Concurrent increases in wet and dry extremes projected in Texas and combined effects on groundwater

The US state of Texas has experienced consecutive flooding events since spring 2015 with devastating consequences, yet these happened only a few years after the record drought of 2011. Identifying the effect of climate variability on regional water cycle extremes, such as the predicted occurrence of La Niña in winter 2017–2018 and its association with drought in Texas, remains a challenge. The present analyses use large-ensemble simulations to project the future of water cycle extremes in Texas and assess their connection with the changing El Niño–Southern Oscillation (ENSO) teleconnection under global warming. Large-ensemble simulations indicate that both intense drought and excessive precipitation are projected to increase towards the middle of the 21st century, associated with a strengthened effect from ENSO. Despite the precipitation increase projected for the southern Great Plains, groundwater storage is likely to decrease in the long run with diminishing groundwater recharge; this is due to the concurrent increases and strengthening in drought offsetting the effect of added rains. This projection provides implications to short-term climate anomaly in the face of the La Niña and to long-term water resources planning.


Introduction
Before Hurricane Harvey hit the US state of Texas in August 2017, heavy precipitation events of nonhurricane origins have already caused multiple floods since April 2015 with devastating consequences. The May 2015 flood resulted from over 400 mm abovenormal rainfall falling on Texas . Then, the May 2016 flood took 12 lives and caused historic river levels, making it the fifth major flood event and the second 500 year flood in the Houston area. As shown in figure 1(a), the precipitation anomaly over the 13 months from May 2015 through May 2016 reveals a marked increase centered around Oklahoma, which is the upstream of many rivers that run through the southern Great Plains (SGP). The timing of these recent spring floods coincides with SGP's rainy season ( figure 1(b)). Starting in April, a 'spring trough' develops west of the SGP and it interacts with the developing low-level jet (LLJ) to form the convergence of moisture fluxes (Helfand and Schubert 1995), leading to the May-June rainfall peak over the SGP (Higgins et al 1997, Wang andChen 2009). The synoptic conditions associated with floods in 2015 and 2016 during April-June (AMJ, defined for spring) are not too distinct from each other, both featuring a quasi-stationary trough west of the SGP (supplemental figure S1 available at stacks.iop.org/ERL/13/054002/ mmedia) generating short waves and squall lines. In May 2015, the developing El Niño further deepened this quasi-stationary trough while enhancing the LLJ . Global warming acted to strengthen the El Niño teleconnection Teng 2007, Stevenson et al 2012) that affects the SGP. Hurricanes such as Harvey also produce excessive rainfall and can lead to flood, but these weather systems are random The succession of the post-2014 floods in Texas could lead the society into overlooking the risk in the comeback of severe droughts. In 2011, Texas underwent an intense drought and associated heat waves that was unprecedented (Nielsen-Gammon 2012) making it the worst 1 year drought on record (Fernando et al 2016). Both the strong La Niña and anthropogenic warming played a role in the severity and increased probability of this record drought (Rupp et al 2012), suggesting that the El Niño-related 2015 flood is the opposite pattern of the 2011 drought. Given the robust influence of the El Niño-Southern Oscillation (ENSO) on the SGP's precipitation (Lee et al 2014, Liang et al 2014, Liang et al 2015 and the role of global warming in strengthening this ENSO influence (Rupp et al 2012, Stevenson et al 2012, it is reasonable to anticipate a drying tendency, or even drought, to occur over the SGP in the face of a (future) La Niña event.
As of November 2017, the NOAA Climate Prediction Center (CPC) indicated a 70% chance for a La Niña to develop through February 2018. Given the emergence of this La Niña event, we decided to analyze the risk of severe drought to occur in the SGP, similar to that of the 2011 condition, and evaluate climate model projections of its water cycle extremes represented by excessive wet season and severe drought.

Data sources
Forty members of climate simulation were produced by the Community Earth System Model version 1's Large Ensemble Project (CESM-LE) with spatial resolution of 0.9 degrees longitude x 1.25 degrees latitude (Kay et al 2015). The simulations cover two periods: (1) 1920-2005 with historical forcing including greenhouse gases, aerosol, ozone, land use change, solar and volcanic activity, and (2) 2006-2080 with RCP8.5 forcing (Taylor et al 2012). The ensemble spread of initial conditions is generated by the commonly used 'roundoff differences' method (Kay et al 2015). CESM-LE was used here partly because it performs well on the depiction of the ENSO cycle and its teleconnection over the North Pacific and North America , Yoon et al 2015. Of note, the land surface model used in the CESM-LE does not include the process of anthropogenic groundwater withdrawal.
For observational data, we use the NCEP/NCAR Reanalysis (R1) (Kalnay et al 1996) that starts in 1948 in order to analyze interdecadal changes. Other datasets include the Extended Reconstructed Sea Surface Temperature (ERSST, v3b) derived from the International Comprehensive Ocean-Atmosphere Dataset (Smith and Reynolds 2003) for the ENSO indexing and the NOAA Precipitation Reconstruction over Land (PREC/L) gridded product (Chen et al 2002) of 0.5 • degree resolution.

