Drought is defined as a lack of water compared to normal conditions which can occur in different components of the hydrological cycle (Palmer, 1965; Tallaksen and Van Lanen, 2004; Sheffield and Wood, 2011). It is commonly subdivided into meteorological drought (rainfall deficit), soil moisture drought (below-normal soil moisture levels), and hydrological drought (below-normal (sub)surface water availability). The normal is often taken as a percentile of the climatology of the variable of interest, and severity (e.g. deficit volume) and duration of drought events can be calculated (Van Loon, 2015).

In the natural sciences, there is a fair understanding of the propagation of drought from meteorological drought to soil moisture drought and hydrological drought (Fig. 2, left side), influenced by catchment properties such as geology and vegetation cover. For example, many hydrological drought types have been recognized, e.g. the classical rainfall-deficit drought, but also hydrological droughts caused by temperature anomalies in snow-dominated areas (Van Loon and Van Lanen, 2012; Van Loon et al., 2015). This is typically regarded as a uni-directional propagation with human receptors at the downstream end. However, in reality, human processes are interlinked with natural processes in various ways (Fig. 2, right side). Soil moisture and hydrological drought (hereafter called drought) are the result of low inputs to the hydrological system (e.g. lack of rain, snow/glacier melt, irrigation, sewage return flows), high outputs (e.g. evapotranspiration, human water use), and limited storage (in soil, groundwater, lakes, and reservoirs). Human activities influence water input, output, and storage and therefore modify the propagation of drought and can even be the cause of drought in the absence of natural drivers of drought. The drought typology based on natural processes should therefore be complemented with drought types based on human processes.

The natural drought types can be grouped together as “climate-induced” droughts and drought types based on human processes can be termed “human-induced” or “man-made” drought (Fig. 3; Van Loon et al., 2016). This parallels an existing widely referenced typology of floods, which includes man-made flood alongside natural floods such as flash flood, snowmelt flood, and ice jam flood (e.g. Yevjevich, 1994; De Kraker, 2015). The distinction between climate-induced and human-induced drought is useful in studies of the attribution of drought to its causes. To further acknowledge the possibly large influence of human activities modifying drought (Fig. 2), we additionally propose the term “human-modified drought” for a drought that is enhanced or alleviated as the result of anthropogenic processes (Fig. 3). For this terminology, we focus on direct human influences on the hydrological cycle such as water abstraction and land use change, although we recognize that anthropogenic climate change indirectly affects the meteorological drivers of drought (e.g. Williams et al., 2015).

With these terms, we actively include humans as drivers and modifiers of drought in the definition. There is no need for rephrasing the general drought definition, in which human processes are implicitly included. Furthermore, the terms we propose are not new (climate-induced drought: Sheffield and Wood, 2011, p. 30; human-induced drought: Wilhite and Buchanan-Smith 2005, p. 10 and Falkenmark and Rockström, 2008, p. 93) and they match well with the flood terminology (Yevjevich, 1994).

Drought is a lack of water compared to a certain normal situation, but what constitutes this normal situation in the Anthropocene? A drought occurs when actual water availability (indicated by water levels or fluxes) is below normal (Fig. 4). In a natural catchment, undisturbed by human activity, both actual and normal water availability are governed by natural processes in response to climate. Normal water availability is determined by the climate (long timescales), for example, a (semi-)arid climate results in low average water availability (aridity; Table 1) and low threshold levels (Fig. 4c). Actual availability is determined by climate variability (here used as term for a combination of weather events; short timescales); for example, a rainfall deficit leading to a climate-induced drought (Table 1; Fig. 4a, c). Even though drought is defined on shorter timescales than aridity, very short periods of below-normal water availability are often not regarded as drought, e.g. drought is defined as “sustained” by Tallaksen and Van Lanen (2004, p. 4), which means it lasts for longer than few days. This makes droughts generally occur on longer timescales than, for example, floods.

