Droughts continue to affect ecosystems, communities and entire economies. Agriculture bears much of the impact, and in many countries it is the most heavily affected sector. Over the past decades, efforts have been made to assess drought risk at different spatial scales. Here, we present for the first time an integrated assessment of drought risk for both irrigated and rainfed agricultural systems at the global scale. Composite hazard indicators were calculated for irrigated and rainfed systems separately using different drought indices based on historical climate conditions (1980–2016). Exposure was analyzed for irrigated and non-irrigated crops. Vulnerability was assessed through a socioecological-system (SES) perspective, using socioecological susceptibility and lack of coping-capacity indicators that were weighted by drought experts from around the world. The analysis shows that drought risk of rainfed and irrigated agricultural systems displays a heterogeneous pattern at the global level, with higher risk for southeastern Europe as well as northern and southern Africa. By providing information on the drivers and spatial patterns of drought risk in all dimensions of hazard, exposure and vulnerability, the presented analysis can support the identification of tailored measures to reduce drought risk and increase the resilience of agricultural systems.
Droughts exceed all other natural hazards in terms of the number of people affected and have contributed to some of the world's most severe famines (FAO, 2018; CRED and UNISDR, 2018). Drought is conceived as an exceptional and sustained lack of water caused by a deviation from normal conditions over a certain region (Tallaksen and Van Lanen, 2004; Van Loon et al., 2016). It can have manifold impacts on social, ecological and economic systems, for instance agricultural losses, public water shortages, reduced hydropower supply, and reduced labor or productivity. While many sectors are affected by drought, agriculture's high dependency on water means it is often the first of the most heavily affected sectors (Dilley et al., 2005; UNDRR, 2019). With nearly 1.4 billion people (18 % of the global population) employed in agriculture, droughts threaten the livelihoods of many and hamper the achievement of the Sustainable Development Goals (SDGs) – notably SDG 1 (no poverty), SDG 2 (zero hunger), SDG 3 (good health and well-being) and SDG 15 (life on land). While there is ambiguity regarding global drought trends over the past century (Sheffield et al., 2012; Trenberth et al., 2013; McCabe and Wolock, 2015), drought hazards will likely increase in many regions in the coming decades (Sheffield and Wood, 2008; Dai, 2011; Trenberth et al., 2013; Spinoni et al., 2017, 2019b; UNDRR, 2019). Identifying pathways towards more drought resilient societies therefore remains a global priority.
Recent severe droughts in southeastern Brazil (2014–2017), California (2011–2017), the Caribbean (2013–2016), northern China (2010–2011), Europe (2011, 2015, 2018), India (2016, 2019), the Horn of Africa (2011–2012), South Africa (2015–2016, 2018) and Vietnam (2016) have clearly shown that the risk of negative impacts associated with droughts is not only linked to the severity, frequency and duration of drought events but also to the degree of exposure, susceptibility and lack of coping capacity of a given socioecological system (SES). Despite this, proactive management of drought risk is still not a reality in many regions across the world. Droughts and their impacts are still mostly addressed through reactive crisis management approaches, for example, by providing relief measures (Rojas, 2018). To improve the monitoring, assessment, understanding and ultimately proactive management of drought risk effectively, we need to acknowledge that the root causes, patterns and dynamics of exposure and vulnerability need to be considered alongside climate variability in an integrated manner (Spinoni et al., 2019a; Hagenlocher et al., 2019).