Results
Based on the SGP domain outlined in figure 1(a), the CESM-LE simulation of the region's monthly precipitation is shown in figure 1(c) for the historical  and a 'near-future' (2010-2060) periods. In terms of climatology, CESM-LE captured the spring rainy season but underestimates the secondary rainfall peak in fall. Overall, precipitation in the SGP is projected to increase throughout the year with a larger increase in spring (April-June or AMJ), in which the precipitation increase amounts to 15%-20%. Subsequently, we examine the El Niño impact on the SGP's spring precipitation based upon the precipitation composites; this is determined by the sea surface temperature anomalies (SSTA) of the NINO3.4 region (170 • -140 • W, 5 • S-5 • N) being greater than 0.5 • C in the spring and 1 • C in the preceding winter (less than −0.5 • C/-1 • C for La Niña). This threshold is applied to each ensemble member and the observation as well, to depict the difference of the precipitation composites between El Niño and La Niña. The observational analysis during 1950-1995 (figure 2(a)) shows anomalous precipitation over southeastern Texas and part of Oklahoma and Louisiana, a known feature. In the near-future simulations (figure 2(c)) and the historical simulations ( figure 2(b)), there is a marked increase in the ensemble mean precipitation composite difference between El Niño and La Niña in the SGP. Furthermore, a similar precipitation analysis using the Coupled Model Intercomparison Project Phase 5 (CMIP5), which is shown in supplemental figure S2 by following Wang et al (2015), reveals corresponding precipitation patterns as well. That both the CESM-LE and CMIP5 simulations capture the general pattern and intensification of the ENSO-related precipitation anomalies in the SGP suggests an impact from anthropogenic climate warming.
In terms of the atmospheric circulation associated with ENSO and its changing teleconnection, figures 2(e) and (f) show the CESM-LE-generated geopotential height differences at 250 hPa; these are generally in agreement with the observed pattern in figure 2(d) including the synoptic trough west of Texas. The El Niño-associated low-pressure anomaly west of Texas is enhanced by about 35% during 2010-2060 (figure 2(f)), while the broad-scale teleconnection pattern remains similar to the historical simulations (figure 2(e)). We should note that a sign reversal in figure 2 indicates the La Niña influence on precipitation reduction in Texas. We should note that seasonal rainfall anomalies in the SGP are not only controlled by ENSO but by other circulation factors as well. Nonetheless, a further sliding correlation analysis between the SGP precipitation and the ENSO cycle, based on different climate models and a different precipitation observation, yields the same outcome of an amplified ENSO impact (supplemental figure S2).
Next, we examine the prevailing circulation pattern modulating the SGP by plotting in figure 3 the anomalous 250 h Pa geopotential height that is regressed with the SGP precipitation during AMJ. A low-pressure anomaly to the west of Texas appears to be the dominant feature embedded in a zonally oriented short-wave train, which is different from the long-wave dominant pattern induced by ENSO (figures 2(d)-(f)). There is not an apparent difference in the anomalous circulation's magnitude over Texas between the two time periods, suggesting that the common circulation features affecting the spring precipitation in Texas will not change. However, the strengthened ENSO teleconnection can become increasingly important to modulate or amplify this circulation feature. We should also note that other major circulation processes may compensate the ENSO teleconnection and that the sample sizes between figures 2 and 3 are different.
To examine the changing association between spring precipitation anomalies in the SGP and the ENSO cycle, we next compute the precipitation's 15 year sliding correlations with the spring NINO3.4 SSTA from the observation and CESM-LE data; this is shown in figure 4(a). There is a clear decadalscale fluctuation in the observation with a more pronounced increase after year 2000. The CESM-LE produces a gradual increase and it becomes persistently significant after 2000, at the 95% level. These results echo Wang et al (2015)'s observational and modeling analyses that the SGP's spring precipitation response to the ENSO teleconnection has intensified and this trend is projected to continue. Other recent studies (Guilyardi et   Zhou et al 2014) also suggest that the changing tropical heat release associated with the ENSO-related teleconnection has amplified its regional impacts worldwide.
Perhaps a greater implication of these analyses can be revealed by the concurrence of a hot summer following a dry spring, loosely describing a flash drought in the central US (Otkin et al 2016). Here, hot summers are defined as the July-September (JAS) surface air temperature exceeding 1 standard deviation (sd.) above the mean, while dry springs refer the AMJ precipitation deficit of 1 sd. below the mean of either era; these are computed within a 15 year running window. This analysis is based on the fact that severe drought in the SGP usually accompanies a below-normal spring/rainy season followed by a hotter-than-normal summer season (Long et al 2013, Nielsen-Gammon 2012. As shown in figure 4(b), the observed drought occurrence has increased after the 1990s, though it appears to be part of a multi-decadal variation. In the CESM-LE, the drought occurrence increases persistently throughout the 21st century. Likewise, the 15 year running occurrence of extremely wet springs (figure 4(c); defined by the AMJ precipitation deficit of 2 sd. below the mean), is projected to increase after 2015 despite a downtrend after 2050. By comparing figures 4(b) with (c), it appears that the disparity between their latter trends is caused by a further increase in dry springs. This downtrend in figure 4(c) is likely due to internal model variability such as that drives the earlier turnarounds in the historical period. Regardless, the ENSO impact on the JAS precipitation change in the SGP is also projected to amplify, as shown in supplemental figure S3 following the composites of figures 2(c) and (d).
To quantify the effect of the changing ENSO teleconnection on the extreme wet/dry spring and summer in the SGP, we further compute the frequencies of (a) back-to-back wet springs consisting of AMJ precipitation anomalies greater than 1 sd. occurring in two consecutive years, (b) same as (a) but only for those occurring with an El Niño winter in between, and (c) reversal of (b) for consecutive two dry summers with the JAS precipitation deficits below 1 sd. occurring with a La Niña winter in between; these are computed within a 30 year window and shown in figure 5. In the observation, the period of 1985-2015 exhibits a distinctly higher frequency in all three cases than the earlier 30 years. Meanwhile, the CESM-LE projects an almost linear increase in both the El Niñoassociated wet springs and the La Niña-accompanied dry summers toward the end of the 21st century. Given the difference between the observation and CESM-LE simulations revealed in figures 4 and 5, one has to take into account that even a sophisticated model like the CESM-LE remains a limited tool to understand the complicated climatological processes involving interannual variation, global warming and associated extremes.

