Hot extremes have become drier in the US Southwest


 The impacts of summer heat extremes are mediated by the moisture content of the atmosphere. Increases in temperatures due to human-caused climate change are generally expected to increase specific humidity; however, it remains unclear how humidity extremes may change, especially in climatologically dry regions. Here, using in situ measurements and reanalyses, we show that specific humidity on low humidity days in the American Southwest has decreased over the past seven decades, and that the greatest decreases co-occur with the hottest temperatures. Hot, dry summers have anomalously low evapotranspiration that is linked to low summer soil moisture. The recent decrease in summer soil moisture is explained by declines in pre-summer soil moisture, whereas the interannual variability is controlled by summer precipitation. Climate models project continued declines in pre-summer soil moisture but increases in summer precipitation, leading to uncertainty as to how summer soil moisture and hot, dry days will change in the future.

quality stations are available across the region, which is topographically diverse; further, all stations are at airports, raising concerns that they are not representative of the region 93 as a whole. Temporally, the relatively short duration of the data record (48 years, 1973-94 2019) provides only a limited view into the potential role of low-frequency ocean-driven 95 variability versus anthropogenic forcings in contributing to the observed trends. 96 To address both of these issues, we first turn to additional sources of information. To 97 allow for intercomparison between data sources in terms of both trends and variability, 98 we define an annual amplification index for the full Southwest region. The index is the 99 average probability across the region of having a dry day given the occurrence of a hot day.

100
A hot day is defined at each location as a day in the 85th-95th percentile range (combined 101 red and orange polygons in Fig. 1b,e) and a dry day has a specific humidity below the 102 temperature-dependent 10th quantile of specific humidity for a GMTA of zero (orange 103 polygon in Fig. 1b,e). The thresholds for hot and dry are less extreme than the percentiles 104 used in the prior analysis because calculating the probability of extreme events through 105 empirical counts is noisier than employing a semiparametric model. Counts are summed 106 across stations, weighted by the area they represent, to produce the annual amplification 107 index. 108 As expected, the amplification index calculated using the ISD data from 1973-2019, ISD 1973 shows an increase in the probability of having low humidity conditions on hot days, par-110 ticularly post-2000 (Fig. 3a). In addition, there is substantial interannual variability, with 111 1979, 1987-1989, and 1993-1995 all exhibiting above-average dryness early in the record. 112 We next calculate the same index using the ERA5 reanalysis [16] in order to assess if the  [18,19,20]. However, the recent uptick is unprecedented in 134 the record, suggesting an additional role of human influence. As an initial estimate of 135 the relative roles of these two factors, we fit a multiple linear regression model for the 136 amplification index using the AMV index and GMTA, both of which are lowpass filtered 137 with a frequency cutoff of 1/10 years. The regression coefficient associated with predicting 138 the amplification index with the AMV is twice that for GMTA (Table S1), but the range 139 of GMTA over the 1950-2019 period is 0.96 • C, compared to 0.44 • C for the AMV, so the 140 total change explained by each factor in the regression model is very similar. A regres-141 sion model that also includes the December-January-February Niño 3.4 index and/or the annual Pacific Decadal Oscillation index does not show a significant contribution of either 143 mode.

144
Summers with hot, dry extremes have low soil moisture 145 We next turn to explaining the physical mechanisms that lead to increases in the probability 146 of dry conditions on hot days. Decreases in near-surface atmospheric humidity can come 147 from three sources: (1) increases in horizontal and/or vertical moisture divergence; (2) 148 decreases in evapotranspiration; and/or (3) increases in precipitation. Due to our focus 149 on the driest days in an already semiarid region, precipitation is not a relevant factor 150 for directly causing humidity levels well below saturation, leaving us to assess (1) and

151
(2). 152 We first consider the potential role of horizontal moisture divergence. Using ERA5, we cal- in the amplification index that we will return to below. 168 We next examine whether near-surface drying is associated with a vertical redistribution of  Having found only weak relationships between moisture divergence and the amplification 177 index, we turn to our second physical mechanism, decreases in evapotranspiration. The  Table S1), except for GLDAS2.0 soil moisture, which has a 210 weaker relationship with the amplification index (p-value: 0.014) because it shows a smaller 211 recent decrease in soil moisture than the other two datasets (Fig. 5a). The predicted ampli-212 fication index using ERA5, JRA-55, and GLDAS2.0 is correlated with the observed average 213 amplification index at 0.63, 0.52, and 0.51 for their respective periods.

Future projections are uncertain due to precipitation
We finally consider the implications of our results for future projections. In our analysis, 216 we have found that June soil moisture has been decreasing since the 1980s (Fig. 5a), which 217 has led to decreased summer soil moisture, decreased evapotranspiration, and an increase 218 in the probability of dry conditions on hot days (recall Fig. 4). While precipitation plays an  In summary, although specific humidity is expected to increase overall with warming due 239 to greater evaporation from the ocean and the increased water vapor holding capacity of 240 the atmosphere, we find that it has decreased during the summer over the semiarid South- q τ (t) = β 0,τ + s 0,τ (T (t)) + β 1,τ G (t) + G (t)s 1,τ (T (t)) where T (t) is the co-occuring temperature and G (t) is the lowpass filtered GMTA. The first 336 two terms on the right-hand side summarize the quantiles of specific humidity anomalies,      Table S1: Regression coefficients, 95% confidence intervals, and p-values associated with two multiple linear regression models for the amplification index. The AMV-GMTA fitted model is shown as the gray line in Fig. 3a. GMTA and the AMV index are lowpass filtered with a cutoff frequency of 1/10 years using a third-order Butterworth filter. The Nino3.4 index and PDO index are not used in the fitting of the AMV-GMTA model. The soil moisture and precipitation fitted model is shown as the colored lines in Fig. 3b. All fitted models use GPCC precipitation; the soil moisture product used in conjunction with the GPCC precipitation is shown in parentheses. Figure S1: The Spearman rank correlation between July-August-September daily average temperature and specific humidity using ERA5 reanalysis (contours) and in situ station data from the Integrated Surface Database (circles). Figure S2: As in Fig. 2, but with the limited number of stations with data beginning in 1950.
30 Figure S3: (Left) The vertical profile of specific humidity on top tercile minus bottom tercile amplification index years. (Right) As in Fig. 4, but for the vertically integrated moisture divergence from ERA5. Vertically integrated moisture divergence is total per day.
31 Figure S4: As in Fig. 4, but for runoff from ERA5. Runoff is the total per day.