Over the past decades, major efforts have been made to improve natural hazard risk assessments and their methodologies across scales, ranging from global risk assessments to local-level assessments. At the global scale several studies have been published in recent years, focusing on the assessment of flood risk (Hirabayashi et al., 2013; Ward et al., 2013, 2014), seismic risk (Silva et al., 2018), cyclone risk (Peduzzi et al., 2012) or multi-hazard risk (e.g., Dilley et al., 2005; Peduzzi et al., 2009; Welle and Birkmann, 2015; Garschagen et al., 2016; INFORM, 2019; Koks et al., 2019; UNDRR, 2019). While major progress has been made regarding the mapping, prediction and monitoring of drought events at the global scale (e.g., Yuan and Wood, 2013; Geng et al., 2013; Spinoni et al., 2013, 2019b; Damberg and AghaKouchak, 2014; Hao et al., 2014; Carrão et al., 2017), very few studies have assessed either exposure to drought hazards (Güneralp et al., 2015) or drought risk at the global level (Carrão et al., 2016; Dilley et al., 2005; Li et al., 2009). The study by Carrão et al. (2016) presents the first attempt to map drought risk at the global scale while considering drought hazard (based on precipitation deficits), exposure (population, livestock, crops, water stress) and societal vulnerability (based on social, economic and infrastructural indicators). While generic drought risk assessments are useful for establishing an overview of the key patterns and hotspots of drought risk, it is increasingly acknowledged that drought risk assessment should be tailored to the needs of specific users so that management plans can be developed to reduce impacts (Vogt et al., 2018; UNDRR, 2019). Impact or sector-specific assessments of who (e.g., farmers) and what (e.g., crops) are at risk as well as what they are at risk of (e.g., abnormally low soil moisture, deficit in rainfall, below average streamflow), where they are at risk and why are needed to inform targeted drought risk reduction, resilience and adaptation strategies (IPCC, 2014). Such analyses are currently lacking. Furthermore, in their exposure analysis, Carrão et al. (2016) do not distinguish between rainfed and irrigated agriculture, although different hazard indicators are relevant when assessing drought risk for these systems. In addition, the vulnerability analysis presented by Carrão et al. (2016) is based on a reduced set of social, economic and infrastructure-related indicators and does not account for the role of ecosystem-related indicators as a driver of drought risk – a gap that was recently highlighted in a systematic review of existing drought risk assessments across the globe (Hagenlocher et al., 2019). A socioecological-system perspective, especially when assessing drought risk in the context of agricultural systems, where livelihoods depend on ecosystems and their services, can help to better understand the role of ecosystems and their services not only as a driver of drought risk but also as an opportunity for drought risk reduction (Kloos and Renaud, 2016).
This paper addresses some of the above gaps by presenting, for the first time, an integrated drought risk assessment that brings together data from different sources and disciplines for rainfed and irrigated agricultural systems considering relevant drought hazard indicators, exposure and vulnerability at the global scale. The spatial variability in drought risk on global and regional scales might help to identify leverage points for reducing impacts and properly anticipate, adapt and move towards resilient agricultural systems.
Today, it is widely acknowledged that risk associated with natural hazards, climate variability and change is a function of hazard, exposure and vulnerability (IPCC, 2014; UNDRR, 2019). Following that logic, Fig. 1 shows the overall workflow of the assessment, while the subsequent sections describe in detail how drought risk for agricultural systems, including both irrigated and rainfed systems, was assessed at the global scale.
Workflow for the overall global drought risk assessment for agricultural systems (including irrigated and rainfed systems).
The composite drought hazard indicators were calculated for irrigated and rainfed systems separately using drought indices based on historical climate conditions (1980–2016), which resulted in integrated hazard maps for both rainfed and irrigated agricultural systems, respectively. The different irrigated and non-irrigated crops by country were considered to be the exposed element. Due to the lack of high-resolution gridded data on an agricultural-dependent population at the global scale, this exposure indicator was not considered. The vulnerability component was assessed through a SES lens, where socioecological susceptibility and a lack of coping-capacity indicators were weighted by drought experts around the world.
