Drought in the United States

Siegfried Schubert and Randal Koster October 2012

Introduction

Droughts have beleaguered the United States throughout its history. The Dust Bowl conditions of the 1930s were caused by a particularly severe, extensive, and long-lasting U.S. drought. The 1950s drought in the Southern Plains was more limited in extent and duration but was equally severe. More recent examples of U.S. drought include those of 1988-1989 and the summer of 2012.

Droughts, of course, are characterized by moisture deficits, but they are also associated with heat waves. The summer of 2012, in addition to being exceptionally dry, was also exceptionally hot over much of the central and eastern part of the country. July 2012 proved to be the warmest month ever recorded in the U.S.

figure 1afigure 1b
click for larger image
Figure 1. a) Short term drought (as of 11 August 2012) reflecting precipitation deficits over the last few months and their impacts on, for example, surface soil moisture and unregulated stream flow. b) Long term drought (as of August 2012) reflecting, for example, reservoir content, groundwater and lake levels. Source: US Drought Monitor. (Plots taken from http://droughtmonitor.unl.edu)

Figure 1 shows how the nature of a given drought differs based on the time period considered. Figure 1a is a representation of short-term drought for summer 2012, reflecting deficits in the "fast responding" components of the water cycle (such as soil moisture and streamflow). Figure 1b provides the corresponding picture of longer-term drought, which considers water storage in, for example, reservoirs, lakes, and groundwater. The longer-term picture for 2012 is thus strongly affected by the exceptionally severe drought conditions that occurred in Texas and surrounding regions during much of 2011.

The disparity between Figures 1a and 1b is a reflection of the ineffectiveness of any single definition of drought. Climate researchers generally consider at least three different types of drought: those associated with precipitation deficits ("meteorological drought"), with soil moisture deficits ("agricultural drought"), and with streamflow or groundwater deficits ("hydrological drought"). Meteorological droughts are the natural precursors to agricultural and hydrological droughts; how the latter two relate to the first is a complex function of a multitude of land surface processes. Societal demand further complicates how drought is defined. A hydrological drought, for instance, that is minor in the climatic sense may have immense societal impacts if demand for streamflow water is high.

What causes drought? Researchers are beginning to understand some of the important large-scale controls on precipitation deficits. Certain spatial patterns of sea surface temperature (SST), for example, appear to be conducive to the generation of meteorological droughts. Relevan spatial patterns of SST variability are dominated by the El Niño – Southern Oscillation (ENSO) on seasonal-to-interannual time scales, while at longer time scales they are associated with the Pacific Decadal Oscillation (PDO), with the Atlantic Multi-decadal Oscillation (AMO), and with a trend related to climate change. Various studies have linked these patterns to drought over the U.S.; Wang et al. (2009), for example, shows how the PDO (and to some extent the AMO) signal appears to be responsible for wetter and cooler conditions over the central part of North America in recent decades, despite the overall drying and warming associated with climate change that has occurred over the rest of the continent. In fact, a key concern is whether the 2011 and 2012 drought conditions reflect a sign change of the decadal (PDO and AMO) SST patterns, a change that would act to reinforce any climate change signal and thus exacerbate drought conditions over the next decade or so. ENSO further complicates the picture by tending to exacerbate drought conditions during a La Niña event or ameliorate drought conditions during an El Niño event on shorter (year-to-year) time scales.

Land surface processes may extend and/or amplify drought through land-atmosphere feedback. A month with low precipitation leads to a drier than average soil, for example, and this in turn can lead to lower than average evaporation. Under certain conditions, this lower evaporation may lead to continued low precipitation. Complicating any consideration of mechanisms producing drought, however, is the fact that weather patterns have a chaotic (random) component — one could argue that some meteorological droughts are not "caused" by anything in particular, but are rather the unlucky outcome of a consecutive string of randomly generated dry periods.

The extent to which identifiable (non-random) factors control droughts affords some hope for accurate drought prediction, and this is the motivation of much research. Recent GMAO work has investigated past droughts (such as the Dust Bowl) and the physical mechanisms that influenced them. Focus here now turns to the 2011 and 2012 summer droughts, putting them in the context of other recent droughts, examining their causes, and assessing our ability to predict them.

