Drier spring over the US Southwest as an important precursor of summer droughts over the US Great Plains

This study addresses the role of atmospheric moisture budget in determining the onset and development of the summer droughts over the North American Great Plains (GP) using two state-of-the-art reanalysis datasets. We identified zonal moisture advection as the major cause of 15 the severe tropospheric drying during the extreme droughts of southern GP 2011 and northern GP 2012. For both events, an eastward advection of anomalously dry and warm air in the free troposphere in spring sets the stage for the summer drought, leading to a sharp drop of relative humidity above the boundary layer, enhancing dry entrainment, and suppressing deep convection. Further breakdown of the zonal advection into the dynamic (caused by circulation 20 anomalies) versus thermodynamic (caused by moisture anomalies) contributions revealed dominance of the thermodynamic advection in the tropospheric drying observed during the onset of both 2011 and 2012 droughts. The dependence of thermodynamic advection on moisture gradient links the spring precipitation in the Rockies and southwestern US, the source region of the anomalous dry advection, to the GP summer precipitation (with correlations > 0.4 using 25 gauged-based data). The results of this study improve our predictive understanding of summer drought onset mechanisms over the GP.


Introduction
The United States (US) Great Plains (GP) is prone to devastating droughts, including the infamous "Dust Bowl" of the 1930s (e.g.Brönnimann et al. 2009;Donat et al. 2016), the extended drought in the 1950s (e.g.Cook et al., 2011), the Texas drought of 2011 (e.g.Fernando et al. 2016), and the record-breaking drought of 2012 (e.g.Hoerling et al., 2013).The projections 35 of the Global Climate Models (GCMs) that participated in the Coupled Model Intercomparison Project Phase 5 (CMIP5) show a robust intensification of dry conditions over the GP under different global warming scenarios in the coming decades (Cook et al., 2015;Teng et al., 2016), which would damage the agricultural and food industries over the region.The current dynamic prediction models have virtually zero prediction skill over the GP in summer (Quan et al., 2012;40 Hoerling et al., 2014).Improvement of our predictive understanding of drought onset and evolution mechanisms would provide scientific foundation for a more accurate and timely prediction of droughts over the region.
The GP drought and its underlying mechanisms have been studied extensively.Numerous studies have shown that, in the early stages of the GP droughts, the upper-level atmosphere has 45 featured an anomalous high and anticyclonic vorticity over central North America (Chang and Wallace, 1987;Namias, 1991;Lyon and Dole, 1995;Cook et al., 2011;Donat et al., 2016;Fernando et al., 2016).A dynamical teleconnection between the height anomalies over the US and the North Pacific SST anomalies has been considered as the main driver responsible for the onset of GP summer droughts in 1980 and 1988 (Trenberth, et al., 1988;Lyon and Dole, 1995;50 Chen and Newman, 1998).Variability of Pacific and Atlantic SSTs has been considered an important driver of droughts in North America with warm SST anomalies in tropical Atlantic and cold SST anomalies in tropical and eastern North Pacific favoring summer droughts over the GP (Namias, 1991;McCabe et al., 2004;Schubert et al., 2004;Kushnir et al., 2010;Wang et al., 2010;Feng et al., 2011;Zhao et al., 2017).However, the role of SST as a main driver of GP 55 precipitation variability has been challenged by numerous studies arguing that the atmospheric internal variability and land-atmosphere feedbacks are the dominant drivers of the GP summer drought for both short (Hoerling et al., 2013;Wang et al., 2014;Fernando et al., 2016;Pu et al., 2016) and long-term (Schubert et al., 2004;Ferguson et al., 2010) time scales.Despite the extensive research, it has remained unclear whether the SST anomalies in winter and spring can 60 significantly influence summer GP droughts.If so, what are the underlying physical mechanisms?This question is central in determining whether the GP summer drought is predictable.As a first step, we need to understand the main cause of the moisture deficit that initiate the summer droughts in the GP.To our knowledge, a systematic moisture budget analysis to determine such causes previously has not been reported.65 Moisture budget analysis has been attempted previously to understand local and largescale sources of moisture (Rasmusson, 1968;Yanai et al., 1973).The climatology, seasonal and diurnal cycles of moisture budget terms have been analyzed in the work of (Rasmusson, 1968) over the US and in studies of (Hao, 1987;Zangvil et al., 1993Zangvil et al., , 2001;;Schubert et al., 1998;Lamb et al., 2012) over the US GP.For the Southern GP (SGP), Lamb et al. (2012) calculated the 70 vertically integrated Moisture Flux Convergence (MFC) terms using the North American Regional Reanalysis in May-June for four selected years (1998, 2002, 2006, and 2007) and identified the horizontal advection and divergence terms, respectively, responsible for the moisture transport to and from the SGP.For the 1980 drought, Hao (1987)  While these studies provide very useful information about the atmospheric moisture sinks and sources over the GP, they only focused on the warm-season vertically integrated budget terms in a few selected years/periods in their analysis.Investigating the vertical structure of the individual moisture tendencies, their seasonal evolution and year-to-year variability, and the relative importance of the moisture transport and evapotranspiration (ET) anomalies on precipitation 85 variability are important questions that need to be addressed yet to understand the processes and feedbacks underlying the GP droughts specially during the onset season (March, April, May).
In this paper, we provide a detailed examination of the atmospheric moisture budget terms using two state-of-the-art reanalysis datasets over the entire period of 1980-2018 (see section 2 for details).Our diagnostic analyses present a comprehensive picture of GP tropospheric 90 moisture sinks/sources by investigating the diurnal cycle and the vertical structure of moisture budget terms and their temporal evolution before and during extreme droughts.A unique contribution of our study is the determination of physical processes that control the variability of moisture tendency over the GP.This was achieved by separating the moisture transport anomalies into their thermodynamic and dynamic contributions, identifying the regional and 95 remote drivers that modulate variability of these contributions, and measuring the relative importance of the individual terms in the onset and development of GP droughts.In the rest of the paper, we provide a detailed explanation of the implemented methods in section 2, present the results and discussion in section 3, and provide a summary of the results and our main conclusions in section 4. 100 = ), Eq. 1 presents the water vapor budget equation for a unit mass of air (Yanai et al., 1973): 105 where t, q, and p stand for time, specific humidity, and pressure respectively; v and  are the horizontal wind vector and vertical wind velocity in pressure coordinate, and e and c are the evaporation and condensation rates of the air parcel per unit mass, respectively.

