Climate Dynamics

, Volume 46, Issue 1, pp 413–426

Extraordinary heat during the 1930s US Dust Bowl and associated large-scale conditions

  • Markus G. Donat
  • Andrew D. King
  • Jonathan T. Overpeck
  • Lisa V. Alexander
  • Imke Durre
  • David J. Karoly

DOI: 10.1007/s00382-015-2590-5

Cite this article as:
Donat, M.G., King, A.D., Overpeck, J.T. et al. Clim Dyn (2016) 46: 413. doi:10.1007/s00382-015-2590-5


Unusually hot summer conditions occurred during the 1930s over the central United States and undoubtedly contributed to the severity of the Dust Bowl drought. We investigate local and large-scale conditions in association with the extraordinary heat and drought events, making use of novel datasets of observed climate extremes and climate reanalysis covering the past century. We show that the unprecedented summer heat during the Dust Bowl years was likely exacerbated by land-surface feedbacks associated with springtime precipitation deficits. The reanalysis results indicate that these deficits were associated with the coincidence of anomalously warm North Atlantic and Northeast Pacific surface waters and a shift in atmospheric pressure patterns leading to reduced flow of moist air into the central US. Thus, the combination of springtime ocean temperatures and atmospheric flow anomalies, leading to reduced precipitation, also holds potential for enhanced predictability of summer heat events. The results suggest that hot drought, more severe than experienced during the most recent 2011 and 2012 heat waves, is to be expected when ocean temperature anomalies like those observed in the 1930s occur in a world that has seen significant mean warming.


Extreme heat Drought Seasonal predictability Climate variability Teleconnections 20th century reanalysis GHCNDEX 

1 Introduction

The climate over much of the US during the 1930s was characterized by exceptionally hot and dry conditions (Peterson et al. 2013), often referred to as the “Dust Bowl”. Despite great scientific interest and doubtless progress in recent years, the mechanisms leading to this extraordinary event are not yet fully understood, a troubling liability as increasingly hot droughts are expected to impact large parts of the world (Collins et al. 2013). A number of observational and modeling studies have suggested Atlantic and Pacific sea surface temperature (SST) anomalies drove the dry Dust Bowl conditions (McCabe et al. 2004; Schubert et al. 2004; Seager et al. 2008; Feng et al. 2011). However, the exact mechanisms linking ocean temperature anomalies to drought and heat over the continent remain unexplained. Further, statistical decomposition shows that while SST patterns can explain the drought conditions to some extent, much of the variance related to the 1930s drought depends on other processes (Cook et al. 2011a). Land-surface interactions and atmospheric dust from land degradation may have played a role in driving the drought, and inclusion of these factors has improved simulations of the Dust Bowl (Seager et al. 2008; Cook et al. 2009, 2011b). Other studies suggest that this drought might have been triggered by random atmospheric variability alone (Hoerling et al. 2009). Most of the studies suggesting mechanisms that contributed to the Dust Bowl drought and heat have used climate model experiments, but relatively little work has used observation-based data to understand atmospheric circulation anomalies related to drought (Brönnimann et al. 2009; Cook et al. 2011b, 2014b). While most Dust Bowl studies focus on hydrological deficit, the nature and role of extreme heat during summer has received little attention. Summer temperature has been shown to be a strong driver of evaporation, potential evapotranspiration and vegetation stress (Feng and Fu 2013; Williams et al. 2013), and might in turn favor drought conditions via land-surface feedbacks.

While previous studies mostly have focused on understanding the 1930s drought conditions, this study puts emphasis on the extreme summer temperatures that occurred during the Dust Bowl years. In fact, many temperature records in North America were set in the 1930s and still hold today (Peterson et al. 2013). Focusing on the individual years with the strongest summer heat, this study investigates the local and large-scale conditions related to the most severe heat events of the past century, and puts them in context with more recent extreme heat in North America. Through the integrated analysis of new observed temperature extremes data (Donat et al. 2013a, also see Sect. 2) and reanalysis data covering the past century (Compo et al. 2011), this paper improves understanding of the Dust Bowl climate. In particular, our study adds to the state of knowledge by providing important insights into the mechanisms linking SST anomalies to drought and heat over North America. Importantly, our results point to possible relationships among large-scale patterns in springtime ocean surface temperatures, suppressed springtime precipitation over the central US, and widespread summer heat and drought, and our analysis of related fields suggests a physical mechanism that may explain the extreme conditions of 1934 in 1936.