In a human-influenced catchment, actual and normal water availability are, besides by climate, also influenced by human activities (Fig. 4b, d). The actual situation is influenced by water use and water management (short timescales), leading to lower or higher water levels, whereas the normal situation is influenced by long-term actions such as groundwater depletion and anthropogenic land use change (long timescales; Table 1). There are different ways to account for this different normal. If we have a long-enough time series to determine the normal situation as influenced by human activities, we can use that as our reference or threshold and only determine our droughts as extreme events relative to this human-influenced normal (Fig. 4b, d: disturbed drought threshold). For example, in the Júcar Basin in Spain drought measures are based on thresholds in measured reservoir levels, groundwater levels, and river flow, which are all heavily influenced by abstraction for irrigation (Andreu et al., 2009). Alternatively, we can use a threshold determined from an undisturbed period or a naturalized model scenario, so using the situation that would have occurred without human activities as reference, as a natural normal (Fig. 4b, d: natural drought threshold; e.g. Van Loon and Van Lanen, 2013). The latter allows for a better identification of human-modified drought, both droughts enhanced and alleviated by human activities (Van Loon and Van Lanen, 2015). Because drought is an extreme event, the normal situation is not characterized by long-term average water levels or fluxes. Instead, a drought threshold (Fig. 3) is used that is calculated as a percentile(s) of a long time series (commonly, the value that is exceeded 80–95 % of the time) or return periods representing rare occurrence (for example, a 50-year drought). Some studies use a variable threshold calculated on daily, monthly, or seasonal timescales to represent seasonality and identify differences between droughts in different seasons (Van Loon, 2015). This is very relevant in the Anthropocene, because humans interact differently with droughts in different seasons. Water abstraction for irrigation, for example, also follows a seasonal pattern and has different effects on summer drought vs. winter drought. On the other hand, in monsoon climates, drought characterized by a prolonged dry season causes different socio-economic impacts than a below-normal wet season.

Drought is often confused with water scarcity and water shortage, which are defined as “less water than needed”, i.e. where demand is greater than supply (Table 1). The demand, or desired level, is included in Fig. 4 to illustrate the difference. In an unpopulated natural region, the desired situation is related to ecosystem requirements. Often these are not different from the normal situation because of the co-evolution of ecosystem and landscape. However, in a human-dominated region, the desired situation or water demand is dependent on population, standard of living, water efficiency, but also on climate. In many areas, the desired situation is out of balance with the normal situation, i.e. average water demand is higher than average water availability, because of rapid population growth, changes in diet, etc. This long-term imbalance leads to water scarcity (see Rijsberman, 2006, for a good overview of water scarcity definitions) and when it coincides with short-term drought it leads to acute water shortage (Table 1 and Fig. 4). If society satisfies its demand by abstracting more water, human-induced drought can occur in the short term (changing the actual situation) and overexploitation in the long term (changing the normal situation; Table 1 and Fig. 4).

Human-induced drought should also not be confused with the term “socio-economic drought” (Wilhite and Glantz, 1985, p. 115), which is used to denote socio-economic impacts of drought. Although socio-economic drought is often mentioned as a type of drought in scientific papers and on websites explaining drought to the general public, a clear distinction should be made between the physical lack of water (drought) and its socio-economic consequences (impacts of drought). These impacts are sometimes used to define the drought threshold (Fig. 3), which then reflects the water level at which ecological or socio-economic impacts are expected to occur, such as ecological minimum flow or minimum reservoir levels.

We have to point out that the definitions of drought and its impacts used here deviate from the definitions used in other scientific disciplines, in particular in the climate community. For example, in the IPCC SREX report, drought, as we define it here, is considered an “impact of extreme (weather or climate) on the natural physical environment” (IPCC, 2012, p. 40 and 167), whereas we see drought as a state of the natural physical environment that can cause ecological and socio-economic impacts. Similar confusion can arise for the terms “attribution”, “mitigation”, and “adaptation”, which are often assumed to be synonymous with attribution, mitigation, and adaptation of (anthropogenic) climate change, but can also be used for the attribution, mitigation, and adaptation of drought.

