The changing extreme values of summer relative humidity in the Tarim Basin in northwestern China

Relative Humidity (RH) in the arid region of the Tarim Basin is crucial for many reasons. The Tarim Basin has experienced a tendency to become wetter in recent decades, and the RH there also shows an increase over the past decade. However, there has been little examination of these RH changes and especially the changes to the extremes. This study investigates the changes in extreme values and the probability density function (PDF) of summer RH using quantile regression during 2006–2018 to understand the possible reasons for the increase in the summer RH anomaly. We find that extremely high values of RH show a consistent significant increase, while extremely low values have no regionally consistent tendency. The overall average value of RH in the Tarim Basin becomes higher, contributed by the upper half of the PDF. To explore the physical mechanism for these changes, we examine the corresponding regional meteorological anomaly patterns. The patterns indicate that the anomalous southwesterly airflow at 500hPa brings ample moisture into the basin and the ground in the middle of the basin significantly cools down when an extreme wet event occurs, promoting the occurrence of the extreme high RH. In this process, the contributions of water vapor transport and temperature are of equal significance though with different relative timing. These corresponding regional meteorological patterns occur more often in the most recent decade, which coincides with the recent increase in RH extremes in this region.


Introduction
Relative humidity (RH) is of great importance for multiple fields. The Tarim Basin is the major source of dust aerosols affecting East Asian countries, and its RH is closely associated with the formation of sandstorms and sand transmission (Mao et al. 2011;Li et al. 2019a;Yang et al. 2019). RH directly affects the formation of dew, which is the most important water source for plants and animals' survival in the desert Gong et al. 2019;Gerson et al. 2014). The changes of RH are associated with the response of the ecological system in desert areas and water cycle to global warming (Wang et al. 2008;Held and Shell 2012;Wright et al 2010). Understanding changes in RH can also provide a deeper understanding of changes in extreme events in this area (Tao et al. 2014;Sun et al. 2014;Zhang et al. 2012).
In the past decade, some researchers have shown that RH over land should have a downward trend with global warming, since the more rapid increase of temperature over land than over ocean should lead to a faster increase in saturation vapor pressure with global warming, while vapor pressure over land cannot increase as rapidly (Sherwood and Fu 2014;Collins et al. 2013;Simmons et al. 2010;O'Gorman 2016, 2018). This trend will lead to a drier climate in the future (Fu and Feng 2014). However, in contrast to the overall situation over the continents, an upward variation of RH in the South Xinjiang, including the Tarim 1 3 Basin region, has been observed in recent decades . Some more observations indicate that in Northwest China, a large region including the Tarim Basin, the precipitation has increased and the climate has become wetter in recent decades (Shi et al. 2007;Han et al. 2019;Peng and Zhou 2017;Chen et al. 2015;Li et al. 2016). Wetness over the Tarim Basin has shown a decadal change, i.e., specifically an increase in recent decades (Tao et al. 2014(Tao et al. , 2016. This decadal variability suggests that large-scale theoretical analysis over land cannot simply be used to understand changes in regional-scale RH. So what caused the increase of RH in the most recent decade? Water transport in some form must be part of the story in such an arid region, but this has not been explored in detail.
To better understand RH changes with global warming, the local conditions must be considered. Analyzing the probability density function (PDF) of RH locally provides a good perspective for interpreting the changes in RH distributions more broadly. This study focuses on summer RH over the Tarim Basin, since the majority of the precipitation falls in summertime (Huang et al. 2015), and the increase in wetness is mainly concentrated in summer (Li et al. 2016;Peng and Zhou 2017). The change in the mean of the time series has a close connection with its PDF and extreme values (Huybers et al. 2014;McKinnon et al. 2016). For example, in some areas, the increase in mean temperature is mainly manifested as the effect of either a decrease in the frequency of extremely low values or an increase in the frequency of extremely high values (Franzke 2013(Franzke , 2015. This indicates that understanding changes in summer RH extremes can also help to understand the mean RH change and to explore the associated physical mechanisms. However, to our knowledge, no studies have focused on the full PDF of RH over the Tarim Basin. The changes to the PDF and particularly to the extreme values of summer RH are the subject of this study.
In this study, we analyze daily summer RH data over the Tarim Basin in order to gain a more complete picture of changes in summer RH. We focus on the most recent decade and explore the mechanisms of changes in RH and the RH probability density function. Existing studies generally agree about the remarkable increase in summer precipitation in this region, but there is no consensus regarding changes in the RH distribution and physical reasons for the changes. Is this an intensification of the hydrological cycle or a change in the regional weather patterns? We study the changes in RH extreme events and the regional anomalous meteorological patterns corresponding to the extreme events to provide a way to better understand the physical mechanisms.
This paper is organized as follows: Sect. 2 describes the data used in this study. In Sect. 3, we examine the decadal change of RH. In Sect. 4 we present the statistical examination of recent changes in RH, including analysis of extremes and probability density function changes. We examine the corresponding regional anomalous meteorological patterns in Sect. 5. We then present discussion and conclusions in Sect. 6.

