A review of drought indices: predominance of drivers over impacts and the importance of local context

. Drought monitoring and Early Warning Systems (DEWS) are seen as helpful tools to tackle drought at an early stage and reduce the possibility of harm or loss. They usually include indices attributed to meteorological, agricultural and/or hydrological drought: physically based drought drivers. These indices are used to determine the onset, end and severity of a drought event. Drought impacts are less monitored or even not included in DEWS. Therefore, the likelihood of experiencing 10 drought impacts is often simply linearly linked to drivers of drought. The aim of this study is to evaluate the validity of the assumed direct linkage between drivers of drought and drought impact. We reviewed scientific literature on both drivers and impacts of drought. We conducted a bibliometric analysis based on 5000+ scientific studies in which selected drought indices (drivers) and drought impacts were mentioned in relation to a geographic area. Our review shows that there is a tendency in scientific literature to focus on drivers of drought, with the preferred use of meteorological and remotely sensed drought 15 indices. Studies reporting drought impacts are more localised, with relatively many studies focusing on Sub-Saharan Africa and Australasia for impacts with regard to food security and water security, respectively. Our review further suggests that drought-impacts studies are dependent on both the physical and human processes occurring in the geographic area, i.e. the local context. With the aim of increasing the relevance and utility of the information provided by DEWS, we argue in favour of additional consideration of drought impact indices oriented towards sustainable development and human welfare. this is to review scientific reporting on drought drivers and drought impacts for affected regions and analyse these two compare. This research aims to improve understanding on the linkage and separation between drought drivers and drought impacts, and to provide directions to further improve the accuracy of the information provided by DEWS. We retrieved scientific studies from categorised geographic areas in which selected drivers of drought and impacts of drought are mentioned. The components of drought drivers and impacts on which the literature focused were explored and compared for different areas of the world.


Introduction
Drought is a threat to a wide range of human activities in virtually all climate zones and countries (Van Loon et al., 2016;Bachmair et al., 2016;Van Lanen et al., 2017). It is an elusive phenomenon without a clear onset and demise. In contrast to other hazards such as floods, landslides or earthquakes, drought has a creeping nature causing impacts to persist for many years (Kim et al., 2019). Consequently, impacts can be cumulative for consecutive periods of droughts, devastating both 25 ecosystems and societies (Bachmair et al., 2016;Van Lanen et al., 2017).
Many concepts exist for defining a drought (Santos Pereira et al., 2009;Lloyd-Hughes, 2014). Definitions of drought are either conceptual or operational. Conceptual definitions of drought are descriptive and highlight the natural hazard element: for example, precipitation below what is expected or normal (Knutson et al., 1998). Operational definitions of drought highlight https://doi.org/10.5194/nhess-2021-152 Preprint. Discussion started: 17 June 2021 c Author(s) 2021. CC BY 4.0 License.
practical implications in an attempt to identify the onset, severity, and cessation of drought periods (Mishra and Singh, 2010). 30 For example, the UN Convention to Combat Drought and Desertification (UN Secretariat General, 1994) defines drought as "when precipitation has been significantly below normal recorded levels, causing serious hydrological imbalances that adversely affect land resource production systems" The numerical value of hydro-climatic variables is associated to three main types of drought: meteorological, agricultural (or soil moisture) and hydrological droughts. These variables are in fact drivers, which refer to the contributing or counteracting 35 factors that affect the development of droughts (Seneviratne, 2012). Those drivers are used by many drought studies as the framework to represent drought propagation. The temporal propagation of drought is often considered to be a sequence occurring in an almost linear order (Wilhite and Glantz, 1985;Zargar et al., 2011;Bachmair et al., 2016), and in which humans have no direct influence. This is a simplification of a complex process, where it is considered that an anomaly (e.g. lower precipitation, higher temperature than average) or a standardised normalisation of the values of those drivers will lead to a 40 cascade reaction influencing the magnitude of other physical variables and leading in turn to the subsequent type of drought.
As such, hydrological drought results from persistence in duration of agricultural (soil moisture) drought, which itself is due to persistence of meteorological drought.
Drought monitoring and Early Warning Systems (DEWS) aim to monitor the drivers of drought to predict drought. They aim to tackle drought at an early stage to reduce the possibility of harm or loss. For assessing the severity of a drought, physical 45 variables are usually translated into indices of drought. The difference between their values and the threshold used to define the level of dryness is considered to depict the severity of a drought (Vogt et al., 2018). Drought impacts, such as water-and food security, are rarely continuously monitored or even included in DEWS. This is understandable as there is already a plethora of definitions for drought and drought types, and there are at least as many possibilities for defining impacts (Mishra and Singh, 2010;Wilhite, 2000;Santos Pereira et al., 2009). Drought impacts are non-structural, difficult to quantify or 50 monetise, and can be direct or indirect due to the extended nature, in time and area, of drought (Wilhite et al., 2007;Logar and Van den Bergh, 2011;Bachmair et al., 2016). But in the current configuration of DEWS, the presumed likelihood of experiencing impacts is mainly linked to the severity of climatic features only (e.g. :(US National Drought Mitigation Center;Princeton Flood and Drought Monitors;Brazilian Drought Monitor).
The aim of this study is to review scientific reporting on drought drivers and drought impacts for affected regions and analyse 55 how these two compare. This research aims to improve understanding on the linkage and separation between drought drivers and drought impacts, and to provide directions to further improve the accuracy of the information provided by DEWS. We retrieved scientific studies from categorised geographic areas in which selected drivers of drought and impacts of drought are mentioned. The components of drought drivers and impacts on which the literature focused were explored and compared for different areas of the world. 60 https://doi.org/10.5194/nhess-2021-152 Preprint. Discussion started: 17 June 2021 c Author(s) 2021. CC BY 4.0 License.