The drought hazard indicators considered here represent the average drought hazard during the period 1980 to 2016 in each spatial unit for which it is computed. Drought hazard is defined as a deviation of the situation in a specific year or month from long-term mean conditions in the 30-year reference period from 1986 to 2015. To quantify drought hazard for such a long period, we used the global water resources and water use model WaterGAP (Müller Schmied et al., 2014) and the global crop water model (GCWM; Siebert and Döll, 2010). The models simulate terrestrial hydrology (WaterGAP) and crop water use (GCWM) for daily time steps on a spatial resolution of 30 arcmin (WaterGAP) or 5 arcmin (GCWM). The most recent version, WaterGAP 2.2d, was forced by the WFDEI-GPCC climate data set (Weedon et al., 2014), which was developed by applying the forcing data methodology developed in the EU project WATCH on ERA-Interim reanalysis data (Table 1). The GCWM used the CRU-TS 3.25 climate data set (Harris et al., 2014) as input. CRU-TS 3.25 was developed by the Climate Research Unit of the University of East Anglia by interpolation of weather station observations and is provided as a time series of monthly values. Pseudo-daily climate was generated by the GCWM as described in Siebert and Döll (2010). Following the definitions of the Intergovernmental Panel on Climate Change put forward in their Fifth Assessment Report (IPCC, 2014), exposure is defined as the elements located in areas that could be adversely affected by drought hazard. The distinct exposure of irrigated and rainfed agricultural systems to drought was considered by weighting grid-cell-specific hazards with the harvested area of irrigated and rainfed crops according to the monthly irrigated and rainfed crop areas' (MIRCA2000) data set (Portmann et al., 2010) when aggregating grid-cell-specific hazards to exposure at a national scale. MIRCA2000 was also used to inform the models used in the hazard calculations about growing areas and growing periods of irrigated and rainfed crops. The data set refers to the period centered around the year 2000; time series information is not available at the global scale. To maximize the representativeness of the land use, the reference period and evaluation period used in this study were centered around the year 2000.
Hazard and exposure indicators used in the analysis and their processed data.
The composite drought hazard indicator is defined as the product of mean
severity and frequency of drought events. For irrigated agriculture (CH_IrrigAg) it combines an indicator for streamflow drought hazard (SH), i.e., for abnormally low streamflow in rivers, with an indicator of an abnormally high irrigation water requirement (IH; Fig. 1). It thus considers the deviations of both demand and supply of water from normal conditions. SH and IH are computed with a spatial resolution of 0.5
IH was calculated by using the GCWM based on a monthly time series of net
irrigation requirements from 1980 to 2016. The net irrigation requirement is
the volume of water needed to ensure that the AET of irrigated crops is
similar to their PET (Fig. 1). The calculations were performed for 487 121 grid cells with a resolution of 5 arcmin, containing irrigated crop areas, and then aggregated to 26 478 grid cells with a 30 arcmin resolution to be consistent with the resolution used by WaterGAP. SH was calculated by using WaterGAP based on a monthly time series of streamflow from 1980 to 2016 in 66 896 grid cells with a
For both IH and SH, drought hazard per grid cell was quantified as the product of the (scaled or transformed) mean severity of all drought events during the evaluation period 1980–2016 and the frequency of drought events during this period. Drought events for IH and SH were determined independently. In the case of IH computation, a drought event starts as soon as the monthly irrigation requirement exceeds the irrigation requirement threshold and ends when the surplus reaches zero. In the case of SH computation, a drought starts if the monthly streamflow drops below the streamflow threshold and ends as soon as the deficit reaches zero. For each grid cell and each of the 12 calendar months, a drought threshold was defined as the median of the variable values in the respective calendar month during the reference period 1986–2015. To avoid spurious short droughts and drought interruptions, it was defined that a drought event starts (1) with at least 2 consecutive months with an IH surplus or a SH deficit and (2) 1 month without an IH surplus or if a SH deficit does not break the event (Spinoni et al., 2019a). The accumulated surplus (IH) divided by the deficit (SH) during each drought event is the severity of the drought event. Mean severity is computed as the arithmetic average of the severity of all drought events during the evaluation period. As in the case of SH, the deficit and thus the severity of streamflow drought are strongly correlated with the mean annual streamflow; mean severity is therefore scaled by dividing the accumulated streamflow deficit by mean annual streamflow. In this way scaled mean streamflow drought severity is expressed as the fraction of the mean annual flow volume that is on average missing during drought events. In the case of IH, mean severity is transformed logarithmically before computation of IH, as in most grid cells the volume of irrigation water needed additionally in drought periods is relatively small (volume in 569 out of the 26 478 irrigated grid cells is lower than 100 m
CH_IrrigAg was then calculated for each grid cell by combining SH and IH. To ensure that both indicators are
weighted equally, their native values were first scaled to a range between 0
and 1 by dividing SH and IH in each grid cell by the maximum SH or IH detected globally. The frequency distribution of the SH values calculated that way was shifted to the left, with a mean of 0.244, while the frequency distribution of IH was shifted to the right, with a mean of 0.664. Therefore,
CH_IrrigAg was calculated for each grid cell as
The exposure of irrigated agricultural systems to drought at the national scale was derived as the harvested-area weighted mean of the CH_IrrigAg across all grid cells belonging to the respective aggregation units.