Drought is often seen from a meteorological perspective (Van Lanen et al., 2016), driven only by meteorological anomalies that disturb the normal water balance in a catchment (Fig. 2, left side). Given the significant human modifications of the terrestrial hydrological cycle, this is too simplistic a perspective (Box 1). If we take a hydrological perspective on drought, a lack of water compared to normal conditions can have a range of drivers (Fig. 2). There are many reasons for adopting a hydrological, rather than meteorological, perspective on drought. Firstly, people mainly use (sub)surface water, not rainfall directly (except for rainwater harvesting), so socio-economic impacts of drought are more related to a lack of (sub)surface water. Secondly, water on and beneath the land surface can be managed and manipulated, in contrast to rainfall, so that hydrological drought can be mitigated. Finally, the direct anthropogenic influences on hydrological drought are probably much larger than climate change influences in many areas of the world. If we adopt a hydrological perspective on drought, it is important to distinguish between the different drivers of drought. This distinction leads to more accurate drought prediction and helps to direct attention and allocate investments between adaptation to climate-induced drought and reduction of human-induced drought. However, distinguishing between climate-induced and human-induced drought is a major scientific challenge.

Human-induced droughts are recognized (Wilhite and Buchanan-Smith, 2005), but there is a large gap in our understanding of the development of human-induced and human-modified drought. We do know that human drivers principally influence soil moisture drought and hydrological drought and generally do not cause meteorological drought (Fig. 2; excluding relatively small-scale land surface feedback, e.g. due to irrigation (Tuinenburg et al., 2014), or the global, indirect effects of anthropogenic climate change). We can also hypothesize that the main process underlying human-induced and human-modified drought is abstraction from groundwater and surface water. There are many scientific studies on the long-term effects of abstraction (decades to centuries) but few on the temporal variability of abstraction on drought timescales (months to years). It is therefore still unclear how important human-induced and human-modified droughts are compared to climate-induced droughts for different areas around the world.

Research questions about drought drivers include

To what extent can observed historic drought events be attributed to different drivers?

Answering these questions requires quantification of the direct human drivers of soil moisture drought and hydrological drought, in absence of meteorological anomalies, for historical drought events. The approach would be to identify droughts in time series of observed hydrological variables and compare those to time series of climate-induced drought (represented by meteorological drought, observed droughts in an undisturbed nearby catchment, or simulated naturalized droughts). This last approach was used successfully in Australia (Van Dijk et al., 2013) and Spain (Van Loon and Van Lanen, 2013) and could be applied in other areas around the world to understand the variability in how human drivers impact drought. Naturalization of disturbed time series is challenging, being very much dependent on accurate modelling or regionalization approaches and data of human disturbances at a sufficiently high spatial and temporal resolution. Many international hydrological databases and data-sharing initiatives, however, have deliberately focused on near-natural systems (e.g. Hannah et al., 2011; Whitfield et al., 2012) in order to discern climate-driven processes from the noise of various human disturbances. We argue for more analysis of the disturbed catchments already included in hydrological databases and promote the extension of these databases with more human-influenced catchments, as suggested previously by Gustard et al. (2004). Perhaps the greatest obstacle to achieving this is the lack of metadata indexing the type and degree of human impact in any one catchment, which is often not known or poorly quantified. There is a pressing need for a bottom-up approach to transfer such knowledge, where it exists, from catchment-, regional- or national-scale archives to the international research community. We also call for more experimental catchments in human-influenced areas in which particular human influences on the hydrological cycle can be isolated and controlled, for example, within the Euromediterranean Network of Experimental and Representative Basins (ERB), the network of Critical Zone Observatories in the USA, and the TERestrial ENvironmental Observatories (TERENO) in Germany. Alternatively, we can make more use of satellite data of hydrological variables, which have become more widely available on the global scale, although still with high uncertainties (AghaKouchak et al., 2015b). Useful satellite products are soil moisture missions (SMAP, SMOS, AMSR-E II, ASCAT) for soil moisture information on high spatial and temporal resolution and NASA's Gravity Recovery and Climate Experiment (GRACE) for total water storage. If these are compared with global precipitation estimates (from satellites, TRMM and GPM, or from re-analysis), human-induced droughts might be identified in the absence of natural drought drivers.