Methodological approach
The methodological approach comprises three steps: Step 1. Exploring which drought drivers are the most recurrent in the scientific literature. We investigated which indices of drought drivers are most frequently used in scientific drought-related studies and to what drought type they were linked. For 65 each of these scientific studies we also retrieved the country of focus. This allowed us to identify: the most frequently mentioned type of drought for different geographic regions, and the prevalent drought indices used in scientific studies.
Step 2. Exploring which drought impacts are the most recurrent in the scientific literature. In contrast with drought drivers, for drought impacts there are no established indices commonly used in DEWS and in scientific studies. We thus retrieved from scientific articles, keywords associated to drought impacts related to water security and food security. This allowed the 70 identification of the most frequently mentioned water-and food-related drought impacts.
Step 3. Comparing the findings of Steps 1 and 2. This enabled evaluation of the alignment between reported drought types and impacts, with regard to the number of publications and differences in geographic focus.

Data
We considered the number of studies about drought indices and drought impacts, respectively, and their geographical 75 distribution as our metrics. We selected commonly used indices to depict operational types of droughts  and the indices commonly used by water managers (Bachmair et al. (2016). This resulted in 32 indices that we linked to three main drought types (Table 1): meteorological (9 indices), soil moisture/agricultural (15) and hydrological (8)  We opted for Scopus to retrieve the scientific publications of interest as it is the database covering the largest range of both, peer-reviewed literature type (scientific journals, books and conference proceedings), and disciplinary fields (science, technology, medicine, social sciences, and arts and humanities ) (Scopus, 2021). We then searched in the Scopus database for 85 queries strictly including "drought" AND "[the indicator]" in the title, abstract and authors' keywords of the studies. We repeated the queries for each indicator individually as we were interested in knowing country-based preferences. The sum of the individual indices linked to drought queries returned 4137 articles for the "meteorological" drought type of indices, 2799 articles linked to "agricultural" drought and 393 articles linked to "hydrological" drought. The title, authors, author's keywords, year of publication, journal name and abstract were retrieved using the Bibliometrix package (Aria and Cuccurullo, 2017) 90 executed on R (version 4.0.0) following Addor and Melsen (2019). In the title, keywords and abstract of each paper, names of countries were identified, corresponding to the area of application of the study. The same approach was followed for the drought impacts. We grouped drought impacts into two focus categories: food security and water security. Their keywords are indicated in Table 1. The queries included "drought" AND selected "[drought impact]". This resulted in 4764 articles linking drought to food security and 805 articles linking drought to water security. 95 All articles were published between 1960 and March 2021 and the exact queries for both drought indices and impacts are included in Table A1. Even though we recognise drought can impact ecosystems, this topic was excluded from the analysis for reasons of brevity. The dataset and the script used for its analysis are both available for consultation (Kchouk et al., 2021).
Many scientific studies are methodological; their goal can be the validation, calibration or improvement of the indices, thus, not all studies have a focus country. We only considered studies mentioning a country in their title, abstract and keywords; 100 this being the only criteria of inclusion or rejection of papers in our analysis. This reduced the number of studies including a name of a country in their title, abstract and keywords by 28% for drought indices and by 44% for drought impacts. We also did a manual verification on some of the scientific studies to see if the association with a country was valid. This allowed us to bring some corrections to the metadata to avoid incorrect associations (e.g. removing mentions of the "Indian Ocean" that led to the incorrect association of the studies to India ; removing the copyrights, generally in the end of the abstract, referring 105 to another country than the one of the study).