Data
In this study, observed daily mean RH data from meteorological stations in the Tarim Basin area are obtained from the China Meteorological Administration (http:// data. cma. cn/) for 1979 to 2018 (as shown in Fig. 1). The quality of the data has been controlled, and after removing stations that are missing data for 7 or more continuous days during the whole period (1979.1.1-2018.12.30), 19 stations remain for this analysis. Linear interpolation has been used to fill periods of missing data less than 7 days.  Fig. 3, is an edge station at the north edge). The center of the basin is the Taklamakan Desert, with tall mountains in the north and south. The map is created from the geographical information using the Google Maps API (http:// code. google. com/ apis/ maps/) with the M_Map mapping package (Pawlowicz 2020) and Matlab code (Bar-Yehuda 2020) In addition to the observed RH records, the 2-m temperature, 2-m dewpoint temperature (from which RH can be calculated) and precipitation from the ERA5-Land hourly reanalysis dataset over 1981-2018 (Copernicus Climate Change Service (C3S) 2019; Muñoz-Sabater et al. 2021) with 0.1 • × 0.1 • resolution are used. We also use 500-hPa geopotential height (Z500), 850-hPa geopotential height (Z850), and zonal (u) and meridional (v) wind speed on 500 hPa and 850 hPa from the ERA5 hourly dataset on pressure levels over 1981-2018 (Copernicus Climate Change Service (C3S) 2018; Hersbach et al 2020), with 0.25 • × 0.25 • resolution. In this study, all data from the ERA5 and ERA5-Land datasets except precipitation are processed into daily data, which is obtained by averaging 4 time points (0:00, 6:00, 12:00, and 18:00 UTC) per day. This averaging procedure is not necessary for precipitation because daily accumulated precipitation is directly available in the ERA5-Land dataset. Summer in this study is defined as June, July and August, and all anomalies are obtained by removing the seasonal cycle, similar to previous studies (Koscielny-Bunde et al. 1998): the climatological seasonal cycle is calculated as the longtime average for each day between 1981 and 2018.

Increasing local tendency of summer RH during the recent decade
We first examine the summer RH anomaly by calculating the mean in each year averaged over the 19 stations, shown in Fig. 2a. From the previous studies, it is clear that RH may show interannual variability (Du et al. 2012), and in our analysis we see distinct interannual and decadal variability of summer RH over the Tarim Basin (Fig. 2a). The recent decade can be easily identified as having an upward tendency (consistent with the "becoming wetter" mentioned in Sect. 1). But we want to understand how RH has changed beyond just the change in the mean. If we understand the recent decadal-scale trend towards higher RH, we may be able to better understand the mechanisms causing interannual variability in RH. To determine the beginning of the apparent trend, the Sequential Mann-Kendall (SQMK) method (Nasri and Modarres 2009) is used to check for the location of a change point in the summer RH anomaly time series. The method is described in the Appendix. The result shows that the transition point year is around 2006 and so we chose 2006-2018 as the period of interest for this study. The spatial distribution of the linear slope in the Tarim Basin is examined in Fig. 2b. It shows that most stations have a tendency towards higher RH, which matches the results in previous studies (Han et al. 2019;Peng and Zhou 2017;).

