Multi-hazard events can be associated with larger socio-economic impacts than single-hazard events. Understanding the spatio-temporal interactions that characterize the former is therefore of relevance to disaster risk reduction measures. Here, we consider two high-impact hazards, namely wet and dry hydrological extremes, and quantify their global co-occurrence. We define these using the monthly self-calibrated Palmer Drought Severity Index based on the Penman–Monteith model (sc_PDSI_pm), covering the period 1950–2014, at 2.5
Natural hazards can interact in diverse ways, leading to multi-hazard events that can exacerbate disaster losses when compared to single-hazard occurrences (Zscheischler et al., 2018). Examples of multi-hazards are the co-occurrence of heavy precipitation or flooding with wind damage from extratropical cyclones (De Luca et al., 2017, 2020; Waliser and Guan, 2017), storm surge combined with fluvial flooding in deltas (Ward et al., 2018), flood episodes along with droughts (Collet et al., 2018), and landslides triggered by earthquakes (Kargel et al., 2016). Such combinations can lead to situations beyond the worst-case scenario planned by emergency managers, (re)insurance companies, businesses, and governments and thus present a critical challenge for disaster risk reduction (Zscheischler et al., 2018). The relevance of multi-hazards has been recognized by scientific and stakeholder communities, and both have devoted significant efforts to the topic over the past decade (e.g. Forzieri et al., 2016; Gallina et al., 2016; Gill and Malamud, 2014; Kappes et al., 2012; Terzi et al., 2019; Zscheischler et al., 2018). Indeed, the United Nations Sendai Framework for Disaster Risk Reduction (UNISDR, 2015) now advocates multi-hazard approaches to disaster risk reduction.
Analysis of multi-hazards is highly relevant given anthropogenic climate change. Events such as floods and droughts already have significant humanitarian and socio-economic impacts (Alfieri et al., 2016; Barredo, 2007; Di Baldassarre et al., 2010; Jonkman, 2005; Naumann et al., 2015; Van Loon et al., 2016; Zhang et al., 2011) and are expected to become more frequent and/or severe in the future (Arnell and Gosling, 2016; Dai, 2011a, 2012; Hirabayashi et al., 2013; Hirsch and Archfield, 2015; IPCC, 2012; Milly et al., 2002), albeit with a large degree of uncertainty (e.g. Orlowsky and Seneviratne, 2013). Numerous studies have investigated the combination of flood and drought events or, more generally, wet and dry hydrological extremes at local and regional scales for both present and future climates (e.g. Berton et al., 2017; Collet et al., 2018; Deangelis et al., 1984; Di Baldassarre et al., 2017; Gil-Guirado et al., 2016; Oni et al., 2016; Parry et al., 2013; Pechlivanidis et al., 2017; Quesada-Montano et al., 2018; Yan et al., 2013; Yoon et al., 2018). Examples include the analysis of abrupt drought–flood transitions in river basins in China (Yan et al., 2013) and in England and Wales (Parry et al., 2013). There is also the dynamical interplay between society and hydrological extremes, intended as the mutual influence of human activities on floods and droughts (Di Baldassarre et al., 2017), and indices assessing the long-term evolution of vulnerability and adaptation to these hazards (Gil-Guirado et al., 2016). Other studies consider wet–dry interactions from a statistical perspective (Collet et al., 2018) or have related these two independent hazards to large-scale modes of climate variability (Cai and Rensch, 2012; Lee et al., 2018; Nobre et al., 2017; Siegert et al., 2001; Ward et al., 2014; Yoon et al., 2018).