Methodology
Assuming a negligible contribution from the moisture tendency term (the 1st term in Eq. 1) for 110 monthly and longer time averages (Trenberth and Guillemot, 1995), vertical integration of Eq. 1 from Pt=0 to Ps results in: where P and E are precipitation and evapotranspiration rates at the surface and g and  A stand for gravitation acceleration of the Earth and water density.Decomposing an arbitrary variable A 115 to a stationary ( F ) and a transient term ( ) ( =  F +  ), and applying the covariance equation ( H =  I J + ̇K ), we can write Eq. 2 as the following: where u and v are the zonal and meridional components of the horizontal wind, v.The first, 120 second, and third terms on the left hand side (LHS) represent the zonal, meridional, and vertical advection, respectively and the last term in the LHS refers to the eddy transient terms of the zonal, meridional, and vertical winds.In our analysis, the transient and stationary terms refer to the monthly mean and six-hourly departure from the monthly mean (see sections 2.3 and 2.4 for more information on the temporal and spatial resolution of the input data and numerical 125 calculations).

Thermodynamic versus dynamic contribution
Breaking up each term in Eq. 3 to a climatological mean and a monthly departure from climatology (e.g. F =  ̅ +  S ), Eq. 3 can be organized as the following (Chou and Lan, 2012;Li et al., 2016;Peng and Zhou, 2017): 130 where, the anomalous precipitation is balanced by the anomalous advection, evaporation, and residual  which accounts for the sub-monthly transient eddy contribution.The transient advection terms in Eq. 4 can be further separated as 135 The first term in the right hand side (RHS) of Eq. 5 is referred to as the thermodynamic contribution of the zonal advection which accounts for the changes in humidity while setting the circulation to climatological wind.The second term in the RHS is referred to as the dynamic contribution which accounts for the changes in wind given the climatological humidity and the 140 third term in RHS is the non-linear term which accounts for the interannual anomalies of both wind and humidity (Seager et al., 2010;Chou and Lan, 2012;Li et al., 2016;Peng and Zhou, 2017).Separating all the advection terms into thermodynamic and dynamic contributions, Eq. 4 can be rewritten as the following:

Data
The moisture budget analysis in this study is based on the European Centre for Medium-Range Weather Forecasts (ECMWF) Interim Re-Analysis (ERA-Interim) (Dee et al., 2011) which covers 6-hourly upper air parameters from 1979 to near-real-time.The atmospheric model has 60 150 levels in a hybrid sigma-pressure vertical coordinate system and a T255 spectral horizontal resolution (~79 km).The data is available online (http://apps.ecmwf.int/datasets/data/interimfull-daily/levtype=sfc/).In addition to ERA-Interim, we also used the Modern-Era Retrospective Analysis for Research and Applications-version 2 (MERRA-2) (Gelaro et al., 2017) and repeated the moisture budget analysis to ensure that our conclusions were not sensitive to the choice of the 155 reanalysis product.MERRA-2 is the latest atmospheric reanalysis of the National Aeronautics and Space Administration (NASA) Global Modeling and Assimilation Office (GMAO) covering the 1980 to near-present time period and is available online at the NASA GMAO website (https://gmao.gsfc.nasa.gov/reanalysis/MERRA-2/). The atmospheric model uses a cubed-sphere horizontal grid with a 0.5°x0.625°resolution and a hybrid-eta vertical coordinate system with 72 160 model levels from the surface to 0.01 mb.
For observed precipitation we used the National