2 Data and methods

2.1 Observational data of temperature extremes

We use global gridded datasets of climate extremes indices calculated from station observations to study extreme summer temperatures. Two such datasets, GHCNDEX (Donat et al. 2013a) and HadEX2 (Donat et al. 2013b), were recently developed, covering the past 60–110 years respectively. HadEX2, a static dataset based on high-quality stations only, covers the period 1901–2010. GHCNDEX is designed to be updated on a regular basis and covers the time period from 1951 to present. A number of long-term stations from the GHCN-Daily archive provide data over North America back to the beginning of the 20th century. These are used to produce an extended GHCNDEX version to cover the period 1901–2012; all results shown here on observed extreme temperature are from this dataset, which allows us to investigate the heat conditions during the 1930s in comparison to the most recent heat waves in 2011 and 2012. Note that the results over North America are almost identical to HadEX2 during the overlapping period 1901–2010. Here, we use the warm day frequency (TX90p), an extreme temperature index that counts the frequency of days when the daily maximum temperature exceeds the 90th percentile of a climatological period 1961–1990. We use the term ‘hot days’ for warm days occurring during July and August—typically the 2 months when the highest temperatures occur in North America. The GHCNDEX gridded fields of TX90p are based on about 4500 stations globally that were used for interpolation, however fewer stations were available in earlier decades of the 20th Century. Over North America, about 2000 stations were used, 480 of which provided data back to 1901 (and about 1000 provided data during the 1930s).

2.2 Gridded observations of precipitation totals

The Global Precipitation Climatology Centre (GPCC) monthly gridded precipitation data (Schneider et al. 2014) are used to investigate observed precipitation totals. GPCC data were interpolated to different horizontal resolutions; we use the global grids at 1° × 1°. For 1901–2010 the full version 6 of GPCC was available which was combined with the not yet final monitoring version for 2011 and 2012. The results are largely identical (see Supplementary Figures S1 and S2 in comparison to Figs. 2, 3) over North America when using other gridded precipitation datasets covering the 20th century, such as CRU (Mitchell and Jones 2005) and GHCN-M (Peterson and Vose 1997).

2.3 Observed sea surface temperatures

For analyses of SSTs we use monthly fields from the Hadley Centre Sea Ice and SST dataset (HadISST1; Rayner et al. 2003). HadISST1 provides monthly global grids of interpolated observations of average SSTs and sea ice concentrations from 1870 to present on a 1° × 1° grid.

2.4 Atmospheric reanalysis

The 20th century reanalysis (20CR, Compo et al. 2011) is a global atmospheric reanalysis product covering the period 1871–2012. It was produced by assimilating daily synoptic surface pressure observations into a global atmospheric model, and using monthly sea-surface temperature and sea-ice as boundary conditions (HadISST1). The horizontal resolution of the NCEP global forecast system (GFS) atmospheric model used to produce 20CR is T62 (approximately 2° × 2°), using 28 vertical hybrid sigma-pressure levels. The 20CR consists of 56 ensemble members generated by an Ensemble Kalman Filter Method to optimally combine the imperfect observational data. We used daily maximum temperatures to calculate the extreme temperature index (i.e., TX90p) from each of the 56 ensemble members, consistent with the index calculation in the observational data. The daily maximum temperature was taken from the 6-hourly forecast runs of 20CR and represents the maximum over all 20-min model time steps. We also use monthly ensemble mean fields to investigate the large-scale seasonal conditions (total precipitation amount, soil moisture content, latent and sensible heat fluxes, mean sea level pressure (MSLP), winds at 850 hPa).

2.5 Scatter plots and regression between variables

To investigate a functional relationship between different variables (e.g. Figs 3, 5), we use scatter plots. Total least squares regression, accounting for uncertainty in both x- and y-direction, is used to illustrate the functional relationships. All time series were de-trended prior to the calculation of the regression and Spearman’s rank correlation coefficients.

3 Results

During the 1930s, the frequency of hot days was anomalously high over large parts of North America (Fig. 1a). This was in fact the warmest decade on record with regard to hot day frequency, with 1934 and 1936 standing out in particular (Fig. 1b). In these years the heat anomaly was most intense over the central US. Most of the North American continent was also affected by unusually positive hot day anomalies in 2012, with strongest excesses again located over the central US; however the 2012 heat was less intense than during the hottest years of the 1930s (Fig. 1c). In contrast, in 2011 the strongest heat was centered further southwest over Texas and northeast Mexico. Similarly, also in 1954 and 1980, 2 years in which anomalously hot summers were also observed, more southerly states were most strongly affected (not shown).
Fig. 1