The severity of droughts is strongly modified by catchment storage and release processes. In the natural situation these modifiers are determined by factors such as soil type, geology, and land cover (Fig. 2, left side). In the Anthropocene, human activities change storage and land properties influencing propagation processes, and modify drought severity directly through anthropogenic inflows or outflows of water (Fig. 2, right side). Just like natural modifiers, human modifiers can have both positive (enhancing) and negative (attenuating) effects on drought severity. The processes underlying direct modification of drought severity by human-influenced inflows or outflows of water are most recognized and understood, whereas the effects of human modification of storage and land properties, although recognized as potentially important, are more elusive.

There are ample examples of how human changes in land properties influence the hydrological cycle. Urbanization, for example, results in less infiltration and more runoff in some cases and in more recharge in others (due to leakage of water supply and sewage systems; Lerner, 1990). Deforestation, afforestation, agricultural practices, and desertification influence evapotranspiration and consequently soil moisture. Some studies focused on the effects of land use change on low flows (Tallaksen, 1993; Hurkmans et al., 2009), but there is very little quantitative research on how these processes influence drought severity and contrasting results are reported between modelling studies (Tallaksen, 1993; Hurkmans et al., 2009) and observation-based studies (Price et al., 2011; Eng et al., 2013).

Research questions about human modifications of drought include

How do human modifications of drought enhance or alleviate drought severity?

Besides new data, new methods are needed to disentangle human modifiers from natural modifiers of drought and quantify how large their effect on drought severity has been for historical drought events and might be for future events. When sufficient data are available, statistical methods, such as multiple regression analysis, can be useful in finding the statistical relationships between drought severity and multiple influencing factors. This approach was used by Van Loon and Laaha (2015) for natural drought modifiers, but can easily be extended to include human modifiers. Paired catchment statistical approaches (as applied to urbanization impacts on floods by Prosdocimi et al., 2015) or upstream (natural) – downstream (disturbed) comparisons (Fig. 5a; López-Moreno et al., 2009; Rangecroft et al., 2016) are other data-driven approaches, although these have yet to be applied extensively for drought and low flows. Another large-scale data analysis method that has great potential for use in drought research is comparative analysis (Wagener et al., 2007), which aims to find patterns by analysing a large set of catchments with a wide range of characteristics, both in terms of natural and human processes (e.g. Price et al., 2011; Eng et al., 2013; Sadri et al., 2016). This method is especially valuable if it is combined with qualitative data to explain the patterns found.

For scenario testing, conceptual models of human-water systems (Di Baldassarre et al., 2013, 2015) are a useful tool. Natural flows are altered by the presence of reservoirs and the resulting outflows depend on (changing) operational rules, i.e. optimized for flood or drought (Fig. 5b; e.g. Mateo et al., 2014). The conceptual model (Martinez et al., 2016) simulates how the occurrence of a flood event might lead to changes in operational rules (e.g. shifting from the “optimized for drought” to “optimized for flood” scenario in Fig. 5b), which will eventually enhance the next drought event (Di Baldassarre et al., 2016).