Computation
The moisture budget terms in Eq. 3 were calculated using 6-hourly ERA-Interim reanalysis on a regular 0.75° grid and 14 selected pressure-levels.The horizontal and vertical gradients were 170 calculated using a centered finite difference approach.The vertical integrals were performed by integrating the product of the moisture tendencies in each layer multiplied by the pressure thickness of each layer (dP) from surface to 50-mb level.The calculations were done at 14 pressure-levels (spanning 1000 mb to 50 mb) where the lowest 6 levels (from 1000 mb to 850 mb), which contain most of the atmospheric moisture, had a 25-mb resolution and the thickness 175 of the remaining levels grew to 50 mb and 100 mb for the mid and upper troposphere.The vertically integrated moisture tendencies were divided by  A and multiplied by a scale factor (24*3600*10 -3 ) to convert m/s to mm/day.To determine the impacts of daytime and nocturnal anomalous circulation, we have separately computed daytime and nighttime composites.The daytime composites for the North American domain were calculated by averaging the reanalysis 180 outputs at 1800 and 0000 Universal Time Coordinate (UTC), and the nighttime composites were obtained by averaging the reanalysis outputs at 0600 and 1200 UTC.More information on the benefits and limitations of the diagnostic computation of the atmospheric moisture budget with reanalysis is provided by Kevin E. Trenberth and Guillemot (1995) and Seager and Henderson (2013), including the impacts of several sources of errors including temporal, horizontal, and 185 vertical resolution, numerical calculation of gradients and vertical integration, and reanalysis initialization.

Significance of correlation coefficients
There are 39 annual samples during our analysis period of 1979-2018.Accounting for the effective sample size by using the (Livezey et al., 1983) method for a lag-1 auto-correlation of 190 0.2 for two time series (r1=r2=0.2and r1r2=0.04)(which is a conservative estimate for the annual time series of standardized anomalies of P, q, and zonal moisture advection) results in significance levels of 7.1% and 1.4% for the correlation coefficients of 0.3 and 0.4, respectively using a two-tailed Student-t distribution with N-2 degrees of freedom.