Frequency of hot days during the summer months (July–August). a observed average annual anomaly during 1930–1939, unit: % of days. b normalized time series (unit: standard deviations) of area-averaged annual frequency of hot days during summer over the central United States (region defined by box in a) based on observational data (GHCNDEX, black line) and the 20th century reanalysis (20CR, red line). Pink shading indicates the spread between the 56 ensemble members of 20CR. Thick lines are the 11-point Gauss-filtered time series for GHCNDEX (black) and the 20CR ensemble mean (red). c Anomalies for 1934, 1936, 2011 and 2012 in hot day frequency during July and August (left) and total precipitation during March–June (right) in observations and the 20CR ensemble mean. Anomalies are calculated relative to the long-term average 1901–2010

The 20CR resembles the observed time series of hot day frequencies reasonably well after around 1910, with a correlation of 0.63 between observed and 20CR ensemble mean time series during 1910–2012 (Fig. 1b). After 1910 the spread is small between the individual ensemble members, suggesting robustness in the simulation of summer heat constrained by the large-scale driving conditions, i.e., SSTs and surface pressure. In particular, during the 1930s the reanalysis also produces exceptionally hot conditions over the central US, although it somewhat overestimates the magnitude of heat in comparison to the gridded observations. There is however some disagreement between observed and modeled summer hot day frequency, with the reanalysis model producing hotter summers than observed in 1942 and 1948 and failing to capture the magnitude of observed summer heat in 1954.

The period of summer heat during the 1930s coincided with extremely dry conditions over the central US in terms of both precipitation deficit and dry soils (Fig. 2). Low soil moisture availability reduces evaporative cooling and increases atmospheric heating from sensible heat flux (Seneviratne et al. 2010; Yin et al. 2014). Extremely low precipitation amounts in the 1930s were observed during summer (accompanying the heat) and the preceding spring. This is replicated in 20CR, although precipitation deficits are stronger during summer than in the observations. However, 20CR fails to properly capture the observed dry anomalies in spring 1954. This is likely related to the failure of 20CR to capture the summer heat that year, as discussed above. Note that 20CR also seems to underestimate the full magnitude of the precipitation deficits in 2012. This is related to a zone of positive precipitation anomalies over the Northwest of the continent, which extends further south than in observations, and into the investigation area used for calculating the area-averages (see Fig. 1c). Anomalously low latent heat fluxes from 20CR are unprecedented in the 1930s during summer but not spring (Supplementary Figure S3), suggesting that the effect of reduced evaporative cooling due to dry soils was strongest in summer (and still apparent in fall), whereas in spring the latent heat fluxes did not show significant anomalies (despite the lack of precipitation and reduced soil moisture). The sensible heat fluxes in 20CR are strongly enhanced during summer in the 1930s, and also slightly during spring and fall, although no unprecedented values are seen in the 1930s transition seasons. The small heat flux anomalies in spring suggest that the moisture available in the soil first may have to evaporate before land-surface feedbacks related to increased sensible heat fluxes and reduced latent heat fluxes become effective in summer.
Fig. 2

Seasonal time series of precipitation and soil moisture content averaged over the central US (see box in Fig. 1a). a GPCC interpolated observed precipitation totals, b 20CR simulated precipitation totals, c 20CR simulated mean soil moisture content. Thin lines are the time series of seasonal values, thick lines are the 11-point Gauss-filtered time series

Both observations and reanalysis show strong precipitation deficits during spring in 1934 and 1936 over large parts of the US with the strongest negative anomalies over the central US (Fig. 1c), while slightly wetter-than-normal conditions are found in the northwestern part of North America. In fact, 1934 was likely the worst drought year of the past millennium in parts of North America (Cook et al. 2014b). In general, in both observations and 20CR, the region of largest spring precipitation deficits compares well with the region of strongest summer heat anomalies; this points to a role of local feedbacks in generating the strong summer heat. The fact that 20CR correctly reproduces the location of the 1930s heat and drought shows the benefit of assimilating pressure observations into 20CR, while climate models forced with SST only often shift the center of the drought more to the south-west (Schubert et al. 2004; Cook et al. 2011b). This implies that we can use 20CR to investigate mechanisms and large-scale features of the 1930s spring drought and summer heat over the central US.

The 20CR is forced by observations of SSTs, sea ice and surface pressure, in addition to prescribed observed greenhouse gas concentrations and volcanic aerosols. The fact that all 20CR ensemble members robustly produce the strong heat and drought observed in the 1930s suggests that clear signals explaining the dry and hot anomalies may be found in the driving fields.