Drought types and indices
The indices mentioned in the drought-related studies were classified according to the categories used in Table 1

125
In addition, not only are MD indices the most investigated, they are also the most associated with a country in studies, in comparison to AD, HD and impacts (Table 1). MD indices represents 53 % of the scientific studies while AD represents 42 % and HD, only 5 %. This indicates that in most of the studies, rainfall and the temperature are the dominant criteria utilised to report the occurrence of drought. Such a result is expected because of the ease of use of MD indices. We further develop this point in Sect. 4.3. 130

Drought-related impacts: food security and water security
Globally, there were five times more studies linking drought to food-security than drought to water-security (Fig. 3). This pattern is the same for most areas of the world. For Sub-Saharan Africa the predominance of food security indices is most pronounced (93%), followed by Asia and Europe (84%). Australia-Oceania is the only region where drought-related water security studies predominate over food security studies (52%), while Sub-Saharan Africa is the region where it is reported the 135 least (6.6%).    In the same way, two geographical patterns appear in the share of drought-related impacts studies. The height of the boxes of SSA and Australia-Oceania for food and water securities, respectively, related to drought is significantly superior to those of 150 the other regions for the same indicator category. This means that food security related to drought is most frequently reported for SSA and that water security related to drought is most frequently reported for Australia-Oceania. Similarly, drought-related water security is least reported for Europe.

Geographic patterns for indices of drivers and impacts
The geographical pattern of drought drivers and impact studies seen in Fig.4 is also present in the cartogram representations in Fig. 5. First, the three drought drivers categories appear to have the same pattern of investigation, all mostly focused on 155 northern high-income countries. The United States and Mexico, North-Mediterranean countries and Australia-Oceania are strongly focusing on drivers in drought-related studies. Middle-income countries with high demographic and economic growth such as China, India and Iran also see a focus on drought-related drivers. They stand out from their geographic neighbours that are almost disappearing from the map.  distribution of agricultural and soil moisture drought studies appears to be more even in African countries, and higher in Sahelian countries.
Looking at the geographical repartition of drought-related impacts studies ( Fig. 5d and 5e), two main observations are notable.
First, the repartition of the impacts studies differs from the drivers studies. Second, both impacts, food and water security, 170 show a different geographic pattern. Water security related to drought is most frequently investigated for Australia, the USA and Mexico, Brazil, the Middle East and South Africa. In contrast, food security is most commonly investigated for India, Ethiopia, Kenya and other African countries.

Discussion
This bibliometric study shows that unbalanced attention is given to drought drivers and impacts across the world. In this 175 discussion section, we start by raising four hypotheses to explain why some features of drought are more frequently reported for some regions or countries than for others. The four hypotheses relate to: physical conditions (Sect. 4.1), socio-economic conditions (Sect. 4.2), data availability (Sect. 4.3), and scientific interests and orientation (Sect. 4.4). We continue by discussing potential limitations in our methodological approach (Sect. 4.5). We posit that these four hypotheses are also the four dimensions that are inherent to the local context of a geographic area. Drought monitoring is influenced by these to accurately 180 predict droughts, their severity and impacts. In that sense, we end by formulating recommendations (Sect. 4.6) about shifting the scope of drought metrics to match the local context of a specific drought events.