Quantifying wet and dry (also extreme) hydrological events at both regional and global scales is a non-trivial task. Some commonly used metrics include the Palmer Drought Severity Index (PDSI) (Dai et al., 2004; Palmer, 1965), the Standardized Precipitation Index (SPI) (McKee et al., 1993, 1995), and the Standardized Precipitation Evapotranspiration Index (SPEI) (Vicente-Serrano et al., 2010). For instance, the PDSI was used to evaluate the combined effect of the Pacific Decadal Oscillation (PDO) and El Niño–Southern Oscillation (ENSO) on global wet and dry changes over land, showing that when these two modes are in phase (e.g. El Niño–warm PDO) wet and dry events are amplified (Wang et al., 2014). The PDSI and SPEI have also been used to quantify wet and dry trends over China, with generally good agreement between the two (Chen et al., 2017). At the global scale, the SPI and SPEI were used to explore wet and dry links with ENSO, PDO, and the North Atlantic Oscillation (NAO) (Sun et al., 2016). The study found that ENSO has the greatest spatial impact for wet and dry changes, followed by the PDO with an effect in North America and eastern Russia as well as the NAO with an affect on Europe and northern Africa. The SPI has also been used in a global multi-model ensemble analysis of future projections in pluvial and drought events (Martin, 2018). This revealed that more severe pluvial events are expected in regions that are already wet, and the same applies for more severe droughts in dry areas, although the overall “wet gets wetter, dry gets drier” paradigm may have some limitations, since when the paradigm is applied over land it does not hold as expected because of changes in atmospheric circulation, horizontal gradients of temperature, and relative humidity (e.g. Byrne and O'Gorman, 2015; Yang et al., 2018).
In this study, we adopt a relatively broad definition of multi-hazard events, i.e. the temporal (yet spatially separate) co-occurrence of wet and dry hydrological extremes at the global scale, quantified following De Luca et al. (2017). We emphasize that the term “hydrological extreme” does not necessarily imply
Notwithstanding their socio-economic relevance, concurrent wet and dry
hydrological extreme events at the global scale have seldom been addressed in the literature. One early study did consider combinations of wet and dry extremes via observed PDSI for two thresholds (wet, PDSI To what extent has the global area impacted by wet, dry, and concurrent wet–dry hydrological extreme events changed? What were the most geographically widespread extreme wet, dry, and concurrent wet–dry events? And what is the associated documentary evidence of extreme conditions during these periods? How comparatively frequent were wet or dry extremes in the past? What is the most likely time interval between opposite extremes at a given location? How are wet and dry hydrological extremes linked with dominant modes of climate variability?
We used the self-calibrated monthly mean Palmer Drought Severity Index based
on the Penman–Monteith model (sc_PDSI_pm) (Dai, 2017; Sheffield et al., 2012) for the 1950–2014 period at 2.5
We further analyse three climate modes of variability known to affect regional and global precipitation patterns: the Niño3.4 (Rayner et al., 2003; Trenberth, 1997), PDO (Mantua and Hare, 2002), and Atlantic Multi-decadal Oscillation (AMO) (Schlesinger and Ramankutty, 1994). All these climate indices are at monthly time resolution from 1950 to 2014, as issued by the National Oceanic and Atmospheric Administration (NOAA).
First, we calculate the percentage of total land area (km
Extreme dry events were calculated in a similar way as extreme wet events
except that in place of AMAX the sc_PDSI_pm annual minima (AMIN), i.e. the lowest monthly sc_PDSI_pm observations within each year, were used to compute the extreme events, provided that they satisfied sc_PDSI_pm
Second, to establish whether the most widespread extreme wet, dry, and wet–dry events were solely due to chance, a bootstrapping analysis of
Lastly, to test whether the sc_PDSI_pm values obtained during the most widespread wet, dry, and wet–dry events were spatially autocorrelated we computed the Moran's
The
Associations between extreme wet–dry hydrological extremes and the three modes of climate variability (Niño3.4, PDO, and AMO) were assessed using
the Spearman's rank correlation test (Corder and Foreman, 2014). Specifically, the correlations were performed for each grid cell only for
monthly wet and dry extreme observations (sc_PDSI_pm
Finally, since Niño3.4 may interact with other modes of climate variability, we removed this signal when correlating the PDO and AMO with
sc_PDSI_pm extreme wet and dry observations by performing partial correlations with the R package ppcor_v1.1. Partial correlations represent the relationship between two random variables after removing the effect of one or more other random variables. Here, the partial correlation between two variables,
The percentage (%) of total global land area impacted by the most widespread extreme wet, dry, and neutral events is shown in Fig. 1 at both monthly and annual resolutions from 1950 to 2014. For extreme wet events
