A new global database of meteorological drought events from 1951 to 2016

Highlights

• •We collected more than 4500 drought events, at global scale, from 1951 to 2016.
• •We identified the biggest 52 mega-droughts occurred in 1951–2016.
• •In the last decades, Amazonia, southern South America, Mediterranean region, most of Africa, north-eastern China stand out as drought hotspots.
• •Over North America, central Europe, central Asia, and Australia, the progressive temperature increase outbalanced the increase in precipitation causing more frequent and severe droughts.
• •The interactive version of the database is hosted by the European Commission's Global Drought Observatory.

Abstract

Study region

This study has three spatial scales: global (0.5°), macro-regional, and country scale. The database of drought events has specific entries for each macro-region and country.

Study focus

We constructed a database of meteorological drought events from 1951 to 2016, now hosted by the Global Drought Observatory of the European Commission's Joint Research Centre. Events were detected at macro-regional and country scale based on the separate analysis of the Standardized Precipitation-Evapotranspiration Index (SPEI) and the Standardized Precipitation Index (SPI) at different accumulation scales (from 3 to 72 months), using as input the Global Precipitation Climatology Centre (GPCC) and Climatic Research Unit (CRU) Time Series datasets. The database includes approximately 4800 events based on SPEI-3 and 4500 based on SPI-3. Each event is described by its start and end date, duration, intensity, severity, peak, average and maximum area in drought, and a special score to classify 52 mega-droughts.

New hydrological insights for the region under study

We derived trends in drought frequency and severity, separately for SPI and SPEI at a 12-month accumulation scale, which is usually related to hydrological droughts. Results show several drought hotspots in the last decades: Amazonia, southern South America, the Mediterranean region, most of Africa, north-eastern China and, to a lesser extent, central Asia and southern Australia. Over North America, central Europe, central Asia, and Australia, the recent progressive temperature increase outbalanced the increase in precipitation causing more frequent and severe droughts.

1. Introduction

The complex nature of droughts makes them one of the most difficult climate hazards to perceive (Wilhite, 2000). Other recurrent weather-related hazards such as heatwaves, floods, and windstorms can be skilfully forecasted in certain cases (Vitart, 2006, Komma et al., 2007, Lowe et al., 2011) and in general unequivocally quantified and classified (Milly et al., 2002, Russo et al., 2015). In addition, a wide spectrum of impacts of such events is usually documented in the scientific and non-scientific literature. Examples are deaths caused by a heatwave (Kovats and Kristie, 2006), tree fall after a windstorm (Schlyter et al., 2006), or damages to infrastructures hit by a flood (Blöschl et al., 2013). Instead, drought develops slowly, its main characteristics such as the onset, duration, and severity, are not easily quantified, and impacts are often indirect, non-structural, and spread over large areas (Dai, 2011). Consequently, drought impacts are often not easily related to a given drought event and difficult to measure in economic terms.

There is no unique definition of drought, as we can distinguish between meteorological, agricultural, hydrological, and socio-economic droughts (Mishra and Singh, 2010), depending on the hydrological compartment affected and/or the related impacts. Drought impacts are often recorded incompletely or in a qualitative manner only (Ding et al., 2011). In addition, drought impacts depend on the exposure and vulnerability of affected societies, and can be exacerbated, for example, by poor land management practices (Cook et al., 2009, Van Dijk et al., 2013), eventually leading to irreversible land and soil degradation (Cook et al., 2014, Reynolds et al., 2007, Wonkka et al., 2016, Marengo et al., 2017).

Given the difficulties to analyze droughts, mentioned above, one consequence is that, at the time of writing, to our knowledge neither a global database of droughts, nor a comprehensive global database of drought impacts exist. One of the objectives of the European Commission's Global Drought Observatory (GDO; http://edo.jrc.ec.europa.eu/gdo/) is to incorporate amongst its features data regarding drought characteristics and their impacts, in order to understand the relationships between both. To do this, the first essential step is to construct a global database of drought events and their characteristics, which is the main goal of this study.

Here we focus on meteorological drought, which owes its name to the meteorological variables (usually precipitation and temperature) analyzed. Therefore, meteorological drought can be defined as a prolonged lack of (or below normal) rainfall (Keyantash and Dracup, 2002), possibly aggravated by hot temperatures causing high evapotranspiration rates (Vicente-Serrano et al., 2015). The role of temperature, which is needed to compute potential evapotranspiration (PET), is still debated in the scientific literature (Dai et al., 2018). In particular, most formulations tend to overestimate PET (compared to actual evapotranspiration) in semi-arid and arid hot areas (Sheffield et al., 2012, Dai, 2013, Trenberth et al., 2014; van der Schrier et al., 2015), leading to uncertainties in the analyses of past and projected droughts (Dai and Zhao, 2017). Consequently, we investigated meteorological droughts driven by precipitation separately from those driven by precipitation and temperature.

This work builds on two previous studies dealing with the collection of meteorological drought events: one focused on the Carpathian Region and over the period 1961–2010 (Spinoni et al., 2013), and the other on the entire European continent and over the period 1950–2012 (Spinoni et al., 2015a). As in those two studies, we used SPI and SPEI at different accumulation scales, however the new global database of events has specific entries for macro-regions and for countries (similar to Spinoni et al., 2015a), while the current period of interest is 1951–2016, with planned updates every two or three years. Consequently in 2019 (or 2020), the period will be extended to 1951–2018 and so on, depending also on input data availability.

As in Spinoni et al. (2015a) for Europe, here we assigned to each event a set of parameters as well as a special score based on duration, severity, intensity, and area involved, in order to compare events in different countries and macro-regions. The scoring system enables the differentiation of three classes – “moderate”, “severe”, and “exceptional” droughts – and a classification of the biggest meteorological drought events occurring at global scale during 1951–2016. The list of the biggest events, together with time series of monthly maps (0.5° spatial resolution) and statistics, are hosted by the European Commission's GDO.

The remainder of this paper is structured into three main sections. Section 2 focuses on methods applied to data collection, computation of drought indicators, and description of parameters for the single drought events. Section 3 focuses on results, and describes separately: the database structure; the biggest drought events collected; global drought trends and hotspots; and the importance of temperature in meteorological drought analyses. Section 4 summarizes the most relevant findings, and anticipates possibilities for further development of the database, such as the incorporation of vegetation-based indicators.

2. Data and methods

2.1. Input data: precipitation and potential evapotranspiration (PET)

The choice of input data was guided by four pre-requisites: medium-high spatial resolution (at least 0.5°, if possible), temporal range (at least as far back as the 1950s), global availability of daily or monthly precipitation and temperature (or PET), and high data quality. Consequently, we selected two global datasets mostly based on station-observed data: the Global Precipitation Climatology Centre (GPCC; Schneider et al., 2008, Becker et al., 2013) dataset of the German Weather Service (DWD), and the Climatic Research Unit Time-Series (CRU TS; Harris et al., 2014) of the University of East Anglia.

From GPCC (version 7), we obtained monthly gridded precipitation data for 1901–2013 at 0.5° spatial resolution, and from CRU TS (version 4.01), we obtained monthly gridded precipitation and PET data for 1901–2016, also at 0.5° spatial resolution. We analyzed data only from 1951 onwards, due to the limited availability of station records before the 1950s, in particular regarding precipitation at high latitudes and in tropical regions as the Amazon rainforest (Schneider et al., 2014, Spinoni et al., 2014). Both datasets have been frequently used in climate studies, for example in Kottek et al. (2006), Rubel and Kottek (2010), Weedon et al. (2011), Morice et al. (2012), Trenberth et al. (2014), and Spinoni et al. (2015b).

Regarding precipitation, we used data from the GPCC for 1951–2013, and from the CRU TS for 2014–2016. To ensure continuity from 2013 to 2014 (and over the entire period), we tested the 1951–2016 series for homogeneity using the newest version (i.e. 3.03) of the Multiple Analysis of Series for Homogenization (MASH; Szentimrey, 1999, Szentimrey, 2006) software. For each area, we created a reference series using a varying number of grid points in the surroundings (up to eight). Then we compared each single series with the reference series using MASH. This introduces a possible bias, but we took it as a good compromise to exclude biased series, given that we used grid point series only.

Only 1.7% of grid points failed the homogeneity tests, more than 90% of which are located in regions that we masked (see next sections and Fig. 2), the remaining are mostly located in Arctic Russia, in semi-desert parts of the Arabian Peninsula, and in the Australian central outback. For grid points that failed the tests and are located in regions of drought interest, we used up to eight surrounding grid points to adjust the series. After this procedure, only 0.2% of grid points failed a second round of homogeneity tests and were excluded from the analyses. The entire procedure was simplified by two facts: both the GPCC and the CRU TS data have the same spatial grid, and share many stations used to create the grids (Becker et al., 2013, Schneider et al., 2014).

Fig. 2.

