serrano et al., 2010), flash droughts (Mo & Lettenmaier, 2016;Otkin et al., 2018), increased wildfire potential (Flannigan & Harrington, 1988), decreased streamflow (Ficklin et al., 2018), and groundwater resilience (Hahm et al., 2019). Persistence and lengthening of dry conditions can force a transition from a system limited by energy to a system that is limited by water, leading to increased vegetation stress (McDowell et al., 2008), plant mortality (Allen et al., 2010), and potential species composition change (Bréda et al., 2006;Trugman et al., 2020). It is important to note, though, that we summarize hydrological intensification at the annual time scale and therefore do not distinguish hydrological intensification within seasons. Regardless, either increasing surplus intensity or lengthening deficit times can impact natural and built systems; the fact that they occur together within the same year for most of the land surface signifies a concerning departure from historical conditions.
An increase in SDI influences the timing, location, and amount of water available for cities, industry, agriculture, and ecosystems, where periods of too much water may be followed by extensive periods without water. Existing research has focused on long-term trends of climate change on water resources (Schewe et al., 2014) and future inter-annual variability for water allocation (Null & Viers, 2013). However, hydrological intensification is likely to challenge water management since water resources management often relies upon or assumes hydroclimate stationarity (Milly et al., 2008).
Generally, water infrastructure and water right/allocation frameworks have been designed to manage seasonal wet and dry periods. For example, dams are built to store wet season runoff for dry season water supply, irrigation, and sometimes environmental flows. In climate types defined by distinct wet and dry seasons like Mediterranean, tropical savannah, monsoon-influenced climates, and some midlatitude and semi-arid climates, variability in seasonal precipitation is more certain than variability among years. Similarly, climate types that historically did not have precipitation seasonality may increasingly experience precipitation surpluses and deficit periods at the end of the 21st century that are challenging to predict and manage (Figures 2  and S4 in Supporting Information S1).
To better quantify water management-relevant shifts, we summarize the GCM ensemble mean SDI results under SSP2-4.5 (moderate emissions scenario) for major river basins throughout the world that have considerable surface reservoir capacity (Lehner et al., 2011;Figure 5;Figures S8-S9 in Supporting Information S1 for SSP5-8.5). For each of these basins, we examine the percent of years in the 2070-2100 time period that have an SDI z-score >1, which gives an indication of the number of hydrologically intense years that may lead to complex or unprecedented water resource management. An example of an SDI year >1 is the Sacramento River Basin in northern California during 2012. In late 2012, in the midst of a prolonged and severe drought, the Sacramento River Basin was hit with a series of major storms that resulted in floods, even though annual precipitation remained well below normal (US Drought Monitor, 2021).
For basins with considerable reservoir storage, we find that, on average, 34.1% of years (10.5 years) during the 2070-2100 time period have an SDI z-score >1 ( Figure 5; the SSP5-8.5 pathway results in 44.5% of years [13.8 years]) compared to only 11% (3.4 years) during the historical baseline. We also find that, on average, SDI z-scores >1 occur 1.7 years in consecutive duration (2.6 years for SSP5-8.5), further complicating water resource management. Of the 10.5 years between 2070 and 2100 with an SDI z-score >1, 3 of those years (4.9 years for SSP5-8.5) have both surplus and deficit time >1 z-score in a given year (baseline value = 0.6 years) and >0.5 z-score for 6 years (8.3 years for SSP5-8.5; baseline value = 2 years). These changes are largely driven by increases in surplus intensity >1 z-score, though basins such as the Amazon River Basin, show large percentages of time during 2070-2100 with a deficit time >1 z-score ( Figure 5).
Basins with the most reservoir capacity are projected to have more hydrologically intense years (significant [p < 0.05] Spearman's correlation of 0.33; Figure 6). These basins with substantial reservoir capacity may be required to spill water from reservoirs during intense wet periods, followed by extended periods when they have numerous empty reservoirs but not enough precipitation to fill them (Hanak et al., 2011;van Dijk et al., 2013). Managing intense and perhaps unprecedented flood events will also challenge existing infrastructure and management (Independent Forensic Team Report-Oroville Dam Spillway Incident, 2018). Overall, managing infrastructure in these large basins to reduce flood risk and store water for extended drought will likely be contentious.