The composite drought hazard indicator for rainfed agriculture (CH_RfAg) was quantified based on the ratio of actual crop evapotranspiration (AET
in m
The GCWM was applied for 24 specific rainfed crops and the two groups
“others, annual” and “others, perennial” to calculate crop-specific AET and
PET on a daily time step. Together, the 24 crops and two crop groups cover
all crop species distinguished by FAO in their database FAOSTAT. The sum of
daily crop-specific AET and PET was calculated for all crops and for each
year in the period 1980–2016 for 927 857 grid cells containing rainfed
cropland and aggregated to 37 265 grid cells with the resolution
The mean ratio between AET and PET (
The combined drought exposure of rainfed and irrigated cropping systems
was evaluated at the country level by averaging the harvested-area weighted drought exposure of irrigated and rainfed cropping systems. As described
before, distinct methods were used to calculate hazard and exposure
of irrigated and rainfed systems so that a direct comparison of the exposure
values is not meaningful. In addition, the frequency distributions differed
considerably, with a harvested-area weighted global mean of the drought
exposure of 0.455 for irrigated systems and 0.189 for rainfed systems. To
ensure a more similar weight of rainfed and irrigated drought exposure,
country-specific exposures were divided by the global mean, and then the
integrated exposure was calculated as harvested-area weighted mean:
According to the Intergovernmental Panel on Climate Change (IPCC, 2014), vulnerability is the predisposition to be adversely affected as a result of the sensitivity or susceptibility of a system and its elements to harm, coupled with a lack of coping and adaptive capacity. The assessment of drought vulnerability is complex because it depends on both biophysical and socioeconomic drivers (Naumann et al., 2014). Due to this complexity, the most common method to assess vulnerability in the context of natural hazards and climate change is using composite indicators or index-based approaches (Beccari, 2016; de Sherbinin et al., 2019). Although their usefulness for policy support has also been subject to criticism (Hinkel, 2011; Beccari, 2016), it is widely acknowledged that composite indicators can identify generic leverage points for reducing impacts at the regional to global scale (De Sherbinin et al., 2017, 2019; UNDRR, 2019).
Following the workflow to calculate composite indicators proposed by the Organisation for Economic Co-operation and Development (OECD, 2008) and Hagenlocher et al. (2018), the methodological key steps on which the vulnerability assessment is based are (1) the definition of the conceptual framework, (2) identification of valid indicators, (3) data acquisition and preprocessing, (4) analysis and imputation of missing data, (5) detection and treatment of outliers, (6) assessment of multicollinearities, (7) normalization, (8) weighted aggregation, and (9) visualization.
Vulnerability indicators used in the analysis and their related expert weights.
An initial set of vulnerability indicators for agricultural systems was
identified based on a recent review of existing drought risk assessments
(Hagenlocher et al., 2019). In total 64 vulnerability indicators, including
social, economic and physical indicators; farming practices; and environmental, governance, and
crime and conflict factors, were selected and classified by
socioecological susceptibility (SOC_SUS, ENV_SUS), a lack of coping capacity (COP) and a lack of adaptive capacity (AC) following the risk framework of the IPCC (IPCC, 2014). Indicator weights, which express the relevance of the identified indicators to characterizing and assessing the vulnerability of agricultural systems to droughts, were identified through a global survey of relevant experts (
Following data acquisition, the data were preprocessed by transforming
absolute to relative values and standardized when necessary (e.g., travel
time to cities
The final drought risk index (DRI; Fig. 1) was calculated by multiplying the indices for drought hazard and exposure by vulnerability. At the pixel level, the presence of hazard and vulnerability point to a certain drought risk, independent of how much crop area is contained in the specific pixel. At the aggregated level, the different crop areas in the specific pixels must be considered; therefore exposure was calculated as harvested-area weighted mean of the pixel-level hazard and then multiplied by vulnerability to calculate drought risk at the country level.