There is also a strong relationship between drought and summer heat in the central US and this agrees with previous observational findings for several regions of the globe (Mueller and Seneviratne 2012) and the US (Durre et al. 2000) in particular. The probability of hot summer temperatures is higher after dry than after wet spring conditions (Mueller and Seneviratne 2012). Based on climate model experiments to investigate land–atmosphere coupling, also Koster et al. (2006) identified the central Great Plains as a coupling hot spot. Figure 3 shows significant anti-correlations between the frequency of hot days during summer (JA) and the precipitation deficits during the preceding 4 months (MAMJ). The anti-correlation based on the gridded observations of heat extremes (GHCNDEX) and monthly precipitation totals (e.g., GPCC) is weaker than in the 20CR but remains significant (p ≤ 0.001). There is also a strong anti-correlation between precipitation during summer and summer heat (i.e., hot summers are usually accompanied by anomalous dry conditions). Several springs during the 1930s were among the driest in the century-long observations and reanalysis. This relationship between spring drought and summer temperatures suggests that explaining the strong spring precipitation deficits may help to understand the large-scale conditions that were leading to the extreme summer heat.
Fig. 3

Scatter plots and total least squares regression between seasonal precipitation totals in spring (a, c), summer (b, d), and hot day frequency in summer over the central US. Precipitation totals and hot day frequency were calculated based on observations (a, b; GPCC and GHCNDEX) and 20CR (c, d), averaged over the boxed area in Fig. 1a. Spearman’s rank correlation (R) and the related p value are also shown

We study the anomalies in MSLP and SSTs, the two fields most closely related to the observations used to drive the 20CR (Fig. 4). Anomalies are investigated for the extended spring seasons (March to June) preceding the summers with the largest numbers of hot days in the 1930s as well as for the most recent heat events in 2011 and 2012. Average large-scale atmospheric flow conditions during spring are dominated by high pressure systems over the subtropical Atlantic and Pacific Oceans. Pressure fields in 1934 and 1936 exhibit positive anomalies over the North Atlantic and over the western part of North America, while negative anomalies are found over the central North Pacific and Hudson Bay. These pressure anomalies mark an exceptional northward extension of the Atlantic subtropical high into the mid-latitudes and an eastward extension of the Pacific subtropical high. During the most recent heat events, pressure was also anomalously high over the North Atlantic, but was lower than normal over the western part of the continent and anomalously high over the central North Pacific.
Fig. 4

MSLP (20CR, left) and SST (HadISST, right) anomalies in 1934, 1936, 2011 and 2012 extended spring (MAMJ). For MSLP the long-term average (1901–2010) is also shown by contour lines

In the hottest Dust Bowl years, the most obvious common concurrent patterns in SSTs are warm anomalies in the western North Atlantic and the northeast Pacific, along with cold anomalies in the central North Pacific. While 2011 and 2012 also showed strong warm anomalies in the North Atlantic, SST anomalies in the Pacific had largely reversed patterns compared to the 1930s.

A large part of the variability in spring rainfall is explained by variations in large-scale atmospheric flow, in particular in the northward flow from the Gulf of Mexico into North America (Fig. 5). The prevailing winds over the US Gulf Coast are typically southerly in spring (Fig. 5a), transporting moisture into North America. Years with particularly dry springs and hot summers have northerly wind anomalies during spring, weakening the moisture transport into the continent (Fig. 5b–d). 1934 and 1936, in particular, had intensified southwesterly flow displaced northeastward along the Atlantic margin, related to the positive pressure anomalies east of Newfoundland. In addition, large-scale flow during spring in 1934 and 1936 was also characterized by enhanced northwesterly flow into the central US, transporting continental (i.e., relatively dry) air into the region. This flow anomaly was related to the enhanced near-surface pressure over the western US and decreased pressure over Hudson Bay. Note that southerly flow was not weakened in 2011 and 2012 (Fig. 5e, f). This is likely related to the finding that 20CR does not simulate strong precipitation deficits in these years over the central US averaging area (although gridded observations show deficits). In particular in 2012 the drought and heat in 20CR are more concentrated on the Southern States and do not extend as far north as in the observations (Fig. 1c). Note that the correlation between southerly wind and rainfall is higher in 20CR than e.g. in the NCEP1 reanalysis (Kistler et al. 2001) for the common data period 1948–2012 (Supplementary Figure S4).
Fig. 5

Winds at 850 hPa in the 20CR dataset during spring (March–June): a averages over the period 1901–2010; b Scatter plot and total least squares regression between the meridional wind component through a line 95W–85W, 30N (see black line in a) and spring precipitation over the central US (boxed area in Fig. 1a); c, f Anomalies from the long-term average (1901–2010) in 1934 (c), 1936 (d), 2011 (e), and 2012 (f), respectively