II. The current drought and comparisons with past droughts

Figure 2 provides one indication of how precipitation has varied over the past few decades. The figure shows total precipitation (for the 3 month period May — July) for the last 30 years in terms of percentile, with, for example, a dark red color indicating precipitation in the lowest 2% of the May — July totals represented. Shown clearly are the exceedingly low precipitation rates centered around western Texas in 2011, and the similarly low rates covering much of the country in 2012. In terms of severity and spatial extent, the precipitation deficits in 2012 are similar to those of 1988.

figure 2
click for larger image
Figure 2. May-July precipitation for each of the last 30 years, expressed in terms of percentile. The precipitation data are from station observations used in the North-American Land Data Assimilation System (NLDAS).

Figure 3 shows corresponding estimates of July 30 soil moisture content across the continental U.S. for each of the last 30 years, again expressed in terms of percentile. (To generate the soil moisture values on which the percentiles are based, the GMAO land surface model was driven with observation-based precipitation and other meteorological forcings over the specified three-decade period.) By focusing on soil moisture, the plots here focus on agricultural drought; corresponding plots focusing on groundwater or reservoir storage would presumably show altered patterns.

In terms of drought spatial extent and severity, the 1988 and 2012 soil moisture deficits clearly stand out in this picture as the worst. Relative to the dryness pattern in 1988, that in 2012 is shifted to the south, with states like Illinois, Indiana, Ohio, and the whole of Missouri being especially hard hit. In these states, according to this analysis, soil moistures in 2012 are the driest they've been in decades.

figure 3
click for larger image
Figure 3. Soil moisture conditions on July 30 for each of the last 30 years, expressed in terms of percentile.

III. What are models and observations telling us about the causes?

A successful prediction of a drought prior to its appearance would indicate that the drought was not a random event, i.e., that it has mechanistic causes that can perhaps be isolated and analyzed. Here the physical mechanisms underlying drought are examined using seasonal forecasts and supplemental idealized simulations from the GMAO modeling system. Understanding the 2012 drought requires an examination of conditions in both 2011 and 2012; deficits in "slower" moisture reservoirs (groundwater, lake levels, etc.) during 2012 have been strongly influenced by what happened in Texas and surrounding regions in 2011.

a) The summer of 2011

The GMAO seasonal forecast system produces an ensemble of 9-month forecasts every month. Figures 4 and 5 show the forecast system's predictions of precipitation and temperature. The last column in each figure, from MERRA, is treated here as observations; the model forecasts, from different start dates, are provided in columns 1–3. The model consistently and correctly predicts warm temperature anomalies in the Texas region, even at longer lead times; for instance, warm anomalies are seen in Texas in the August forecast when the model is initialized in May. It also generally tends to predict reduced precipitation in the middle of the country, though with much less consistency. The differences seen between the observations and the forecasts reflect, in part, the nature of an ensemble forecast, as discussed further below.

figure 4
click for larger image
Figure 4. Column 1: precipitation anomalies forecast by the GEOS-5 system, for a start date of early May. Column 2: precipitation anomalies forecast by the GEOS-5 system, for a start date of early June. (Note that May forecasts for this column are irrelevant.) Column 3: precipitation anomalies forecast by the GEOS-5 system, for a start date in early July. Column 4: MERRA estimates of 2011 precipitation anomalies for May, June, and July.
figure 5
click for larger image
Figure 5. Same as Figure 4, but for 2m temperature.

The partial success of these forecasts, even several months in advance, suggests that the model was able to take advantage of drought-inducing physical mechanisms that were operating in nature. Presented now are results from a series of supplemental model runs performed using a simpler version of the GEOS-5 model in which the ocean model was replaced with observed SSTs either globally or in selected regions (with SSTs outside these regions set to their climatological values), and in which land-atmospheric feedback processes were either enabled or disabled. The next two plots focus on temperature rather than precipitation, i.e., on the extent to which SST anomalies and soil moisture feedbacks were responsible for the development of the extreme heat suffered by the southern Great Plains during the summer of 2011.