The total drought risk score for irrigated and rainfed systems combined (DRI
The outcomes of the risk assessment for irrigated and rainfed systems
combined (DRI
This section presents the results of the global drought risk assessment for agricultural systems (irrigated and rainfed) at the pixel level (Figs. 2 and 3) and for the total risk of both systems combined at national resolution (Fig. 4). The dark-red patterns show high levels of the different risk components, while dark blue reflects low scores of the different risk components.
Drought risk
Drought risk
Drought risk
The drought risk for irrigated agricultural systems varies significantly among continents and countries. Especially large countries such as the USA, Brazil, China and Australia show a high variation at the country level due to varying climatic conditions. Drought hazard and exposure was highest in regions with a high density of irrigated land and high irrigation water requirements such as the western part of the USA, central Asia, northern India, northern China and southern Australia. Vulnerability was high particularly in sub-Saharan Africa but also in some countries in central Asia and the Middle East and low in general for industrialized and high-income countries. The combination of hazard and vulnerability to risk resulted in the highest values for large parts of western, central and southern Asia; eastern Africa; and the eastern part of Brazil. Low-risk areas include western Europe, the USA, Australia and most parts of China (Fig. 2).
High levels of risk (dark yellow to red color scheme) for rainfed agricultural systems are observed in southern Africa, in southeastern Europe, in northern Mexico, in northeastern Brazil, at the western coast of South America, in southern Russia and in western Asia. The vulnerability to drought highlights the relevance to increasing the coping capacity of the countries in order to reduce their overall drought risk. For instance, Australia, despite being highly exposed to drought hazard, has low socio-ecological susceptibility and high enough coping capacities to considerably reduce the overall drought risk.
The hazard and exposure maps shown in Fig. 4 are slightly different to the ones shown in Figs. 2 and 3 due to the aggregation at the country level. The analysis shows that regions with low hazard and exposure of rainfed and irrigated crops to drought tend to be tropical and subarctic regions following the Köppen–Geiger climate classification (1980–2016; Beck et al., 2018). There are significant regional differences when comparing irrigated and rainfed drought hazard and exposure. For instance, the northern parts of Latin America and central Africa have low hazard and exposure levels, given the humid climate conditions resulting in a low total risk, even though those regions are characterized by high vulnerability levels. Southern Africa, however, has a high amount of drought-exposed rainfed crops but lower vulnerability compared to other African countries. Despite this, risk scores in that region are very high. Very high drought hazard and exposure and vulnerability levels can be found in the Middle East and northern Africa.
Although the drought hazard was computed differently for the different
agricultural systems, the countries with high risk of drought to both
farming systems are Botswana, Namibia and Zimbabwe (Figs. 2 and 3). These
countries share the same relevant indicators that define their high
vulnerability: a high soil and land degradation rate, a low literacy rate and
low total renewable water (Fig. S3). Table 3 shows the top and bottom 10 countries with the highest and lowest total drought risk (DRI
Rank of countries with the highest and lowest risk of drought for combined agricultural systems (rainfed and irrigated).
Seven out of the 10 countries with the highest overall drought risk are located on the African continent. However, Kosovo, East Timor and Kazakhstan also possess high risk levels (Table 3). Zimbabwe ranks as the country with the highest drought risk, mainly due to its high exposure combined with its high vulnerability (Fig. S1 in the Supplement).
In general, the countries that present higher drought risk have a high amount of exposed crops. Vulnerability varies among them, with Zimbabwe being the country with the highest vulnerability. The lack of coping capacity and socioecological susceptibility were determinant factors for countries like Botswana and Zimbabwe (Fig. S1). There were cases where countries such as Namibia presented high socio-ecological susceptibility in contrast with high coping capacity, reducing its overall vulnerability. The drought risk in countries such as Lesotho and Mauritania that have, in contrast, limited coping capacities is notably higher (Fig. S1). The analysis also reveals that, although risk is currently close to zero in several countries (e.g., Fiji, Central African Republic, Guinea-Bissau, etc.), this could rapidly change once these countries are affected by droughts given their very high vulnerability.