Tendencies of the extreme high and extreme low RH anomalies using quantile regression
To understand changes in extreme values of summer RH anomaly during the past decade, a non-parametric technique to estimate the slope in any percentile of a distribution, quantile regression, is employed (Koenker and Bassett Jr 1978;Cade and Noon 2003;Gao and Franzke 2017;Huybers et al. 2014). Linear slopes of the 5th, 50th and 95th percentiles of the summer RH anomaly time series at station 51639 are shown as examples (Fig. 3). This method can effectively show the year-by-year local trend of a specific percentile value. (For the total 92 days in the summer of each year, there is no impact from the day-to-day ordering on the percentile slopes.) The different slopes of the 5th and the 95th percentiles imply a change in RH intraseasonal variability. A block bootstrap is used here for the estimation of significance of the signs of the linear slopes, following McKinnon et al. (2016). The fitted linear slope in the data is removed first and then the residuals are resampled with replacement using a 92 day block, which is chosen based on the assumption that the RH is correlated within any given summer, but has negligible interannual correlation. After adding these values back to the linear trend removed in the first step, the trend is re-estimated. The process is repeated 1000 times. As in McKinnon et al. (2016), the resulting distribution of bootstrap trends is used to determine whether the trend is significant, with a somewhat unusual definition of significant as when 95% of the bootstrap slopes are of the same sign as the observed trend (regardless of the magnitude of the bootstrap slopes). This method will determine whether there is a detectable non-zero linear slope in the interannual change.
The quantile regression method is applied at all 19 stations in the Tarim Basin to investigate the changes in the summer RH extreme values. The extreme dry and wet days are defined when the corresponding RH is lower than the 5th percentile and higher than the 95th percentile for each summer, respectively. Figure 4 shows that there is an increasing local tendency of the 95th percentile of summer RH anomaly across nearly the entire Tarim Basin area (Fig. 4c) and a weaker trend that is nevertheless largely spatially coherent in the 50th percentile ( Fig. 4b). (Of course this is very similar to Fig. 2a.) There is no equivalent positive tendency for the 5th percentile (Fig. 4a); the extreme dry values of RH anomaly demonstrate a complex distribution of changes with few significant tendencies. In general, the high values become higher, while the low values have no uniform tendency. The average value of RH in the Tarim Basin increases, which is contributed by increases in the upper part of the distribution. We thus expect that the intraseasonal variance of RH has also increased, and we analyze the PDF in the next section.

Changes in the probability density function
The PDF provides a more comprehensive picture of summer RH variability and its changes. In order to observe the changes of extreme values and mean values, it is necessary to study the change of the PDF over time. We divide the data of the past 13 years into two time periods (2006-2012 and 2012-2018), and then qualitatively examine the PDF in both periods. Figure 5a shows a case study (station Fig. 3 Quantile regression example for RH anomalies in JJA at station 51639 in the Tarim basin. Daily RH anomaly data is shown as a function of year (black circles). The dashed lines are the trends in the different percentiles, with red corresponding to the 95th percentile, blue corresponding to the 5th percentile, and green corresponding to the 50th percentile The 95th percentile shows a consistent and increasing tendency during this period, while the 5th percentile does not have a consistent tendency 51639), and the PDF of 2012-2018 is wider than the PDF during the previous time period. This case study suggests an increase in the variance of RH variability over time. To test whether it is the case in the entire region, the year-byyear standard deviations of RH anomalies are calculated. Figure 5b shows that the mean standard deviation of each year (that is, the average of the seasonal standard deviations of 19 stations) is increasing in the past decade, as we expected from Fig. 5a.
The PDF of summer RH anomaly in the Tarim Basin is non-Gaussian with a long wet side and a short dry side. Because the changes of non-Gaussian distributions are more complicated than those of normal distributions, the changes of extreme values are not necessarily simply shifts with the mean (Huybers et al. 2014;McKinnon et al. 2016;Loikith and Neelin 2019). Compared with the PDF in the previous period, the long tail on the wet-side of the PDF during 2012-2018 obviously moves towards higher values, while there is little change on the dry-side (Fig. 5a). A slight shift in the distribution peak can also be observed (Fig. 5a). Although they differ in some details, the PDFs of the other 18 stations show similar and significant movement across the two periods, which corresponds to the results of the mean change (Fig. 2) and quantile regression (Fig. 4) analyses. There is only one station where the deviation of the second period is not obvious. Combined with the analysis above, the 95th percentile of RH anomaly demonstrates a consistent increasing trend in the past decade. This increase exhibits remarkable regional consistency, implying that more extreme wet events have occurred throughout this arid area over 2006-2018. In the next section, we aim to understand this extremum change.