Physical conditions
The most notable result from Sect. 3 is the more abundant investigation of MD over  Oceania), with the SPI being the most used indicator in drought-related studies.
By focusing on MD, it is mainly the deficit of precipitation that is investigated. In humid areas, tropical, continental or temperate climates, a deficit of precipitation is less likely to affect the overall physical water scarcity and cause water shortage.
In that sense, the occurrence of a drought is only statistically-based and not reflecting a true water deficit for the demand, only a below average situation (which is, however, in line with formal definitions of drought). In arid and semi-arid climates with 190 lower levels of precipitation, it is recommended to use SPI cautiously because it can fail to indicate drought occurrence (Wu et al., 2007) and opt instead for indices that include evapotranspiration like the SPEI (Salimi et al., 2021). In such areas where evapotranspiration plays a larger role with regard to evaporative demand, water shortage is more common. For arid and semiarid areas with low average rainfall and a higher risk of water scarcity, it may be more appropriate to determine water deficit at the crop, field or farm scale. This could explain the more frequent use of AD indices in the more arid  and Sub-Saharan regions (Fig. 2 & 4) that mainly monitor vegetation (NDVI, LAI) and soil water content (SWS) (Fig.1). For some AD indices, there is both an upper and a lower limit that is independent of whether the climate of the area is arid or humid: vegetation health or soil water content are or are not frequently deteriorated or in deficit, respectively. In that sense,

Socio-economic conditions
SSA combines the lowest number of studies about drought indices with the highest proportion in terms of drought impacts 205 (Fig. 4). Even though SSA is known to experience a rise of temperatures and an increase of aridity in the past, present and future by observation and model projections (Niang et al., 2014;Serdeczny et al., 2017) the reported impacts in the Emergency Database (EM-DAT) are scarce (Harrington and Otto, 2020). Yet, the International Disaster Database (EM-DAT) run by the Centre for Research on the Epidemiology of Disasters (CRED), has the most complete and global records of past natural and human-made disasters events (Guha-Sapir et al., 2012). 210 Most of SSA is in a situation of economic water scarcity (Molden, 2013), implying a lack of human, institutional and financial capital to satisfy the demand for water, even in areas where the physical availability of water is not limited. The symptoms described by Molden (2013) associated to economic water scarcity include scant infrastructure development, either small or large scale, meaning that populations experience difficulties obtaining sufficient water to meet agricultural or domestic needs.
Applying the same reasoning, drought mitigation or monitoring bodies and scientific publications are a product of human, 215 institutional and financial capital. Thus, it is likely that drought drivers are under-investigated in SSA, leading to the same effects of economic water scarcity: water and food insecurities. Also, the report of impacts of extreme weather in SSA to disaster databases as EM-DATA are predominantly conducted by non-governmental organisations rather than governments, often as a side product of their main task to identify the location with the greatest need for humanitarian aid (Harrington and Otto, 2020). 220 In some areas, food insecurity can be a cumulative result of a dry climate and high pressure on natural resources enhanced by rapid demographic growth. Countries such as Bangladesh, China, Ethiopia, India, Indonesia and Pakistan, have some of the highest number of drought-related food security publications (Fig. 5). Most of these countries have high fertility rates and rapid population growth (United Nations, 2019; Vollset et al., 2020). According to the Food and Agriculture Organization food production depends on croplands and water supply, which are under strain as human populations increase. This suggests that countries with arid climates and a high demography are more exposed to food security impacts. 230 Moreover, populations of low income countries are the most exposed to drought-related food insecurity. In the world's poorest countries, around 30 percent of GDP comes from agriculture; those countries are mostly concentrated around the Sahelian region: Mali (37.4% of GDP), Niger (35.4%), Chad (46.1%), Central African Republic (31.9%), Sudan (31.2%), Kenya (31.1%) and Ethiopia (34.7%) (World Bank, 2016). As we can see from Fig.5 FS, those countries are most commonly reporting food security impacts related to drought. In contrast, in OECD economies -regarded as developed and high-income countries 235 agriculture accounts for less than 1.5 percent of GDP (World Bank, 2016). In the same way, we note the fewest amount of publications related to food security in those OECD countries. Also, in these Sahelian countries, agriculture accounts for more than 80% of the livelihoods (FAO, 2021). As more people rely on agriculture for their livelihood, they are more exposed to natural hazards like drought and thus vulnerable to food-insecurity and the poverty trap.