Implication on groundwater
At face value, an increasingly wet climate in the SGP (cf. figure 1(c)) would suggest a net increase in the water storage, but how the concurrent increase in dry summers affect the groundwater storage in the SGP in the future remains unclear. Analyzing the CESM-LE's groundwater outputs, the trends of the 1956-2080 period are computed for groundwater storage anomalies and groundwater recharge. The The 30 year frequencies of (a) consecutive two wet springs, (b) same as (a) but with an El Niño winter in between, and (c) consecutive two dry summers with a La Niña winter in between, computed from the observation (orange) and CESM LE data (blue). simulated groundwater storage anomalies are shown in figure 6(a) from 1956-2080 averaged every 30 year, relative to the mean of 1956-1985. The result indicates a significant decrease with a rate of approximately −1.5 cm per decade in the future. Figure 6(b) shows the simulated groundwater recharge rate from 1956-2080, in which the decline in the recharge rate becomes negative after around 2015. While an increase in annual precipitation is projected in the SGP, it seems that the increase in surface evapotranspiration plays a substantial role in affecting the terrestrial water budget; this leads to a smaller infiltration and a reduction in the groundwater recharge and soil water.

Concluding remarks
The US southern Great Plains has experienced an alternation of severe drought and extreme flood since 2010 with devastating consequences. We have analyzed large-ensemble simulations that project the future of extreme wet and dry seasons in the SGP and assessed their association with the changing ENSO teleconnection under global warming. Both intense drought and excessive precipitation are projected to increase towards the middle of the 21st century and this projection is associated with a strengthened relation with ENSO teleconnections. The findings presented here echo the documented effect of El Niño in strengthening the anthropogenic warming-induced increase in summer rainfall elsewhere in the world, such as central China (Yuan et al 2018). Despite the projected increase in precipitation over the SGP, groundwater storage is anticipated to decrease with diminishing groundwater recharge; this is primarily due to the surface warming and projected increase in summer drought that reduces infiltration. These, subsequently, offset the effect of increased precipitation. The analysis presented here may be model-dependent and requires further verification using more sophisticated land surface models and/or subsurface observations.