Between 1951 and 2016 and according to the SPEI-3 August 2015 is the month with the largest area in drought conditions at global scale (27.4%). Extreme drought is shown in red, severe drought in orange, and moderate drought in yellow.

We opted for precipitation data from the GPCC (instead of using the CRU data only) because we already successfully used them in previous studies at global scale (Spinoni et al., 2014, Spinoni et al., 2015b). Moreover, the GPCC uses an interpolation algorithm over space and time which results in dynamic values over areas poor of stations (Becker et al., 2013), while the CRU sometimes uses the long-term climatology (for example over the Amazon Forest in the mid-20th century), causing critical problems for drought indicators based on statistical distributions. We highlight that the overall number of stations (in particular those with precipitation data) used to create the grids shows a decrease from the 1970s onwards (Becker et al., 2013), consequently the analyses over the periods with the highest number of records (from the 1960s to the 2000s) are the most accurate.

Regarding PET, this is provided by the CRU TS dataset, which estimates it using Penman-Monteith's formulation, and based on sunshine duration, temperature, vapour pressure, humidity, and wind speed data (Allen et al., 2006). To be precise, PET data in the CRU TS dataset are not estimated using real wind data, but a fixed monthly climatology, as well as sunshine duration and humidity data derived from other variables as temperature, cloud cover, and latitude (Harris et al., 2014). The Penman-Monteith approach requires many variables that are not easily retrievable from historical records over large areas. However it is often considered the most suitable for drought-related analyses (Trenberth et al., 2014, Dai and Zhao, 2017), although it is not free from criticism and other formulations are accepted (Van der Schrier et al., 2011, Spinoni et al., 2017). In particular, the assumptions in the Penman-Monteith approach refer to a surface of well-clipped grass with a sufficient amount of water. These conditions are not met everywhere and at any time, moreover under water stress the actual evaporation (AET) tends to decrease while the potential evapotranspiration tends to increase (Brutsaert and Parlange, 1998), so when the conditions are very or extremely dry the Penman-Monteith equation tends to overestimate PET and consequently the SPEI overestimates drought. This limitation could be partly overcome with the use of the self-calibrated Palmer Drought Severity Index (sc-PDSI; Wells et al., 2004), which is less affected by this issue, as it is based on an estimated actual evapotranspiration, and might be included in the updates of the database.

As in the case of precipitation, we tested the PET series for homogeneity before using them to compute the drought indicators: in this case, only 0.3% of grid points falling into the area of drought interest were rejected.

2.2. Drought indicators: the SPI and the SPEI

As anticipated in the introduction, we decided to account for the temperature-related effects on drought events, without discarding the droughts forced by rainfall deficits only. Thus, we split our database of drought events in two sections, depending on the indicator used: the Standardized Precipitation Index (SPI; McKee et al., 1993, Guttman, 1999), which is based on precipitation only, and the Standardized Precipitation-Evapotranspiration Index (SPEI; Vicente-Serrano et al., 2010, Beguería et al., 2014), which is based on precipitation and PET (which includes temperature). Following this approach, we made our database more flexible and suitable for sectors sensitive to droughts and high temperatures (Williams et al., 2013, Naumann et al., 2015, Shukla et al., 2015). In addition, this approach allows us to exploit the database to consider how the past meteorological drought trends change – at global scale – including or excluding temperature effects, as already evaluated at European scale (Spinoni et al., 2017).

Both SPI and SPEI are standardized indicators that fit the input variable(s) with a statistical distribution over a baseline and classify the drought conditions with a simple scheme related to standard deviations from the median. In order to compute the standardized indicators, we fitted the Gamma (for SPI) and log-logistic (for SPEI) probability distributions over the baseline 1951–2016. The applied distributions are the most widely used in literature and recommended by the indicators’ original developers. Cressie (2015) presents a detailed description on the theoretical background of the distributions. Both indicators have been computed at multiple temporal accumulation periods (from 3 to 72 months), in order to let the user choose the accumulation period which best accounts for the sector of interest. In this paper, we focus on 3- and 12-month periods. However in GDO the user will find results derived also for all the other accumulation periods.

As a baseline, we decided to use the entire period (1951–2016) in order to fit the distribution on the longest possible time-series and to derive robust results (Guttman, 1999). Moreover, semi-arid and arid areas can be problematic with meteorological indicators computed at short accumulation periods, especially with SPI. In fact, a notable number of zero values (no cumulated precipitation, for example for more than half of 3-month periods) would result in unreliable estimations of the indicator, for statistical reasons such as a Gamma distribution fitted with too few non-zero values (Cressie, 2015). On the other hand, fitting the difference between precipitation and PET over a short baseline period can result in problems computing SPEI due to climate change effects (i.e. rising temperatures) that are manifest in PET over recent decades (IPCC, 2014). In fact, with a baseline of 1951–1980, for example, SPEI values of recent years (or decades) may be unrealistic, because recent differences between precipitation and PET fall outside the range recorded during the baseline period, thus biasing the fit of the statistical distribution.

2.3. Meteorological drought events: definition and parameters

In this section, we use SPEI-3 as an example, however all definitions and explanations apply to both SPI and SPEI at any accumulation period. In this study, a drought event starts once the analyzed indicator falls below the value corresponding to a given negative standard deviation (e.g. SPEI-3 < −σ) for at least two consecutive months, and ends when the indicator rises above 0. Usually, the threshold for the start of the event is −1 (McKee et al., 1993). In this study this is valid when a drought event is evaluated for individual grid points, since −1 corresponds to −1σ for the indicators. However, this is not valid for a country or region, which contains a non-unitary number of grid points. So, how do we define whether a country or region is affected by a drought event? Firstly, we computed country time series by averaging the time series at grid point level in that country. Secondly, we calculated the standard deviation over the averaged time series (e.g. σ = 0.75). Thirdly, we used the new standard deviation as a threshold; in this example a drought event starts once the indicator is below −0.75 for at least two consecutive months. With the two months criterion we excluded the very short droughts (one month only), which are rarely impacting (differently than flash floods) and could outstandingly increase the number of events of low importance in the database. As for grid points, also at country or regional scales the drought event ends when the indicators become positive. In this study, after the end of a drought event, at least 2 months should pass before a new event is considered.

Fig. 1 shows as an example the country series of SPEI for four accumulation periods for Argentina. It demonstrates that at country scale there is no need that the drought indicator falls below −1 for a drought event (in red) to start. However, when the database is implemented in GDO, the user will also have access to the indicator values at grid point scale (0.5°), as shown for example in Fig. 2, which represents the month of August 2015, which between 1951 and 2016 and according to the SPEI-3 shows the largest percentage of global land areas under meteorological drought conditions (27.4%).

Fig. 1.

Time series of SPEI at different accumulation time scales for Argentina. The drought events are marked in red. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

All numbers reported in this paper and statistics included in GDO do not consider the masked areas in Fig. 2. Other than Antarctica (not shown in this paper), we excluded deserts (condition: average precipitation/PET ratio for 1951–2016 below 0.05) and cold areas (condition: average PET for 1951–2016 below 365 mm), as was done previously in Spinoni et al. (2014).

The database collects drought events for countries and, at a larger spatial scale, for macro-regions. We focus on countries because most of the drought impacts registered in open online (e.g., the EM-DAT database, see: https://www.emdat.be/) and proprietary (e.g., MunichRe, 2019) databases deal with country data. Macro-regions are included because we aim to classify the mega-droughts of the last decades, possibly involving multiple countries or even different climatological regions. Regarding countries, we included only those having an extent of at least three grid points and at least 95% of valid data (over 1951–2016) for both the SPI and the SPEI at 3- and 12-month accumulation periods. Thus, we discarded small islands and very small countries and collected drought events for 171 countries. Moreover, we performed the analyses and collected events for 23 macro-regions, following the same procedure for countries (see Fig. 3 and Table 1 for the corresponding acronyms). The macro-regions are similar to those described in Giorgi (2006), with some exceptions: we excluded high-latitude cold areas (e.g. Greenland) and the Sahara. The borders do not perfectly match with those in Giorgi (2006), as we aimed at macro-regions with roughly comparable dimensions. Excluding Alaska (ALA), Northern Europe (NEU), and Northern Asia (NAS), all regions range between approximately 3 and 8 million km².

Fig. 3.

The 23 macro-regions analyzed in this study. They represent a modified version of the regions described in the IPCC 5th Assessment Report (AR5). We excluded the Arctic, Antarctica, and the Sahara. For the list of the acronyms, see Table 1.

Table 1.

Acronyms used for the macro-regions. See Fig. 3 for their geographical domains.

