1. Introduction
We define flash drought (FD) as a drought with a rapid onset and intensification and with substantial impacts on agriculture (Svoboda et al. 2002; Otkin et al. 2018). The term was coined in 2002 (Svoboda et al. 2002). It is the rapidity of onset that is described as its key feature, with FD onset on the order of weeks, in contrast to the traditionally slow development of drought, which usually occurs over months. Thus, FD classifies as a subseasonal-scale drought, posing a new challenge for subseasonal to seasonal (S2S; from weeks to a few months) prediction (Pendergrass et al. 2020). In essence, a location showing no sign of drought can develop agricultural drought conditions within several weeks through the rapid depletion of soil moisture due to coincident low precipitation and above average evaporation (Otkin et al. 2018).
Otkin et al. (2018) and Pendergrass et al. (2020) outline a comprehensive framework for future research on FD, suggesting the use of suitable drought indices that reflect the rapidity of FD onset. The rapid intensification of drought conditions during an FD is particularly detrimental to the agricultural industry, where it limits time for mitigating measures such as additional irrigation, delayed seeding of crops, or adaptation for livestock (Nguyen et al. 2019). Past FDs have been accompanied by significant economic damage: for example, the 2012 drought in the U.S. Midwest had an estimated loss of US$30 billion (Rippey 2015), while the 2018 FD event in South Queensland, Australia, resulted in high livestock mortality (Nguyen et al. 2019).
As with other drought types, difficulties arise in establishing a definition for FD identification (Lisonbee et al. 2021). Since FDs are often seen as rapidly developing agricultural droughts, soil moisture is the primary proxy for FD. Ford and Labosier (2017) proposed a criterion whereby root-zone soil moisture has to decline from above the 40th to below the 20th percentile within 20 days. Studies from Mo and Lettenmaier distinguish between two types of FD: precipitation-driven (Mo and Lettenmaier 2016) and heatwave-driven FDs (Mo and Lettenmaier 2015), where the latter are defined by rapid increasing evaporation rather than a sudden reduction in precipitation. Even though Otkin et al. (2018) argue that Mo and Lettenmaier’s percentile-based thresholds for soil moisture of below the 40th percentile over a 5-day period is not dry enough to be defined as drought, their separation highlights the two important drivers for FD. Pendergrass et al. (2020) proposed two definitions for operational use, research, and prediction. First, for the United States only, a two-category decrease in the U.S. Drought Monitor (USDM; https://droughtmonitor.unl.edu/AboutUSDM) over 2 weeks and sustained over 2 more weeks has to happen. The second definition is for global application and requires a 50-percentile increase in the evaporative demand drought index (EDDI; Hobbins et al. 2016) within 2 weeks and sustained drought conditions for the following 2 weeks. Other studies are based on standard anomalies or rates of change within a given period of the individual index used (Otkin et al. 2015; McEvoy et al. 2016; Otkin et al. 2018; Nguyen et al. 2019; Noguera et al. 2020).
Based on the above, drought indices that account for the processes responsible for drying conditions have been used to examine FD in several regional studies in the United States (Otkin et al. 2013; Hunt et al. 2014; Ford et al. 2015; McEvoy et al. 2016; Otkin et al. 2016; Ford and Labosier 2017), China (Wang et al. 2016; Zhang et al. 2017; Liu et al. 2020a,b), Spain (Noguera et al. 2020), and Australia (Nguyen et al. 2019). While only McEvoy et al. (2016) compared common indices used in these studies, this was confined to the United States: a global assessment is still lacking.
The simplest and most widely used index in drought research and operational monitoring is the standardized precipitation index (SPI; McKee et al. 1993). The SPI compares precipitation with its climatological average. However, reliance on the SPI alone ignores the evaporative component, which has been described as a key factor for the rapid intensification in FDs, at least for the United States (Ford et al. 2015; Otkin et al. 2018; Pendergrass et al. 2020) and China (Liu et al. 2020a,b). In contrast, the EDDI (Hobbins et al. 2016) calculates the evaporative demand of the atmosphere E0, an estimate of the crop reference evapotranspiration. It incorporates influences from temperature, radiation, wind speed, and humidity and thus accounts for the moisture demand in the atmosphere and reflects moisture conditions at the surface, through land–atmosphere feedbacks encoded in the complementary relationship of actual evapotranspiration (ET) and E0. Another index, which accounts for both ET and E0, is the evaporative stress index (ESI; Anderson et al. 2007). By incorporating E0, the ESI is connected to atmospheric drying conditions, yet also takes the surface moisture supply into account by using ET. The ESI is simply defined as the standardized anomaly of the ratio of actual ET to E0. McEvoy et al. (2016) showed that EDDI successfully identifies drought over the southern continental United States and is consistent with identification from the SPI and ESI.
However, whether the connection between low precipitation (e.g., SPI), ET (ESI), E0 (EDDI), and FD is replicated globally is unknown. Nguyen et al. (2019) noted in their observational study over Australia that certain regions can be strongly affected by factors other than rainfall and temperature, such as soil memory, water capacity, plant types, vapor pressure, and wind speed.
A global comparison of ET and non-ET based indices is needed to identify their efficacy for FD detection, to provide insight to the driving mechanisms of FD, and to identify regional differences. Global coupled climate models provide a useful tool for this comparison because they provide consistent data globally and demonstrate the relationships between precipitation, evaporative demand, and soil moisture. Limitations in the models’ land–atmosphere coupling are well known (Orlowsky and Seneviratne 2013; Seneviratne et al. 2013; Lorenz et al. 2016; Yuan and Quiring 2017; Ukkola et al. 2018a,b), but knowing how these models simulate FD and represent relevant processes allows us to further assess FD in a changing climate (Yuan et al. 2019; Pendergrass et al. 2020).
This study examines the occurrence of FD in coupled climate models from phase 5 of the Coupled Model Intercomparison Project (CMIP5; Taylor et al. 2012), which are described in the next section. To do so, we first define FD based on percentiles in the methods section and apply those percentiles to a standardized measure of the upper layer soil moisture (SSI) from the CMIP5 model output to produce a global climatology of FD-event frequency, which will be our “truth.” Drought index values at the onset of an FD detected in the soil moisture and their rates of change prior to that will provide information on how sensitive the chosen drought indices are to upper-layer soil moisture variability. In addition, we apply the same percentile-based FD definition on the drought indices themselves. From this we calculate the detection skills for each index to infer their prediction capabilities. Looking at the intermodel spread provides insight as to how the individual CMIP5 models differentiate from each other in terms of FD frequency, sensitivity, and detection skills. The findings from these analyses are then discussed and summarized.
2. Data
The model outputs from six CMIP5 models (Taylor et al. 2012) listed in Table 1 were analyzed as they provide all the daily data necessary to calculate the drought indices on a global scale. The following variables are required to calculate the indices: 10-cm soil moisture (mrsos), maximum air temperature (tasmax), minimum air temperature (tasmin), surface wind speed (sfcWind), humidity (huss), surface pressure (psl), downward shortwave solar radiation (rsds), and precipitation (pr).
CMIP5 models and variables used.
The selection of CMIP5 models was limited by the availability of daily soil moisture output. One major caveat of all CMIP5 models is that daily soil moisture data are only available for the 0–10-cm layer, which does not fully represent the root-zone soil layer. However, the highly dynamic changes in moisture availability during the development of an FD necessitates analysis at the daily time scale. Additionally, not all models that provide daily soil moisture do so for the full historical run. Consequently, only the six CMIP5 models listed in Table 1 offer all necessary variables for the historical simulation with a common time period of 139 complete years (1867–2005).