Data availability 240
The SPI is the most widely used index in drought-related studies (Table 1 and Fig. 1). This can be explained by its ease of use: First, it only requires (monthly) precipitation data, easy to monitor by use of rainfall gauge networks or satellite estimation.
Second, SPI reference values exist so they can be compared and are applicable in all climate regimes. Finally, SPI can be computed for different periods of time including periods of record containing missing-data, even though it ideally needs at least 30 years of monthly precipitation data (WMO, 2012). 245 However, all these strengths are at the same time weaknesses. The SPI will provide in all cases an output whatever inputs are used (Svoboda and Fuchs, 2016). As an example, a significant quantity of zero precipitation values at short time scales may lead to biased values of the SPI, because the rainfall might not fit for the recommended gamma distribution, which is a fundamental first step of the SPI calculation (Wu et al., 2007). This scenario is applicable to dry climates with a distinct dry season when calculated for periods shorter than 12 months. As mentioned in section 4.1, an index including an additional 250 temperature parameter to account for evapotranspiration is more suitable for such areas. As we can see in Figure 6, many countries with dry climates (Iran, Australia and Pakistan) commonly use the SPI in their drought-related studies. In those dry contexts, it has been proposed to focus on the duration of the drought rather than only its severity (Wu et al., 2007). However, even short-lived dry spells often combined with heatwaves of a few days, characteristic of dry climates, when occurring during the reproductive stage of crop development can be enough to ravage an entire harvest leading to food insecurity (Hatfield and 255 Prueger, 2015).  represented, areas of greater aridity are severely underrepresented (Walker et al., 2016). Consequently, reanalysis rainfall products are also less reliable for these more arid regions due to a lack of ground truthing data (Walker et al., 2016). The availability of data seems to be closely tied with the socio-economic condition of a country. As mentioned in Section 4.2, countries exposed to economic water scarcity generally experience a lack of capital to satisfy the demand for water and a lack of an extensive and well-maintained hydro-climatic monitoring network. Therefore, most of the countries of SSA are 270 underrepresented or absent from publications related to drought indices, while high-income countries commonly report them (Fig.5).
The same applies for HD indices studies that are under-reported in SSA (Fig 3, 4 & 5). River flow monitoring networks in SSA are experiencing a similar decline to meteorological monitoring networks (Walker et al., 2016). However, globally, little attention seems to be given to the monitoring of HD indices (Fig. 1 &2). One reason could be that a "hydrological" drought is seen by many as a drought impact, and there is a demonstrable lack of extensive monitoring of drought impacts (Lackstrom et al., 2013;Pozzi et al., 2013;Wilhite et al., 2007).
As Table 1 shows, NDVIa remotely sensed indexis the most commonly used in agricultural drought-related studies. Only 3 out of the 15 agricultural drought indices are not remotely sensed. Just like the HD indices, this can reflect (i) the lack of hydrometric (field) observations or (ii) if they exist, a lack of sharing and access to them (Bachmair et al., 2016). Bachmair et 280 al. (2016) highlight how "the scarcity of water status observations, especially for groundwater, reflects the common focus on drought seen through the lens of rainfall and soil moisture that can be easily (remotely) monitored and/or modelled". Indeed, the data needed to calculate AD indices seem more accessible. The most used index is the NDVI and requires land surface imagery containing both red and infrared bands and processing software; global NDVI datasets are available open source at relatively high spatiotemporal distributions. As there are no requirements for historical data for calibration or a monitoring 285 network, this could explain why the African continent more prominently reports AD than MD and HD (Fig. 5).
It is important to realise that data availability may be closely tied to the year of implementation of the drought indices. Indeed, hydro-climatic databases have different ages and dataset quality according to the country, but it can also be possible that the implementation of drought indices is a precursor of hydro-climatic data monitoring.