(sc_PDSI_pm
Percentage (%) of total land area with
For extreme dry events (sc_PDSI_pm
The neutral events (
Finally, the area with 1-month concurrent wet–dry hydrological extreme events (Fig. 1g) shows an increasing and statistically significant trend (Sen's slope
We next consider the single most extensive wet, wet–dry, and dry events and
show that they match reports of severe flood and drought events. The most
widespread global extreme wet event was also the most widespread wet–dry event and occurred in December 2010 (Fig. 2a). Recorded events matching
this occurrence include the devastating Queensland floods in Australia (BBC,
2010a; Smith et al., 2013; Trenberth and Fasullo, 2012; Zhong et al., 2013),
heavy floods and landslides in south-east India which killed more than 180 people (Reliefweb, 2010), widespread flooding and landslides in Colombia and Venezuela causing about 300 deaths and leaving thousands homeless (BBC, 2010b; Telegraph, 2010; Trenberth and Fasullo, 2012), and flooding affecting the north-western USA (NWRFC, 2010). We also find anomalously wet conditions in central–eastern Europe (Fig. 2a), although in this region no significant
damage was reported by the literature and the media. Such a widespread wet
event impacted 7.8 % of the total global land area. December 2010 was
characterized by a very strong negative Niño3.4 phase within the
2010–2012 La Niña event (Luo et al., 2017). Moreover, the PDO and AMO were respectively in their cold and positive phases. The same phases occurred during November 2010 (not shown), and these antecedent conditions may have contributed to the extreme wet and dry events in the sc_PDSI_pm series (Lee et al., 2018). At the same time, droughts were recorded in central Asia, Madagascar, the Horn of Africa (BBC, 2011), South America, the eastern USA (NOAA,
2011), and northern Canada, covering a total of 5.9 % of land area. Both the
extreme wet and dry percentages (%) of land area impacted (Fig. 2a) are
significant at the 5 % level (
Wet–dry (WD) ratio derived for every grid cell. Blue (WD ratio
The most widespread extreme dry hydrological event occurred during January 2003, with 8.6 % of total land area impacted by drought and 3.8 % of land experiencing wet hydrological extremes and floods (Fig. 2b). During this event, eastern Australia was the most affected region, with the worst
drought in 20 years driven by an El Niño event that led to severe dust
storms and bushfires (Gabric et al., 2010; Horridge et al., 2005; Levinson and Waple, 2004; McAlpine et al., 2007). This episode belongs to the so-called “millennium drought” (Van Dijk et al., 2013), which affected Australia between 2001 and 2009. Other regions experiencing severe drought during January 2003 were north-east China, India (Sinha et al., 2016),
Scandinavia (Irannezhad et al., 2017), western Africa, parts of Brazil, and a few scattered areas in Mexico, the USA, Canada, Russia, and Indonesia. January 2003 was an El Niño month, with the Niño3.4 index being in a positive phase along with a warm PDO phase. On the other hand, the AMO registered an almost neutral phase. As for the December 2010 episode in Fig. 2a, such climate patterns also occurred in the previous month (not shown). Meanwhile, other regions experienced wet hydrological extremes and floods, such as south-east China, central Russia, Europe, southern Great Britain (BBC, 2003; Marsh, 2004), Madagascar (Reliefweb, 2003), Argentina, Chile, and scattered parts of Africa and Canada (DFO, 2008). As for Fig. 2a, the percentage of land area impacted by both extreme wet and dry events during January 2003 (Fig. 2b) was significantly different at the 5 % level (
Moran's spatial correlation results are shown in Table 1. Wet and dry
extremes in December 2010 (Fig. 2a) show
Moran's
The WD ratio highlights the 65-year propensity for more or less wet or dry hydrological extremes on a cell-by-cell basis (Fig. 3). Hotspots for extreme wet propensity emerge in the USA, northern Mexico, Colombia, Venezuela, Argentina, Bolivia, Paraguay, northern Europe, North Africa, western China, and western and central Australia. On the other hand, regions with higher frequencies of extreme dry events are found in Canada, central South America, central and southern Europe and Africa, eastern China, and south-eastern Australia. Other regions, such as Russia, display mixed patterns. These WD ratio patterns agree with global trends in drought over the period 1950–2010, identified using the sc_PDSI_pm dataset (Dai, 2012).
In Fig. 4a we show the average time intervals (months) of extreme transitions (ETs) from wet-to-dry and dry-to-wet extremes during the period 1950–2014 plotted against the percentage of total global land area affected. The ET from wet to dry (blue curve) exhibits a modal value of 22 months, associated with 4.3 % of the total global land area. On the other hand, ET from dry to wet (red curve) peaks at 18 months, with
Extreme transition (ET) time intervals between extreme wet to dry (blue) and between extreme dry to wet (red).