From these models, the r1i1p1 ensemble member was used for the analysis, which represents the first initial conditions (r1) for the first initialization method (i1) using the first set of physics (p1). All models’ outputs were interpolated to the coarsest common grid (2.8° × 2.8°) using bilinear interpolation.
3. Methods
As introduced before, the SPI, EDDI, and ESI are used for this study. For evaporative demand E0, reference evapotranspiration is used for both EDDI and ESI, which is based on the FAO-56 international standard (Allen et al. 1998). Reference evapotranspiration requires input data of minimum and maximum temperature, downward shortwave radiation, wind speed, and humidity—all from the land surface. The actual ET needed for the ESI was derived from the latent heat flux directly output from the model using a conversion factor of 28.35 W m−2 for water with a density of 1000 kg m−1 at 20°C. The SPI input uses only precipitation and so no preprocessing of model data is required. The indices are computed using daily input data.
Because daily rainfall, E0, and ET undergo naturally strong fluctuations on short time scales, their input data are aggregated (SPI and EDDI) or averaged (ESI) over 30 days to achieve a 1-month time scale. Strong daily fluctuations are also present in the top 10-cm soil moisture layer. This can be limiting for our analysis as this top level can be too responsive to the atmosphere and less indicative of changes in soil moisture in the root-zone layer at a depth of ~1 m (Entekhabi et al. 1996). Depending on the soil type and quantity of precipitation and its partitioning into runoff and percolation, the lag between top-layer and root-zone soil moisture can vary significantly and soil moisture variability is dampened deeper in the soil column (Ford and Quiring 2014). Consequently, soil moisture was also aggregated over the same time period of 30 days as EDDI and SPI to achieve a more similar variability to the root-zone layer soil moisture.
The resulting values of all indices were then translated into drought categories according to the EDDI guidelines (Lukas et al. 2017), which is based on the USDM (Svoboda et al. 2002) to quantify their severity and to simplify comparisons with other studies. The categories range from “ED4” (exceptionally dry) to “ED0” (abnormally dry), for drought severity, “None” representing normal conditions, and from “EW0” to “EW4” for classifying severities of wetness (Table S1 in the online supplemental material). FD events in soil moisture are treated as “truth” and are detected using a percentile definition, defined next.
Because an FD is an extreme event, its occurrence should be relatively rare in a time series. To establish an appropriate definition, we initially use that of Ford and Labosier (2017), who define FD as “periods when the pentad-average 0–40-cm volumetric water content declines from at least the 40th percentile to most the 20th percentile in 4 pentads [20 days] or less.” Their percentiles are based on the “non-drought” and “moderate drought” conditions according to the USDM (Svoboda et al. 2002).
However, some adjustments to this definition were necessary for this study as FD frequency exceeded more than 40 FD events per decade for much of the global domain, which is too frequent for an extreme event relative to, for example, a maximum of 5 FD events per decade in Koster et al. (2019). Although we attempted to imitate the dynamics of deeper layer soil moisture using CMIP5 10-cm soil moisture content by smoothing the time series with a 4-week running mean, the variability in the SSI still far exceeds that of the deeper soil layers. We have compared time series of the aggregated 10-cm SSI with deeper layer monthly soil moisture of the top 1.5-m column, the closest to 1 m provided by the procured CMIP5 models, and we found that the variability of the aggregated 10-cm SSI still exceeds that of the 1.5 soil layers. Thus, the high frequency of FD events is likely associated with the responsiveness of the upper 10-cm soil moisture to the atmospheric layer above (Entekhabi et al. 1996). Consequently, we adjusted the FD definition by lowering the percentile defining the onset of an FD to the 10th (USDM’s “severe drought”) and the time period in which the rapid decline has to occur to 14 days (from 20 days), which is closer to the proposed definition for operations in the United States by Pendergrass et al. (2020), requiring a two-category change in the USDM in 2 weeks.
As a result of these changes, most locations registered only a few FD events per decade, closer to what is expected from regional observational studies (Ford and Labosier 2017; Zhang et al. 2018; Koster et al. 2019). The FD terminates once soil moisture is restored to the 40th percentile or above. These thresholds were derived empirically as there is no previous study that applied the Ford and Labosier (2017) formulation on a top 10-cm soil moisture level. Additionally, only FD events with a duration longer than 4 weeks were examined since shorter events in the upper layer do not necessarily penetrate deeper into the soil profile on shorter time scales (Ford and Quiring 2014). Effectively, the root-zone layer soil moisture (~40–100 cm) should be affected under FD conditions (Otkin et al. 2018). We acknowledge that these different, equally justifiable definitions will give somewhat different results, however, our study is not a comparison with that from Ford and Labosier (2017), but rather seeks to apply a globally appropriate definition in regard to the used data.
To examine the differences in FD dynamics within the CMIP5 model group, we look at the intermodel spread, defined as the ratio of ensemble standard deviation to ensemble mean. A value ≪1 implies that the intermodel differences are much smaller than the mean, and vice versa for ≫1. This measure was applied to the covariability of drought indices to soil moisture and the FD event frequency and will provide information as to what extent the models agree or disagree on their relationships between soil moisture, precipitation, and ET0.
Once the time series of FD events were created for all drought indices, the ability of the index to detect FD was assessed. This was achieved using a contingency table commonly used to assess the accuracy of weather forecasts. The statistical measures of hit rate (HR) and false alarm ratio (FAR) were determined. The HR represents the success of the index to capture modeled FD events. Higher HRs mean that more events are correctly detected by the particular index. In contrast, the FAR shows the proportion of times the indices incorrectly predicted FD to occur, that is when the index yields an FD but it does not register in the SSI. A high FAR shows that the index detects many FDs that are not observed in the SSI. A contingency table gives the required parameters to calculate those skills, which is a 2 × 2 matrix of forecast–event pairs for the dichotomous, nonprobabilistic verification situation.
4. Results
a. Representation of flash drought in drought indices
The SSI represents the “true” occurrence of FDs in the CMIP5 models, and the ESI, EDDI, and SPI are tested as indices for capturing these events. Correlating the latter three indices with the SSI using a gridpoint Pearson correlation, performed globally, indicates that the relationship between them during all conditions, not just drought. The correlations were computed as cross correlations where each drought index was shifted in time relative to the soil moisture from −14 to +14 days in 2-day increments.
Figure 1 shows the maximum explained variance, and Fig. 2 displays its corresponding lag or lead. The CMIP5 multimodel mean of all indices represent the soil moisture fluctuations very well with explained variances of 0.4–0.8. The SPI and ESI (Fig. 1, left and right column) have generally the strongest explained variances of 0.49 and 0.57, respectively, averaged across all seasons. The explained variance between soil moisture and the EDDI is, with 0.22, significantly weaker. For all three indices, the strongest relationship mainly encompasses regions from the subtropics to the mid-/high latitudes where variability in precipitation and E0 are largest and environments are often water limited (Koster et al. 2004). Natural seasonal variations in soil moisture, such as a frozen state, affect the correlations across the northern higher latitudes (45°–60°N) whereas subtropical regions, especially in the Southern Hemisphere, have a far more consistent relationship between soil moisture and precipitation/E0 throughout the year.