Scientific interest and orientation 290
As mentioned previously, in DEWS, the indices linked to the three categories of drought are seen as drivers as they are used to determine the occurrence and severity of a drought. However, as shown in Sections 3 and 4.3, the distinction between drought drivers and impacts, based on hydro-climatic variables, is not always clear. First, the linear representation of drought implies that AD and HD are an impact of MD. Yet the indices used for MD have a different scope to those used for AD and HD. Taking the example of the most used indices, the SPI has a temporal focus with a strong statistical perspective on drought. 295 Whereas for AD, the NDVI has a "spatial distribution" focus as it uses remote sensing to indirectly determine water-limitation in the vegetation at a specific time, like a snapshot of the vegetation health. In that sense, the NDVI measures a drought impact.
Moreover, water security is often confounded with hydrological drought. However, as we can see from Fig. 5d and Fig. 5e, the areas where each HD and WS are reported in scientific studies are not the same, suggesting that the occurrence of the first does not imply the other. In that sense, the literature seemingly indicates that HD is not the only driver of WS. 300 The scientific reporting about drought suggests its risk of occurrence in an area and potentially an initiative of preparation for related damages. Though for each country, it is likely that drought is investigated according to: (i) a determined scientific approach, more physical or social; (ii) a purpose, in the sense of what is at greatest risk of being impacted by drought.
As shown in Table 1, most of the drought-drivers indices are investigated under the domain of environmental, Earth and agricultural sciences, suggesting a more physically-based approach. Food and water securities related to drought, respectively 305 more reported in SSA and in Australia-Oceania (Fig. 5), are also studied through the scope of physical sciences but unlike the drivers, also through the lens of social sciences (Table 1). Food security is a complex concept that is looked at in a holistic way. Food systems underpin food security and they are the result of the production, processing, distribution, preparation and consumption of food. These steps are themselves the results of dynamic interactions between and within the bio-geophysical and human environments (Gregory et al., 2005). Thus, its 310 study requires the intervention of different specialists. Food systems encompass three main components : "(i) food availability (with elements related to production, distribution and exchange); (ii) food access (with elements related to affordability, allocation and preference) and (iii) food utilisation (with elements related to nutritional value, social value and food safety)" (Gregory et al., 2005). Hence, when food systems are stressed, food security is affected. As food security depends on many components, it stands vulnerable to the disturbance of any of them. These components can be disturbed by a range of factors 315 that can be environmental, like droughts, but also circumstantial like conflict, changes in international trade agreements and policies, HIV/AIDS (Gregory et al., 2005). Food insecurity can be enhanced when these factors are combined. SSA is an area particularly prone to extreme heat-related impacts heatwaves, as we mentioned in Sect. 4.2, but also to these circumstances. all classified as in fragile and conflict-affected situations (Corral et al., 2020). Thus, food security related to drought studies in SSA may also be related to the implication of these social processes.
Australia, known to be the driest inhabited continent (Hill, 2004), has a "National Plan for Water Security"(Government of Australia, 2007) that comprises a variety of mechanisms addressed by national and state governments (Cook and Bakker, 2012). Water security is also aimed to be addressed in an integrative and multi-scale way by "taking action on climate change, 325 using water wisely, securing water supplies and supporting healthy rivers and wetlands"(Government of Australia, 2007).
Besides Australia, the fact that water security is reported for countries with extreme differences in socio-economics, such as countries in the Sahel and the USA (Fig. 5), suggests the experience of different types of water security. The definition of "water security" by UN Water (2013) is quite holistic. A population's access to adequate quantities of acceptable quality water has the goal to sustain three areas: livelihoods, human well-being, and socio-economic development (Montanari et al., 2013). 330 Countries at different stages of development are more likely to focus on one of those three areas. Human well-being related to water-security can have many different understandings (Jepson et al., 2017;Hoekstra et al., 2018). Those can vary from one extreme to the other, as enough water for sanitary purposes, e.g. sanitation and showers, to indulgent leisure (E.g. swimming pools and gardens (Savelli et al., 2021;Bradley and Bartram, 2013;Willis et al., 2010)). In South Africa, experiences of Cape Town Day Zero's water crisis were diametrically different amongst the wealthy elite and the township dwellers. The first went 335 through restrictions to water their garden and fill up their swimming pools while the second had insufficient water to take showers and go to the toilet (Savelli et al., 2021). Livelihoods and socio-economic development can also be understood and applied in different ways: from subsistence farming (Makurira et al., 2011) to agrobusiness and irrigation of crops meant for export (e.g. California (Morris and Bucini, 2016)). The same can apply to food security: from malnutrition (Belesova et al., Therefore, not only can areas be exposed to food and/or water insecurities, but they can be exposed to different declinations and severity within each. Water and food insecurities are very context specific, not even attributable to the country scale but to smaller areas. They are the result of complex and multi-disciplinary mechanisms, including social processes in addition to the physical ones. Thus, to be accurately monitored, drought-related water and food insecurities also need multi-disciplinary metrics. This comes in contradiction with drought indices that measure drought severity by looking only at the hydro-climatic 345 component. Consequently, by eluding (the monitoring of) social processes that can trigger and enhance drought impacts, drought indices seem to be formulating an incomplete forecast of the severity of droughts.