Code	Macro-region
ALA	Alaska
AMZ	Amazon
CAM	Central America and Caribbean Islands
CAS	Central Asia
CNA	Central North America
CSA	Central South America
EAF	East Africa
EAS	East Asia
ENA	East North America
EQF	Equatorial Africa
MED	South Europe, Mediterranean
NAS	North Asia
NAU	North Australia and Pacific Islands
NEE	Northeastern Europe
NEU	Central and North Europe
SAF	Southern Africa and Madagascar
SAS	South Asia
SAU	South Australia, New Zealand
SEA	Southeast Asia
SSA	Southern South America
TIB	Tibetan Plateau
WAF	West Africa
WNA	West North America

For each drought event, we assigned a set of parameters, describing its key characteristics (Table 2): start and end month and year, duration, severity, intensity, average area in drought conditions during the event, lowest indicator value and the widest area in drought conditions (the last two parameters with corresponding month and year). As in Spinoni et al. (2015a), we calculated severity as the sum (in absolute values) of indicator values during the drought event, and intensity as the ratio between severity and duration. Finally, we calculated a special score, allowing for an overall comparison of events (see Section 2.4).

Table 2.

Parameters assigned to a drought event as included in the database. Regarding the special score, see Table 3.

Ind	c/r	Ev	EvC	SM	SY	EM	EY	DD	DSa	DI	DA	P	PM	PY	PA	AM	AY	S
Spei-3	AFG	1	1	11	1952	5	1953	7	5.55	0.79	18.3	−0.94	12	1952	39.1	12	1952	3

Ind: indicator; c/r: country or region; Ev: number of event (entire database); EvC: number of event for that country or region; SM: start of the event (month); SY: start of the event (year); EM: end of the event (month); EY: end of the event (year); DD: duration (months); DS: severity (sum of all indicator values, in absolute values, during the event); DI: intensity (severity/duration); DA: average area in drought (%) during the event; P: peak – lowest indicator value during the drought event; PM: peak of the event (month); PY: peak of the event (year); PA: largest area in drought during the event (%); AM: peak area of the event (month); AY: peak area of the event (year); S: special score to classify droughts (0–25).

^aDrought severity is computed over the drought threshold for that country or region.

The drought duration, severity, and average area were then used to investigate drought trends at various spatial scales (see Section 3). Moreover, to each event we assigned information on the indicator (and accumulation period) used, a country code, and two numbers referring to the number of the event for that country (or macro-region) in the entire database. In GDO, each drought event has a unique 30-character code that helps searching for a particular event.

2.4. A special scoring system to compare droughts at different scales

In the last few years, whether in the newspapers, on the internet, or in the media in general, it is not uncommon to find popular claims that the latest event – for example the California drought of 2014 (Science News, 2014), Moscow's heatwave in summer 2016 (Russia Beyond, 2016), or the South Asia flood of 2017 (The Guardian, 2017) – broke century or millennial records. Sometimes these claims are based on robust data and analyses, but sometimes not. One of the goals of this study is to define a methodology that can compare and classify drought events at different locations. We emphasize that our approach focuses on drought events, not drought conditions, thus it differs from drought classifications such as in the United States Drought Monitor (Svoboda et al., 2002) and in the European Drought Observatory (Sepulcre-Canto et al., 2012).

Table 3 shows how our special scoring system for drought events derives from the aggregation of six parameters. The first two deal with the severity and intensity, and assign a score from 0 to 5 points depending on how the event compared with all events included in our database (for the selected drought indicator and accumulation period) at national or macro-regional scale. Severity and intensity, as defined in this study, are often mutually exclusive, as the typical severe event is slowly developing and long lasting, whilst an intense event such as a flash drought (usually lasting just a few months) has a sudden severe breakthrough and ends in a few months. Severity and intensity are important features of droughts, as they relate to impacts affecting different sectors; for example, a severe event is more likely to affect the hydrological cycle (Van Loon, 2015), and an intense event is more likely to affect soil moisture (Hayes et al., 1999). According to SPEI-12, in our database one example of an exceptionally severe drought is the event in the mid-1950s in the United States - also known as the 1950s Texas drought (Woodhouse and Overpeck, 1998). On the other hand, an example of an exceptionally intense drought is that in the Philippines during the infamous 1997–1998 El Niño event (McPhaden, 1999).

Table 3.

Special score assigned to each drought event contained in the database.

Event feature	Range	Points	Condition
Severity (DS)	0–5	5	DS ≥ 90%ile over all the events recorded
(normalized)		4	70%ile ≤ DS < 90%ile over all the events recorded
		3	50%ile ≤ DS < 70%ile over all the events recorded
		2	30%ile ≤ DS < 50%ile over all the events recorded
		1	10%ile ≤ DS < 30%ile over all the events recorded
		0	DS < 10%ile over all the events recorded
Intensity (DI)	0–5	5	DI ≥ 90%ile over all the events recorded
		4	70%ile ≤ DI < 90%ile over all the events recorded
		3	50%ile ≤ DI < 70%ile over all the events recorded
		2	30%ile ≤ DI < 50%ile over all the events recorded
		1	10%ile ≤ DI < 30%ile over all the events recorded
		0	DI < 10%ile over all the events recorded
Area (DA)	0–5	5	DA ≥ 90% of the country (region) area
		4	70%ile ≤ DA < 90%ile of the country (region) area
		3	50%ile ≤ DA < 70%ile of the country (region) area
		2	30%ile ≤ DA < 50%ile of the country (region) area
		1	10%ile ≤ DA < 30%ile of the country (region) area
		0	DA < 10%ile of the country (region) area
Top event	0–4	1	Longest event for the country (region)
		1	Most severe event for the country (region)
		1	Most intense event for the country (region)
		1	Widest (largest area) event for her country (region)
Peak intensity	0–3	3	Lowest indicator value ≤ −2.5σ (σ is drought threshold)
		2	2.5σ ≤ Lowest indicator value < -2σ
		1	2σ ≤ Lowest indicator value < -1.5σ
		0	Lowest indicator value > −1.5σ
Peak area	0–3	3	Largest area ≥ 1M km2 for countries (3M km2 for regions)
		2	500k km2 (2M km2) ≤ Largest area < 1M km2 (3M km2)
		1	250k km2 (1M km2) ≤ Largest area < 500k km2 (2M km2)
		0	Largest area < 250k km2 for countries (1M km2 for regions)
Score (total)	0–25	12–25	Exceptional drought event
		8–11	Severe drought event
		0–7	Moderate drought event

The wider the area, the bigger the chance that the event caused remarkable impacts (Sheffield et al., 2009). Thus, the third parameter – the average percentage of area in drought conditions during the event – assigns up to 5 points. The Philippines drought in 1997–1998 was not only exceptionally intense, but also covered on average more than 80% of the national territories, thus this event receives 4 points in this category. Extra points (one for each feature) are given if, for a given country or macro-region, the drought event is the longest, most severe, most intense, or widest in the period 1951–2016. We underline that these thresholds, though based on percentiles, are necessarily subjective choices.

The last two categories in Table 3 regard the drought peak, which in some occasions could be linked with the largest impacts, as occurred for the Central European drought and heatwave in the summer of 2003 (Fink et al., 2004; Ciais et al., 2005). We considered two different peaks, the maximum intensity and the largest area hit by a drought. For example, the 1950s drought in the United States receives 3 points in the “peak area” category, as its maximum extension exceeded 1 million km² (Stahle and Cleaveland, 1988).

Based on the total number of points, the drought events are divided into three classes: moderate, severe, and exceptional. The definition of an extreme event is therefore different from that applied in previous studies (Spinoni et al., 2013, Spinoni et al., 2015a). Using the SPEI-12, our database contains, for example, 72 exceptional events at macro-regional scale (without considering events involving more than a macro-region), 85 severe events, and 112 moderate events. At macro-regional scale, the event with the highest total score (20) is the drought over southern South America in 2008–2009, which caused more than four billion dollar of damages to Argentinian farmers (BBC News, 2009, Inter Press Service, 2009; United States Department of Agriculture, 2008). This event also features in our database at country scale for Argentina (score: 21), Uruguay (score: 16), and Paraguay (score: 16).

In this study, the special scoring described above has been applied to each detected drought event, for both drought indicators, and at all spatial scales and accumulation periods. We then used the special score derived from the SPEI-12 indicator at macro-regional scale in order to list the biggest mega-droughts from 1951 to 2016. Depending on the indicator and accumulation period analyzed, this list might change to some extent. However, extreme droughts will reflect significantly in all indicators and accumulation periods. It should be also borne in mind that the proposed classification system should be viewed as a first attempt, which may be refined in planned updates of the database. Although the three main classes should not be used in a quantitative way, they do however follow an objective classification system, partly inspired by round-table discussions with experts during the 2017 European Drought Observatory User Meeting (Vogt et al., 2017).

In this study, we focused on country and macro-regional scales to assign the single parameters and the special score to drought events. This choice has limitations when, for example, a country is large and a drought event occurs in a relatively small part of the country only. In such a case, it can be missed or the statistics at country scale might not represent its characteristics in a proper way. In Fig. 4 we show, some country and macro-regional time series for the SPEI-3 and how the user can investigate single events (month by month) in the web platform of GDO.

Fig. 4.