Pressure variations over North America, the Atlantic, and the Pacific are related to both precipitation over the central US and southerly wind from the Gulf of Mexico into North America (Fig. 6a, b). Over Western North America and the Labrador Sea, where pressure was above normal in 1934 and 1936, sea level pressure is significantly anti-correlated with both precipitation over the central US and meridional flow from the Gulf of Mexico.
Fig. 6

Correlation maps of total precipitation over the central US during spring (March–June, boxed area in Fig. 1a) and meridional flow at 850 hPa from the Gulf of Mexico into North America (March–June, winds crossing the black line in Fig. 5a) with concurrent large-scale driving fields. a precipitation and MSLP, b 850 hPa meridional winds and MSLP, c precipitation and SST, d 850 hPa meridional winds and SST. Correlations are calculated over 1901–2012. Stippling indicates significant correlations (p ≤ 0.05). Precipitation, wind and MSLP from 20CR, SST from HadISST1. All fields were detrended prior to the calculation of correlations. The black boxes indicate the areas for which Pacific and Atlantic area-average SST anomalies are used in the composite analysis of relationships between ocean surface temperatures and dry and hot anomalies

Sea surface temperatures also show distinct correlation patterns with precipitation over the US and large-scale meridional flow into the continent that match the specific anomalies during the 1930s (Fig. 6c, d). Similar correlation patterns are found for both precipitation and southerly winds out of the Gulf of Mexico, although there is a stronger correlation of SSTs with atmospheric flow compared to precipitation. In particular, both seasonal precipitation over the central US and southerly flow into the US Gulf Coast are significantly negatively correlated with SSTs in the Labrador Sea and northeast Pacific, indicating that precipitation is reduced when the North Atlantic and Northeast Pacific are warm.

Strong positive correlations with central North Pacific SSTs (Fig. 6c, d) indicate that precipitation over the central US is reduced when the Pacific is cooler in this area—as was the case for 1934 and 1936. Although there are also positive correlations with West Atlantic SSTs and negative correlations with SSTs in the tropical Atlantic, SST anomalies in those regions were small during the hot years of the 1930s.

Anomalies observed for the extreme hot years of 1934 and 1936 (see Fig. 4) are consistent with these long-term relationships between large-scale pressure variations and both precipitation over the central US and southerly winds from the Gulf of Mexico into North America. The correlation maps (Fig. 6) suggest that spring precipitation deficits over the central US are generally associated with enhanced pressure over western North America and the Northwest Atlantic, and decreased pressure over the subtropical West-Atlantic and eastern Pacific Oceans. Hence, the long-term correlation patterns suggest an eastward extension of the Pacific Subtropical High and a northward extension of the Atlantic Subtropical High being related to dry springs in the central US. As shown for 1934 and 1936, those pressure patterns led to the weakened transport of moist air from the Gulf of Mexico combined with enhanced northwesterly flow moving dry air into the Dust Bowl region of North America. The correlation maps also show that warm anomalies in the Northeast Pacific and North Atlantic (regions of negative correlations in Fig. 6c, d) are often related to weakened southerlies at the Gulf coast and hence reduced precipitation over the continent. Positive correlations in the central Pacific and Gulf of Mexico indicate that these regions are often cooler than normal when dry spring seasons occur over the central US. For the Pacific, the correlation patterns show similarity to SST anomaly patterns related to the Pacific Decadal Oscillation (PDO).

While there was no strong El Niño-Southern Oscillation activity in the mid-1930s that might have contributed to drought in North America, the Dust Bowl occurred at the beginning of a long period of warm anomalies in the North Atlantic Basin that lasted from 1930 to around 1960. The Atlantic Ocean has been shown to be an important driver of multi-decadal variations in summertime climate over North America (Sutton and Hodson 2005). Precipitation variability over the Great Plains in particular shows some link to the Atlantic Multi-decadal Oscillation (Nigam et al. 2011) and this relationship seems to be strongest during spring and summer (Cook et al. 2014a). However, our study shows that a few individual years had particularly hot summer conditions, whereas other years during the 1930s were closer to average. This suggests that anomalous atmospheric flow conditions were also necessary to generate the extraordinarily hot summers of the Dust Bowl years.