Results 195
The Great Plains Summer Drought The US GP, located east of the Rocky mountains and west of the Mississippi River, are characterized by a semi-arid climate with a land surface covered primarily by farmlands and temperate grasslands.On average, the region has an annual precipitation of 1-2 mm/d, approximately half of which occurs during the boreal summer (Figure 1b).The climatology of 200 observed summer precipitation during 1979-2018 features a zonally asymmetric pattern with JJA precipitation less than 1 mm/d over the Rockies and US southwest, between 1 and 3 mm/d over the central plains, and greater than 3 mm/d over the US Midwest and eastern US (Figure 1a).The GP have been subject to recurrent severe drought and heat waves with two extreme droughts in 2011 and 2012 occurring during the most recent decade (Figure 1e and also Cook et al., 2011;205 Hoerling et al., 2013;Fernando et al. 2016).Summer droughts over the region usually develop in previous spring and peak in mid-to late summer.As indicated by the maps of JJA standardized precipitation anomalies (Figures 1c, 1d and ET in May (Figures 2c and 2d).The negative precipitation and ET anomalies extended through the 2012 summer and early fall with a gradual recovery of drought conditions closer to winter.As shown in Figure 2, the amplitude of the anomalies and their temporal evolution is consistent between the two reanalysis datasets.
The atmospheric profiles of specific humidity (qv) and cloud liquid and ice water content during 220 the 2011 SGP and 2012 NGP droughts are compared against the 1979-2018 climatology in Figure 3 and 4 (see Figure S1 for relative humidity (RH) and Fraction of Cloud Cover (FCC)).
The qv climatology for both the SGP and NGP indicates that the largest annual values occurred during summer where the maximum humidity (larger than 10 g/kg) was confined to the lower troposphere (1000-800 mb) and gradually decreased to ~5 and ~3 g/kg in the mid-and upper-225 troposphere (Figures 3a and 4a).The climatological value of specific humidity at all levels start decreasing in fall with the lowest annual rates (<3 g/kg) during winter.The annual average values of the specific cloud ice and water content peak in fall and spring, and reach the minimum values during summer over both the SGP and NGP, respectively (Figures 3d and 4d).Over the NGP, annual minimum values also occur in winter (Figure 4d).The spring-time peak of specific 230 cloud liquid and ice water is much greater than the peak values in fall with the largest values (>10 g/kg) confined to the low-and mid-troposphere (850-500 mb) in the NGP and the lowerfree troposphere (850-650 mb) in the SGP.
The specific humidity and cloud liquid and ice water in both dry years were generally much smaller than their climatological values.However, the two variables reveal distinct temporal 235 evolutions and vertical structures.As revealed by Figures 3c and 4c, large negative anomalies of qv extending from the surface to the mid-troposphere persisted year-around for the SGP 2011 event.The maximum dry anomalies of qv were located in the near surface levels and peaked during the May-June and August-September periods, indicating intensive drying of the boundary layer air during the drought peak.The dry anomalies in summer was preceded by an extended 240 drier lower troposphere in the spring season.For the 2012 NGP event, however, the spring-time anomalies of qv remained reasonably wet during March-April until the drought intensified rapidly in May.The qv anomalies remained negative during the entire summer and fall 2012 with the largest negative anomalies occurring near the surface in August and September.Figure 4f shows that the negative anomalies of cloud liquid and ice water content over the NGP developed 245 in winter, and persisted during the entire 2012.The negative anomalies of cloud liquid and ice water content started four months earlier than the negative qv anomalies, highlight the impact of warmer temperature during 2012 winter and spring, which reduced relative humidity (Figure S2c) and consequently, cloud liquid and ice water (Figure 4f) and fractional coverage (Figure S2f), as well as depleted soil moisture (Sun et al., 2015;Mo et al., 2016).250 For both 2012 NGP and 2011 SGP drought years, the cloud liquid and ice water content of dry years were much lower throughout the depth of the troposphere and over the course of the year with the largest decline (~40%) occurring in the lower-and mid-tropospheric levels in spring and early summer (Figures 3f and 4f).The drying of the low-and mid-troposphere was linked to a sharp drop of mid-and upper-troposphere RH in spring as shown in Figures S1c and 255 JJA for both calculations of MFC (Figures 5e and 5f).The imbalance is partially due to the (neglected) atmospheric moisture storage and in part due to the unclosed moisture budget in the reanalysis (Trenberth et al., 2011;Seager and Henderson, 2013).
To investigate the GP summer droughts from a moisture budget perspective, we looked at the individual moisture tendencies, their vertical structure, annual cycle, and diurnal variability for 290 both the 2011 and 2012 events compared with the 1979-2018 climatology in ERA-Interim.For the SGP (Figure 6), all climatological tendencies indicate strong seasonal variability with the Over the NGP, zonal advection is the dominant moisture source year-round (vertically integrated 315 values > 0.5 mm/d) with positive tendencies extending from above the PBL to 300mb (Figure 7a  and 7c).The climatology of meridional advection reveals negative tendencies year-round throughout the troposphere, except during summer in lower tropospheric levels where the moisture convergence is noticeably larger overnight highlighting the northerly moisture transport by the GP LLJ (Figures 7d and 7f).The vertical advection term is moderately positive during 320 April and May and strongly negative during the rest of the year (Figures 7g and 7i).The transient terms reveal a vertical structure similar to that of the SGP with the strong negative tendencies confined to the May-September period (Figures 7j and 7l).Similar to the 2011 SGP drought, the 2012 NGP drought onset is marked by strong advection of dry air in April and May concentrated in the lower free-troposphere, which lead to the large (<-1.5 mm/d) decline in vertically 325 integrated moisture tendencies during that period (Figures 7b and 7c).Besides the horizontal advection, all other terms during the 2012 spring indicate normal or greater than normal moisture tendencies characterizing the zonal advection term as the large-scale source of tropospheric drying during the 2012 drought onset.In the 2012 summer, both the vertical and meridional advection terms indicate large moisture divergence mostly due to considerable strengthening of 330 the dry tendencies in the mid-and upper-troposphere from July onward.
To identify potential drivers of the spring-time tropospheric drying shown in sections 3.1 and 3.2, we decomposed the zonal, meridional, and vertical advection anomalies into their thermodynamic, dynamic, and non-linear contributions (see section 2.2).The results for the zonal advection term are presented in Figures 8 and 9 for the SGP 2011 and NGP 2012 events, 335 respectively.For both events, the contribution of dynamic and nonlinear terms to the anomalies of zonal advection are considerably small (as compared to the thermodynamic term) during spring and early summer and nearly zero throughout the year in both ERA-Interim and MERRA2 reanalysis.During spring and early summer, the thermodynamic term reveals large negative moisture tendencies for both the SGP 2011 and NGP 2012 cases with the vertical structure of the 340 anomalous tendencies in the two reanalysis consistently agreeing with one another (Figures 8c, 8d, 9c, and 9d).Since the thermodynamic contribution is defined as the product of the climatological zonal wind (featuring large westerlies at 700mb) and the gradient of anomalous humidity, its variability is entirely controlled by the zonal gradient of qv anomalies.As a result, the strong advection of dry lower-and mid-tropospheric tendencies during the 2011 and 2012 345 drought onsets were almost entirely forced by the zonal gradient of specific humidity, or more simply, by a relatively drier troposphere in the US SW and Rockies located upwind of the SGP and NGP.