Meteorology associated with recent changes in extreme high RH
The previous analyses demonstrate that the changes in the mean are contributed by the upper part of the distribution, and there is an increase in the wet extremes of summer RH over the Tarim Basin during the recent decade. We now examine the large-scale meteorological conditions associated with the increase in extremes.
To analyze the weather associated with extreme events, one useful method is to make composites of the atmospheric fields (Gao and Franzke 2017). Since the distribution of meteorological station data is uneven, reanalysis data are greatly advantageous for investigating the spatial pattern of climate variables.

Comparison between ERA5-Land reanalysis and observed records over the Tarim Basin
We use ERA5 reanalysis data to examine the weather patterns when extremely high RH occurs in the observations. To verify that this is providing an adequate representation of the moisture, we first compare the summer RH anomalies in observations with those in ERA5 data, as shown in Fig. 6. The mean value for each summer averaged over all grid boxes in the research area using ERA5 data shows similar variability to that from the 19 meteorological stations, especially in the recent decade (Fig. 6a). We also compare extremely high values of summer RH in ERA5 data to those in the observations, similar to the temperature analysis done by Mao et al. (2010). We find that values of 95th percentile RH for each summer averaged over the research area using ERA5 data agree reasonably well with those averaged from the 19 meteorological stations (Fig. 6b). Thus, ERA5 data show a good coincidence with the observations not only for the mean values but also for extremely high values, and so it can provide a relatively reliable analysis for extreme wet days. This result is not surprising, because ERA5-Land assimilates near-surface temperature and humidity data (Hersbach et al 2020;Muñoz-Sabater et al. 2021).

Dry and hot climatology of the Tarim Basin
Before analyzing the possible causes of extreme RH events, we show the climatology in the Tarim Basin region to help contextualize the mechanism for extreme events. The Tarim Basin maintains a high-temperature climate in summer (Lu et al. 2019) and is an extremely arid desert area where drought occurs often (Zhang et al. 2015;Wang and Qin 2017). Water is extremely scarce in this area, with annual precipitation less than 200 mm (Wang and Qin 2017). The center of the basin is the vast Taklamakan Desert, located in northwestern China. The climatological seasonal cycle is shown in Fig. 7a, showing that the RH of this region in summer is relatively low (Wang and Gaffen 2001). The soil is also relatively drier than in other seasons (Su et al. 2016).
All these indicate that summer is a dry season. In this case, the climatological 2-m temperature of the entire basin in summer is above 300 K and the climatological daily accumulated precipitation for JJA is less than 1mm (Fig. 7). Hot and dry conditions are the normal state in JJA for the Tarim Basin, corresponding to low RH. It is difficult for moisture to be transported into the desert region. The 38-year climatological mean wind in summer shows the prevailing westerlies, with weak southwesterly airflow entering the basin at 500 hPa and leaving the south side of the basin without penetrating the center of the desert (Fig. 7). Furthermore, near the surface, the airflow from the southwest is mainly blocked by the Himalaya Mountains, as well as the Kunlun Mountains at the south edge of the basin, which is one of the reasons for the formation of this desert (Hartmann 2015). The Tianshan Mountains are also important for the formation of the climatological characteristics in this region (Baldwin and Vecchi 2016). The climatology at 850 hPa demonstrates a dry and hot current bypassing the Tianshan Mountains and entering the basin area from the northeast, where the Gurbantünggüt Desert is located. The mountains to the north and south of the basin inhibit the transport of humid air. We seek to understand the occurrence of extremely high RH conditions in such an arid area.

