Hydrological extremes, i.e., droughts and floods, occurred frequently throughout the world in the past decade, with annual economic losses exceeding $8 and $30 billion, respectively (International Disaster Data Base, http://www.emdat.be). Droughts and floods barely occur at the same time, but a rapid succession of these two hazards can aggravate adverse impacts caused by individual hazards alone1–5, as exemplified by recent events in Queensland (Australia, 2010)6, United Kingdom (2012)7, India (2016)8, California (United States, 2017)9, Peru (2017)10, and Mozambique (2019)11. Australia suffered an infamous Millennium drought (2001−2009), with economy and environment hit hard, but the drought ended with devastating floods that led to riverbank collapse, 35 known deaths, and significant displacement6. In order to improve resilience and adaptation to such consecutive hazards, it is necessary to understand how the abrupt shift from drought to flood will change in the future12–17.
The abrupt shift from drought to flood has been studied at local and global scales18–21, but most research analyzes the consecutive drought-flood hazards based on the commonly adopted drought indices that use the multi-month or yearly cumulative precipitation as input19,22,23. Such an evaluation paradigm can be problematic since floods are generally short-lived and episodic, not as long and cumulative as droughts24,25. Heavy rainfall events can occur during dry spells and cause floods, but the rainfall amount may not be sufficient to break the drought. For instance, England and Wales experienced a multiyear intense drought from 2010 to early 2012, which was abruptly terminated by a series of floods. Specifically, a heavy rainfall event caused flash flooding in late September 2012, but the corresponding monthly rainfall anomaly was negative because the first half of September was very dry26. Therefore, previous work based on the multi-month cumulative precipitation may fail to represent the post-drought flash flooding due to heavy rainfall, whose frequency and intensity are generally expected to increase under climate change27,28, thereby leading to a large underestimation on the possible risks of consecutive drought-flood hazard29,30.
In a warming climate, droughts are expected to intensify due to elevated evapotranspiration and the change in rainfall patterns31,32. Numerous studies have also revealed regionally increased flood risks with warming33,34, as a result of more extreme precipitation over short time scales due to the increase in atmospheric moisture and convective cloud feedbacks28. However, it is unclear whether a rapid succession of drought and flood events will increase or decrease with warming. The underlying processes that cause the abrupt shift from major droughts into destructive floods are also not fully understood, and they might be closely linked to the large-scale environmental conditions conducive to intense storms near the termination of persistent droughts. For example, California recently experienced a historically unprecedented drought starting in 2011 with disastrous consequences and was then hit by significant floods caused by a series of strong rainstorms in December 20149. Specifically, 40% of droughts in California ended abruptly with the arrival of intense storms35. The drought-flood abrupt shift has fundamental impacts on landscapes and ecosystems, such as the morphological development of ridge-trough pairs in the Brazos Delta (Texas) and the seasonally flooded forest and communities in Amazonia36. In order to improve resilience and adaptation to such consecutive hazards, it is desired to develop a comprehensive and robust assessment of the abrupt shift from drought to heavy rainfall and its response to climate change.
Here, we seek to explore and understand projected changes in the probability of rapidly succeeding drought and heavy rainfall over the global land area under a moderate-emission mitigation policy (RCP4.5) and a high-emission scenario (RCP8.5). Drought is quantified through the self-calibrated Palmer drought severity index (scPDSI) while extreme precipitation is defined as daily rainfall exceeding the 95th and 99th wet-day percentiles (R95p and R99p, respectively). Although our estimates should not be interpreted as an actual calculation of the consecutive drought-flood hazard, extreme precipitation plays a dominant role in triggering floods, especially flash floods, over short- and medium-sized rivers37.
Historical abrupt shift from drought to heavy rainfall
Historical trends of drought and extreme precipitation are detected to identify the respective vulnerability hotspots (Fig. 1a,b) and are consistent with previous studies32,38. The scPDSI is significantly decreasing in Africa, East Asia, and Southern Europe, while total precipitation exceeding the 95th wet-day percentile (R95pTOT) is significantly increasing in South America East Coast and Southeast Africa. Fig. 1c presents the probability of abrupt shift from drought to R95p heavy rainfall within a month under present climate. The probability of abrupt shift from drought to heavy rainfall is statistically significant over 42% of the global land area (with p-value less than 0.05), with average values of 53% and 20% for R95p and R99p heavy rainfall, respectively (Figs. 1c and S1). This indicates that quite a few regions may frequently experience consecutive drought-flood hazards during 1955−2004 as a result of the high probability of abrupt shift from scPDSI drought to R95p heavy rainfall. Hotspots where the probability of abrupt shift exceeds 80% include the eastern United States (US), Amazonia, Southern Brazil, Central Africa, and Southern Europe; these are also regions where multiple high-impact real-world events have been reported. For example, Southern Brazil is one of the most vulnerable regions to droughts and floods and experienced remarkably dry conditions from January 2014 to February 2015. However, the drought was followed by a heavy rainfall event in 2015, which forced the Guaíba River to overflow its banks and thus affected over 200,000 people across the states of Rio Grande do Sul and Santa Catarina in Southern Brazil39,40.