The relation between anomalous moisture advection and the spring and summer dry/wet conditions 350
The relationship between thermodynamic zonal moisture advection and anomalous dry/wet conditions were investigated using single point lag/lead correlation maps between the tendency term over the SGP and NGP and multiple atmospheric variables over the US (Figures 10 and  11).For both regions, the correlation between the MAM anomalies of the zonal thermodynamic advection and specific humidity at 700 mb features a dipole pattern with strong positive 355 (negative) correlations over the US west and southwest (east and northeast) highlighting the zonal gradient of humidity anomalies as the main driver of variability of the moisture term.At the surface, the correlation maps for MAM precipitation and ET indicate a similar pattern with significant positive correlations over the Rockies and US southwest and relatively weak negative correlations over the eastern US for both regions (the magnitude of positive correlations are 360 stronger for SGP than NGP; Figures 10c, 10e, 11c and 11e).The positive correlations indicate that the dry (wet) anomalies of ET and P over the upwind region are linked to the anomalous moisture divergence (convergence) over the SGP and NGP.
The spring-time variability of thermodynamic advection over the GP is linked to the summertime surface and atmospheric conditions over the US interior plains.The correlation maps of JJA 365 qv for both SGP and NGP, indicate positive correlations over the central US, east of the Rockies, and near zero correlations elsewhere over the US.The correlation between the MAM moisture tendency in the SGP and JJA ET are strongly positive over the Rockies and central plains (Figure 10d).A similar correlation pattern exists for the NGP tendency and JJA ET with the band of significant positive correlations extending from the eastern Rockies and central US to the US 370 Midwest and East (Figure 11d).Similar to ET maps, the correlations between the MAM moisture term over both the south and north GP and the JJA precipitation anomalies are strongly positive (> 0.45) over the US northern plains and Midwest and weakly positive over the southern plains and northwestern US (Figures 10f and 11f).Similar correlation patterns were reproduced using the CPC-gauged based precipitation as an independent observational data set in the lag/lead 375 correlations with the MAM moisture term anomalies in the SGP and NGP (see Figures S3 and  S4).
The strength and spatial patterns of the correlations between the moisture term and both MAM and JJA precipitation (shown in Figures 10 and 11) signals a potentially significant relationship between the MAM precipitation in the US SW and JJA precipitation in the GP.Using the CPC 380 precipitation, we calculated single-point correlations between the standardized anomalies of MAM precipitation in the US SW and the JJA precipitation at each grid cell (Figure 12a).The results indicate strong positive (> 0.3) correlations over the US west coast, Rockies, and northern GP, weak positive correlations over the US mid-west, near zero correlations over Arizona and SGP, and weak negative correlations over the US east and southeast.The contours of positive 385 correlations are especially strong over the NGP.The comparison of time series of JJA precipitation anomalies over the NGP against the MAM precipitation anomalies in the US SW (Figure 12b) indicates a strong coverability between the two time series during the 1979-2018 period with a correlation coefficient of 0.41 (significant at 1%).The correlation magnitude is surprisingly large as compared to the near zero correlation between the standardized anomalies 390 of MAM and JJA precipitation in the NGP.