ETs from dry to dry and wet to wet were also computed (Fig. S3). Dry-to-dry time intervals peak at 27 months, with 3.2 % of global land area taking
this value, whereas wet-to-wet time intervals peak at 30 months with 3.1 % of land area. Half of all dry-to-dry ETs occurred within
In Fig. 5, we show global correlations between hydrological extremes (wet and dry) and three of the major modes of climate variability (Niño3.4, PDO, and AMO; see Sect. 2.4). We also computed the same correlation tests for the NAO (Barnston and Livezey, 1987), Pacific North American (PNA) pattern (Barnston and Livezey, 1987), and Quasi-biennial Oscillation (QBO) (Baldwin et al., 2001). However, these results had low statistical significance (Fig. S4). Generally, the correlations shown in Fig. 5 are consistent with the concurrent wet–dry spatial patterns observed in Fig. 2.
Correlations between monthly wet (sc_PDSI_pm
ENSO is one of the modes with the most widespread global impacts and is
represented here by the Niño3.4 index (Fig. 5a). The positive phase of
Niño3.4 (associated with El Niño events) is negatively correlated
(
Correlations for PDO (Fig. 5b, with the Niño3.4 signal removed) partly resemble the spatial patterns found for Niño3.4. Here, negative correlations are also found in north-western North America, equatorial Africa, and eastern Russia, although most significant correlations over Australia, China, and India vanish. Moreover, positive correlations are found in the central–western USA, southern South America, and Kazakhstan. The fact that Niño3.4 and PDO correlations show similar spatial patterns (Fig. 5a and b) suggests that when these two indices are in phase (i.e. El Niño–warm PDO and La Niña–cold PDO), wet and dry extremes are amplified (Wang et al., 2014). The correlation patterns shown in Fig. 5b also agree with season-ahead peak river flow correlations with the PDO (Lee et al., 2018). The PDO significantly impacts a smaller area (12 % of total global land) compared to Niño3.4. Niño3.4 and PDO correlations also tend to resemble the WD ratio patterns (Fig. 3). For instance, we note that both Niño3.4 and PDO show positive rank correlations with the extreme sc_PDSI_pm over the southern and western USA (Fig. 5a and b), which are reflected by the predominance of wet extremes (over dry extremes) in Fig. 3. Similar patterns are also observed over south-eastern Brazil and Argentina. In addition, Fig. 5a and b show negative correlations with wet extremes over central and eastern Russia, a pattern matched by the predominance of dry extremes (over wet extremes) in Fig. 3. A similar coherence in patterns also applies to eastern Australia and central–southern Africa.
The pattern of AMO correlations (Fig. 5c, with the Niño3.4 signal removed) differs from the Niño3.4 and PDO indices and returns more significant (
Wet and dry extremes can coincide in time and/or space, creating multi-hazard events that lead to significant socio-economic losses. Geographically remote yet temporally coincident extremes potentially impact stakeholders with global assets and/or supply chains. For instance, knowledge of recurrent patterns of coincident hydrological extremes could be used to hedge losses in regional hydropower production (Ng et al., 2017; Turner et al., 2017) and agricultural yields (Leng and Hall, 2019; Xie et al., 2018; Zampieri et al., 2017) or to manage crop planting dates (Sacks et al., 2010). Rapid successions of extremes at the same location pose challenges for disaster preparedness, event management, and long-term risk reduction. Floods and droughts are also expected to become regionally more frequent and severe in the future due to anthropogenic climate change (Arnell and Gosling, 2016; Dai, 2011a, 2012; Hirabayashi et al., 2013; Hirsch and Archfield, 2015; IPCC, 2012; Milly et al., 2002), underscoring the importance of research on concurrent wet–dry hydrological extremes.