Regional anomalous meteorological patterns for extreme wet events
A detailed study of the evolution of the local weather system when extreme events occur provides a way to better understand the underlying physical mechanisms, as is commonly done to study extreme events in the extratropics (e.g., Westby and Black 2015; Smith and Sheridan 2018; Loikith and Neelin 2019; Risbey et al. 2019;Gershunov et al. 2009;Teng et al. 2013). We look at composites of regional anomalous meteorological patterns during and leading up to extreme wet events, defined as those exceeding the 95th percentile of an averaged RH anomaly time series. This time series is defined during summer over  (Fig. 8), but days −1 and −3 have no surprising features. Figure 8 shows composites of anomalies in precipitation, temperature, 500 hPa geopotential height (Z500), 850 hPa geopotential height (Z850) and the corresponding anomalous Fig. 7 a The climatology of RH averaged over 1981-2018. The climatological mean in JJA for b 2-m temperature (K), c precipitation (cumulative daily amount; mm/day), d 500-hPa geopotential height (m) (Z500), e 850-hPa geopotential height (m) (Z850). The red vector arrows indicate the climatological wind velocity (m/s) at 500 and 850 hPa, respectively. For Z850, the area where the time mean sur-face pressure is lower than 850-hPa is set to white to avoid attempting to interpret data on underground pressure surfaces. Note that the 500 and 850 hPa wind vectors have different scalings. Overall, these climatological fields show that the Tarim Basin maintains a dry and hot climate wind fields over the Tarim Basin for the extreme wet days (day 0), two days preceding the extreme wet days (day −2 ), and 4 days preceding the extreme wet days (day −4 ). Before extreme events occur, at day −4 , a small amount of anomalous precipitation occurs in the oasis and mountains on the edge of the desert. Although the desert area in the center of the basin is relatively dry (Fig. 8a), the surrounding atmosphere is getting wet. At this time, temperature anomalies are weakly negative over the basin. A low pressure anomaly appears on the western side of the basin (with the weak anomalous airflow from the south at 500 hPa). Z850 anomalies do not change within the basin (Fig. 8a). Approaching the day when the extreme event occurs, these phenomena become more pronounced. At day −2, positive precipitation anomalies and negative temperature anomalies take place in the basin (Fig. 8b), and the temperature decreases rapidly in the area of cold anomalies, coinciding with high pressure appearing near the surface of the Tarim Basin at 850 hPa. High vapor pressure and relatively low temperature are the conditions for high RH. The declining temperature Fig. 8 Composite time-evolution maps for a day −4 (first column), b day −2 (second column) and c day 0 (third column) for extreme wet days exceeding the 95th percentile of the distribution of an averaged RH anomaly, which is computed from 4 stations (51639, 51810, 51839, 51765, shown as red stars) at the edges of the Tarim Basin during JJA. Composites are for anomalies of climate variables: total precipitation anomalies (Precipitation, mm/day); 2-m temperature anomalies (T2m, K); 500-hPa geopotential height anomalies (Z500, m) and 500-hPa wind anomalies (vectors, m/s); 850-hPa geopotential height anomalies (Z850, m) and 850-hPa wind anomalies (vectors, m/s). For Z850, the area where the surface pressure is lower than 850-hPa is set to white. The wind speed intensity is indicated in the upper right corner of the relevant panels and increasing moisture in the area suggest a change towards high RH conditions for this region.
The changes in temperature and precipitation are exacerbated at day −2. A large-scale cyclone centered at the west of the Tarim Basin has its east half over the desert (Fig. 8b). Compared with the composites of the Z500 anomalies and wind vector anomalies at day −4, Z500 anomalies have decreased substantially as a low pressure center has developed on the west of the Tarim Basin at day −2 (Fig. 8b). In this case, the relatively humid air current enters the basin from the south with strong winds. We note how unusual this pattern is by recalling that the mean wind pattern is just westerly (Fig. 7d).
The extreme wet day composite shows positive precipitation anomalies throughout the basin (Fig. 8c). The cold anomalies have also intensified. These strong negative temperature anomalies coincide with high pressure developed at 850 hPa near the surface. At day 0, the cyclone at 500 hPa amplifies considerably with the center moving eastward. Z500 anomalies increase from west to east over the Tarim Basin, implying stronger wind from the direction of the Indian Ocean entering the basin region. We note that upperlevel moisture transport from the south plays an important role in causing heavy precipitation in the summer (Huang et al. 2015), and this transport has happened more with the recovery of the Indian monsoon beginning in the early 2000s (Jin and Wang 2017;Huang et al. 2020). Although the increase in the RH that we observe begins in 2006 and not the early 2000s, the strengthening of the Indian summer monsoon is conducive to more water vapor transport into the Tarim Basin. This mechanism may contribute to the increase in extreme RH in the past decade. These conditions work together to cause the extreme wet events.