Discussion
Our analyses of the variability and vertical structure of MFC components during the SGP 2011 and NGP 2012 extreme droughts identified severe lower-free tropospheric drying over the US SW, and the resultant dry zonal advection anomalies to the US GP in spring as the major drought 395 onset mechanism for both events.The influence of lower-tropospheric humidity on GP precipitation grows continually in spring as the GP precipitation regime begins to shift from a dominantly frontal precipitation regime in winter toward convective precipitation in summer.
Our results indicate that a drier lower free troposphere in the US GP, due to strong zonal advection of dry air in spring, is associated with a sharp drop of RH above the PBL which 400 increases dry entrainment and decreases the buoyancy of a rising moist plume.The increased dry entrainment would decrease precipitation during spring and early summer by limiting the convective penetration depth and shifting the convection structure from predominantly deep convective towers toward frequent shallow cumulus clouds (Derbyshire et al., 2004;Zhang et al., 2010;Del Genio, 2012).For the SGP 2011 and NGP 2012 events, the suppressed convection in 405 spring and early summer was supported by the severe decrease (~30-40%) of specific cloud liquid and iced water content above the PBL (Figures 3f and 4f) and the FCC in the upper troposphere (Figures S1f and S2f).The strong control of the free-tropospheric humidity on convective precipitation has already been demonstrated in both cloud-resolving model (CRM) simulations as well as observational studies (Derbyshire et al., 2004;Sherwood et al., 2010;410 Zhang et al., 2010;Zhuang et al., 2018).Meanwhile, the conventional convective parameterization schemes tend to severely underestimate the sensitivity of moist convection to environmental humidity largely due to underestimation of the turbulent entrainment of drier air into the rising convective cells (Derbyshire et al., 2004;Del Genio, 2012).This underestimation would lead to overestimation of deep convection in climate models implementing convection 415 schemes, and could be a potentially major source of uncertainty responsible for poor performance of the current dynamic models in predicting summer drought in the GP.
The temporal evolution of RH during the SPG 2011 and NGP 2012 droughts reveals a transition of the maximum dry anomalies of RH from the free-tropospheric levels in spring to the lower troposphere and boundary layer in summer.A positive land-atmosphere feedback could facilitate 420 this shift by perpetuating the initial dry land surface conditions in spring to the severe drying and warming in summer.In this mechanism, an anomalously lower precipitation and lower FCC would lead to a relatively drier surface and enhanced insolation in late spring.As a result, ET would decline steadily in the following months leading to a significant decrease in surface latent heat flux (estimated about 50 w.m-2for the 1988 summer by Lyon et al. 1995), which is largely 425 balanced by an increase in upward sensible heat flux and air temperature.The hotter-drier surface would intensify the decline of boundary layer and lower tropospheric humidity causing further decrease of precipitation in summer.This feedback mechanism was found to be responsible for intensification of several extreme cases of summer drought and heat waves over the US interior plains (Chang and Wallace, 1987;Hao, 1987;Namias, 1991;Lyon and Dole, 430 1995).The anomalous warming of the PBL in summer can also increase the difference between the surface temperature and dew point (T-Td) resulting in elevation of the level of free convection (LFC), increase of convective inhibition energy (CIN), and suppression of deep convection (Hao, 1987;Myoung et al., 2010).
The breakdown of total MFC into its zonal, meridional, and vertical advection terms in our 435 analysis shows the meridional and zonal advection terms to be the dominant sources of incoming moisture over the SGP and NGP, respectively.This is clear from the year-round strong positive tendencies of meridional advection over the SGP (confined to the lower troposphere; Figure 6d) and zonal advection in the free-tropospheric levels over the NGP (Figure 7a).While the role of meridional advection of moisture from the Gulf of Mexico to the US interior plains has received 440 extensive attention in the literature (Schubert et al., 1998;Weaver et al., 2008;Berg et al., 2015), the importance of zonal advection as a major moisture transport mechanism has been overlooked.In the case of the NGP 2012 drought, for example, the severe moisture divergence during the drought onset has been attributed to the dry anomalies of meridional moisture advection as a result of weakening of the GP LLJ (Hoerling et al., 2013 and2014).Our close 445 examination of the moisture budget terms, however, rejects this suggestion by revealing higher than normal moisture convergence for the meridional term during both 2011 and 2012 events and attributing the observed tropospheric drying for the two events to the zonal advection term.
Further breakdown of moisture advection anomalies into their dynamic and thermodynamic contributions suggests that the thermodynamic contribution was almost entirely responsible for 450 the extreme dry anomalies of zonal advection during the SGP 2011 and NGP 2012 droughts.By definition, the thermodynamic contribution is driven by the gradient of qv and the dominance of zonal thermodynamic advection in the onset of the 2011 and 2012 events signifies the The role of zonal thermodynamic advection in linking the dry/wet conditions over the GP and its upwind region is further supported by the lag/lead correlation analysis between the moisture term and multiple atmospheric and surface parameters in ERA-Interim reanalysis.Similar correlation 470 analysis applied to the CPC observed precipitation provided additional independent evidence indicating that the MAM precipitation anomalies in the US SW region lead the variability of JJA precipitation over the NGP (statistically significant at the 1% level).

Conclusions
We investigated the GP summer drought from a moisture budget perspective and looked at the 475 sub-daily, monthly, seasonal, and interannual variability of the moisture tendencies in two stateof-the-art reanalysis.For the two extreme droughts (the SGP 2011 and NGP 2012) in our study period, we found that a severe moisture divergence forced by the zonal advection tendency at the lower free troposphere (850 mb to 600 mb) dominated the anomalously dry MFC at the early stage of the droughts.The severe free-tropospheric drying resulted in a sharp drop of RH above 480 the boundary layer and an increase of dry entrainment which suppressed the deep convection during spring, setting the stage for extremely dry summers.The anomalies of moisture convergence were further decomposed into their dynamic and thermodynamic contributions with the former isolating the impact of humidity gradient and the latter isolating the impact of wind circulation.The results from ERA-Interim and MERRA2 consistently attributed the observed dry 485 anomalies of tropospheric moisture during the SGP 2011 and NGP 2012 drought onsets to the thermodynamic contribution of the zonal advection tendency.The thermodynamic advection tendency itself was strongly modulated by the spring-time conditions over the upstream region (the US west and southwest) and significantly linked to the JJA precipitation and ET over the US GP.The GP summer precipitation anomalies were found to be strongly correlated with MAM 490 precipitation anomalies in the US SW, suggesting the spring-time dry or wet anomalies over the US SW and the Rockies to be a precursor of the drier or wetter summer over the US GP.The results of this study provide a comprehensive picture of atmospheric moisture supply over the GP as well as the major drivers of strong moisture divergence during drought onset in the GP.The identified relationship between spring conditions over the US SW and GP summer 495 conditions can facilitate a better understanding of the hydrologic extremes over the GP and potentially improve the seasonal and sub-seasonal prediction skill.