CMIP5 ensemble mean of seasonal explained variance of (left) SPI, (center) EDDI, and (right) ESI with 0–10-cm SSI over the entire historical run length of 139 years. Stippling indicates statistically significant correlations at the 95% level. The lag/lead at which this highest correlation was detected is shown in Fig. 2, below. (bottom) The annual intermodel spread in the CMIP5 models.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
CMIP5 ensemble mean of seasonal explained variance of (left) SPI, (center) EDDI, and (right) ESI with 0–10-cm SSI over the entire historical run length of 139 years. Stippling indicates statistically significant correlations at the 95% level. The lag/lead at which this highest correlation was detected is shown in Fig. 2, below. (bottom) The annual intermodel spread in the CMIP5 models.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
CMIP5 ensemble mean of seasonal explained variance of (left) SPI, (center) EDDI, and (right) ESI with 0–10-cm SSI over the entire historical run length of 139 years. Stippling indicates statistically significant correlations at the 95% level. The lag/lead at which this highest correlation was detected is shown in Fig. 2, below. (bottom) The annual intermodel spread in the CMIP5 models.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
Lag/lead at which the maximum explained variances shown in Fig. 1 occurred for (left) SPI, (center) EDDI, and (right) ESI. Stippling indicates where the explained variances are significant at the 95% level. (bottom) The annual intermodel spread in the CMIP5 models. Lead/lag refers to the timing of the behavior of the drought indices relative to SSI: an n-day lead or lag time respectively indicates that the index responds n days prior to or after the SSI.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
Lag/lead at which the maximum explained variances shown in Fig. 1 occurred for (left) SPI, (center) EDDI, and (right) ESI. Stippling indicates where the explained variances are significant at the 95% level. (bottom) The annual intermodel spread in the CMIP5 models. Lead/lag refers to the timing of the behavior of the drought indices relative to SSI: an n-day lead or lag time respectively indicates that the index responds n days prior to or after the SSI.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
Lag/lead at which the maximum explained variances shown in Fig. 1 occurred for (left) SPI, (center) EDDI, and (right) ESI. Stippling indicates where the explained variances are significant at the 95% level. (bottom) The annual intermodel spread in the CMIP5 models. Lead/lag refers to the timing of the behavior of the drought indices relative to SSI: an n-day lead or lag time respectively indicates that the index responds n days prior to or after the SSI.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
The intermodel spread of the six CMIP5 models (Fig. 1, bottom row) is large, predominantly in arid regions for SPI and EDDI, and in the tropics and higher latitudes for ESI. This shows that the models employed here have a different covariability of indices and soil moisture resulting in insignificant explained variances for the ensemble mean.
The high correlations of the ESI are often associated with a lag time from −2 to −10 days, meaning that its signal is delayed relative to the SSI and the ESI responds to changes in soil moisture. The EDDI often has a small lead time of 2–4 days in significantly correlated regions but also sometimes lags soil moisture changes, e.g., in southern Australia during MAM. Physically, E0 (i.e., EDDI) drives changes in ET, and so should lead in energy-limited conditions. Correspondingly, in water-limited conditions, E0 is driven by changes in ET, and so should lag soil moisture. Precipitation shows a lead time of 2–12 days for the SPI with longer leads occurring during the colder season outside the tropics.
The intermodel spread in the bottom row of Fig. 2 highlights further discrepancies between the CMIP5 models. The lead time of precipitation relative to soil moisture is consistent within the CMIP5 model suite, as indicated by an intermodel spread of 0–0.4 for the global domain, which means that the intermodel spread is only a maximum of 40% of the magnitude of the mean lag/lead. In other words, the standard deviation is much smaller than the mean. In contrast, large differences exist for the EDDI in the subtropical Southern Hemisphere; this is especially true for the ESI, which has intermodel spread values exceeding 2, meaning that the difference between the models is twice as large as the mean.
The large intermodel spread highlights the differences in land–atmosphere processes in CMIP5 models as outlined in the introduction. However, to which degree the intermodel spread is a result of the partitioning and timing of energy and water fluxes that drive E0 and actual ET remains unknown since issues with pedotransfer functions in land surface models are well known (Pitman 2003; Vereecken et al. 2016; Van Looy et al. 2017) and contribute to the intermodel spread.
The climatology of FD events in the CMIP5 ensemble mean was produced using the SSI (Figs. 3a–d). The FD events are fairly evenly distributed across the globe and occur in almost all environments throughout the year, though predominantly during the warm seasons. Seasonal hot spots are present in southern Africa, South America, and Australia during the austral summer season while southeast Asia, Africa’s Sahel zone, and parts of northern North America are more affected by FD during the boreal spring. Koster et al. (2019) have identified a similar pattern but with slightly lower frequency of FD events in the Northern Hemisphere due to a different definition during the April to September 1980–2017 period using MERRA-2 reanalysis data of root-zone soil moisture. This similarity provides some confidence that the definition of FD that we have set is reasonable. However, the differences in soil moisture depth of 10 cm in our data versus 100 cm in Koster et al. (2019) prohibits us from simply applying their definition.
CMIP5 ensemble mean of seasonal FD frequency expressed as FD events per decade for (a)–(d) SSI, (e)–(h) SPI, (i)–(l) EDDI, and (m)–(p) ESI over the entire historical 139-yr period. Stippling indicates that at least 50% of the CMIP5 models listed in Table 1 identified nonzero FD.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
CMIP5 ensemble mean of seasonal FD frequency expressed as FD events per decade for (a)–(d) SSI, (e)–(h) SPI, (i)–(l) EDDI, and (m)–(p) ESI over the entire historical 139-yr period. Stippling indicates that at least 50% of the CMIP5 models listed in Table 1 identified nonzero FD.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
CMIP5 ensemble mean of seasonal FD frequency expressed as FD events per decade for (a)–(d) SSI, (e)–(h) SPI, (i)–(l) EDDI, and (m)–(p) ESI over the entire historical 139-yr period. Stippling indicates that at least 50% of the CMIP5 models listed in Table 1 identified nonzero FD.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
Next, we applied the FD thresholds previously described to the other indices to create an FD climatology for each of the SPI, EDDI, and ESI drought indices (Figs. 3e–p). These climatologies were then compared with the SSI (Figs. 3a–d).
In general, the ESI shows a similar event frequency to SSI. Yet, it has some incongruences in the NH around 40°–50°N where it detects a high event frequency in comparison with the rest of its global domain, especially during the boreal winter. Tropical regions between 10°N/S are characterized by a low event frequency by the ESI, presumably because rainfall in the tropics is relatively reliable. Subtropical regions have the highest event frequency (Figs. 3m–p).
The SPI FD frequency (Figs. 3e–h) is larger by a factor of approximately 2.5 than the SSI. Similarly, the EDDI (Figs. 3i–l) records consistently higher FD frequency to differing degrees across the globe relative to the SSI. The EDDI shows only relatively few areas with fewer than 8 events per decade and exceeds 12 events per decade in North Africa and high latitudes. This means that an FD occurs on average once every year. Both the SPI and EDDI show similar climatologies of FDs for all seasons and even detect FDs in the high latitudes during the boreal winter. This is in contrast to the ESI, which barely detects FDs in the high latitudes in any model as the frozen ground does not release moisture for evaporation and the environment is energy limited. In general, FD event frequency shows a similar range in the EDDI, ESI, and SPI climatologies but with different base values, where the climatology for FD events is highest for EDDI, lowest for ESI, and SPI between them.