The comparison of the drought risks of rainfed and irrigated cropping systems (Fig. 5) shows that several countries such as Zimbabwe, Iraq and Algeria are exposed to high risk for both cropping systems. These countries are frequently hit by drought and similarly have a high vulnerability to drought (Figs. 2 and 3). In contrast, countries such as Switzerland, Finland and New Zealand are characterized by low drought hazard and exposure of irrigated and rainfed systems and low vulnerability to drought (Figs. 2 and 3). In countries such as Botswana, Oman and the United Arab Emirates, drought risk is high for rainfed cropping systems but low for irrigated cropping systems (Fig. 5). These countries are defined by arid climate conditions, exposing rainfed crops to high risk, while the drought risk for irrigated cropping systems is low because of relatively low interannual variability in climatic conditions resulting in low variability in the irrigation water requirement and streamflow. Their risk is also determined by their different vulnerability dynamics (e.g., hydroelectric sources, retaining renewable water). In contrast, drought risk for irrigated cropping systems is high and drought risk of rainfed cropping systems is small in countries such as Burkina Faso, Madagascar and Côte d'Ivoire (Fig. 5). In these three countries, there is a big variability in climatic conditions, with irrigated crops being cultivated in the more arid parts of the country and rainfed crops being cultivated in more humid parts. In addition, aquatic crops with high water demand, such as rice and sugarcane, are the most commonly cultivated irrigated crops in these countries (Frenken, 2005).
Country profiles contrasting the drought risk of irrigated and rainfed agricultural systems. The size of the bubbles indicates the crop growing area (sum of rainfed and irrigated areas per country in Mha.
Comparison of total risk against drought impact data.
The comparison of drought risk (DRI
The present study performs, for the first time, a separate global drought
risk analysis for irrigated and rainfed cropping systems, including regions
that indicate a high vulnerability to droughts and are particularly exposed.
In previous assessments, the share of irrigated cropland was either ignored
or considered to be a vulnerability indicator (Carrão et al., 2016). The
drought hazard analysis is based on three indicators: SH, IH and CH_RfAg, which quantify drought as a deviation from normal conditions consistent with common definitions. In agreement with the results for drought hazard obtained by Carrão et al. (2016), the largest drought hazard is obtained for arid and semi-arid regions such as northern and southern Africa, northern Mexico, along the coastline of Peru and Chile, the Arabian Peninsula, and Mongolia for rainfed systems; Italy, Turkey and western Mexico for irrigated systems; and the western USA, northeastern Brazil, western Argentina, central Asia, the Middle East, western India, northern China and southern Australia for both
irrigated and rainfed systems. In contrast, previous studies based on
standardized indices such as the standardized precipitation index (SPI) have detected the highest drought
hazard mainly in humid regions such as central Europe, southeastern Asia,
southern Brazil and tropical Africa (Geng et al., 2016). The reason for this difference could be that deviations from normal conditions should not be treated similarly for arid and humid regions, as not every precipitation or streamflow deficit in humid regions will automatically become a hazard for cropping systems. In fact, in humid regions, crops often perform better in relatively dry years (Holzkamper et al., 2015). We account for these effects by normalizing streamflow deficits with long-term mean annual river discharge (SH) or by calculating the probability of reductions in the AET
In the present study, the rainfed hazard is computed as the probability of
a 10 % decline in the AET
The calculation of the drought hazard of irrigated cropping systems in this study is based on the two components SH and irrigation IH reflecting the water supply and water demand, respectively, of irrigated systems. Therefore we do not consider specifically in our approach the availability and use of groundwater resources for irrigation. It is well known that dynamics in streamflow are usually larger than dynamics in groundwater storage so that groundwater is used by many farmers to substitute temporary deficits in surface water supply for irrigation systems. In general, access to groundwater should therefore be considered to reduce drought hazard and vulnerability of irrigated cropping systems. Consideration of groundwater resources would, however, require dynamic quantification of groundwater storage and groundwater levels, which is challenging for global-scale analyses and not possible with the models applied in this study. In addition, more conceptual work is needed to decide which degree of temporal variability in groundwater levels constitutes a hazard and how to treat long-term depletion of groundwater resources (negative trends) in drought risk studies.