Limitations
The inability to deduce a cause-and-effect relationship between two variables, solely on the basis of an observed association or correlation between them is common to all disciplines. The same applies for drought drivers and drought impacts even in 350 drought prone areas. It is difficult to state if droughts are the cause of impacts, aggravator of existing impacts, or if those "impacts" were pre-existing conditions and not linked to the occurrence of drought events. Without continuous and widespread monitoring of drought impacts, the societal pattern enabling understanding of how drought is experienced differently and why, will not be identified. Therefore, the attempt of explaining the geographical repartition of drought-related impact studies by linking some features of drought to one or many of the four hypotheses detailed above, as per this study, 355 remains then purely hypothetical.
Our approach separated studies by geography, principally at sub-continental scale. Other divisions on which to base our analysis could have been applied, like climatic or income levels, and may have led to additional insights. However, separating studies by geographical region allowed highlighting of: (i) both physical and socio-economic similarities expected in homogenous geographic areas; (ii) countries standing out. This enabled the investigation of potential justifications. 360 Disparities exist inside countries, where, for example, areas are wealthier than others or experience different climates. This applies particularly to larger countries such as the United States, China, Brazil and India. Therefore, it is possible that in some large countries, more prosperous areas might be centralising the background work and the studies on drought indices. It is also possible that physical, socio-economic, data availability and interest disparities may also exist inside countries. However, because our drought metrics investigation and analysis is at the country level, our discussion is also generalised to that scale. 365 Getting rid of that aggregative propensity and grasping those regional disparities would have required an investigation at the scale of within-country regions (e.g.: California Central Valley, Brazilian semi-arid). Yet, it is mostly the name of the countries that are used in publications on Scopus. Moreover, that level of detail and analysis would be more appropriate for comparative studies between chosen semi-arid regions of the world rather than a broader study, like this one, where similar focus on drought and drought impacts indices are examined. 370 The studies we obtained and analysed were a result of using Scopus, rather than another abstract and citation database, and of how we formulated our queries. Our search was constrained to articles having their title, abstract and keywords in English, potentially excluding important articles written in other languages. Additionally, the queries of the drought drivers were per indices, individually, while the queries of the impacts were regrouped by two themes. We justified the approach of grouping drought impacts keywords due to the lack of metrics existing for water and food insecurities related to drought, as is the case 375 for drought drivers indices.

Recommendations
It has to be recognised and highlighted that DEWS have achieved the goal of providing timely and reliable information to decision makers for drought management and mitigation. The value of this study being to increase the relevance and utility of drought-related variables, we also acknowledge that the metrics they rely on are mostly conceptual and descriptive which 380 contradicts their operational purposes. Their structure also tends to exclude the human influence on drought and drought influence on humans. The emphasis is on the natural effects on the hydrological system. Subsequently, the accuracy and efficiency of drought mitigation measures can be sub-optimal if it is based only on information lacking consideration of observed (local) drought impacts.
A recommendation would be to also consider a drought mitigation focus shifted on the optimal human welfare that drought is 385 obstructing, rather than only measuring on which hydrosystem compartment there is a deficit. In humanitarian approaches, a human welfare approach makes sense as the addressed disaster damages, in the short and long-term, can adversely affect basic human safety through malnutrition, displacement, livestock or even human mortality. But this can also be applicable in drought management. Indeed, there is a lack of consensus in defining a drought and its impacts resulting in difficulty on agreeing on coherent and accurate drought metrics. Therefore, shifting the focus of drought mitigation to observable, graspable and 390 quantifiable goals, such as human welfare, could overcome the uncertainty around drought and drought impacts definition.
In a drought mitigation perspective, the human welfare proxy could be considered as an optimal situation without water shortage, e.g. zero hunger, poverty, conflicts and water insecurity. Thus, it could be aligned with the Sustainable Development Goals (SDG) as they (i) represent the development priorities of both low-or high-income countries; (ii) benefit from existing and improvable metrics. Also, similarly to drought indices, SDGs have a global nature inclined to overlook the local context. 395 By taking into account local particularities, the SDGs could be reached at the local level even it is through a drought mitigation scope. Instead of the linear and still conceptual driver-focused "meteorological-agricultural-hydrological" droughts, the disaster scope could shift to more societally relevant goals linked to "poverty, water security, and food security". Thus, operational approaches of drought management would be the equivalent of determining the extent to which drought is hampering the achievement of one or many of these defined goals. 400 Some studies have already been arguing in favour of considering other approaches than the two main top-down and bottomup approaches for climate change adaptation strategies (Ludwig et al., 2014;Conway et al., 2019). Both approaches come with their strengths and weaknesses and conciliating them represents a challenge and many complexities, often unsuitable for integrating into water management (Ludwig et al., 2014). The issues complicating the decision-making are well known: the https://doi.org/10.5194/nhess-2021-152 Preprint. Discussion started: 17 June 2021 c Author(s) 2021. CC BY 4.0 License.
top-down approach is too broad and presents too much uncertainty; the bottom-up approach focuses too much on socio-405 economic vulnerability and too little on developing (technical) solutions (Ludwig et al., 2014). Thus, a risk-oriented approach that focuses more on "systems of receptors rather than conventional sectors" (Warren et al., 2018), where research identifies vulnerability to different extreme events rather than only analysing their probabilities of occurence (Bliss and Bowe, 2011), is an alternative.