The FD event frequency can vary considerably between the CMIP5 models, shown as the intermodel spread in Fig. 4. A value of >1 means that the intermodel differences are larger than the ensemble mean. Regionally, the intermodel spread exceeds the number of events of the ensemble mean for ESI and SSI, highlighting large intermodel differences relative to mean number of events. The discrepancies in arid regions and high latitudes are most prominent, with an intermodel spread of 2–3 times the magnitude of the mean event frequency per decade in each season of 0.5–1.5. This is mainly driven by the fact that in these areas, fewer than 50% of the models detect an FD at all, indicated by a lack of stippling. Slightly better model agreement with a spread of around 1 is present in the subtropics and midlatitudes where stippling is present showing where at least three of the six CMIP5 models detected at least one FD event. The intermodel spread for EDDI is slightly lower with a value of around 0.8 with small differences globally when compared with the ESI and SSI since all models detect at least one EDDI FD in all grid cells. The intermodel spread in the ESI and SSI cannot be attributed to any particular model or models that show distinct differences when compared with the other models. However, the spread for the EDDI is heavily weighted by a significantly higher FD event frequency in the CanESM2, which is a factor-of-4 higher than the other models. When the CanESM2 is excluded, the mean event frequency and the intermodel spread are lower and closer to those observed in the SPI FD frequency. Of all indices, the FD frequency computed from the SPI is most similar between models (i.e., it has the lowest intermodel spread) across the global domain and seasons. This result implies that a significant source of the intermodel spread is coming from the land surface models (LSMs). Another possibility is that the intermodel spread is caused by those atmospheric variables (e.g., humidity, wind, etc.) that are used to generate ET0. However, the intermodel spread in the EDDI is less than for ESI and SSI, suggesting that atmospheric variables make less contribution to the spread than land surface variables.
Seasonal CMIP5 intermodel spread of FD frequency for (a)–(d) SSI, (e)–(h) SPI, (i)–(l) EDDI, and (m)–(p) ESI. The intermodel spread is defined as the ratio of ensemble standard deviation to ensemble mean. Stippling indicates that at least 50% of the CMIP5 models listed in Table 1 identified an FD.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
Seasonal CMIP5 intermodel spread of FD frequency for (a)–(d) SSI, (e)–(h) SPI, (i)–(l) EDDI, and (m)–(p) ESI. The intermodel spread is defined as the ratio of ensemble standard deviation to ensemble mean. Stippling indicates that at least 50% of the CMIP5 models listed in Table 1 identified an FD.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
Seasonal CMIP5 intermodel spread of FD frequency for (a)–(d) SSI, (e)–(h) SPI, (i)–(l) EDDI, and (m)–(p) ESI. The intermodel spread is defined as the ratio of ensemble standard deviation to ensemble mean. Stippling indicates that at least 50% of the CMIP5 models listed in Table 1 identified an FD.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
The coincidence of the FDs in the climatologies of EDDI, ESI, and SPI with those detected in the SSI shows whether the index is appropriate for agricultural FD monitoring and detection. We examined this by selecting the FDs identified in the soil moisture by the SSI, our reference, and looked at the representation of these events by EDDI, ESI, and SPI in two ways. First, the index value at the onset of the FD, which is when the 10th percentile in soil moisture is surpassed and second, the rate of change (RoC) in the drought index 2–8 weeks prior to the onset, both averaged across all models and all FDs detected for each grid cell. The RoC shows the index change per week within the 2–8-week window, e.g., if the index changes as much in 2 weeks as it does in 6, the change per week is stronger for 2 weeks. Figure 5 shows the first metric for SPI, ESI, and EDDI as absolute values for each season. A category of ED2 (severe drought) corresponds to the 10th (90th for EDDI) percentile, which is observed for all indices across the globe and all seasons where FD is detected by the SSI.
Index categories by season at the time of FD onset in the SSI for (left) SPI, (center) EDDI, and (right) ESI. For example, a category of ≤ED2 correspond to the ESI’s and SPI’s 10th and the EDDI’s 90th percentiles, respectively. Stippling indicates that at least 50% of the CMIP5 models listed in Table 1 identified an FD.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
Index categories by season at the time of FD onset in the SSI for (left) SPI, (center) EDDI, and (right) ESI. For example, a category of ≤ED2 correspond to the ESI’s and SPI’s 10th and the EDDI’s 90th percentiles, respectively. Stippling indicates that at least 50% of the CMIP5 models listed in Table 1 identified an FD.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
Index categories by season at the time of FD onset in the SSI for (left) SPI, (center) EDDI, and (right) ESI. For example, a category of ≤ED2 correspond to the ESI’s and SPI’s 10th and the EDDI’s 90th percentiles, respectively. Stippling indicates that at least 50% of the CMIP5 models listed in Table 1 identified an FD.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
Here, the SPI shows the closest relationship to the FDs detected in the soil moisture. The SPI has the strongest negative values for both the RoC and index value at the onset of FD in soil moisture, which is a consistent result throughout all seasons. While ESI (EDDI) also shows a consistent decline (increase) of close to one category per week in the 2 weeks prior to the FD onset in the soil moisture, their absolute values at the onset itself are regionally much weaker than those of the SPI. This includes subtropical regions where FD is an imminent threat to agriculture, for example, Australia and eastern South America. Overall, this implies that all indices capture FD identified by soil moisture. Yet, in conjunction with the lag/lead times in the correlation, it seems that FDs are primarily driven by precipitation, while the signal in the evaporation does not occur until later, after the drought onset.
The RoC shows how quickly the drying intensifies several weeks before the onset of an FD. Figure 6 displays that change over the 2 weeks prior to FD onset. The change of each index per week over the 2-week period is compared with the change over 4 (Fig. S1 in the online supplemental material), 6 (Fig. 7), and 8 weeks (Fig. S2 in the online supplemental material). Again, the SPI has consistently stronger changes prior to the onset of the FD event than ESI and EDDI. These changes are fastest for all indices from 2 weeks to the onset and decay when longer periods are considered. This indicates that the movement into an FD is rapid in the models, with the greatest drying in the 2 weeks prior to drought onset. While the RoC is consistently negative across all models, showing a drying signal in all indices, the actual index values for ESI and EDDI are regionally indicating wetter than average conditions, especially for the GFDL models in subtropical regions of the Southern Hemisphere, which modifies the ensemble mean accordingly to around normal conditions (ED0–EW0) (GFDL models represent three of the six models of the ensemble). Here, the processes that influence ET differ most between GFDL models and CSIRO Mk3.6.0, CanESM2, and MIROC5 when it comes to the timing of the ET response signal to the soil moisture.
Rates of change (RoC) per week from 14 days (2 weeks) prior to the onset of FD in the SSI for (left) SPI, (center) EDDI, and (right) ESI. Stippling indicates that at least 50% of the CMIP5 models listed in Table 1 identified an FD.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
Rates of change (RoC) per week from 14 days (2 weeks) prior to the onset of FD in the SSI for (left) SPI, (center) EDDI, and (right) ESI. Stippling indicates that at least 50% of the CMIP5 models listed in Table 1 identified an FD.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
Rates of change (RoC) per week from 14 days (2 weeks) prior to the onset of FD in the SSI for (left) SPI, (center) EDDI, and (right) ESI. Stippling indicates that at least 50% of the CMIP5 models listed in Table 1 identified an FD.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
As in Fig. 6, but for 42 days (6 weeks) prior to the onset of FD in the SSI.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
As in Fig. 6, but for 42 days (6 weeks) prior to the onset of FD in the SSI.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
As in Fig. 6, but for 42 days (6 weeks) prior to the onset of FD in the SSI.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
b. Detection skill of drought indices
Using the FD climatologies from all indices we estimate the event-detection skill for EDDI, ESI, and SPI by determining the HR and FAR. Figure 8 (A section) shows the HR for each index split by season, with an 8-week tolerance window prior to the observed FD event applied. The stippling shows where for each grid cell at least 50% of the CMIP5 models have marked at least one FD in the SSI. Only these regions are discussed.