Several studies have discussed the combined effects of Pacific and Atlantic SSTs in shaping drought in North America (e.g. McCabe et al. 2004; Schubert et al. 2009; Seager and Hoerling 2014), although mostly focusing on tropical or basin-wide SST anomalies. We have identified strong correlations of spring precipitation deficits particularly with extra-tropical Pacific and Atlantic SST anomalies, close to the North American continent [note that Fig. 6 also indicates correlations with tropical SSTs, however anomalies in extra-tropical waters were stronger in the 1930s (see Fig. 4)]. Therefore, we explore the effects that Northeast Pacific and North Atlantic SST anomalies have in association with drought and heat over the continent, separately and in combination. We identify the 25 warmest spring seasons (MAMJ) in the Northeast Pacific (35N–55N, 135W–125W) and North Atlantic (37.5N–52.5N, 57.5W–42.5W, see black boxes in Fig. 6c, d) and plot composites of spring (MAMJ) precipitation, wind and MSLP fields and hot day frequency during summer (JA), as differences of each set of 25 warmest years from all other years in the record 1901–2012 (Fig. 7, left and middle column). We also consider the combined effect of Northeast Pacific and North Atlantic SSTs by plotting the composites for those years that are among the warmest 25 in both SST regions. The warm North Atlantic composite contains all four heat years (1934, 1936, 2011, 2012) discussed in this study, and the warm Northeast Pacific and combined composites include 1934 and 1936 but not 2011 and 2012.
Fig. 7

Composites of a MSLP (during MAMJ), b 850 hPa winds, c precipitation totals (during MAMJ), and d frequency of hot days (during JA) for the 25 warmest SST spring seasons (MAMJ) in the Northeast Pacific (left) and North Atlantic (middle). The right column shows composites for years when SSTs (during MAMJ) are among the 25 warmest in both North Atlantic and Northeast Pacific. Composites are calculated from the respective 20CR fields. Stippling indicates where differences are significant (Student’s t test, p ≤ 0.05)

The composites clearly show how spring seasons over the central Great Plains are drier than normal when the Northeast Pacific is warm. There are also significantly more hot days during summer when the Northeast Pacific is warm during spring, pointing to potential seasonal-scale predictability of summer heat from spring SST anomalies. Similarly, large parts of the central US are dry in spring and hot in summer when the North Atlantic is warm during spring. Pressure field anomalies that steer the large-scale atmospheric flow are also related to SSTs, whereby the Pacific pressure anomalies (eastward extension of the subtropical Pacific high) are most significantly associated with warm Northeast Pacific SSTs and North Atlantic pressure fields (northward extension of the Atlantic subtropical high) are associated with warm North Atlantic surface waters.

In combination, warm Atlantic and Pacific SSTs amplify the dry and warm anomalies over the central and eastern US (Fig. 7, right column). The pressure field anomalies are similar to those observed for the hottest individual years (Fig. 4), showing the anticyclonic anomaly over western North America and the northward shift of the Atlantic subtropical high. In combination, these pressure anomalies reduce the flow of moist air into the continent, leading to reduced precipitation over the central and eastern US during spring. The subsequent summer seasons show strong exceedances in the frequency of hot days over much of North America, and most strongly over the central US/Great Plains region. Overall, the pressure anomalies for the warm Northeast Pacific composites are very similar to anomalies during warm PDO phases in general (Trenberth and Hurrell 1994). Respectively, the warm North Atlantic composites resemble the pressure anomaly patterns during warm Atlantic multidecadal oscillation (AMO) phases (e.g. Knight et al. 2006).

While the composites shown in Fig. 7 use seasonal precipitation totals and hot days from 20CR, the composite fields look qualitatively similar when using gridded observations (Supplementary Figure S5). In particular for spring precipitation we also find negative anomalies over the central US in the composites of warm Pacific and warm Atlantic SST regions, with amplified dry anomalies when both ocean regions are warm. The JA hot day anomalies from the observations have a similar pattern as in 20CR but the warm anomalies are weaker compared to 20CR. This is likely related to the finding that the relationship between spring precipitation deficits and summer hot day frequency was stronger in 20CR than in the observational data (compare Fig. 3).

4 Discussion

While previous studies documented that SST forcing (Schubert et al. 2004; McCabe et al. 2004) and land effects (Cook et al. 2009, 2011b), at least in part, drive drought conditions in North America, our study highlights the more complete relationships that likely existed among SSTs, atmospheric flow anomalies, and the strong precipitation deficits and subsequent heat of the Dust Bowl. In contrast to the discussion in Feng et al. (2011), our results clearly indicate that the Atlantic Subtropical High was extended to the mid-latitudes during the 1930s drought. As a consequence, southerly winds were shifted further northeast along the Atlantic margin, while the northward transport of warm, moist air into the continent was weakened. In addition, the eastward extension of the Pacific High over the western part of the continent led to enhanced transport of dry, cool air from the Northwest into the central US. Anti-cyclonic anomalies over western North America, as identified here for the hottest individual years, are also apparent in multi-year averages of lower troposphere geopotential height during the 1930s (Cook et al. 2011b), associated with an upper-tropospheric ridge. Atmospheric dust was shown to enhance the high-pressure anomaly over western North America in a climate model simulation (Cook et al. 2011b). Previous work has also suggested that this anti-cyclonic anomaly weakened a local summer wind system (the “Great Plains Low Level Jet”) that transports moisture from the Gulf of Mexico (Cook et al. 2011b; Brönnimann et al. 2009). Our results point to spring flow anomalies on a larger scale, related to shifts in the major pressure patterns over both the North Atlantic and Pacific oceans, associated with precipitation deficits during spring. It is worth noting that land degradation effects and atmospheric dust loading, while not explicitly used to drive the 20CR, might be reflected in the pressure observations used to drive the atmospheric reanalysis.