We find that over 40% of the global land area experience no trend neither in drought nor heavy rainfall occurrence over the last five decades, but 11% of it may suffer from a high probability (e.g., over 80%) of abrupt shift from drought to heavy rainfall. For example, the eastern US has shown no statistically significant change in drought and heavy rainfall over the past 50 years, but it is very likely to experience heavy rainfall immediately after long-term droughts, with a mean probability of nearly 78%. Amazonia is an area where extreme precipitation even shows a decreasing trend but has a nearly 79% probability of experiencing heavy rainfall immediately after droughts. Such a high probability of abrupt shift indicates a strong lagged dependence between drought and heavy rainfall. We quantify the impact of the dependence on the probability of abrupt shift from drought to heavy rainfall (see Fig. 1d and Methods). The global mean probability of abrupt shift would reduce from 53% to 23−26% if drought and heavy rainfall occurred independently (Fig. 1d). This implies that the lagged dependence between drought and heavy rainfall may double the probability of consecutive drought-flood hazards that would be assumed from the independent occurrence of both hazards. For instance, the probability of abrupt shift in Amazonia would be 38−49% instead of 79% if drought and heavy rainfall occurred independently. Such increase would be up to three times if heavy rainfall was defined as R99p (Fig. S1d). The disaster emergency system may be stretched thin as a result of ignoring such lagged dependence and the resultant consecutive hazards, thereby leading to disproportionate impacts41,42.
Abrupt shift from drought to heavy rainfall in a warmer climate
We confirm that climate models well reproduce the spatial distribution of the observed probability of abrupt shift, albeit with a large model spread (Figs. 2a and S2). The probability of abrupt shift from scPDSI drought to R95p heavy rainfall within a month is projected to robustly increase under the high-emission scenario RCP8.5. Specifically, the fraction of locations experiencing a high probability of more than 80% is projected to more than double (from 10% to 23−46%) by the end of the 21st century (Fig. 2a,b). Such an increase will be further magnified if the 99th percentile is used to detect heavy rainfall events, with the probability of abrupt shift increasing from 0.2% to 5−12% (Fig. S3). The mean probability of abrupt shift under future climate is also projected to decrease from 46−71% to 44−68% significantly (p < 0.01) when the emission scenario is changed from RCP8.5 to RCP4.5 (see Figs. S4 and S5 as well as Methods).
Approximately 39−72% of the global land area, particularly mid-latitude areas, are projected to experience increases in the probability of abrupt shift from scPDSI drought to R95p heavy rainfall (Fig. 2c). The detected hotspots where the future probability of abrupt shift exceeds 80% are the western and eastern US, Southern Africa, and Northern Europe (Fig. 2b). For example, the western US is expected to experience an increase in the probability of abrupt shift from 56% on average to 66−77%, which may suggest an increased risk of another 2017 California flood after a 6-year drought. This recent drought-flood abrupt shift led to an estimated loss of over $1.05 billion and the evacuation of nearly 188,000 residents43. In 2019, Southern Africa (Mozambique) was hit by flash floods due to intense tropical cyclones after prolonged droughts, thereby having devastating impacts on huge areas of farmland and leading to severe food shortages11. Such an episode is also projected to occur more frequently since the probability of abrupt shift is expected to increase from 53% to 37−78%.
The projected increase in the regional probability of abrupt shift is statistically significant over Northern Europe, Southern Africa, and Central Asia, where most climate models agree on the change (see the projected probability of abrupt shift outside the baseline uncertainty ranges for NEU, SAF, and CAS in Fig. 2e). Climate models do not agree on the trend of future changes in the probability of abrupt shift over most tropical areas (Fig. 2c). This can be attributed to the large uncertainty of extreme precipitation simulated by climate models as reported in previous studies28.
Figure 3 presents the regions where changes in drought (scPDSI) and heavy rainfall (R95p) have opposite trends with the change of the probability of abrupt shift (pAS) in the multi-model mean CMIP5 simulations (Fig. S6 for R99p heavy rainfall). The projected increase in the probability of abrupt shift is consistent with that of the heavy rainfall occurrence over more than half the global land area (Fig. 3a,c), but the tropics are projected to experience a lower probability of abrupt shift and more occurrence of heavy rainfall. 4−13% of the global land area, such as the southwestern US, Northern Mexico, Southern Africa, Australia, and Southwest coast of South America, are projected to experience a higher probability of abrupt shift but less occurrence of heavy rainfall (see the cyan areas in Fig. 3a,c). This indicates that heavy rainfall is projected to occur more frequently after drought, albeit with a decreased overall frequency of heavy rainfall. On the other hand, 27−41% of the global land area is projected to experience fewer drought events but to suffer higher probabilities of consecutive drought-flood hazards due to the higher probability of abrupt shift, such as Northern China under RCP4.5 (Fig. 3b) and Southern Africa under RCP8.5 (Fig. 3d). This indicates that substantial increases in the risks associated with abrupt shift from drought to heavy rainfall may emerge in regions with decreased risks of droughts and floods.