Conclusions 410
We conducted a bibliometric analysis on 5000+ scientific studies in which drought was associated to an index and water and food securities, with the aim of comparing how drought drivers (e.g. precipitation, temperature, evaporative demand) and drought impacts (food and water insecurities) were reflected in the literature. Our results revealed that drought is mainly depicted through a conceptual lens, focusing on precipitation-based and remotely sensed indices. It is the SPI, a single-variable index, that is the most broadly used in different climatic and geographic contexts, despite being the one including the least 415 local contextual information. Drought is regularly approached merely as a rainfall statistical anomaly and equated to meteorological drought.
Drought drivers studies tend to focus on particular geographical regions, especially northern countries, whereas studies reporting impacts related to food and water securities are more commonly located in Sub-Saharan Africa and Australia-Oceania respectively. Moreover, the areas where drought drivers are reported in scientific studies are different from the drought impacts 420 ones. There is also a difference in the geographic repartition of drought-related food security and water security scientific studies. This suggests that drought-impacts studies are certainly dependent on both the physical and human processes occurring in the geographic area, i.e. the local context.
Because "local context" can have different meanings, we raised four hypotheses that can be attributed to local context and that can contribute to drought drivers resulting in drought impacts. First, the physical availability of water; drought drivers indices 425 measure the water deficit in one or several of the components of the hydrological cycle, implying that the severity of drought is the same in arid or humid climates. Second, the socio-economic conditions in the countries, as the income per capita and the demography that affect, respectively, the capital involved in research and the vulnerability to hazards. Third, the data availability, related to the second point concerning socio-economic conditions, affects the selection and accuracy of an index, especially if the chosen index is unsuitable for the particular climate. Fourth, the scientific approach and the interest in the 430 country that determines from which physical and/or social sciences scope drought will be looked at and for what purpose. It seems that drought impacts are considered more through social sciences lenses than drought drivers. Drought drivers indices seem to remain conceptual metrics depicting climate features and do not seem to be linked to human-centered solutions. Also, both water and food securities are scientific concerns mostly in arid and semi-arid regions, from high to low income and whether drought drivers are investigated or not. This suggests many variants of the same type of impact according to what or 435 who is likely to be most impacted by drought in the area. Thus, more research is needed where the scope of drought mitigation is widened to the vulnerability to drought events rather than only their probability of occurrence. DEWS would then more accurately predict the severity of a drought by also including drought indices that are people-centered. In this way, drought metrics would also better align with SDSs. These drought metrics could become more useful in monitoring the negative role of drought in achieving human welfare, and with that, the SDGs.

Code and data availability 445
Both code and data are available in the 4tu.ResearchData platform.

Author contribution
SK has designed and conducted the research in collaboration with DWW, supervised by LAM and PRvO. SK has written the 450 manuscript with input from all co-authors. The final version has been approved by all co-authors.

Competing interests.
Authors have declared no competing interests.

Acknowledgements
This work is part of the research program Joint SDG Research Initiative with project number 07.30318.016, which is (