The multi-dimensional nature of vulnerability of agricultural systems is
represented by a set of 26 expert-weighted indicators. One of the major
limitations of this data-driven approach is the spatial detail information
for computing the model; however, at a global level it is not feasible to
get a harmonized data set of all the proxy variables, but some caution must
be advised when zooming in at the subnational level (Naumann et al., 2018).
When interpreting the results, it is necessary to consider that some highly
correlated indicators were maintained in the analysis, as they present
different drivers of vulnerability and hence different entry points for
vulnerability reduction. The selected indicators comprise social, economic,
environmental, physical and governance-related factors contributing to
socioecological susceptibility and the lack of coping capacity. In doing
so, the present study goes beyond existing global drought risk assessments
(Carrão et al., 2016), which are based on equal weights and do not
consider relevant environmental vulnerability indicators to be a driver of
drought risk. The latter, however, is relevant when assessing drought risk
for agricultural systems, where factors such as land degradation and soil
erosion are shown to exacerbate drought risk (Hagenlocher et al., 2019). In
future assessments an alternative to the expert-based weighting of
vulnerability indicators chosen here could be the use of statistical
approaches (e.g., principal component analysis – PCA) to identify relevant
indicators. However, given the high number of experts who participated in
the weighting exercise (
The findings of the drought risk assessment presented here correspond to a certain degree to the findings of Carrão et al. (2016). Although the focus of the current paper is more explicitly on agriculture, both studies present methodological similarities. In Carrão et al. (2016) the percentage of crop land per grid cell is one factor in the exposure analysis, and the percentage of irrigated agricultural land is one of the vulnerability factors. Although Carrão et al. (2016) include other factors such as population density, livestock density and baseline water stress in the analysis, the results give a high weight to the risk for agriculture. In both studies the regions less affected by droughts correspond to the regions with little or no exposure of agriculture and population (e.g., deserts and tropical forests). This is mainly the case in Amazonia and central Africa. Also, similarities between areas of high levels of risk are evident, including southern and eastern Europe, the Eurasian steppe, northern Africa and the Middle East, northeastern Brazil, and southeastern South America.
Similarities are also found for the risk of irrigated agricultural systems. Examples are irrigated croplands in India, the US and Australia. Differences in the overall patterns are due to the separation of irrigated and non-irrigated agriculture in the current study and the aggregated exposure information in Carrão et al. (2016). In an updated version of the risk map from Carrão et al. (2016), using a higher-resolution population database and grid-level exposure information, as shown in Vogt et al. (2018, Fig. 7), similarities are even more evident.
However, the present study includes a spatially explicit model of AET for the main crop types of two different agricultural systems (irrigated and rainfed agriculture) and includes a specialized vulnerability index for this sector according to expert judgment. These differences revealed the importance of focusing more clearly on distinct impacts (e.g., on irrigated vs. rainfed systems) when conducting drought risk assessments, even within the same sector. For instance, irrigated agricultural systems in Latin America are highly exposed to droughts, whereas the probability of droughts occurring in rainfed agricultural systems in that region is comparably low.
Despite these advancements, the presented analysis does have limitations. First, due to the lack of up-to-date land use data on irrigated vs. rainfed agriculture at the global scale, the exposure analysis is based on MIRCA data from the year 2000 (Portmann et al., 2010). Given that cropping systems are subject to change, this adds uncertainty to the results. Second, data used for the vulnerability analysis stem from different sources, which makes it difficult to evaluate the inherent uncertainties in the data. Third, the data are not consistently available for all countries for the same years (Table 2). Fourth, the vulnerability analysis is based on nation–state-resolution data, which do not allow for mapping spatial variability in vulnerability at the subnational level. Fifth, applying expert opinions to weight drought vulnerability indicators according to their relevance brings subjectivity to the assessment, which necessitates a strong network of relevant experts. Sixth, preventive or adaptive planning requires going beyond evaluating drivers of risk and mapping current patterns of risk. Future scenarios of drought risk, considering both changing environmental and climate conditions as well as possible future socioeconomic development pathways, are needed in order to anticipate future challenges.