Example of three drought events (A, B, and C) as they appear in the country and macro-regional time series and on the maps included in the Global Drought Observatory.

An example are the United States, which suffered multiple droughts between 2011 and 2017 (Los Angeles Times, 2017), in particular over the South-West and Central Plains. According to our data, the most severe period started in autumn 2011 and ended early 2013 (A in Fig. 4; special score for the United States is 16: exceptional drought), but another peak was observed in spring 2015 (B in Fig. 4; special score is 9: severe drought), which corresponds to the exacerbation of drought in California (USC, 2015). In our database, the events A and B can be found for both the United States and the macro-region WNA, but no information on the location can be inferred from the statistical parameters associated to them. However, on the GDO website, the user can visualize not only the country or macro-region series and the associated droughts with their parameters, but also the gridded indicators. In this way, the geographical features of the drought events can be visualized (Fig. 4, right column).

One of the planned improvements for our database is the choice of more refined macro-regions and the splitting of large countries. In the current version, we already did this for Russia, divided into European and Asian Russia. In this way, one can better characterize events like the one that hit western Russia in summer 2010 (Russo et al., 2015; C in Fig. 4).

3. Results and discussions

3.1. The database of drought events

The new global database of meteorological drought events is subdivided according to the indicator and accumulation period considered. In this section, we focus on the SPEI-12 and SPI-12 indicators, providing also some statistics on the events derived from SPEI-3 and SPI-3. Data and events derived from other accumulation periods are available in GDO.

For countries, at the global level, and based on SPEI-3, the overall number of drought events collected is 4827, while based on SPI-3 it is slightly smaller (4504). For longer accumulation periods (i.e. SPEI-12 and SPI-12) the numbers are smaller: 1992 and 1947 respectively. Of course, any event that spread over two or more countries is assigned to more than one country. Such large-scale events are also included in the database dedicated to macro-regions, where, based on SPEI-3 and SPI-3, the user will find 617 and 614 events respectively, and based on SPEI-12 and SPI-12, 269 and 266 events respectively.

In Section 3.2, we validate the list of the 52 detected mega-droughts between 1951 and 2016, checking their effective reporting in the scientific literature and media. Having a limited number of exceptional events to check makes it a feasible validation exercise, but looking for a confirmation of more than 4000 events is barely impossible in reasonable time. Thus, we focus on Europe – which is made of three macro-regions, see Table 1 – and as reference we used the independent European Drought Reference (EDR) and European Drought Impact Report Inventory (EDII) databases (http://www.geo.uio.no/edc/droughtdb/index.php), including events and impacts collected as part of the Drought-R&SPI project (Stahl et al., 2016). We highlight that this database was created based on impact reports, so it might suffer from a distribution of events biased towards the later decades. We also highlight that no real quantitative comparison can be done, so the validation is qualitative (presence or absence of the events).

Table 4 details how many of the 37 drought events included in the EDR-EDII database (period: 1951–2012) can be found in our database, depending on which input indicator and time scale we used. To make the comparison, we used the events per country (in our database) if the event in the EDR-EDII is labelled with a single country and we used the events per macro-region if the event in the EDR-EDII is labelled with a region (e.g., Scandinavia in EDR-EDII corresponds to Northern Europe in our database). 30 out of 37 events (i.e. the 81%) are included in our database in at least one of the four subsets analyzed, and 30 out of 36 events (i.e. 83%), if we consider the drought in Europe and Northern Europe in spring 2012 as a single event.

Table 4.

Drought events reported in the EDR-EDII database and in our database (according to different indicators and timescales). In bold, the events that are not present in our new global database.

Drought event (EDR-EDII)		Drought event in our database
Region or country	Period/indicator	SPI-3	SPEI-3	SPI-12	SPEI-12
Denmark	1954		√
Scandinavia	1955	√	√	√
Scandinavia	1959	√	√	√	√
Norway	1972	√	√
Scandinavia	1975	√	√	√	√
Europe	Summer 1976	√	√	√	√
Norway	1985	√	√	√	√
Norway	1989
Greece	1989	√	√	√	√
Greece	1990	√	√	√	√
Iberian Peninsula	1991–1995	√	√	√	√
Norway	1991				√
Scandinavia	1992	√	√	√	√
Norway	Winter 1993			√	√
Norway	1994			√	√
Netherlands	1995–1997	√	√	√	√
Scandinavia	1996	√	√	√	√
Northern Europe	1997	√	√	√	√
Greece	1999–2002	√	√	√	√
Germany	2000
Scandinavia	2002–2003	√	√	√	√
Europe	Summer 2003	√	√	√	√
Iberian Peninsula	2004–2007	√	√	√	√
Germany	2005
Northern Europe	Summer 2006		√		√
Greece	2006	√	√
Po Plain, Italy	2006–2007	√	√	√	√
Europe	2007	√	√	√	√
Northern Europe	2008		√
Northern Europe	2009
Central Europe	2010
Netherlands	2011	√	√
Europe	2011	√	√		√
Europe	Spring 2012		√	√	√
Northern Europe	Spring 2012
Netherlands	Summer 2012		√
England	Winter 2012

In our database, however, many more than 37 drought events are reported for Europe in the period 1951–2012. As an example, we take the case of Greece. According to the EDR-EDII, five droughts occurred in Greece in that period (1989, 1990, 1999–2002, 2006, and 2007, labelled as Eastern Europe in EDR). In our database and according to the SPEI-12, one can find eight events (1957, 1977–1978, 1985–1986, 1989–1991, 1992–1995, 2000–2002, 2007–2009), including all those in EDR-EDII–with the exception of the one in 2006, which is contained in our database, if the SPEI-3 is selected as input indicator. This example suggests that a combined use of the EDR-EDII database with ours is a viable option for analysing past drought events over Europe.

Users can exploit our database to discover, for example, which countries experienced the largest increase or decrease in drought frequency between the periods 1951–1980 and 1981–2016. According to SPEI-12, four of the top 10 countries showing the largest increase in drought frequency between these periods, are in the Mediterranean region (Italy, ranked first; Tunisia; Montenegro; Spain), while five are in the Sahel and central-western Africa (Liberia; Chad; Niger; Cameroon; Central African Republic). The same areas are hotspots according to SPI-12, because the corresponding top-10 countries include two Mediterranean countries (Albania; Spain) and six sub-Saharan countries (South-Sudan, ranked first; Central African Republic; Chad; Ethiopia; Ghana; Cameroon). Conversely, both SPEI-12 and SPI-12 highlight north-eastern Europe as the region with the largest decrease of drought frequency during the last decades. According to SPEI-12, four of the top-10 countries with the largest decrease in drought frequency are: Iceland; Sweden; Finland; European Russia (also Georgia and Armenia), while using SPI-12, we find the same four countries, plus Latvia and Poland.

Table 5 shows the difference in frequency and severity of drought events between the periods 1951–1980 and 1981–2016 for macro-regions. Five macro-regions show an increase in both frequency and severity (Mediterranean/MED; equatorial and southern Africa/EQF & SAF; central and eastern Asia/CAS & EAS), and four show a corresponding decrease in both quantities (central North America/CAN; northern and north-eastern Europe/NEU & NEE; southern Asia/SAS). The largest combined increases are in the Mediterranean region and equatorial Africa. Using the classification shown in Table 3 and applied to macro-regions, the largest increases of extreme drought events are in western North America, the Mediterranean region and eastern Africa, based on SPEI-12, and in eastern, western, and southern Africa, and eastern Asia, based on SPI-12.

Table 5.

Macro-regional differences in frequency (ΔDF, events/10y) and average severity (ΔDS, score) of drought events between 1951–1980 and 1981–2016, according to SPEIi-12 and SPI-12. In bold, the values that show opposite tendencies using the SPEI-12 and the SPI-12.

Region	1981–2016 minus 1951–1980
	SPEI-12		SPI-12
	ΔDF	ΔDS	ΔDF	ΔDS
ALA	0.7	−9.3	0.4	−8.3
WNA	−0.3	17.6	−0.9	15.0
CNA	−0.6	−8.1	−1.2	−11.2
ENA	−0.1	8.0	0.3	−13.9
CAM	0.9	−6.2	0.8	−7.0
AMZ	1.3	−10.6	1.3	2.7
CSA	0.7	−25.9	1.7	−24.4
SSA	0.0	−0.3	−0.9	6.5
NEU	−0.7	−4.7	−1.0	−16.3
NEE	−1.3	−4.8	−1.9	−13.3
MED	1.9	21.0	0.6	13.7
WAF	1.7	−4.1	1.0	9.1
EAF	2.5	1.4	1.7	−6.1
EQF	1.3	18.1	0.8	15.8
SAF	1.9	7.4	1.3	0.5
NAS	−0.6	8.5	−0.6	0.3
CAS	0.3	12.3	0.3	4.2
TIB	−1.2	4.6	−1.5	−8.4
EAS	1.9	13.8	0.9	6.0
SAS	−0.3	−2.5	−0.3	−1.1
SEA	−0.3	13.4	−0.3	10.6
NAU	0.0	−10.6	0.7	−17.5
SAU	0.0	19.2	0.0	−0.2

To summarize, Fig. 5 shows all of the detected drought events at macro-regional scale between 1951 and 2016, according to the SPEI-12. In the 1950s and 1960s, North and South America and northern Europe were hit by frequent extreme events. The 1970s saw the overall smallest number of droughts. In the 1980s, North America and the Sahel were the most hit by droughts. While the 1970s and early 1990s saw fewer droughts than other periods, from the late 1990s up to the present, southern Europe, Africa, most of Asia and southern Australia show a tendency towards more frequent droughts, with a couple of huge events over eastern Asia and Australia that lasted three years or more. One example is the Millennium Drought over south-east Australia that lasted from the late 1990s until 2009 (Heberger, 2012; Van Dijk et al., 2013).