We investigate the probability of abrupt shift while considering the impacts of elevated CO2 concentration on the potential evapotranspiration (PET) estimation44. Although the PET can be overestimated as a result of neglecting the vegetation response to elevated atmospheric CO2 concentration (Fig. S7), the global trends of scPDSI differ only slightly (Fig. S8). To further enhance the robustness of results, we also investigate the probability of abrupt shift from drought to heavy rainfall without taking into account the impacts of elevated CO2 concentration (Figs. S9−S14). Results show that the probability of abrupt shift is also projected to increase, and the increase is similar to that with the consideration of the impacts of elevated CO2 concentration.
Changes in processes linked to drought-flood events
The abrupt shift from drought to heavy rainfall is linked to large-scale atmospheric conditions, especially high convective available potential energy (CAPE) and high convective inhibition (CIN) at the drought demise as the high CAPE suggests a high potential of moist convection in the atmospheric boundary layer and the high CIN inhibits the onset of weak-moderate convection, thereby enhancing the probability of intense rainstorms after drought termination15,16. We confirm that the historical CAPE and CIN patterns simulated by climate models are similar to reanalysis data (Fig. S15). The high-CAPE and high-CIN situations during drought demise (Fig. 4a,c) coincide with historical hotspots for the high probability of abrupt shift, such as South Central Africa and South Asia (Fig. 1c).
We find that under a high emission scenario (RCP8.5) the future drought demise is linked to stronger anomalies in atmospheric conditions than those in the past. In particular, the convection potentials at the drought demise are generally higher than those in the past (Fig. 4a,b), especially at low-latitude areas. The change in the CIN anomalies, however, shows a different spatial distribution (Fig. 4c,d). Specifically, the central US and the Mediterranean region show decreased CIN anomalies but the southwestern US, Southern Africa, South Asia, and Australia show enhanced CIN anomalies. Such enhanced CAPE and CIN anomalies further suggest a higher probability and a higher intensity of heavy rainfall after drought terminations, given that previous work using the CMIP5 models has suggested a general increase in the mean CAPE and CIN intensity under global warming45.
The atmospheric moisture transport also plays a crucial role in generating the abrupt shift from drought to heavy rainfall35. We confirm that climate models can capture the historical patterns of vertically integrated moisture flux convergence (VIMFC) (Fig. S16). Though VIMFC anomalies are rather small under present-day conditions (Fig. 4e), large increases are observed during future drought demise, especially for low-latitude areas (Fig. 4f). Although increasing atmospheric water vapor is not surprising−given the well-understood thermodynamic expectations with global warming−we point out that droughts are expected to terminate under RCP8.5 in an atmosphere with higher moisture convergence than those during all other periods in the future (Fig. 4f). Such converging moist air supplies more water for precipitation after drought termination than normal periods, which occurs mostly at mid- and low-latitude areas, such as South Central Africa, Amazonia, North Australia, East Asia, South Asia, and Southeast Asia. The intense water vapor convergence along with the high-CAPE and high-CIN atmosphere significantly increases the probability and risk of “double-whammy” weather events in the future, such as Southern Africa, where the risk of another deadly flood following a severe drought in Zimbabwe in 2017 and in Namibia in 200846 might be reinforced.
Societal implications of abrupt hydrological shifts
Collectively, our findings suggest that the lagged dependence between drought and heavy rainfall may double the probability of consecutive drought-flood hazards that would be assumed from the independent occurrence of both hazards. Strong anthropogenic warming will render droughts more likely to abruptly shift to heavy rainfall, with a doubling in probability for the areas that already have a high probability (i.e., 80%) in present climate, suggesting that consecutive drought-flood hazards are likely to become a common occurrence. Traditional risk assessments of droughts and floods individually may provide guidance for adapting to future extreme events, but an increased probability of abrupt shift from drought to heavy rainfall may emerge in 27−41% and 4−13% of the areas with projected decreases in the number of droughts and heavy rainfall, respectively, such as the southwestern US, Northern Mexico, Southern Africa, Australia, and Southwest coast of South America. These changes signify a critical threat to well-developed prevention strategies and pose challenges to tradeoffs between water storage and flood control.
Our findings also suggest that future droughts will more likely terminate along with intense convection and strong water vapor convergence strongly exceeding those during normal periods in the future, implying that the intensification of future heavy rainfall after droughts might continue beyond thermodynamic expectations with global warming. Such intensification would render global water infrastructure more vulnerable to the rapid drought-flood cycles since structural degradation of levees, dams, and canals, besides others, may be enhanced by extended periods of drought. For instance, the Millennium Drought in southeast Australia (2001–2009) triggered at least sixty riverbank failures, which was followed by widespread floods in Queensland in 20106. The recurrence of such consecutive hazards may lead to considerable socioeconomic damages and loss of life. Therefore, it is particularly important to understand risks associated with such abrupt shift events in the areas with little change or even decreases in individual drought and flood occurrence.