Future research should address these challenges by also investigating subnational patterns in vulnerability and developing future drought risk scenarios in all dimensions of drought hazards, exposure and vulnerability. In addition, attempts to investigate changes and trends in drought risk and risk components are highly needed to better understand trajectories of drought risk in different countries and for the whole world. Further, inherent uncertainties, as well as the sensitivity of the risk assessment outcomes towards changes in the input parameters (e.g., indicator choice and weighting), should be investigated and validated statistically. This gap has also been highlighted in a recent review of climate vulnerability assessments (de Sherbinin et al., 2019) in general as well as in a recent review of drought risk assessments (Hagenlocher et al., 2019) in particular.
The comparison conducted in this study has shown that there are limited data available on agricultural losses and impacts caused by droughts at the global level. Furthermore, impacts are not always direct, as droughts can have cascading indirect impacts (Freire-Gonzáles et al., 2017; Van Lanen et al., 2017) which are difficult to assess. In addition, for countries where we find high drought risk (e.g., Mongolia, Iran, Kazakhstan and the countries in southeastern Europe), no or very few drought events are registered in EM-DAT. The reason for this mismatch could be that drought events in these countries were not registered in EM-DAT. For example, in Romania, EM-DAT reports two drought events, while according to other reports, 12 years between 1980 and 2012 were classified as drought years, with 48 % of the agricultural land affected (Lupu et al., 2010; Mateescu et al., 2013). On top of this, in Iran, EM-DAT reports one drought event while other sources recounted several droughts during 1980–2005, with the most extreme drought lasting for 4 years, from 1999 to 2002 (Javanmard et al., 2017; Zoljoodi and Didevarasl, 2013). These examples suggest that it cannot be concluded from missing drought records in EM-DAT that specific countries were not affected by drought. Once improved and reliable impact data are available at the global scale, future research should also focus on the statistical validation of drought risk assessments with drought events and impact data. Ongoing efforts of countries to report their losses and impacts due to natural hazards (e.g., as part of the Sendai monitoring) are considered to be a first important step towards that direction.
Lastly, while this study presents the first attempt to assess drought risk for agricultural systems, more work is needed to analyze drought risk for other sectors, such as public water supply, tourism, energy production and waterborne transport, among others.
This paper presents, for the first time, a global-scale drought risk assessment for both irrigated and rainfed agricultural systems from a socioecological perspective by integrating drought indicators for hazard, exposure and vulnerability. It goes beyond previous studies by including a separated and spatially explicit analysis of the drought hazard and exposure of irrigated and rainfed agricultural systems as well as an empirically based weighting of vulnerability indicators. The latter is based on the judgment of drought experts around the globe. The presented methodology can serve as a blueprint for the analysis of other affected sectors, such as water or energy. Findings from this study underscore the relevance of analyzing drought risk from a holistic perspective (i.e., including the sector-specific hazard, exposure and vulnerability) and are based on a spatially explicit approach. By providing information on high-risk areas and underlying drivers, this approach helps to identify priority regions as well as entry points for targeted drought risk reduction and adaptation options. While this first attempt provides valuable information at the global level, improvements could be achieved with the availability of more spatially explicit vulnerability information (i.e., at subnational levels) and the availability of standardized drought impact information that can serve as a quantitative validation of risk levels.
Data can be accessed under the following link:
The supplement related to this article is available online at:
IM conducted the risk assessment with the support of SS, EER under the supervision of MH. SS, PD, JK, HG, CH, EP, EER and HN developed the hazard indicators and SS conducted the exposure analysis. IM, MH, GN, and JVV conducted the global expert survey. IM, GN, JVV and MH were responsible for the vulnerability assessment. All authors have contributed to drafting the manuscript, the interpretation of the results, and approved the final manuscript.
The authors declare that they have no conflict of interest.
This article is part of the special issue “Global- and continental-scale risk assessment for natural hazards: methods and practice”. It is a result of the European Geosciences Union General Assembly 2018, Vienna, Austria, 8–13 April 2018.
The authors would like to thank the 78 experts for their participation in the global expert survey, the two reviewers for their valuable comments and Harrhy James for proofreading the revised paper.
This research is part of the project GlobeDrought and has been supported by the German Federal Ministry of Education and Research (BMBF; grant no. 02WGR1457A-F) through its Global Resource Water (GRoW) funding initiative.
This paper was edited by Philip Ward and reviewed by Veit Blauhut and one anonymous referee.