The heavily managed river basins with >40% of hydrologically intense years during the 2070-2100 time period include the Amazon River Basin, Danube River Basin, St. Lawrence River Basin, and the Yukon River Basin.  Basins such as the Murray-Darling River Basin, Niger River Basin, and Nile River Basin have hydrologically intense years <20% of the time during the 2070-2100 time period, but still more than the historical time period ( Figure 5). The reasons for these changes in hydrologically intense years vary from basin-to-basin. For example, hydrological intensification in the Amazon River Basin is largely a result of increased deficit time, while hydrological intensification in the Yukon River Basin is driven by large increases in surplus intensity ( Figure 5). Previous research has shown that observed seasonal surface water storage variability is concentrated in arid and semiarid regions and south of 45° north latitude (Cooley et al., 2021). Here, we nuance this finding by showing that many heavily managed river basins in northern Europe, Asia, and North America will experience considerable hydrological intensification by the end of this century ( Figure 5). Some basins that are hotspots for hydropower development, like the Amazon and Mekong River Basins (Barbarossa et al., 2020), are likely to have reduced generation as a result of hydrological intensification, particularly when intensification is driven by increased deficit time. River basins with considerable reservoir capacity are also basins with large populations (significant [p < 0.05] Spearman correlation with reservoir volume = 0.42), land irrigated with ground or surface water (correlation = 0.19), and threatened and endangered aquatic species (significant correlation = 0.32), suggesting that water in these basins must be allocated amongst these three sectors ( Figure 6). Similarly, the basins with >40% of hydrologically intense future years currently have, on average, greater populations, irrigated land, and threatened and endangered species, compared to basins with <20% of hydrologically intense years. This implies that hydrological intensification among urban, agricultural, and environmental water uses will likely lead to conflict among water sectors.
Depending on water rights structures, water shortages may be shared among users, or water may be allocated based on water right priority. Agricultural water users generally receive the most water, while urban water users value water the highest. Water markets sometimes transfer water from agricultural to urban or environmental water uses (Grafton et al., 2013). Urban water suppliers value water reliability and often stress test systems so they have a good understanding of future risk and promising options to obtain extra water before droughts (Brown et al., 2012). Managed environmental water allocations are typically small and are further reduced during droughts, so environmental flows may be inadequate to preserve species and ecosystems during prolonged or persistent droughts (Grafton et al., 2013;Null & Viers, 2013). Ecosystem management tends to be reactive to hydrological intensification, with little planning before floods and droughts (Lund et al., 2018).
Our findings beg the question, to what extent is water management infrastructure ready for a more hydrologically intense future? Hydrological intensification could be alleviated to some extent in basins with large reservoirs that provide multiyear storage, although reservoirs sometimes lead to counter-intuitive outcomes like increasing water demand and vulnerability (Di Baldassarre et al., 2018). The dependence on mountain water storage in the form of snow may further complicate water resource management (Livneh & Badger, 2020), as the shift from snowto-rain may decrease snowpack water storage in mountainous regions where hydrological intensification is the highest. Promising solutions to recurrent drought from increasing deficit time are likely varied but may include combinations of conjunctive use of surface and groundwater (Scanlon et al., 2016), prioritizing water conveyance to improve flexibility in basins with substantial surface storage (Null, 2016), stormwater and floodplain management (Palmer et al., 2008), and perhaps new surface or underground storage (Ehsani et al., 2017). Incorporating weather forecasts into reservoir operations may improve water management with precipitation surplus events (Alexander et al., 2021). In general, building flexibility into water infrastructure and management is an adaptation for hydrological intensification (Gersonius et al., 2013) and could include modifying dam operations (Ehsani et al., 2017), revising flood operating rules (Willis et al., 2011), or incorporating climate projections into flood risk mitigation (Sims & Null, 2019).
In highly managed river basins, hydrological intensification could result in flood and drought events that are beyond infrastructure design specifications (Raymond et al., 2020). Historic floods and droughts will be invaluable to understand and mitigate risk, although we show that future events may be more intense, more frequent, or longer. Understanding the drivers behind hydrological intensification, quantified here as increasing deficit time or an increase in surplus intensity, will also be useful to plan and assess risk. Hydrological intensification alternatives should be incorporated into regional or watershed-scale water resources management research to evaluate specific climate adaptation strategies and understand the limitations of existing infrastructure to manage more intense flood or drought periods. Our study identifies major river basins that are priorities for this future work. Overall, our findings suggest that the common practice of modeling effects of climate-induced annual temperature and precipitation changes for water management may underprepared us for more severe climate intensification events that will stress water infrastructure and management.

Conclusions
Our results indicate that the hydrological cycle will continue to intensify into the future. Globally, these changes are driven largely by increases in precipitation surplus events (+11.5% for a moderate emissions scenario by the end of the century; +18.5% for a high emissions scenario), with increases in deficit time overall playing a smaller role (+5.1%; +9.6%). This leads to major river basins in northern latitudes, sub-Saharan Africa, and other regions that are projected to experience large increases in intensification due to precipitation surplus events. However, some important water resource regions (such as the Amazon River Basin) show that increases in deficit time will dominate hydrological intensification in the future. In sum, these changes are expected to increase the number of hydrologic extremes within a given year, resulting in increased stress on existing water resource infrastructure. Consequently, we find that the river basins with the largest water storage are also projected to have the strongest hydrological intensification with some basins (generally found in the northern latitudes) having three times more hydrologically intense years than the historical baseline. This indicates that in the future, water resource managers in areas with extreme hydrological intensification must balance between storing and/or releasing water for agricultural, ecological, and urban sectors in the face of increasing hydrologic extremes and must therefore incorporate flexibility into water resource infrastructure and management.