Authors Contribution
AE designed the study and conducted the analysis in close collaboration with RF.Both authors equally contributed in writing the manuscript, reviewing the results, and editing the paper.500 compared the vertically integrated horizontal advection and divergence terms of 1980 and 1979 summers 75 (calculated from radiosonde data over the SGP) and indicated that the horizontal divergence was the dominant contributor to the extreme drying in the summer of 1980.Schubert et al., (1998) identified the GP low-level jets (LLJ) as the dominant contributor to the summer mean moisture influx to the US interior and indicated a strong link between sub-seasonal variability of moisture influx from the Gulf of Mexico and warm-season precipitation over the central and eastern US. 80 Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2019-154Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 8 April 2019 c Author(s) 2019.CC BY 4.0 License.
), the drought of 2011 was confined to the SGP, especially Texas, while the 2012 event covered nearly the entire US GP with the drought epicenter over the NGP.The temporal evolution of the 2011 drought shows a steady decline of 210 precipitation and ET started in February, extending throughout the spring and peaking during summer (-2 mm/d) in both MERRA2 and ERA-Interim reanalysis data (Figures 2a and 2b).The Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2019-154Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 8 April 2019 c Author(s) 2019.CC BY 4.0 License.dry anomalies started recovering in the fall and the drought finally ended in late fall/early winter.Similarly, the NGP 2012 drought developed (somewhat rapidly) in spring as noted by a sharp decline in precipitation in March followed by a normal April and a large drop in precipitation 215 Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2019-154Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 8 April 2019 c Author(s) 2019.CC BY 4.0 License.
S2c.A sharp decline of free tropospheric RH intensifies the entrainment of dry air into the rising moist air above the boundary layer limiting the convective penetration depth and shifting the convection structure from predominantly deep convection to frequent shallow cumulous clouds(Derbyshire et al., 2004;Zhang et al., 2010; Del Genio, 2012).The FCC difference fields during the spring and early summer of both 2011 over the SGP and 2012 over the NGP indicate large 260 negative anomalies extending from above the Planetary Boundary Layer (PBL) to the upper troposphere suggesting a strong suppression of deep convection during the onset season (FiguresS1f and S2f).Moisture budget analysisSummer in the GP is the warmest season of year with the highest rate of seasonal 265 evapotranspiration (ET).Despite its highest share of annual rain, JJA precipitation minus evapotranspiration (P-E) is negative with the maximum deficit (1-3 mm/d) over the GP and US Midwest.Such a P-E deficit is balanced by the atmospheric moisture flux convergence (MFC) over monthly and seasonal time scales (see section 2.1).Using ERA-Interim 6-hourly data over 1979-2018, we calculated the individual moisture tendencies in Eq. 3 and compared the sum of 270 vertically integrated terms with the ERA-Interim reported vertically averaged moisture convergence (-1*divergence) to evaluate the accuracy of our numerical calculations.The spatial patterns of the JJA climatology of the MFC are very similar between our calculated values and those reported by ERA-Interim, for example, over the inter tropical convergence zone (ITCZ) between the equator and 15°N and over the regions of sub-tropical anticyclones (Figure5aand 275 5b).Over land, the JJA climatology in the ERA-Interim MFC and that of numerically calculated MFC from the 6-hourly atmospheric fields indicate near zero differences over much of Alaska, western Canada, and central and eastern US, except for over the complex terrain of western US and north western Atlantic (Figure5d), where a relatively larger difference (between 0.5 and 1.5 mm/d) occurs.These differences originate from multiple sources including the vertical resolution 280 (14 pressure levels in our calculation vs the 60 model levels in ERA-Interim) and numerical calculation of the divergence and gradient terms (seeTrenberth et al., 2011 andSeager andHenderson, 2013 for more details).Overall, our numerically calculated MFC maintains a desirable accuracy in comparison to the ERA-Interim MFC.The difference fields between MFC and P-E reveal moisture budget imbalances as large as 1.5 mm/d over the US central plains in 285 Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2019-154Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 8 April 2019 c Author(s) 2019.CC BY 4.0 License.vertically integrated tendencies (blue line) revealing positive values (moisture convergence) as large as 1 mm/d for the zonal advection during summer, meridional advection year-round, vertical advection during spring, and horizontal transient term during winter (Figures 6c, 6f, 6i, 295 and 6l).The major sources (<-1 mm/d) of negative tendencies (moisture divergence) are the transient term in spring and summer and the vertical advection term in fall and winter.For the summer season in particular, the eddy transient term features strong moisture divergence extending from the surface to the upper troposphere (Figures 6j and 6l) whereas the meridional advection reveals strong positive tendencies that are confined to the lower troposphere and much 300 stronger during night featuring the moisture transport from the Gulf of Mexico through the GP LLJ (Figures 6d and 6f).The zonal advection also indicates moderate to strong positive tendencies during summer that are confined to the mid-and upper-troposphere (Figures 6a and 6c).The difference between the day-time and night-time moisture tendencies is quite large for the meridional advection term during spring and summer and the vertical advection term during 305 late summer and fall, and negligible year-around for the horizontal advection and transient terms.The annual cycle and the vertical structure of all moisture terms for the SGP in 2011 remained near or greater than the corresponding climatological values, with the exception of zonal advection.The zonal advection in 2011 indicates a major increase in dry tendencies (vertically integrated values <-3 mm/d) extended from the 900mb to the upper troposphere that persisted 310 from March to June (Figure 6b and 6c).Meanwhile, all other moisture transport sources in Figure 6 remained wetter than normal during the 2011 spring up until late summer, making the zonal advection of dry air solely responsible for the severe tropospheric drying during the drought onset, previously identified in Figure 3.
Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2019-154Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 8 April 2019 c Author(s) 2019.CC BY 4.0 License.importance of the west-east gradient of tropospheric moisture.The spatial patterns of MAM climatology of qv indicate a relatively large meridional gradient where qv decreases sharply 455moving northward from Mexico toward the GP and a smaller zonal gradient with higher qv values over the Rockies gradually decreasing in the eastward direction toward the NGP and US-Midwest (FigureS5).Despite a relatively larger magnitude of the meridional gradient of humidity, the zonal advection tendency becomes much larger (2 to 3 times over the SGP and 4 to 5 times over the NGP) than the meridional advection in the free tropospheric levels mainly due 460 to the large zonal (westerly) and the near zero meridional vectors of the horizontal wind over GP at those levels.However, since the zonal gradient of moisture at the free tropospheric levels is small, an anomalous dipole pattern (drier west-wetter east) or even a severe decline of qv over the Rockies can change the direction of the climatological west-east moisture gradient diverting the zonal thermodynamic advection tendency from its climatological values (strongly positive 465 over the NGP) to strong negative anomalies as large as those observed in the SGP 2011 and NGP 2012 MAM season.