(Aa)–(Al) Hit rate and (Ba)–(Bl) false alarm ratio for (left) SPI, (center) EDDI, and (right) ESI. The tolerance window for valid detection is from 8 weeks prior to the onset of an FD in soil moisture. Stippling indicates that at least 50% of the CMIP5 models listed in Table 1 identified an FD in the SSI.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
(Aa)–(Al) Hit rate and (Ba)–(Bl) false alarm ratio for (left) SPI, (center) EDDI, and (right) ESI. The tolerance window for valid detection is from 8 weeks prior to the onset of an FD in soil moisture. Stippling indicates that at least 50% of the CMIP5 models listed in Table 1 identified an FD in the SSI.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
(Aa)–(Al) Hit rate and (Ba)–(Bl) false alarm ratio for (left) SPI, (center) EDDI, and (right) ESI. The tolerance window for valid detection is from 8 weeks prior to the onset of an FD in soil moisture. Stippling indicates that at least 50% of the CMIP5 models listed in Table 1 identified an FD in the SSI.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
The SPI consistently shows the highest HR globally throughout all seasons, with around 60%–80% of FDs detected in the SSI also detected by the SPI. The EDDI and ESI have a significantly lower HR in the ranges 40%–60%, and around 20%, respectively. The higher HR for the SPI could be the result of the high responsiveness of the top soil layer to rainfall. Deeper soil layers might reveal differences in the HR between the SPI and the two evaporation-based indices. However, as previously outlined, these data are not available at the temporal resolution required for the analysis.
The high HR for the SPI is consistent across all models. The EDDI HR is generally weaker in the CanESM2, despite its much higher EDDI FD frequency. Theoretically, an overestimation of the frequency should increase the chances of an FD in the EDDI coinciding with one from the SSI, but this is not the case. This suggests that many high RoC periods in the EDDI (i.e., sudden, high E0) do not coincide with similarly rapid depletion in soil moisture. Instead, they must occur during times when soil moisture is not showing severe drying conditions either illustrating issues with the land–atmosphere processes in extreme conditions or E0 spikes more often upward despite ample soil moisture for this particular model. For the ESI, the HR is very low in all GFDL models contributing to the overall low HR in the ensemble mean. CanESM2, CSIRO Mk3.6.0, and MIROC5 generally show HRs of greater than 50% for the ESI.
The above results are also embodied in the FAR (Fig. 8, B section) as a high HR usually means a low FAR, and vice versa. While the FARs for EDDI are very similar to those of the SPI (around 40%), the FAR for the ESI is considerably higher with around 80% in stippled areas. The FAR for the EDDI is largely driven by the high departures of FD frequency in CanESM2. The seasonally varying FAR for the SPI and EDDI are caused by their relatively uniform distribution of FDs over all seasons, whereas FDs in the soil moisture mainly occur during the warm season. During the colder season, EDDI and SPI might still detect the soil moisture FD but their event frequency in general is too high. The overdetection of FD in the cool season is less apparent in the HR and FAR for the ESI. While the FD frequency identified in the ESI is very similar to the SSI (and would yield a low FAR if the HR was good), they tend to happen slightly after the FD occurs in the soil moisture, for which the high responsiveness in top 10-cm soil layer is likely to be responsible. As a result, both detection skills are very weak. However, this varies between the individual CMIP5 models more than HR and FAR for EDDI and SPI.
The results for the ESI improve when the detection window is extended to 2 weeks after the FD occurred in the soil moisture (Fig. 9). This time, FDs detected in the ESI are a result of an FD in the SSI and are captured as well, though showing a lag. The ESI HRs improve by approximately 20%–30% for the CMIP5 ensemble (Fig. 9, right column), again showing the highest HRs in CanESM2, CSIRO Mk3.6.0, and MIROC5. Similarly, there is a concurrent drop in the FAR. The HR for the EDDI also improves by about 10%–20%, but there are no differences for the SPI due to its continuous leading signal. While the ESI’s HR now closely matches that of the EDDI, both are still lower than the SPI’s HR. The marked improvement in skill for the ESI, and to some extent for the EDDI, for the 2-week extension of the detection window to post-SSI FD onset implies that both indices are less skillful for FD prediction but remain useful for FD monitoring. For the top 10-cm soil moisture, the SPI is the best index for FD prediction.
As in Fig. 8, but the tolerance window has been extended to 2 weeks after FD onset in soil moisture.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
As in Fig. 8, but the tolerance window has been extended to 2 weeks after FD onset in soil moisture.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
As in Fig. 8, but the tolerance window has been extended to 2 weeks after FD onset in soil moisture.
Citation: Journal of Hydrometeorology 22, 6; 10.1175/JHM-D-20-0262.1
5. Discussion
The goal of this study was to examine the occurrence of flash drought (FD) in CMIP5 models and to evaluate how well various drought indices represent FDs in comparison with FD computed directly from the models’ soil moisture output.
The characteristic behaviors of the EDDI, ESI, and SPI provide information about the processes related to FD in the upper 10-cm soil moisture level. As the ESI uses the latent heat or evaporation flux, it has a direct link to the available moisture; this explains why its FD event frequency is similar to that of the SSI. When available moisture in the land surface depletes, incoming energy from shortwave radiation is transformed into sensible heat, which increases temperature and drives E0 and thus creates a water-limited condition, a dynamic that drives the complementary relationship between ET and E0 (Bouchet 1963). This is when the ESI starts to decrease as ET can no longer match E0. As long as available moisture is sufficient in an energy-limited condition, the ESI cannot decrease because ET and E0 are in a parallel relationship (Anderson et al. 2007). However, the response in the ESI is delayed as an increase in E0 caused by increasing temperature (e.g., by advection) initially results in an increase of ET due to their parallel relation during energy-limited conditions (Hobbins et al. 2017). The lag in the correlation as well as the weak rates of change are evidence of this delay.
The EDDI is also coupled to available moisture due to the complementary relationship between E0 and ET. The initial response of E0 to changing synoptic conditions is amplified due to the decrease in cloudiness increasing surface energy availability and thus by an increase in sensible heat flux during the development of a drought when ET decreases. This is especially the case in regions of high net radiation and where there are seasonal transitions between water-limited and energy-limited conditions (Koster et al. 2009b). On the other hand, in conditions of sufficient soil moisture supply, high E0—due to increased wind speed, for example—can also occur without increasing the risk of drought; this can offset the increased moisture demand (Otkin et al. 2018). The EDDI is also less reliable in energy-limited environments where ET and E0 are positively correlated (McEvoy et al. 2016). As a result of different driving mechanisms, EDDI shows the highest frequency for the occurrence of FD.