Modelling tools are also indispensable for prediction of drought under human modification. There are many types of models and many options to use these models for drought in the Anthropocene. Large-scale hydrological models are being adapted to include more anthropogenic processes (e.g. WaterGAP and PCR-GLOBWB; Wada et al., 2011; Döll et al., 2012; Nazemi and Wheater, 2015a; Veldkamp et al., 2015). Analysing these models specifically during drought periods has given some encouraging results (Fig. 5c; e.g. Van Lanen et al., 2004; Verbeiren et al., 2013; Wada et al., 2013; De Graaf, et al., 2014; Forzieri et al., 2014; Wanders and Wada, 2015), although model uncertainties during low flow and drought remain high. Since many human influences on the hydrological cycle are on the local scale, hyper-resolution modelling might be needed to explicitly represent all relevant human activities (Wood et al., 2011). For parameterization of these models, however, a thorough understanding of the processes is essential (Beven and Cloke, 2012). Most predictions on the local scale, however, are done with lumped or semi-distributed hydrological models. It is often not straightforward to incorporate dynamic human processes into these models and more work is needed to adapt these lumped hydrological models for use in the Anthropocene. An example of a new lumped hydrological model that incorporates man-made extraction and supply of water to both surface and subsurface water is WALRUS by Brauer et al. (2014). Current physically based models are better fitted to simulate human responses to drought, e.g. SIMGRO (Querner et al., 2008; Van Lanen et al., 2004). Once again, an important limiting factor is availability of data and metadata on the human modifiers. If information on human pressures is available, modelling can be a key tool in separating human and natural drivers (thus paving the way to attribution) through a multiple working hypotheses approach (see, for example, the work of Harrigan et al., 2014).

On the other side of the propagation diagram are the environmental and socio-economic impacts of drought (Fig. 2). Drought impacts, compared to the impacts of other hazards, are mostly non-structural and difficult to quantify. Drought impacts also have a high diversity, ranging across agriculture, water supply, industry, energy production, human health, aquatic ecology, forestry, and other sectors (Stahl et al., 2016). Impacts are sometimes characterized into direct and indirect or tangible and intangible impacts (Wilhite and Vanyarko, 2000). Thus, the quantification of drought impacts depends on the affected sector and on the level of impact (direct or indirect, and perhaps cumulative). Direct impacts on the agricultural sector are often documented as losses or reductions in crop yields. However, associating indirect economic losses directly to drought is not always straightforward (Ding et al., 2011). Indirect negative consequences are often quantified by the number of people affected or by number of people who died as a result of related food security or health issues, but factors other than a direct association to drought may play an important role as well. Especially drought impacts on (mental) health are complex and dependent on a multitude of factors (Stanke et al., 2013; Obrien et al., 2014).

Whether a drought event has negative consequences on one of these sectors also depends strongly on people's perception and thus on the vulnerability of affected sectors (Knutson et al., 1998; Iglesias et al., 2009). Understanding a particular sector's vulnerability can benefit from specific information and quantification of drought impacts in addition to knowledge on the general vulnerability factors that describe the sensitivity and adaptive capacity of the considered community or region.

For drought characteristics, ample data sources exist. However, as noted before, they rarely specify the level of human modification to the drought signal.

Research questions that need to be addressed thus include

How should drought impacts be monitored and quantified?