The end of 2010 was marked by a transition to La Niña conditions, which are known be conducive to drought (and associated warm temperatures) in the U.S. Great Plains. The panels on the left side of Figure 6 show that the remnants of that La Niña event indeed extended into the summer of 2011. By this time, the main negative anomalies are off the equator and actually appear to reflect more of a negative PDO pattern. Smaller positive temperature anomalies occur over the western tropical Atlantic. The very hot conditions over much of the U.S. Southern Plains during this time are shown in the last column of Figure 5.

What temperature anomalies does the GEOS-5 model produce over the U.S. as a result of the imposed SST forcing? The third column of Figure 6 shows results of the model simulations initialized (from MERRA reanalysis) on 1 January 2011 and forced with the global time series of observed SSTs. (All results shown in the third column of Figure 6 are averages over 20 parallel "ensemble" simulations, each simulation representing a potential trajectory of the forecast.) The model produces a general warming over the southern and southeastern U.S. similar to that found in the observations, even reproducing the observed "bulls-eye" of warming over the Southern Plains. Because the SSTs are the only observed anomalies imposed in the model, they are necessarily responsible for the simulated anomalies of temperature shown here, and because the patterns agree reasonably well with the observed patterns, it can reasonably be inferred that the SSTs were also responsible for the warm conditions that occurred in nature.

The model results do differ from the observations, however, in their magnitude and in the details of their spatial structure. Such differences are not unexpected. Due to chaotic atmospheric dynamics (popularly considered in terms of a butterfly flapping its wings and the resulting impact on future weather systems), nature can take a number of different trajectories. Accurate predictions are possible for as long as the potential trajectories remain similar. The results in the third column of Figure 6 are averages over the ensemble members (i.e., averages over multiple potential trajectories), and the averaging necessarily mutes the anomaly amplitudes that can be achieved with any single trajectory. The observations, however, represent a single trajectory, and thus they will show some differences with the "muted" ensemble mean. Some of the individual ensemble members do in fact show temperature anomalies with amplitudes comparable to those of the observations (e.g., the last column of Figure 6).

figure 6
click for larger image
Figure 6. Left column: The global distribution of the observed surface temperature anomalies for May — July 2011. Second column from left: The observed (MERRA) surface temperature anomalies over the United States. Third column from left: Same as the second column, except for the ensemble mean from model simulations initialized on the 1 January 2011, and forced with observed SSTs. Right column: Same as third column but for a single ensemble member, chosen for being similar to the observed.

Figure 7 shows the impact of the different ocean basins on warm conditions in the Southern Great Plains. Of key interest are the tropical Pacific (TPac), northern Pacific (NPac), tropical Atlantic (TAtl), and northern Atlantic (NAtl) basins. The second column shows that 2011 SST anomalies in the tropical Pacific by themselves act to induce warm conditions there in the early part of the summer, and the fourth column shows that, later in the summer, the warm conditions were further encouraged by SST anomalies in the tropical Atlantic. The last column of Figure 7 shows the difference in the temperatures obtained with and without land-atmosphere feedback enabled. The magnitudes of these differences are similar to those presented in the first column, and thus this feedback is critical for amplifying the temperature anomalies induced by SST anomalies to sizable levels.

figure 7
click for larger image
Figure 7. Left column: The ensemble mean simulated surface temperature anomalies over the United States. Second column: Impact of tropical Pacific SST. Third column: Impact of North Pacific SST. Fourth column: Impact of tropical Atlantic SST. Fifth column: Impact of North Atlantic SST. Last column: Impact of soil moisture feedbacks.

b) The summer of 2012

Forecasts for the summer of 2012 again show some semblance of the drought, suggesting some skill in its prediction. Shown in the first column of Figure 8 are the ensemble mean forecasts (5-7 ensemble members) of July 30 soil moisture content, in terms of percentile, for three different forecast start dates (early May, early June, and early July). The top panel shows what was predicted for July 30 in forecasts initialized in early May: wetter-than-average soil in parts of the far West and East and a swath of drier-than-average conditions extending from Texas to the Northeast. The position of this swath is roughly similar to that shown in Figure 3 for the estimated actual conditions on July 30, suggesting some skill in forecasting the drought even three months out. The middle panel of the first column shows July 30 soil moisture as forecast in early June, and the bottom panel shows these moistures as forecast in early July. As should be expected, the forecasts become more accurate as the lead time of the forecast decreases.