Fig. 5.

World macro-regional drought events in 1951–2016 according to the SPEI-12. Moderate events are shown in yellow, severe events in orange, extreme events in red. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

3.2. The biggest drought events from 1951 to 2016

Using the scoring system reported in Table 3 and selecting SPEI-12 as the input indicator, we found 52 exceptional drought events at macro-regional scale (score ≥11), which we named “mega-droughts”. For this task, we chose SPEI-12 as it includes temperature and detects droughts forced by mega-heatwaves, such as that in summer 2003 over Central Europe (Rebetez et al., 2006). Moreover, a medium accumulation scale filters the flash droughts but detects seasonal and annual drought events (Spinoni et al., 2015a). The most important drawback is that some events might be recorded in our database with a few months delay, especially when they break out suddenly. Mega-droughts encompassing two or more macro-regions can show smaller scores in one of the regions, as the event can be centred over the borders between regions.

The list of the mega-droughts during 1951–2016 (Table 6) has been validated through a careful search of scientific papers and reliable media reports: out of 52 events, only one could not be traced, thus we can estimate an accuracy of 98%. Of course, this validation does not check if an event has been omitted over all possible droughts, as this would be unfeasibly complex and time-consuming. However, we highlight the fact that other regionally widespread droughts, such as the Amazon drought of 2005 (Zeng et al., 2008), are classified as severe and consequently are included in the database, but not in Table 6, because this is a list of exceptional events at very large scale. Thus, missing well-known droughts may be included in our database under a different class. Furthermore, as Table 6 is based on input data until December 2016, there are a few droughts that were still ongoing at the end of 2016, consequently the parameters of such events could change in updated versions of the database. We should also mention that, for recent droughts, it was not always easy to find scientific information for validation, as usually publications based on research projects or PhD theses focusing on specific drought events, are published with considerable delays.

Table 6.

List of macro-regional extreme drought events from 1951 to 2016 according to the SPEI-12. An event is defined extreme if the special score is larger than 12 (see Table 3). DS means drought severity, DI drought intensity, DA the average percentage of area in drought, and Sc is the special score. Due to space constraints, we report only the name of the leading author of the references, but in the references section we include the entire citation. The events “ending” on December 2016 could extend into 2017 or 2018.