Retrospective analysis of the physical characteristics of past droughts (through some drought indicator) and the impacts that they have triggered have moved this search for a link function forward, especially if compared across different societal contexts, in particular different degrees of vulnerability. However, methods to link physical indicators and societal impacts have only recently been explored more in depth and still require more systematic appraisal. Figure 6 gives an overview of the different methods. The most widely adopted approach to relate drought indicators to impacts is to link commonly used hydrometeorological drought indicators to agricultural yield (Lobell et al., 2008; Vicente-Serrano et al., 2012, 2013; Bachmair et al., 2016a). Most of these studies are based on correlation and as summarized by Stagge et al. (2015a), thus are useful for screening relationships, but they measure the response of a variable, such as crop yield, across its entire range of values including typical or even productive years. A further complicating factor is the non-linearity of the climate-yield relation, which can show ambiguous relations with positive effects during drought or threshold behaviour for reductions in yield (Fig. 6a). Report-based impact data cover a wider range of impact types, but are tedious to gather and have many biases. So far, they have mostly been converted to binary or counts of “impact occurrences” for indicator-to-impact studies (Fig. 6b). Data-driven statistical models have used time series or spatial variability of these impact occurrences as a response variable in regression and classification tree models (Fig. 6c; Stagge et al., 2015a; Bachmair et al., 2016b; Blauhut et al., 2015). These studies have also shown that impact generation is more complex than previously assumed and may be caused by the co-occurrence of several extremes, lagged effects, and seasonality (Stagge et al., 2015a). A useful outcome of these modelling exercises was the objective determination of “best-indicators” for impacts in particular sectors that are strongly influenced by human factors. For example, when using the Standardized Precipitation Index (SPI) or Standardized Precipitation Evapotranspiration Index (SPEI) the best accumulation period suited to predict agricultural impacts clearly differs for irrigated and rain-fed agriculture (Stagge et al., 2015a); similarly, the best accumulation period to predict drought impact on public water supply differs depending on the relative contributions of groundwater vs. surface water resources and the type of reservoirs available (Bachmair et al., 2016b). These examples show that human perception of drought impacts can differ from the occurrence of drought in the natural hydrological system, depending on the prevailing water management framework and thus the vulnerability. Future analysis could use impact information to better characterize impacts of human-modified or human-induced drought.

The interaction between natural hydroclimatological processes and human influences is not a simple addition of both effects, but instead comprises complex and dynamic feedbacks resulting in a strongly non-linear response of the hydrological system (Fig. 2). There are negative feedbacks, when human management responses to drought (impacts) lessen drought; and positive feedbacks, where management responses exacerbate drought. There is growing knowledge of climate feedbacks (also called land–atmosphere feedback), in which drought influences evapotranspiration rates positively or negatively (Teuling et al., 2013), dependent on geographic situation and time frame. There is, however, only very limited understanding of human feedbacks during drought.

Short-term human feedbacks are responses to drought situations (whether observed, or at least perceived, or predicted) that influence water storages and fluxes within a particular water system in a catchment over timescales of days to years. These influences can include reductions in water use, implementation of water-saving technologies, planting of less water-demanding crops, using other water sources (e.g. from surface water to groundwater; from clean to grey water), short-term increases in groundwater abstraction because of surface water shortage, and water transfer from wetter areas or areas where water demand is lower (e.g. Andreu et al., 2005).

There is strong non-linearity in the reaction of the water system to these short-term influences (Sivapalan et al., 2012). Timescales often do not match; for example, the societal response might be in the order of weeks, but the reaction of groundwater can be in the order of years (Gleeson et al., 2010; Castle et al., 2014). Consequently, there is a difference between short- and long-term droughts, where longer droughts show a more complex interaction of natural and human processes (Van Dijk et al., 2013). Societies can also learn from historic droughts and adapt drought policy in the long-term to be more proactive, rather than reactive, when the next drought comes (McLeman et al., 2014). Crucially, however, human activities are not only influenced by climate and the drought state of the system but are also strongly dependent on domestic water behaviours (Pullinger et al., 2013), national policy styles (Gober, 2013), existing public policies (particularly for agriculture; Campos, 2015), water law and governance (Maggioni, 2015), and even indirectly by international food markets and geopolitics.

Research questions about human feedbacks include

Are there commonalities in the response of different societies to different drought events?

Qualitative data are essential in our quest for increased understanding of this topic. One novel type of qualitative data is the use of drought narratives (i.e. stories of historic drought events), which can give new insights into societal responses and feedbacks (e.g. Daniels and Endfield, 2009). This is an example of how citizen science can help harvest data. It is especially interesting to study paired drought events, i.e. drought events of similar magnitude that occurred in the same region, to investigate whether societies learn from drought events and what the effect of this learning is on the next drought. Despite the obvious uncertainties of such an approach, it can provide information on drought responses and feedbacks from one drought event to the next, as was shown for paired flood events by Kreibich et al. (2016).