To some extent, however, the similarity between the (estimated) actual July 30 soil moisture fields and forecast July 30 soil moisture fields reflects the persistence of the model's soil moisture. Soil moisture in the model (as in nature) is characterized by some implicit memory, so a dry condition at the start of a forecast will tend to produce a dry condition during the forecast. Because the land model used for the forecasts is the same as that used for the soil moisture estimates presented in Figure 3, some of the successful capture of the drought by the forecasts may be artificial; true success is indeed limited by the degree to which the model represents soil moisture memory accurately, and this level of accuracy is very difficult to ascertain quantitatively, though the ongoing Soil Moisture Active Passive (SMAP) mission may help shed light on the character of soil moisture persistence.

figure 8
click for larger image
Figure 8. Left column: Forecast soil moisture anomaly (ensemble mean) on July 30, as generated by forecasts starting in early May (top), early June (middle), and early July (bottom). Middle column: Same, but for a single ensemble member, chosen for being relatively wet. Right column: Same, but for a single ensemble member, chosen for being relatively dry.

The impact of different potential trajectories, or realizations, of weather — and thus an indication of the uncertainty of forecasts in general — is illustrated in the second and third columns of Figure 8, which show the forecast soil moisture for different individual ensemble members: one that is generally wetter than the others (second column) and one that is generally drier than the others (third column). Forecasts initialized in June, for instance, display a rather large spread in results. Also notice that of the two shown simulations starting in early May, one essentially predicted no drought at all, whereas the other predicted a drought with a magnitude and extent similar to that which actually occurred. This latter single realization of the model thus turns out to be similar to the single realization represented by nature. Soil moisture memory becomes relatively more important as the forecast lead decreases, and as a result, even the "wetter" forecast simulation initialized in early July shows some degree of agricultural drought on July 30.

As shown in Figure 9, the GMAO May-initialized forecasts for precipitation show (generally) low precipitation rates for 2012 in the center of the country, particularly in June and July. The same holds true for the forecasts starting in June, and the forecasts starting in July were fairly confident that July would have low rain rates in the middle of the continent. Again, it must be emphasized that the forecasts shown are averages of multiple ensemble members, whereas nature, in contrast, provides only one realization.

figure 9
click for larger image
Figure 9. Column 1: Precipitation anomalies (ensemble means) produced in forecasts initialized in early May, for May (top), June (middle), and July of 2012. Column 2: Same, but for forecasts initialized in early June. Column 3: Same, but for forecasts initialized in early July. Column 4: MERRA estimates of the precipitation anomalies.

The temperature forecasts (Figure 10) provide a similar picture of forecast skill, though the July temperatures from forecasts initialized in early June (bottom row, second panel from the left) are considerably more realistic than the corresponding forecasts of precipitation. Curiously, for 2012, the temperature forecast skill stems more from the initialization of the land and atmosphere states than from the initialization of the SST states, as indicated by the analysis of a supplemental set of long-term simulations with prescribed observed SSTs, a set of simulations similar to that discussed in the context of Figure 6. Figure 11 shows that the impact of SSTs in that simulation was confined to the western US, where it produced a modest warming, similar to that shown in the bottom left panel of Figure 10.

figure 10
click for larger image
Figure 10. Same as Figure 9, but for 2m temperature.
figure 11
click for larger image
Figure 11. Left column: The global distribution of the observed surface temperature anomalies for May — July 2012. Second column from left: The observed (MERRA) surface temperature anomalies over the United States. Third column from left: Same as the second column, except for the ensemble mean from AMIP simulations forced with observed SSTs. Right column: Same as third column but for a single ensemble member, chosen for being similar to the observed.

IV. What can we say about the future?

Figure 12 shows that, as of the summer of 2012, the consensus product of the NMME predicts warm anomalies for the tropical Pacific for the period January through March (JFM) 2013. (NMME stands for National Multi-Model Ensemble, a seasonal forecasting project spanning multiple institutions.) Each of the contributing models is predicting the same, though the amplitude of the warming varies considerably amongst them. The models thus predict a transition from the La Niña conditions that developed in late 2010 to El Niño conditions.