#	Name	Reg	St	En	DS	DI	DA	Sc	Main reference(s)	Remarks
1	Scandinavia, Russia 1952	NEE	12/51	09/52	15.9	1.6	46.1	13	Briffa et al., 1994	Pan-European drought in the early 1950s.
2	U.S. (Texas) and Mexico 1951–57	CAM CNA ENA	12/51 09/52 07/55	06/54 05/57 06/56	35.0 83.8 19.6	1.1 1.5 1.6	30.8 40.7 32.5	13 17 14	Woodhouse and Overpeck, 1998 Liverman, 1999 Cook et al., 2007 Seager et al., 2009 Méndez and Magaña, 2010	Southern Texas, Northern Mexico above all, low precipitation. Severe as the Dust Bowl in the 1930s, involved mostly Central and Southern Great Plains, Central U.S. but also Eastern US and Florida.
3	Central Australia 1957–58	SAU	09/57	09/58	20.4	1.6	60.9	14	Heathcote, 1969	Central Australia, especially the outback, little long-term damages to vegetation.
4	UK, Northern Europe, Russia 1959–60	NEU NEE	06/59 06/59	08/60 06/61	30.7 39.1	2.0 1.6	60.0 40.7	18 17	Marsh et al., 2007 Meshcherskaya and Blazhevich, 1997	Excessive sunshine, very low cloudiness, low humidity, heatwave, impacts on crops especially in the U.K. In Russia and Eastern Europe, yield decrease (especially grains) was remarkable.
5	Argentina, Uruguay 1961–64	CSA SSA	11/61 03/62	10/64 08/63	53.4 26.3	1.5 1.5	34.6 45.8	15 14	Sheffield and Wood, 2007 Boulanger et al., 2005	La Niña event caused dry conditions over Eastern Southern South America. This event was similar to the one in 1988, droughts linked with La Niña are uncommon
6	Tibetan Plateau 1962–64	TIB	07/62	01/64	25.7	1.4	32.9	12	Zhu et al., 2011	Low precipitation from 1961 onwards, effects in 1962. Anomalous circulations over Pacific and Indian Ocean, together with anomalous monsoon season partly drove the drought.
7	North-Eastern U.S., Great Plains 1962–64	ENA CNA	06/62 06/63	07/64 01/65	34.1 28.4	1.3 1.4	32.4 33.5	12 14	Seager et al., 2012 Schubert et al., 2004	Northeastern North America involved, New York City experienced problems. This event was a break during wetting trend years. Unusually long dry-inducing atmosphere circulation, extreme negative Northern Atlantic Oscillation, and Northern Pacific sea surface temperature anomalies.
8	Argentina, La Plata Basin 1967–72	CSA SSA	01/67 05/68	01/72 10/72	77.1 49.8	1.3 0.9	31.5 33.8	16 13	Rivera and Penalba, 2014 Minetti et al., 2007	Central and Western part of Southern South America, involved in particular semi-arid areas, but also the La Plata river basin. Negative sea surface temperature anomaly due to El Niño, corn productivity decreased in many areas in this region.
9	Alaska, Canada 1969–70	ALA	01/69	08/70	31.5	1.6	40.7	16	Xiao and Zhuang, 2007	1969 was an exceptional year for forest fires, reinforced by drought in Alaska and Canada
10	Australian Queensland 1969–70	NAU	08/69	03/71	25.3	1.3	51.6	14	Queensland Government, 2009	Queensland drought in 1969–70 was mainly due to prolonged lack of rain.
11	India 1972–73	SAS	07/72	07/73	26.4	2.0	43.0	18	Kanamitsu and Krishnamurti, 1978	Circulation anomalies (hypothesis), summer monsoon was anomalous.
12	European Russia 1972–73	NEE	06/72	09/73	33.3	2.1	47.9	16	Bradford, 2000	Very dry winter in Eastern Europe and Russia, where some rivers recorded lowest levels of the 20th century.
13	Tibetan Plateau, Mongolia 1974–76	TIB	05/74	07/76	33.3	1.2	31.7	12	-	-
14	U.K., Baltic Republic, Russia 1975–77	NEE NEU	05/75 12/75	08/76 06/77	34.6 35.8	2.2 1.9	54.9 56.7	17 17	Sheffield et al., 2009 Perry, 1976	The drought involved the Baltic countries and moved eastwards in 1976 until the Caspian Sea. Drought due to low/very low rainfall, especially over England and Wales.
15	U.S. Great Plains 1976–77	CNA	07/76	11/77	33.0	1.9	46.5	17	Diaz, 1983	Spring drought (1997 and afterwards), due to very dry weather conditions over Central and Western United States
16	Western U.S., Central Plains 1977–82	ALA CNA	09/76 07/80	09/81 04/82	54.4 31.0	0.9 1.4	29.1 38.1	12 14	Karl and Quayle, 1981 Namias, 1982, Namias, 1983	Extensive drought over the Central Plains, in summer 1980 an exceptional heatwave (up to + 4.5 °C), lack of cloud cover caused relative humidity to be very low during the day also over Central U.S. However, precipitation were not so low.
17	Southeast Asia, Pacific Islands 1982–83	SEA	10/82	11/83	24.7	1.8	45.2	15	Quiroz, 1983 Gibbs, 1984 Kiladis and Diaz, 1986	Coupled with a strong El Niño event (anomalous climate conditions), this drought hit Indonesia, Papua, the Philippines, Borneo, northern Australia.
18	Sahel drought 1983–88	WAF EAF	03/83 02/84	07/88 08/85	99.2 39.9	1.5 2.1	45.7 49.0	18 17	Henricksen, 1986 Gommes and Petrassi, 1996 Hulme, 2001	The mid-1980s Sahel drought was driven by low precipitation from the early 1980s and resulted in extended desiccation and degradation of large areas. The drought hit Ethiopia in 1983–84 with important impacts.
19	Conterminous United States 1985–91	ENA CNA WNA	05/85 10/87 10/87	05/89 11/90 05/91	61.9 46.6 50.9	1.3 1.2 1.2	29.1 36.2 31.6	14 15 12	Andreadis et al., 2005 Kogan, 1995 Peters et al., 2002	Very long drought in the conterminous U.S., sources report 4–7 years of consecutive drought conditions. This is also considered a soil moisture and runoff drought. Vegetation impacts were outstanding and are clearly visible from satellite images.
20	Patagonia 1988–90	SSA	11/88	03/90	31.6	1.9	55.9	16	Rivera and Penalba, 2014	Drought over Patagonia and Central Southern South America, with large impacts on grain production and hydroelectric power production. Coupled with La Niña event.
21	Balkans, Greece 1989–91	MED	03/89	06/91	35.4	1.3	33.0	14	Tselepidaki et al., 1992	Low precipitation over Greece and the Balkans triggered drought conditions.
22	Philippines, Indonesia 1992–93	SEA	10/91	07/95	47.1	1.0	30.2	13	Salafsky, 1994 Hilario et al., 2009	Strong (but not exceptional) 1991–92 El Niño event, drought involved the Philippines (large impacts) and Indonesia.
23	Northeastern Brazil 1992–93	AMZ	01/92	10/93	30.9	1.4	35.7	12	Rao et al., 1995	Drought followed a dry period in northeastern Brazil and the Amazon forest.
24	Southern Africa 1992–93	SAF	02/92	09/93	27.3	1.4	43.3	12	Unganai and Kogan, 1998 Eldridge, 2002 Munro, 2006	Rivers dried, millions of cattles died, cereal production dropped, 86 million people affected, malnutrition problems in Zimbabwe.
25	Western U.S., Mexico 1994–97	ALA WNA	08/94 07/94	06/97 05/95	50.7 15.0	1.5 1.4	37.4 36.5	15 13	Hayes et al., 1999 Chávez, 1999	Drought hit more Western and Southwestern U.S. in 1995–96. Over Northern Mexico, the Rio Bravo region suffered the largest impacts.
26	South Africa, Botswana 1995–96	SAF	12/94	01/96	23.5	1.7	52.4	14	Mussá et al., 2015	Drought over South Africa and Botswana, named the worst drought in 70 years in the Crocodile river catchment. Groundwater was used as emergency source.
27	U.K., France, Denmark 1996–97	NEE	02/96	04/97	22.3	1.5	35.7	12	Fleig et al., 2011 Parry et al., 2012	United Kingdom and Denmark involved above all, hydrological issues, many river catchments in England recorded low levels.
28	Southeast Asia 1997–98 (El Nino)	SEA TIB	09/97 08/97	12/98 07/98	34.0 18.3	2.1 1.5	53.7 32.9	19 13	Nakagawa et al., 2000 Ronghui et al., 2000	Drought coupled with El Niño event in 1997–98. Relevant forest impacts, summer climatic anomalies. In southeastern Asia and the Pacific Islands, El Niño 1997–98 was a super event. Over Borneo, Indonesia, and Eastern Asia fires aggravated drought and vice versa.
29	Central Amazonia 1997–99	AMZ	10/97	03/99	29.3	1.6	40.6	13	Williamson et al., 2000 Nepstad et al., 2004	Increased mortality rate of trees in Central Amazon.
30	Central Asia, Pakistan, China 1999–03	CAS EAS	02/99 08/99	03/02 08/03	71.0 61.1	1.9 1.3	52.5 31.9	19 12	Barlow et al., 2002, Barlow et al., 2016 Kogan, 2002 Zhang and Zhou, 2015	Cold sea surface temperature anomalies over the Indian Ocean and the Western Pacific. Worst drought in Pakistan and Afghanistan in 50 years. Drought peaked over northern China and the Yangtze river basin was heavily impacted, causing great crop losses.
31	Western U.S. and Canada 2000–04	WNA	07/00	08/04	58.8	1.2	31.7	15	Seager, 2007 Woodhouse et al., 2010 Schwalm et al., 2012	This drought peaked in 2002, lasted from 2000 to 2004 and was defined the worst in 800 years regarding impacts on forests and empty rivers. Carbon sequestration cut by 51%.
32	Balkans, Greece, Cyprus 2000–02	MED	06/00	09/02	40.1	1.4	38.9	14	Pashiardis and Michaelides, 2008 Stagge et al., 2013	Drought over southeastern Europe, the Balkans, Greece, and countries over the Aegean Sea. A combination of hot summer temperatures and low rainfall was the cause.
33	India 2002–03	SAS	06/02	09/03	24.5	1.5	38.9	14	Dutta et al., 2015	All-India drought year, big problems for agriculture in Rajasthan, Uttar Pradesh, Madhya Pradesh, Chhattisgarh, Karnataka, and Tamil Nadu. Anomalous monsoon caused floods after drought.
34	Tropical Africa, Congo river basin 2002–06	EAF EQF	05/02 11/03	07/03 11/06	21.9 54.1	1.5 1.5	34.7 36.1	13 18	Calow et al., 2010 Zhou et al., 2014	More than one African region affected, wide surface involved, groundwater affected over central Africa. Forest degradation and decreased greenness could be observed with remote sensing images over forests in the Congo river basin.
35	European heatwave-drought 2003	NEU	03/03	08/04	31.5	1.8	51.0	16	Fink et al., 2004 Ciais et al., 2005 Rebetez et al., 2006	Summer 2003 heatwave over central Europe caused a lot of deaths (especially in France), primary productivity reduction, crop failures, impacts also on natural vegetation. The heatwave lasted from March to September.
36	SW China, Yangtze River basin 2004–10	EAS TIB	07/04 08/06	04/10 01/10	118.0 52.2	1.7 1.2	34.0 34.3	18 16	He et al., 2011 Barriopedro et al., 2012, Yang et al., 2012 Zhang et al., 2013	In 2006–07 an extreme drought hit Sichuan and the Yangtze river basin, causing a fall in production of rice, potatoes, and beans. The drought moved to Yunnan and Southwestern China in 2008–09. It was assumed that an anomalous monsoon season over India pushed extreme drought conditions over southwestern China.
37	U.S. Great Plains and Canada 2006–07	CNA	06/05	05/07	20.4	1.7	43.2	14	Dong et al., 2011 Basara et al., 2013	Drought in the U.S. Great Plains and Canada, causing impacts in many sectors. Positive pressure anomalies over Southwestern deserts combined with negative anomalies over the Great Lakes.
38	The Millennium Drought (Aus) 2006–10	SAU	10/06	08/10	51.7	1.1	41.9	13	Van Dijk et al., 2013 Heberger, 2012	The Millennium Drought lasted 5–8 years (in some areas 10 years). Declines in rainfall and runoff led to widespread crop failures, livestock losses, dust storms, and bushfires.
39	Southwestern Europe, Turkey 2007–08	MED	01/07	01/09	42.3	1.7	36.3	15	August et al., 2008 Simsek and Cakmak, 2010	This drought had agricultural impacts over Turkey during a hot-dry summer. Also Cyprus and Greece were impacted, but with a lower degree of severity.
40	Argentina, Chile 2008–10	SSA	03/08	01/10	42.9	1.9	56.2	20	NASA EO (2009),Müller et al., 2014	Considered one of the top droughts in the last five decades over Southern America. It resulted in vegetation, soil, and crop impacts.
41	Horn of Africa 2008–10	EAF	08/08	08/10	45.3	1.8	44.5	16	Zaitchik et al., 2012 Masih et al., 2014 Nicholson, 2014	Due to prolonged low rainfall regimes, this drought pushed another drought (2011), but this time only over the easternmost part in the Horn of Africa.
42	Middle-East, Central Asia, India 2008–10	CAS SAS	03/08 06/09	09/09 08/10	30.2 27.3	1.8 1.8	54.2 38.8	16 15	USDA, 2008 Neena et al., 2011 Cook et al., 2016 FAO, 2017	Middle-East worst drought since decades: Iran, Iraq, Syria, and Middle-East reduced grain production, wheat production (-23%). Also Tajikistan was severely involved in 2009–10. India 2009 drought occurred during the summer monsoon season. Internal circulation and weather dynamics had the leading role for India.
43	Russian heatwave-drought 2010	NEE	07/10	07/12	32.3	1.3	34.4	13	Wegren, 2011, Trenberth and Fasullo, 2012 Russo et al., 2015	The Russian mega-heatwave in 2010 forced drought and food insecurity. Large tundra and forest fires.
44	Central U.S. 2011–13	CNA	09/11	09/13	47.0	1.9	48.5	18	Hoerling et al., 2014	Very scarce spring precipitation caused this drought with no early warning. A dry summer prolonged the drought conditions.
45	U.K., Central-Southern Europe 2011–13	MED	10/11	08/13	33.3	1.5	37.4	15	Bissolli et al., 2012	Drought in 2011 over U.K. and then southern Europe in spring 2012. Main causes: dry periods in 2010–12 and high-pressure conditions. Lot of impacts on crop yields, water supplies, waterways, and health.
46	Southern U.S., Mexico 2011–13	CAM	08/11	10/13	29.0	1.1	33.9	12	Seager et al., 2014 Brower et al., 2015	Drought in Southern U.S. and Mexico started in winter 2010 and went on in 2011. Possible drivers: La Niña event in the tropical Pacific ocean, sea surface temperature anomalies, negative North Atlantic Oscillation in winter 2010–11.
47	China spring-summer 2011	EAS	04/11	06/12	24.4	1.6	33.9	12	Lu et al., 2014	Spring drought in 2011 over China, defined as once-in-a-50-year drought over the Yangtze river basin and the southern region. Shortage of drinking water for people and livestock, tremendous losses in agriculture and shipping industry.
48	U.S. and California drought 2011–14	ENA WNA	05/12 05/12	06/13 04/14	21.4 31.1	1.5 1.3	33.2 35.4	13 13	Griffin and Anchukaitis, 2014 Seager et al., 2015 Otkin et al., 2016	Flash drought over Central and Northeastern U.S., with soil moisture conditions changing rapidly. This drought includes the 2011–14 California drought, an unusual drought spread over New Mexico and northern Mexico. It was driven by high temperatures and low (but not exceptionally low) rainfall.
49	East Australia 2013–16	SAU	01/13	08/16	58.0	1.3	45.4	15	ABC, 2013 Aus Gov (BOM), 2015	After the Millennium drought, in 2012–14 drought came back to Australia, especially over the East. It was due to long-term rainfall deficits. Also Queensland was hit.
50	South Africa 2015–16 (18)	SAF	02/15	12/16	43.8	1.9	54.4	19	Archer et al., 2017; Phys. Org. (2018) Masante et al., 2018 Yuan et al., 2018	This drought caused the Cape Town water crisis 2018, with reservoir volumes down to 19%. People collaborated to save water to avoid “day zero”. After an agricultural drought in 2017–18, South Africa declared the state of national disaster in March 2018 (lifted in June 2018)
51	Mediterranean 2015–16 (17)	MED	11/15	12/16	24.6	1.8	41.5	17	Van Lanen et al., 2016 Di Giuseppe et al., 2017 Garcia-Herrera et al., 2018	Drought started in 2016 due to low rainfall, especially in Southern Europe; it was prolonged by a summer heatwave over southern Europe in 2017, especially over Italy and Northern Africa (up to 48 °C).
52	Amazonia, central S-America 2015–16 (17)	AMZ CSA	10/15 04/16	12/16 12/16	35.8 12.7	2.4 1.4	49.7 47.8	16 12	Jiménez-Muñoz et al., 2016 Erfanian et al., 2017 Reliefweb, 2017	El Niño event in 2015–16 forced drought over the Amazon forest with an unprecedented warming in eastern Amazonia (western Amazonia was wet). Eco-hydrological consequences from the 2016 drought are more severe and extensive than the 2005 and 2010 droughts. Human factors potentially contributed to drought severity.