Schubert et al. (2004) highlighted the role of tropical SSTs in shaping the Dust Bowl drought conditions. We also find that the spring precipitation deficits show correlation with tropical surface waters. However, we focus on extra-tropical SSTs because this is where anomalies are largest in the hottest Dust Bowl years 1934 and 1936. It is however possible that these extra-tropical SST anomalies have tropical origins. Further, it should be noted that these SST anomalies do not necessarily drive the circulation anomalies described, but rather the pressure and circulation patterns may also affect the SST anomalies (Johnstone and Mantua 2014). Also note that our study focuses on SSTs in individual spring seasons of the hottest years, whereas Schubert et al. (2004) present multi-year averages when investigating the drivers of drought conditions.

Schubert et al. (2009) investigate influences of Pacific and Atlantic SSTs on drought in the US and conclude that, based on simulations with atmospheric general circulation models (AGCMs), Pacific SST anomalies appear to play a dominant role, whereas the response to Atlantic patterns is less robust. Our results, using observation-based datasets, highlight that both Atlantic and Pacific SSTs have significant correlations with drought and summer heat over the central US. In particular the superposition of warm anomalies in both ocean regions appears to be associated with amplified dry and hot anomalies. This is in line with McCabe et al. (2004) who, also based on observational data, report a combined relationship of Pacific and Atlantic SSTs with drought in North America.

While SST anomalies do explain some of the variability in atmospheric flow patterns, much of the variance in pressure fields may also be related to other processes (Cook et al. 2011a). Atmospheric noise may also play a role in setting up the strong flow anomalies related to the extreme dry and hot conditions (Hoerling et al. 2009). However, our results suggest that the severity of heat and drought in the hottest Dust Bowl years was associated with a combination of Pacific and Atlantic ocean temperature forcings, land surface feedbacks and atmospheric flow anomalies. Summer heat is of great importance not only in terms of impacts, but also due to its potential contribution in shaping drought conditions. High temperatures cause vapor-pressure deficits that are a major component of drought itself and contribute significantly to vegetation stress and tree mortality (Williams et al. 2013). As a consequence, dying plants lead to de-stabilization of soil and induce dust feedbacks (Cook et al. 2009, 2011b), which in turn may amplify drought.

Some of the key features identified in association with the 1930s heat extremes were also observed in the most recent hot summer of 2012 (Hoerling et al. 2013): the spring of 2012 was characterized by drier-than-normal conditions, and the North Atlantic showed basin-wide warm anomalies. However, the Pacific Ocean had reversed anomaly patterns compared to the 1930s, suggesting that there was no Pacific contribution to the large-scale flow anomalies that favor spring drought and summer heat in the central US. Further, Kumar et al. (2013) demonstrated that the 2012 event did not require extreme SST forcing and might have arisen from atmospheric noise alone, consistent with Wang et al. (2014) who concluded that the contribution from SST forcing was rather small in 2012. The large-scale flow conditions in 2012 were different compared to the 1930s hot years, suggesting that this event was of a different nature. For example, 2012 was characterized by strong positive temperature anomalies already in spring, pointing to a possibly larger contribution of evaporation effects in shaping this particular hot drought event.

Our study sheds new light on a combination of mechanisms inducing the hottest summers of the past century and shows how extremely hot and dry summers may be linked to spring precipitation deficits that in turn are strongly related to large-scale flow and SST conditions. Enhancing our understanding of how the different factors combine may lead to better predictions of future droughts and heat waves, improving on dynamical predictions based on SST initialization only (Quan et al. 2012; Kumar et al. 2013). Different extreme heat and drought events are likely to develop from different large-scale contributions. Therefore, we would like to emphasize that the strongest predictability probably comes from the combined consideration of all three factors discussed throughout this manuscript, that is, the combination of Atlantic and Pacific SSTs, large-scale atmospheric flow, and land-surface feedbacks.