A transition from La Niña to El Niño would reduce the tendency for drought in the Great Plains. This is borne out in the forecasts from the GEOS-5 system, which predict higher-than-average precipitation in the south central US for winter 2012-2013 (Figure 13). (Note, however, that the warm season is when the region gets most of its rain, so that the impact of the El Niño event on 2013 warm season precipitation may be key to breaking the drought.) GEOS-5 also, however, predicts precipitation deficits for the northwest and northeast. Additionally, the GEOS-5 system predicts that the central US will remain warmer than usual into the fall (September – November) but will, along with the eastern U.S., become cooler than usual by winter (December – February). Warm anomalies remain in the forecast over much of the northwest and throughout western Canada and Alaska. While the forecasts over North America are generally consistent with the canonical response to an El Niño event, it must be pointed out that the average skill in forecasts of temperature and precipitation at these lead times is quite low.

figure 12
click for larger image
Figure 12. Surface temperature forecasts initialized in early August for JFM of 2013 from the various models included in the National Multi-Model Ensemble (NMME) project. NMME in the figure refers to the consensus (Plots from http://www.cpc.ncep.noaa.gov/products/NMME/)
figure 13
click for larger image
Figure 13. Left panels: Forecasts of 3-month mean surface temperature (ensemble mean) at various lead times, initialized in early August 2012. Right panels: Same as left panels but for precipitation. (Plots from http://www.cpc.ncep.noaa.gov/products/NMME/)

V. Concluding remarks

The droughts of 2011 and 2012 and the widespread heat waves associated with them have had major impacts in the U.S., with the greatest impacts occurring over the south/central U.S. (especially Texas) during 2011 and over the central and northern Great Plains during the summer of 2012. Experiments with the GEOS-5 forecast system show that SST anomalies (e.g., those in the tropical Pacific and tropical Atlantic during 2011) helped to establish these droughts and/or the warm conditions associated with them, though during both years there was a considerable stochastic (unforced by SST) element, as evidenced by the spread of parallel ensemble members.

We have reason to expect that our predictions of drought will improve in the coming years. We find, for example, that the aforementioned stochastic element can be reduced in the first few months of the forecast by initializing the land (i.e., the model's soil moisture state variable); we thus expect that new and improved soil moisture estimates, in particular those from the Soil Moisture and Ocean Salinity (SMOS) and Soil Moisture Active/Passive (SMAP) satellite missions, will translate into improved drought prediction. Improvements in the forecast model itself should lead to better simulations of ENSO and other low frequency variability (e.g., decadal variability linked to the AMO and PDO) that should translate (together with better estimates of the subsurface ocean) into improved seasonal and longer-term predictions. Recent decadal predictions carried out with GEOS-5 and other models as part of the CMIP5/IPCC project provide tantalizing evidence for predictability that extends well beyond that associated with the ENSO cycle.

Up to date information on the latest forecasts can be found at:

GMAO website: http://gmao.gsfc.nasa.gov/cgi-bin/products/climateforecasts/GEOS5/index.cgi
NMME website: http://www.cpc.ncep.noaa.gov/products/NMME/
The National Integrated Drought Information System (NIDIS) website: http://www.drought.gov/drought/content/welcome
The Global Drought Information System (GDIS) website: http://www.clivar.org/organization/extremes/activities/GDIS-workshop

A list of recent GMAO publications regarding drought, as well as their summaries, can be found at:

GMAO website: http://gmao.gsfc.nasa.gov/research/climate/US_drought/US_drought_pubs.php

References

Schubert, S.D., M. J. Suarez, P. J. Pegion, R. D. Koster, J. T. Bacmeister, 2004: On the Cause of the 1930s Dust Bowl. Science, 33, 1855-1859. doi: 10.1126/science.1095048

Schubert, Siegfried, and Coauthors, 2009: A U.S. CLIVAR Project to Assess and Compare the Responses of Global Climate Models to Drought-Related SST Forcing Patterns: Overview and Results. J. Climate, 22, 5251.5272. doi: 10.1175/2009JCLI3060.1

Wang, H., S.D. Schubert, M. J. Suarez, J. Chen, M. Hoerling, A. Kumar and P. Pegion, 2009: Attribution of the seasonality and regionality in climate trends over the United States during 1950-2000. J. Climate, 22, 2571-2590. doi: 10.1175/2008JCLI2359.1

« Climate Diagnostics and Prediction
« GMAO Research Highlights