Table 6 provides an overview of the periods that experienced big droughts during 1951–2016. Four exceptional droughts occurred in the 1950s, one of which encompassed three macro-regions (Central America, eastern and central North America). The 1960s were characterized by six exceptional events (two ending in the 1970s), none having a special score higher than 16. Also the 1970s saw six exceptional events, while the 1980s had five such events. An increasing trend is evident from the early 1990s onwards: eight exceptional events in the 1990s, thirteen in the 2000s, and ten during 2010–2016. If we compare the number of exceptional events during 1951–1980 (16) versus 1981–2016 (36), the tendency towards more frequent exceptional droughts is evident.

Table 6 also includes some details on the drivers and sectorial impacts of the 52 mega-droughts. A recurrent feature of the events listed in Table 6 is the correlation between strong El Niño events and mega-droughts, such as 1982–1983, 1997–1998, and 2014–2016 (Wang et al., 2017). On the other hand, extreme droughts and La Niña events are less frequently correlated (e.g. 1988–1989 and 2010–2011). More details about El Niño and La Niña events can be found in Trenberth (1997).

3.3. Global drought tendencies from 1951–1980 to 1981–2016

We used time series of the SPEI-12 and SPI-12 indicators to investigate drought trends. Specifically we compared the frequency (Fig. 6) and average severity (Fig. 7) of drought events during the periods 1951–1980 and 1981–2016. Some areas experienced increased drought frequency based on both indicators (but larger based on SPEI-12): the U.S. East Coast; Amazonia and north-eastern Brazil; Patagonia; the Mediterranean region; most of Africa; north-eastern China. The most relevant decrease is in northern Argentina, Uruguay and northern Europe. Compared to 1951–1980, during 1981–2016 the meteorological droughts were more severe for both SPEI-12 and SPI-12 over north-western U.S., parts of Patagonia and southern Chile, the Sahel, the Congo River basin, southern Europe, north-eastern China, and south-eastern Australia. Conversely, eastern U.S., south-eastern Brazil, northern Europe, and central-northern Australia experienced less severe droughts.

Fig. 6.

Average frequency of drought events in 1951–1980 (upper boxes), in 1981–2016 (central boxes), and difference between the two periods (lower boxes), according to the SPEI-12 (left boxes) and to the SPI-12 (right boxes).

Fig. 7.

Average severity of drought events in 1951–1980 (upper boxes), in 1981–2016 (central boxes), and difference between the two periods (lower boxes), according to theSPEI-12 (left boxes) and to the SPI-12 (right boxes).

Combining the results in Figs. 6 and 7, we can identify the “hotspots” hit by more frequent and more severe meteorological droughts in the last decades: Patagonia; the Mediterranean region; the Sahel; the Congo River Basin; north-eastern China. The same hotspots were indicated by Sheffield et al. (2012), Dai, 2011, Dai, 2013, Spinoni et al. (2014), and Dai and Zhao (2017), though the spatial borders do not perfectly overlap because of the use of different indicators – e.g. the Palmer Drought Severity Index (Palmer, 1965) – and time periods.

Patagonia (and southern South America), however, represents an exception as it is not considered a hotspot in Sheffield et al. (2012) and only limited areas in southern Chile and Argentina show a trend towards more frequent droughts according to Dai (2013), Spinoni et al. (2014), and Dai and Zhao (2017). The main reason for the discrepancy concerns the climate data that were included in the different studies: in Dai (2013) and Spinoni et al. (2014) these extended only until 2010, in Dai and Zhao (2017) only until 2012, in Sheffield et al. (2012) only until 2008, and in our study until 2016. According to Table 6, an extreme drought hit Argentina and Chile in 2008–2010 and this event could not be taken into account in Sheffield et al. (2012). Moreover, in our database (for SPEI-12), out of eleven droughts over southern South America in 1951–2016, three occurred in 2008–2016, specifically 2008–2010, 2011–2012, and 2013–2014 (Naumann et al., 2019).

Regarding the areas in drought (Fig. 8), our results agree with the previously cited studies: at global level and according to SPEI-12, a small increase is observable especially after the mid-1990s. Since no significant trend is found according to SPI-12, the trend documented by SPEI-12 is likely temperature-driven. The macro-regions with a tendency towards a higher percentage of areas in drought are: southern South America (SSA); the Mediterranean region (MED); central and eastern Asia (CAS and EAS); the entire African continent south of the Sahara (WAF, EAF, EQF, and SAF).

Fig. 8.

Percentage of areas in drought conditions from 1951 to 2016 according to the SPI-12 (blue) and to the SPEI-12 (red) for 23 macro-regions and at global level. For the list of the acronyms, see Table 1. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

3.4. The role of temperature in meteorological drought

Table 7 shows the average global shifts between 1951–1980 and 1981–2016. According to the SPEI-12 indicator, the increase of drought frequency is approximately 9.7% and the absolute increase of areas in drought is 3.3%. According to SPI-12, however, an opposite slight decrease of drought frequency and areas in drought can be observed. On average, both indicators point towards longer, more severe, and slightly more intense droughts, but they diverge on drought frequency and areas in drought (Table 7).

Table 7.

Global statistics computed at grid point scale. DF stands for drought frequency, DD for drought duration, DS for drought severity, DI for drought intensity, and DA for area in drought conditions. DD, DS, and DI refer to the “average” drought event, DA to the area in drought conditions averaged over all the drought events.