We note that some of our results regarding the relationship of drought and heat with SST anomalies lead to somewhat different conclusions from previous studies that used AGCMs forced with prescribed SSTs (e.g. Schubert et al. 2004, 2009; Wang et al. 2014). 20CR is also forced by prescribed SSTs but in addition assimilates pressure observations, which ensures that the atmospheric circulation more closely resembles observed characteristics. While AGCM experiments allow testing the sensitivity to specific SST changes, for example, isolated by ocean regions, it is likely that these simulations are affected by biases in the atmospheric circulation that are typical for atmospheric GCMs [such as bias in atmospheric mean flow and underestimation of blocking (Scaife et al. 2010) and in latitude of storm tracks (Kidston and Gerber 2010)]. Therefore, it is possible that GCMs do not correctly reproduce the observed teleconnections between ocean temperature anomalies and climate events over land (e.g. King et al. 2014), and their representation of large-scale atmospheric circulation features and relevant teleconnections should be evaluated when studying the impact of SST forcing on remote climate events. Versions of the GFS model, albeit without assimilation of surface pressure observations, showed a relatively strong connection between Pacific SSTs and US surface temperatures compared to other models (Schubert et al. 2009), and a relatively weak land–atmosphere coupling strength (Koster et al. 2006). However, assimilation of pressure observations in the GFS model used for 20CR is likely constraining the atmospheric teleconnections of the model towards observed teleconnections.

5 Summary and conclusions

Unprecedented hot summers occurred over the central US in the 1930s. We use novel observational datasets and atmospheric reanalysis to investigate this period of extreme heat in the context of the last century. We find that the extraordinarily hot summers of the 1930s were preceded by anomalously dry spring seasons. The spring precipitation deficits in turn were strongly related to large-scale atmospheric flow anomalies that weakened the transport of moist air from the Gulf of Mexico into the Great Plains region. Our results show that these precipitation and flow anomalies were associated with warm SSTs in the Northeast Pacific and North Atlantic.

Composite analysis over the 1901–2012 period shows that warm spring SSTs in both regions favor occurrence of spring precipitation deficits and warm summers over the central US, suggesting predictability of summer heat from spring-time ocean temperatures and large-scale atmospheric flow. On average, the spring precipitation deficits and summer heat anomalies are strongest when both ocean regions are warm in spring—as was the case in the hottest years of the 1930s, but not in 2011 and 2012 when the most recent severe heat events occurred. These results suggest that the coincidence of warm North-Atlantic and Northeast-Pacific surface waters and the related flow anomalies, as in the 1930s, might generate hot drought conditions more severe than experienced during the most recent heat waves that occurred in the early 21st century. In a warming world, today already 0.6 °C warmer on average than in the 1930s, such conditions might have the potential to break those heat records that still remain from 80 years ago.


This study was supported by the Australian Research Council grants CE110001028, LP100200690 and DE150100456, the U.S. National Science Foundation Grant AGS1243125 (JTO), and a Victorian Centre for Climate Change Adaptation Research Fellowship (JTO). We are grateful to Gil Compo for providing daily data from all 20CR ensemble members to calculate the extremes indices. Support for the 20CR Project is provided by the U.S. Department of Energy, Office of Science Innovative and Novel Computational Impact on Theory and Experiment (DOE INCITE) program, and Office of Biological and Environmental Research (BER), and by the National Oceanic and Atmospheric Administration Climate Program Office. We also express our thanks to two anonymous reviewers for their constructive comments that helped to improve the manuscript.

Supplementary material

382_2015_2590_MOESM1_ESM.pdf (491 kb)
Supplementary material 1 (PDF 491 kb)

Copyright information

© Springer-Verlag Berlin Heidelberg 2015

Authors and Affiliations

  • Markus G. Donat
    • 1
  • Andrew D. King
    • 1
    • 4
  • Jonathan T. Overpeck
    • 2
  • Lisa V. Alexander
    • 1
  • Imke Durre
    • 3
  • David J. Karoly
    • 4
  1. 1.ARC Centre of Excellence for Climate System Science and Climate Change Research CentreUniversity of New South WalesSydneyAustralia
  2. 2.Department of Geosciences and Department of Atmospheric Sciences, Institute of the EnvironmentUniversity of ArizonaTucsonUSA
  3. 3.NOAA’s National Climatic Data CenterAshevilleUSA
  4. 4.ARC Centre of Excellence for Climate System Science, School of Earth SciencesUniversity of MelbourneParkvilleAustralia

Personalised recommendations