Precipitation alone, represented by the SPI, sits between EDDI and ESI in terms of its ability to capture FD event frequency. A sharp decrease in precipitation may not necessarily lead to an FD, especially not during the cold season when ET is low or when a surplus in precipitation was received prior to the decline. The SPI can produce large oscillations on short time scales of just a few days due to the episodic nature of rainfall. Thus, each prominent drop in precipitation will likely result in an FD event even though the surface conditions do not indicate drought. Since both a precipitation deficiency and high anomalies of E0 can occur independently without causing the soil moisture to decline significantly (Otkin et al. 2018), especially in the root-zone layer (for which CMIP5 does not provide a daily measure), the absolute number of FD events is likely overestimated by SPI and EDDI.
The strong negative values expressed in categories according to the USDM and strong RoC indicate that precipitation has a primary role for priming an FD, as shown by the SPI. Arguably, a precipitation deficiency is a common feature in all drought types and might not be unique enough to characterize an FD (Otkin et al. 2018). This is supported by the overestimation of FD in the SPI itself. While Fig. 5 just represents the average index value at drought onset suggesting that precipitation is the primary driver for FD, the RoC for those indices can vary for individual FD events. This is consistent with findings from Koster et al. (2019), who ascribed the importance of ET to the development of FD but identified its overall contribution as “small relative to the contribution of precipitation deficits.” It would also support the separation into heatwave- and precipitation-driven FD (Mo and Lettenmaier 2015, 2016), with FDs primarily caused by a precipitation deficit occurring more frequently. In contrast, studies from Liu et al. (2020a,b) with focus on China found temperature and E0 to be the driving factors for the development of flash drought. In the context of a warming climate, Yuan et al. (2019) show that the increasing risk of future flash droughts is attributed to increases in temperature and ET, yet also manifested on the Chinese domain. Droughts in central Asia have been found to be primarily driven by heat waves (Schubert et al. 2014), which would support this anomaly for parts of China. The differences in the relative importance of E0 in these Chinese studies in comparison with here indicate the dependence of temperature and E0 on location.
The relationship of precipitation, E0, and ET with the top 10-cm soil moisture dictates their utility for FD prediction. Precipitation as the moisture supply is closely correlated with top 10-cm soil moisture variability and therefore must lead the soil moisture signal. Consequently, the prediction skill of the SPI is highest (HR ~70% during warm seasons). EDDI detects only half of the soil moisture FD (HR ~50%), showing the primary contribution of precipitation to the development of FD and the longer response time of increasing E0 from the sensible heat flux. Nevertheless, a lead time of up to 2 months was observed for the EDDI to detect FD in the summer of 2012 in the United States (McEvoy et al. 2016).
The weak skill of the ESI (HR 30%), undercuts the initial positive impression based on the very similar FD event frequency and covariability to the SSI. The results suggest that the ESI is rather a tool for FD monitoring than for early warning. However, this is inconsistent with what is found in the literature of observed case studies. Nguyen et al. (2019) detected a lead time of about 1 month for the ESI in Queensland, Australia, and similar lead times were described in studies from Otkin et al. (2013, 2015) in the central United States. The difference from previous studies (Otkin et al. 2013, 2015; McEvoy et al. 2016; Nguyen et al. 2019) is that we apply a stricter FD definition. The indices might start showing a drying signal in the time series earlier than it is picked up by our thresholds.
However, the fact that we are limited to the upper 10-cm soil layer shortens the response time of the soil moisture signal to atmospheric conditions and hinders the comparison with these studies, which either focus on the root-zone layer of at least 1-m depth or only investigate specific FD cases and do not explicitly determine detections skills across a population of FDs. Consequently, the lead of EDDI and ESI in reference to the deeper soil layers is offset by the fast response of the upper most soil layer. Moreover, FDs in the upper 10 cm do not necessarily penetrate deep enough into the ground to affect the root-zone layer. Our chosen minimum FD duration of 4 weeks is based on Ford and Quiring (2014), which is determined in Oklahoma and therefore might not be applicable to other climates and soil types across the globe.
The results described herein need to be seen in the light of CMIP5 models and their underlying limitations, with the imposed LSMs in mind. Their formulations will have geographically and seasonally varying biases dependent on the land fluxes and states they produce (Koster and Suarez 2001; Koster et al. 2009a). The overall lowest spread for SPI and EDDI FD event frequency relative to ESI and SSI confirms again the models’ limited ability to model land–atmosphere interactions that reflect the latent heat flux and available moisture correctly relative to their ability to model atmospheric phenomena such as precipitation. However, issues with pedotransfer functions in LSMs are well known (Pitman 2003; Vereecken et al. 2016; Van Looy et al. 2017) and contribute to the intermodel spread. The consistent higher frequency occurrence of FD identified by the EDDI might also be indicative of an overestimation of E0 caused by an excessive sensible heat flux (Seneviratne et al. 2010). An overestimation of evaporative droughts similar to our findings has been observed in offline LSMs by Ukkola et al. (2016). Their results show that LSMs poorly parameterize the partitioning of sensible and latent heat fluxes. These discrepancies are also embedded in CMIP5 models (Ukkola et al. 2018a,b). An overestimation of sensible heat would directly lead to an increased E0 causing EDDI to increase more rapidly (Seneviratne et al. 2010). An underestimated latent heat flux, and thus ET, during water-stressed conditions would in turn decrease the ESI disproportionally but slow down the drying of soil moisture. The former would also yield an FD more rapidly in the ESI whereas the latter does not in the SSI.
6. Conclusions
This study has examined the flash drought (FD) event frequency globally in CMIP5 models using three drought indices representing both the supply and demand drivers of drought and were tested against the integrating state of soil moisture. FDs were identified using soil moisture and the detection of FD was compared using indices that measure precipitation, evaporative demand (E0) from reference ET, and evapotranspiration as included in the SSI, SPI, EDDI, and ESI respectively. This analysis determined the detection capabilities of the indices. We provided information on the use of these indices as early warning tools for FD detection, based on the improvement in our metrics using leads and lags between the indices and the SSI.
Our findings show that the SPI and EDDI overestimate FD event frequency relative to that determined from soil moisture, while the ESI shows a similar frequency. The data from the CMIP5 models show less agreement in the results of event frequency for SSI and ESI than for EDDI and SPI, which is caused by the model biases associated with different land–atmosphere interactions. Consequently, the results show that the ESI, while a useful index for the detection and monitoring of FD, cannot provide an early warning capability. On the other hand, SPI, and to some extent EDDI, are capable of early detection in the models at lead times of over a week. However, they also have relatively high FAR associated with their overestimation of FD event frequency. These results vary between the individual models, especially for the ESI, highlighting that these results might be a function of the different LSMs and coupling.
The results presented here are for the top 10-cm layer of soil only, due to limited availability of daily time scale data from the selected CMIP5 models; this limits the result’s application to actual FD events, which are defined by the impact on the root-zone layer soil moisture (Otkin et al. 2018). Further investigation using observational data is warranted to determine whether the detection capabilities of ESI, EDDI, and SPI in the CMIP5 models are representative of reality.
Acknowledgments
This work was partially supported by the Australian Research Council Centre of Excellence for Climate Extremes (Grant CE170100023). Author Gallant was supported by the Australian Research Council project DE150101297. Author Hobbins was supported in part by the NOAA Cooperative Agreement with CIRES, NA17OAR4320101. We acknowledge the National Computational Infrastructure at the Australian National University (NCI), which is supported by the Australian Government, and the Earth System Grid Federation for making the CMIP5 model outputs available. We also gratefully acknowledge the World Climate Research Programme’s Working Group on Coupled Modeling, which is responsible for CMIP, and thank the climate modeling groups (listed in Table 1) for producing and making available their model output. We also acknowledge Julie M. Arblaster for providing feedback on the paper.