Global	SPEI-12				SPI-12
	1951–1980	1981–2016	Absolute shift	Percent shift	1951–1980	1981–2016	Absolute shift	Percent shift
DF (ev/10y)	1.75	1.92	0.17	9.7	1.76	1.71	−0.05	−2.8
DD (months)	17.44	18.96	1.52	8.7	18.00	18.41	0.41	2.3
DS (score)	19.14	21.82	2.68	14.0	20.37	21.82	1.45	7.1
DI (score)	1.09	1.15	0.06	5.5	1.13	1.17	0.04	3.5
DA (%)	15.24	18.56	3.32	21.8	15.41	15.35	−0.06	−0.4

The difference between the two indicators is small for drought frequency in 1951–1980, with SPEI-12 showing a smaller number of events compared to SPI-12, over semi-arid areas in Kazakhstan, Uzbekistan, and northern China (Fig. 9). In 1951–1980, SPEI-12 shows a larger drought severity (compared to SPI-12) over northern Colombia and Venezuela, Bolivia, and semi-arid areas in western Argentina, whilst fewer severe events are shown especially over southern central Australia. In 1981–2016 the differences between the indicators are greater, due to the progressive temperature increase in the last decades (IPCC, 2014) with the largest differences over southern Africa and western Australia regarding drought frequency, and over western Amazonia, the Sahel, central Asia, Mongolia, and southern Australia regarding drought severity.

Fig. 9.

Comparisons between frequency (DF) and average severity (DS) of drought events in 1951–1980 (left boxes) and in 1981–2016 (right boxes) as derived by using the SPEI-12 and to the SSPI-12.

The role of temperature in meteorological droughts is crucial (Dai et al., 2018) and the methodology to estimate evapotranspiration has been frequently debated (Van der Schrier et al., 2011, Trenberth et al., 2014, Spinoni et al., 2017). In this study, we used the Penman-Monteith approach (Allen et al., 2006), which is generally considered more robust than two of the most used approaches – Hargreaves–Samani (Hargreaves and Samani, 1985, Samani, 2000) and Thorntwaite (Thornthwaite, 1948). However, particularly for short-accumulation periods, SPEI can be problematic in semi-arid areas and tends to overestimate drought severity compared with SPI, because of the difficulty to estimate potential evapotranspiration (PET) in such conditions (e.g., Hernandez and Uddameri, 2014). Examples are north-western Chile, the Arabian Peninsula, mountainous regions surrounding the Aral Sea, the cold Gobi desert, and the Australian outback (Fig. 9).

Similarly to Spinoni et al. (2015c) for Europe, we analyzed the change in precipitation and PET between 1951–1980 and 1981–2016, as shown in Fig. 10, to evaluate whether the drying or wetting tendency reflected by drought indicators was driven by precipitation, temperature (as included in PET), or both.

Fig. 10.

Drought tendencies from 1951–80 to 1981–2016 according to the SPI-12 (top) and the SPEI-12 (bottom) and corresponding increase or decrease of precipitation (RR) and potential evapo-transpiration (PET), per country.

The only input variable for SPI is precipitation, so an increase or decrease in precipitation is very likely to result in, respectively, a wetting or drying tendency for SPI. However, over some countries the variability of precipitation plays a more important role than the mean change: in Guyana, Poland, and Libya, for example, SPI-12 shows a slight wetting tendency despite a small decrease in precipitation; conversely, in Armenia, Georgia, Oman, Tajikistan, and Bangladesh, SPI-12 suggests a drying tendency despite a small increase in precipitation. This occurs in particular in small countries such as Oman and Tajikistan, where arid and semi-arid areas are predominant, and consequently the computation of SPI may suffer from biases.

Fig. 10 also provides a visualization of the geographical patterns: according to SPI-12, and with few exceptions, in the last decades North America, southern South America (excluding Chile), central and northern Europe, central, northern, and eastern Asia (excluding North Korea and Japan), and Australasia experienced a wetting tendency. Conversely, tropical South America, southern Europe, most of Africa (excluding the Horn of Africa), and South Asia experienced a drying tendency.

The influence of temperature on drought trends is clear from Fig. 10. Switching from SPI-12 to SPEI-12, Mexico, Canada, central Europe, central and eastern Asia, and Australia change from a wetting to a drying tendency, and, in most of the countries characterized by this change, this is due to the increase in PET (i.e. hotter climate), which outweighs the wetting trend. Over the United States, northern Europe, Russia, and south-eastern Asia, the warming trend is not strong enough to outweigh the precipitation increase. According to SPEI-12, the worst drying combination (i.e. precipitation decrease and PET increase) occurs over Brazil, the Mediterranean region, most of Africa, the Middle East, South Asia, and Japan.

Combining the information in Fig. 10 with the drought hotspots listed in previous sections, we distinguish three different situations. Over the Mediterranean Region, the Sahel, and the Congo River basin, the increase of drought frequency and severity is due to both precipitation decrease and hotter climate conditions. Over north-eastern China, the outstanding increase in PET due to hotter temperatures is the main driver of more frequent and severe meteorological droughts. Argentina (with the exception of Patagonia) shows, on average, a decrease in drought frequency and severity (for both SPI-12 and SPEI-12), due to a precipitation increase and a (very small) decrease in PET, especially at southernmost latitudes. Patagonia, which was hit by recurrent long and severe drought events in the last decades, exhibits a different behaviour.

4. Conclusions

The presented global database of meteorological drought events during the period 1951–2016 contains thousands of events organized according to country and macro-region. Each featured event is assigned a set of parameters that may be used as “proxy” variables for analysing potential drought impacts. The database is also divided according to the SPI and SPEI indicator, and for various accumulation periods, ranging from 3 to 72 months. All entries in the database – together with dedicated maps, time series, and charts – are available to the public through the European Commission's Global Drought Observatory. It can be exploited to look for past events, to perform historical analyses, and to compare current events with those that have previously occurred. The information in the database can support the development of drought management strategies (Estrela and Vargas, 2012), and thus drought mitigation and preparedness efforts (Wilhite et al., 2007, Vogt and Somma, 2000). Especially countries with a scarcity of long-term high-quality climatic data (Mirza, 2003) can benefit from this country-based historical dataset of drought events.

We have furthermore presented the first scientific results derived by exploiting the database, such as trends in global drought frequency and severity between 1951 and 2016, which enables the identification of drought hotspots around the globe. It is planned to update the drought database approximately every two years, although this depends also on the updating frequency of the input datasets (i.e. GPCC and CRU TS). A most recent update, which extends the time interval of the GPCC dataset from 1891 to 2016, will be included in the next version of our database.

A limitation in our study is the use of two input datasets only. An evaluation is currently underway of two other updating options: the TerraClimate dataset (Abatzoglou et al., 2018), already available until 2017, and the ERA5 reanalyses dataset of the European Centre for Medium-range Weather Forecasts (ECMWF), whose complete release is scheduled for 2019. With a higher spatial resolution than 0.5°, these datasets enable analysis of drought patterns at finer scales. A note of caution is that the introduction of other datasets not only provides information that is more complete, but can also increase uncertainty, especially if different input data lead to contradicting results (e.g. occurrence or non-occurrence) regarding single drought events.

Despite the large number of drought indicators currently in use (Heim, 2002, Mishra and Singh, 2011), we deliberately limited our analysis to SPI and SPEI, as the major objective of our study was to analyze meteorological droughts due to rainfall deficits, and (separately) those forced by climate conditions that are both dry and hot. This choice has its limitations and no sector-based drought modelling was included. The use of two indicators and multiple time scales, however, enlarges the potential applicability of the database: while the events based on a 3- to 6-month accumulation periods are often used to investigate on agricultural drought impacts (Rhee et al., 2010), those based on 12-month or longer time scales are often linked with hydrological drought impacts (Nalbantis and Tsakiris, 2009), for example. In the near future, we plan to consider also vegetation indicators, to examine, for example, if the detected meteorological droughts have affected forests and green areas (McCormick and Gobron, 2016), and how far changes in land use and land cover, such as deforestation processes in the Amazonia and the Congo basins (Malhi et al., 2008, Megevand et al., 2013) affect drought events and are interconnected with drought patterns. Candidate indicators are the fraction of Absorbed Photosynthetically Active Radiation (fAPAR; Myneni and Williams, 1994), the Normalized Difference Vegetation Index (NDVI; Peters et al., 2002), the Leaf Area Index (LAI; Carlson and Ripley, 1997), and the Vegetation Health Index (VHI; Bento et al., 2018).

In this study, we have detected the following drought hotspots that have experienced a robust increase in drought frequency and severity between the periods 1951–1980 and 1981–2016: Patagonia, the Mediterranean region, the Sahel, the Congo River basin, north-eastern China, and (though not unanimously agreed by all indicators) central Asia and southern Australia. Using the same indicators, methodology, and regions that are described in this paper, we are currently investigating the drought hotspots as the 21st century progresses. Preliminary results, obtained using more than a hundred simulations from the Coordinated Regional Climate Downscaling Experiment (CORDEX), indicate that the hotspots of the last decades are projected to see a further increase in frequency, duration, and severity of meteorological droughts. Consequently, it is important to implement efficient adaptation strategies to avoid severe economic impacts, or even triggering irreversible land degradation processes in these areas. These results and the likely increase of drought frequency and severity under a changing climate, underline the need for efficient adaptation strategies.