For quantitative prediction of the effect of feedback on drought, water management models could be adapted to include more hydrology and feedbacks. The modelling tools that are used in water management generally take water availability as external forcing and do not include the feedbacks of the water use on the hydrological system (e.g. Higgins et al., 2008; Borgomeo et al., 2014). Like some global and lumped hydrological models mentioned before, many water management models are capable of simulating the effect of the allocation of water on hydrological processes also during drought, as was shown by Querner et al. (2008) and Van Oel et al. (2012), or simulating the influence of water management decisions on the evolution of a given drought scenario (e.g. Watts et al., 2012).

Socio-hydrology models aim to account explicitly for the two-way feedbacks between social and hydrological processes (e.g. Sivapalan et al., 2012). Di Baldasarre et al. (2013, 2015) have applied this approach to flooding, and the development of a similar modelling framework for drought is underway (Kuil et al., 2015). As the interplay between water and people is still poorly understood, socio-hydrological theory is still to be developed via an iterative process of empirical study, comparative analysis, and process-based modelling. Thus, while the current studies do contribute to improve the current understudying of water–society interactions, their predictive power is still very limited (Viglione et al., 2014). Modelling approaches are most successful when people themselves are actively involved in the modelling process; stakeholders can, for example, guide scenario analysis (Loucks, 2015). In contrast to modelling studies, environmental social science epistemologies, such as grounded theory building, offer alternative means of understanding water resource use and human behaviour (Pearce et al., 2013), potentially enabling more holistic insights into the role of drought feedbacks in the hydrosocial cycle (Linton and Budds, 2014, p. 170).

We now live in a fast-changing environment; both climate change and long-term human influences on the water cycle are changing the reference normal situation, even within 30-year time blocks that are traditionally being used to determine a climatology or a drought threshold. This is important from a drought perspective because the normal situation is our reference to determine the occurrence and severity of drought events (Fig. 4). There are many uncertainties in dealing with extreme events like drought under conditions of change. Some model studies of future hydrological drought commented on the assumption of using the same threshold for the historic and the future period (e.g. Giuntoli et al., 2015; Wanders et al., 2015). Two aspects should be mentioned. Firstly, regime changes trigger methodological considerations, because they can result in detection of drought events that should otherwise not be classified as drought, such as earlier snowmelt resulting in a drought in the normal snowmelt period (Lehner et al., 2006; Van Huijgevoort et al., 2014). Secondly, ecological and societal systems might adapt to a changing normal situation, but it is unclear how fast these adaptations will take place and whether tipping points will be passed (Mera et al., 2014).

Research questions related to a changing normal situation include

Is the normal situation actually changing or do we not have the data or understanding of natural variability to say anything about what is normal?

These methodological explorations on how to deal with changes of the normal situation in drought analysis are urgently needed, but we should also get a better understanding of long-term changes in the perception of drought impacts and vulnerability. This perception drives adaptation to extreme events like drought and influences feedbacks between the physical and social system. Societies might be able to adapt to a changing mean, but they are more likely to be triggered by extreme impacts of a severe drought, resulting in long-term adaptations aiming to reduce impacts of drought in the future (Fig. 7; Smit et al., 2000; Dillehay and Kolata, 2004). More research is needed to understand trajectories of social development that lead to adaptation to drought.

We can benefit from the work done on long timescales, regarding long-term climate change, long-term human influence on the water cycle (overexploitation), and long-term water demand and scarcity (Table 1). Research on groundwater depletion (Aeschbach-Hertig and Gleeson, 2012) and water scarcity (Rijsberman, 2006) has been carried out on large temporal and spatial scales (annual and country level), because that is the level of relevance and the level of available data. Accounting for temporal variability and increasing spatial resolution can close the gap with drought research (Savenije, 2000; Hoekstra et al., 2012; Hering et al., 2015; Vörösmarty et al., 2015). Veldkamp et al. (2015) and Mekonnen and Hoekstra (2016) were the first to explore sub-annual timescales of water scarcity.