Data availability statement
The CMIP5 outputs used in this study are available from the Earth System Grid Federation (https://esgf-node.llnl.gov).
REFERENCES
Allen, R. G., L. S. Pereira, D. Raes, and M. Smith, 1998: Crop evapotranspiration: Guidelines for computing crop water requirements. FAO Irrigation and Drainage Paper 56, 300 pp., http://www.fao.org/3/X0490E/X0490E00.htm.
Anderson, M. C., J. M. Norman, J. R. Mecikalski, J. A. Otkin, and W. P. Kustas, 2007: A climatological study of evapotranspiration and moisture stress across the continental United States based on thermal remote sensing: 1. Model formulation. J. Geophys. Res., 112, D10117, https://doi.org/10.1029/2006JD007506.
Arora, V. K., and H. D. Matthews, 2009: Characterizing uncertainty in modeling primary terrestrial ecosystem processes. Global Biogeochem. Cycles, 23, GB2016, https://doi.org/10.1029/2008GB003398.
Arora, V. K., and G. J. Boer, 2010: Uncertainties in the 20th century carbon budget associated with land use change. Global Change Biol., 16, 3327–3348, https://doi.org/10.1111/j.1365-2486.2010.02202.x.
Bouchet, R. J., 1963: Évapotranspiration réelle et potentielle, signification climatique. IAHS Publ., 62, 134–142.
Entekhabi, D., I. Rodriguez-Iturbe, and F. Castelli, 1996: Mutual interaction of soil moisture state and atmospheric processes. J. Hydrol., 184, 3–17, https://doi.org/10.1016/0022-1694(95)02965-6.
Farahmand, A., and A. AghaKouchak, 2015: A generalized framework for deriving nonparametric standardized drought indicators. Adv. Water Resour., 76, 140–145, https://doi.org/10.1016/j.advwatres.2014.11.012.
Ford, T. W., and S. M. Quiring, 2014: Comparison and application of multiple methods for temporal interpolation of daily soil moisture. Int. J. Climatol., 34, 2604–2621, https://doi.org/10.1002/joc.3862.
Ford, T. W., and C. F. Labosier, 2017: Meteorological conditions associated with the onset of flash drought in the eastern United States. Agric. For. Meteor., 247, 414–423, https://doi.org/10.1016/j.agrformet.2017.08.031.
Ford, T. W., D. B. McRoberts, S. M. Quiring, and R. E. Hall, 2015: On the utility of in situ soil moisture observations for flash drought early warning in Oklahoma, USA. Geophys. Res. Lett., 42, 9790–9798, https://doi.org/10.1002/2015GL066600.
Hobbins, M. T., A. Wood, D. J. McEvoy, J. L. Huntington, C. Morton, M. Anderson, and C. Hain, 2016: The evaporative demand drought index. Part I: Linking drought evolution to variations in evaporative demand. J. Hydrometeor., 17, 1745–1761, https://doi.org/10.1175/JHM-D-15-0121.1.
Hobbins, M. T., D. McEvoy, and C. Hain, 2017: Evapotranspiration, evaporative demand, and drought. Drought and Water Crises, D. Wilhite and R. Pulwarty, Eds., CRC Press, 259–288.
Hunt, E. D., M. Svoboda, B. Wardlow, K. Hubbard, M. Hayes, and T. Arkebauer, 2014: Monitoring the effects of rapid onset of drought on non-irrigated maize with agronomic data and climate-based drought indices. Agric. For. Meteor., 191, 1–11, https://doi.org/10.1016/j.agrformet.2014.02.001.
Jeffrey, S. J., L. D. Rotstayn, M. A. Collier, S. M. Dravitzki, C. Hamalainen, C. Moeseneder, K. K. Wong, and J. I. Syktus, 2013: Australia’s CMIP5 submission using the CSIRO Mk3.6 model. Aust. Meteor. Oceanogr. J., 63, 1–13, https://doi.org/10.22499/2.6301.001.
Koster, R. D., and M. J. Suarez, 2001: Soil moisture memory in climate models. J. Hydrometeor., 2, 558–570, https://doi.org/10.1175/1525-7541(2001)002<0558:SMMICM>2.0.CO;2.
Koster, R. D., and Coauthors, 2004: Regions of strong coupling between soil moisture and precipitation. Science, 305, 1138–1140, https://doi.org/10.1126/science.1100217.
Koster, R. D., Z. Guo, R. Yang, P. A. Dirmeyer, K. Mitchell, and M. J. Puma, 2009a: On the nature of soil moisture in land surface models. J. Climate, 22, 4322–4335, https://doi.org/10.1175/2009JCLI2832.1.
Koster, R. D., S. D. Schubert, and M. J. Suarez, 2009b: Analyzing the concurrence of meteorological droughts and warm periods, with implications for the determination of evaporative regime. J. Climate, 22, 3331–3341, https://doi.org/10.1175/2008JCLI2718.1.
Koster, R. D., S. D. Schubert, H. Wang, S. P. Mahanama, and A. M. DeAngelis, 2019: Flash drought as captured by reanalysis data: Disentangling the contributions of precipitation deficit and excess evapotranspiration. J. Hydrometeor., 20, 1241–1258, https://doi.org/10.1175/JHM-D-18-0242.1.
Lisonbee, J., M. Woloszyn, and M. Skumanich, 2021: Making sense of flash drought: Definitions, indicators, and where we go from here. J. Appl. Serv. Climatol., 2021, 1–19, https://doi.org/10.46275/JOASC.2021.02.001.
Liu, Y., and Coauthors, 2020a: Flash droughts characterization over China: From a perspective of the rapid intensification rate. Sci. Total Environ., 704, 135373, https://doi.org/10.1016/j.scitotenv.2019.135373.
Liu, Y., and Coauthors, 2020b: Two different methods for flash drought identification: Comparison of their strengths and limitations. J. Hydrometeor., 21, 691–704, https://doi.org/10.1175/JHM-D-19-0088.1.
Lorenz, R., and Coauthors, 2016: Influence of land-atmosphere feedbacks on temperature and precipitation extremes in the GLACE-CMIP5 ensemble. J. Geophys. Res. Atmos., 121, 607–623, https://doi.org/10.1002/2015JD024053.
Lukas, J., M. Hobbins, and I. Rangwala, 2017: The EDDI user guide. NOAA, 35 pp., https://psl.noaa.gov/eddi/pdf/EDDI_UserGuide_v1.0.pdf.
McEvoy, D. J., J. L. Huntington, M. T. Hobbins, A. Wood, C. Morton, M. Anderson, and C. Hain, 2016: The Evaporative Demand Drought Index. Part II: CONUS-wide assessment against common drought indicators. J. Hydrometeor., 17, 1763–1779, https://doi.org/10.1175/JHM-D-15-0122.1.
McKee, T. B., N. J. Doesken, and J. Kleist, 1993: The relationship of drought frequency and duration to time scales. Preprints, Eighth Conf. on Applied Climatology, Anaheim, CA, Amer. Meteor. Soc., 179–184.
Milly, P. C., and Coauthors, 2014: An enhanced model of land water and energy for global hydrologic and Earth-system studies. J. Hydrometeor., 15, 1739–1761, https://doi.org/10.1175/JHM-D-13-0162.1.
Mo, K. C., and D. P. Lettenmaier, 2015: Heat wave flash droughts in decline. Geophys. Res. Lett., 42, 2823–2829, https://doi.org/10.1002/2015GL064018.