Figure 5 .
Figure 5. JJA climatology (1979-2018) of the vertically integrated MFC (mm/d) calculated 535 diagnostically from the 6-hourly ERA-Interim output (a) and the monthly-mean MFC reported by ERA-Interim (b).JJA climatology of P-E (mm/d) has been also calculated from the ERA-Interim monthly outputs over the same period (c).The difference between the ERA-Interim reported and the calculated MFCs (d) represents the bias introduced by the numerical calculation of the budget terms in our analysis.The moisture budget imbalance is represented by subtracting 540 the P-E climatology from those of the calculated MFC (e) and the ERA-Interim reported MFC (f).

Figure 6 .
Figure 6.Hovmoller diagram of the vertical profile of the atmospheric moisture budget components averaged over the US Southern Great Plains for the 1979-2018 climatology (a, d, g, and j) and 2011 (b, e, h, and k).The 1 st to 4 th rows respectively represent the zonal advection, meridional advection, vertical advection, and horizontal transient terms in mm/day.The 3 rd 550 column (c, f, j, and l) represents the annual cycle of the corresponding terms (vertically integrated) for the climatology (blue) and 2011 (red) during the day-time (solid) and night-time (dashed) steps using 6-hourly ERA-Interim reanalysis.

Figure 8 .Figure 9 .
Figure 8.The Hovmoller diagram of the dynamic (a and b), thermodynamic (c and d), and nonlinear (e and f) contributions to the monthly anomalies of the zonal advection (mm/d) in ERA-560 Interim (a, c, and e) and MERRA-2 (b, d, and f) over the SGP in 2011.The monthly anomalies were calculated in respect to the 1979-2018 climatology for ERA-Interim and 1980-2018 climatology for MERRA2.

Figure 10 .
Figure 10.Single point correlation maps between the standardized time series (1979-2018) of the MAM zonal thermodynamic advection at 700 mb averaged over the SGP (the box in a) with the 570 standardized anomalies of ERA-Interim specific humidity at 700 mb (a and b), evapotranspiration (c and d), and precipitation (e and f) for the MAM (a, c, and e) and JJA (b, d, and e) seasons.The correlation coefficients greater than 0.3 and 0.4 are statistically significant at the 10% and 2% levels, respectively (see section 2.6).575

Figure 11 .
Figure 11.Same as Figure 10 but for the NGP.

Figure 12 .
Figure 12.Single point correlation maps between standardized anomalies of MAM precipitation 580 over the US SW region (30°-45° N and 105°-115° W) and JJA precipitation at each grid cell (a).The time series of standardized anomalies of JJA precipitation over the NGP (blue) and MAM precipitation in the US SW (yellow) are shown in panel b.The CPC gauged-based precipitation from 1979 to 2018 was used to derive both the correlation map and time series.585