We found that the land area affected by extreme dry and geographically remote wet–dry events is increasing, with statistically significant trends at both monthly and annual timescales (Fig. 1). This matches the expectation that such hazards are likely to increase in the future (Güneralp et al., 2015; Hirabayashi et al., 2013) and is in agreement with previous studies (Dai, 2012; Dai et al., 2004). However, we applied a more stringent definition of extreme events (De Luca et al., 2017) in order to capture well-known flood
and drought episodes. We further showed that these extremes can have global-scale impacts, corresponding to documented flooding and drought
events, by detecting the most widespread wet, dry, and wet–dry events (Fig. 2). As a limitation of our study, we recognize that the coarse horizontal resolution of the dataset used (sc_PDSI_pm at 2.5
We introduced two new metrics: the wet–dry (WD) ratio (Figs. 3 and S1–S2) and the average time between extreme transitions (ETs) for wet-to-dry and dry-to-wet extremes (Fig. 4). The former reveals the local frequency of extreme wet relative to extreme dry observations. Areas experiencing more wet than dry extremes were detected in the USA, northern and southern South America, northern Europe and North Africa, western China, and most of Australia. More dry than wet extremes were experienced in most of the remaining areas. The ET metric estimates for every grid cell the average time interval between opposing extremes (i.e. transitions from wet to dry and from dry to wet). The median time between wet-to-dry transitions is on average slower than the one between dry to wet. Monitoring long-term changes in ET intervals between wet-to-dry and dry-to-wet hydrological extremes could provide valuable information on loss accrual and socio-economic impacts.
To this end, it is important to identify possible climate drivers of the observed hydrological extremes. In this study, we computed correlations between wet–dry hydrological extremes and corresponding values of the Niño3.4, PDO, and AMO indices (Figs. 5 and S5–S6). Our results confirm previous findings about the effect of ENSO, PDO, and AMO on global flood hazard and global season-ahead correlations with river peak flows (Emerton et al., 2017; Hodgkins et al., 2017; Lee et al., 2018; Mallakpour and Villarini, 2015; Tootle et al., 2005; Wang et al., 2014; Ward et al., 2010, 2014), while presenting a useful tool for interpreting the most widespread wet–dry events and WD ratio and ET metrics. PDO spatial correlations with hydrological extremes generally match those of Niño3.4, which supports the view that when Niño3.4and PDO are in phase they amplify the global wet and dry changes (Wang et al., 2014). Niño3.4 and PDO correlations also tend to reflect the patterns found for the WD ratio. In other words, when Niño3.4 and PDO are in a positive (negative) phase this leads to extreme wet and dry conditions in some regions, and these wet–dry patterns also occur in areas which have in the past respectively experienced more wet–dry conditions. The AMO shows different, and in some cases opposite, correlation patterns when compared to Niño3.4 and the PDO. During the most widespread wet, dry, and wet–dry hydrological extremes, the AMO was weak, and indeed the geographical footprint of these events does not closely match that of the AMO. Hence, we assert that the most widespread wet and wet–dry (dry) hydrological extreme events were driven by a La Niña (El Niño) event coupled with a strong negative (positive) PDO phase. However, we note that modes of climate variability cannot entirely explain the physical mechanisms driving these multi-hazard events. Indeed, when selecting similar phase values of the Niño3.4, PDO, and AMO indices occurring during the most widespread events (Fig. 2), it is not possible to recover similar events in terms of overall impacted areas (Figs. S7 and S8).
The analysis was conducted using the self-calibrated monthly mean Palmer
Drought Severity Index based on the Penman–Monteith model (Dai, 2017; Sheffield et al., 2012). Future research opportunities include using other indices, such as the Standardized Precipitation Index (McKee et al., 1993, 1995) or the Standardized Precipitation Evapotranspiration Index (Vicente-Serrano et al., 2010), to validate our findings and to account for uncertainty in the observations of concurrent wet–dry extremes. Additionally, there is scope to use more recently developed soil moisture metrics emerging from the ESA Soil Moisture CCI Project (Gruber et al., 2019) and the NASA Soil
Moisture Active Passive (SMAP) mission (
The sc_PDSI_pm dataset is freely available from Dai (2017)
The supplement related to this article is available online at:
PDL conceived the methods, performed the analyses, created the figures, and wrote the first paper draft. GM and RW contributed to the methods. All the authors contributed to the writing.
The authors declare that they have no conflict of interest.
This article is part of the special issue “Hydro-climate dynamics, analytics and predictability”. It is not associated with a conference.
The authors would like to thank the two anonymous reviewers and the editor for their constructive comments. Paolo De Luca would also like to thank Venugopal Thandlam for the useful discussions and comments.
This research has been supported by CENTA NERC (UK) (grant no. NE/L002493/1) and the Swedish Research Council Vetenskapsrådet (grant no. 2016-03724).
This paper was edited by Valerio Lucarini and reviewed by two anonymous referees.