Mo, K. C., and D. P. Lettenmaier, 2016: Precipitation deficit flash droughts over the United States. J. Hydrometeor., 17, 1169–1184, https://doi.org/10.1175/JHM-D-15-0158.1.
Nguyen, H., M. C. Wheeler, J. A. Otkin, T. Cowan, A. Frost, and R. Stone, 2019: Using the evaporative stress index to monitor flash drought in Australia. Environ. Res. Lett., 14, 064016, https://doi.org/10.1088/1748-9326/ab2103.
Noguera, I., F. Dominguez-Castro, and S. M. Vicente-Serrano, 2020: Characteristics and trends of flash droughts in Spain, 1961-2018. Ann. N. Y. Acad. Sci., 1472, 155–172, https://doi.org/10.1111/nyas.14365.
Orlowsky, B., and S. I. Seneviratne, 2013: Elusive drought: Uncertainty in observed trends and short- and long-term CMIP5 projections. Hydrol. Earth Syst. Sci., 17, 1765–1781, https://doi.org/10.5194/hess-17-1765-2013.
Otkin, J. A., M. C. Anderson, C. Hain, I. E. Mladenova, J. B. Basara, and M. Svoboda, 2013: Examining rapid onset drought development using the thermal infrared–based evaporative stress index. J. Hydrometeor., 14, 1057–1074, https://doi.org/10.1175/JHM-D-12-0144.1.
Otkin, J. A., M. C. Anderson, C. Hain, and M. Svoboda, 2015: Using temporal changes in drought indices to generate probabilistic drought intensification forecasts. J. Hydrometeor., 16, 88–105, https://doi.org/10.1175/JHM-D-14-0064.1.
Otkin, J. A., and Coauthors, 2016: Assessing the evolution of soil moisture and vegetation conditions during the 2012 United States flash drought. Agric. For. Meteor., 218–219, 230–242, https://doi.org/10.1016/j.agrformet.2015.12.065.
Otkin, J. A., M. Svoboda, E. D. Hunt, T. W. Ford, M. C. Anderson, C. Hain, and J. B. Basara, 2018: Flash droughts: A review and assessment of the challenges imposed by rapid-onset droughts in the United States. Bull. Amer. Meteor. Soc., 99, 911–919, https://doi.org/10.1175/BAMS-D-17-0149.1.
Pendergrass, A. G., and Coauthors, 2020: Flash droughts present a new challenge for subseasonal-to-seasonal prediction. Nat. Climate Change, 10, 191–199, https://doi.org/10.1038/s41558-020-0709-0.
Pitman, A. J., 2003: The evolution of, and revolution in, land surface schemes designed for climate models. Int. J. Climatol., 23, 479–510, https://doi.org/10.1002/joc.893.
Rippey, B. R., 2015: The U.S. drought of 2012. Wea. Climate Extreme, 10, 57–64, https://doi.org/10.1016/j.wace.2015.10.004.
Schubert, S. D., H. Wang, R. Koster, M. Suarez, and P. Groisman, 2014: Northern Eurasian heat waves and droughts. J. Climate, 27, 3169–3207, https://doi.org/10.1175/JCLI-D-13-00360.1.
Seneviratne, S. I., T. Corti, E. L. Davin, M. Hirschi, E. B. Jaeger, I. Lehner, B. Orlowsky, and A. J. Teuling, 2010: Investigating soil moisture–climate interactions in a changing climate: A review. Earth-Sci. Rev., 99, 125–161, https://doi.org/10.1016/j.earscirev.2010.02.004.
Seneviratne, S. I., and Coauthors, 2013: Impact of soil moisture-climate feedbacks on CMIP5 projections: First results from the GLACE-CMIP5 experiment. Geophys. Res. Lett., 40, 5212–5217, https://doi.org/10.1002/grl.50956.
Svoboda, M., and Coauthors, 2002: The Drought Monitor. Bull. Amer. Meteor. Soc., 83, 1181–1190, https://doi.org/10.1175/1520-0477-83.8.1181.
Takata, K., S. Emori, and T. Watanabe, 2003: Development of the minimal advanced treatments of surface interaction and run-off. Global Planet. Change, 38, 209–222, https://doi.org/10.1016/S0921-8181(03)00030-4.
Taylor, K. E., R. J. Stouffer, and G. A. Meehl, 2012: An overview of CMIP5 and the experiment design. Amer. Meteor. Soc., 93, 485–498, https://doi.org/10.1175/BAMS-D-11-00094.1.
Ukkola, A. M., M. G. De Kauwe, A. J. Pitman, M. J. Best, G. Abramowitz, V. Haverd, M. Decker, and N. Haughton, 2016: Land surface models systematically overestimate the intensity, duration and magnitude of seasonal-scale evaporative droughts. Environ. Res. Lett., 11, 104012, https://doi.org/10.1088/1748-9326/11/10/104012.
Ukkola, A. M., A. J. Pitman, M. G. De Kauwe, G. Abramowitz, N. Herger, J. P. Evans, and M. Decker, 2018a: Evaluating CMIP5 model agreement for multiple drought metrics. J. Hydrometeor., 19, 969–988, https://doi.org/10.1175/JHM-D-17-0099.1.
Ukkola, A. M., A. J. Pitman, M. G. Donat, M. G. De Kauwe, and O. Angélil, 2018b: Evaluating the contribution of land-atmosphere coupling to heat extremes in CMIP5 models. Geophys. Res. Lett., 45, 9003–9012, https://doi.org/10.1029/2018GL079102.
Van Looy, K., and Coauthors, 2017: Pedotransfer functions in earth system science: Challenges and perspectives. Rev. Geophys., 55, 1199–1256, https://doi.org/10.1002/2017RG000581.
Vereecken, H., and Coauthors, 2016: Modeling soil processes: Review, key challenges, and new perspectives. Vadose Zone J., 15, 1–57, https://doi.org/10.2136/vzj2015.09.0131.
Vicente-Serrano, S. M., S. Beguería, and J. I. López-Moreno, 2010: A multiscalar drought index sensitive to global warming: The standardized precipitation evapotranspiration index. J. Climate, 23, 1696–1718, https://doi.org/10.1175/2009JCLI2909.1.
Wang, L., X. Yuan, Z. Xie, P. Wu, and Y. Li, 2016: Increasing flash droughts over China during the recent global warming hiatus. Sci. Rep., 6, 30571, https://doi.org/10.1038/srep30571.
Yuan, S., and S. M. Quiring, 2017: Evaluation of soil moisture in CMIP5 simulations over the contiguous United States using in situ and satellite observations. Hydrol. Earth Syst. Sci., 21, 2203–2218, https://doi.org/10.5194/hess-21-2203-2017.
Yuan, X., and Coauthors, 2019: Anthropogenic shift towards higher risk of flash drought over China. Nat. Commun., 10, 4661, https://doi.org/10.1038/s41467-019-12692-7.
Zhang, Y., Q. You, C. Chen, and X. Li, 2017: Flash droughts in a typical humid and subtropical basin: A case study in the Gan River Basin, China. J. Hydrol., 551, 162–176, https://doi.org/10.1016/j.jhydrol.2017.05.044.
Zhang, Y., Q. You, C. Chen, J. Ge, and M. Adnan, 2018: Evaluation of downscaled CMIP5 coupled with VIC model for flash drought simulation in a humid subtropical basin, China. J. Climate, 31, 1075–1090, https://doi.org/10.1175/JCLI-D-17-0378.1.