Contribution of moisture transport and temperature to extreme wet events
Since RH is controlled by both water vapor transport and temperature, we further explore their separate evolution during the development of this regional weather system for extreme wet events. The definition of RH is where e is vapor pressure, and e s is saturation vapor pressure. e s is a function of temperature T (e.g. by the Teten formula, Xu et al 2012). Assuming that the air pressure is constant, e is a function of specific humidity q, which is defined as the mass of water vapor in a unit mass of moist air. For efficient decomposition, the logarithmic form is adopted: ln RH = ln e − ln e s .
To calculate the anomaly, we use ln RH = ln RH + (ln RH) � , ln e = ln e + (ln e) � and ln e s = ln e s + (ln e s ) � . ln RH is the logarithmic climatological average of RH, and similarly for ln e and ln e s . Then the local derivative to analyze the change of (ln RH) � in time using Eq. (2) is calculated: Using daily data, the daily tendency term, t (ln RH) � , represents the daily change of RH anomaly. (The use of daily data avoids the influence of unwanted diurnal signals.) Using Eq. (3), the daily change of RH anomaly, t (ln RH) � , computed as a finite difference according to t (ln RH(t)) � = 1 day −1 × ((ln RH(t)) � − (ln RH(t − 1)) � ) , can be represented as the contribution of water vapor (the first term on the right side) and temperature (the second term on the right side). Figure 9 shows the composites of the local daily change of (ln RH) � , −(ln e s ) � and (ln e) � for extreme wet days (day 0) and 1-4 days prior to the extreme wet days (days −1, −2, −3, and −4 ). t (ln RH) � increases incrementally until day −1 (Fig. 9b-e), when fields in the regional anomalous meteorological patterns are more similar between day −1 and day 0 than between any other 2 days (not shown). This reveals a more rapid development of the weather system at the beginning, approaching its mature state on day −1, then the rate of changes slows down, with the anomalies reaching their maximum size on day 0. Compared with − t (ln e s ) � , the contribution of t (ln e) � exhibits stronger growth at day −4, providing a considerable contribution of 70% to the total growth rate and thus suggesting a fairly important role of water vapor transport in the initial development. Then the vapor pressure contribution becomes almost the same as that of − t (ln e s ) � at day −3, indicating the equal importance of water vapor transport and temperature at this time. Subsequently, the contribution of − t (ln e s ) � is comparable with t (ln e) � at day −2, becoming considerably more important at day −1 to day 0 and reaching 96% at day 0. This progression shows that the impact of temperature grows gradually with the increasing contribution of daily changes in saturation vapor pressure. Evaporation after precipitation will cool the surface, and along with the increased water vapor in the air, these conditions favor the increase of RH. The relationship between precipitation and surface cooling over desert is complex (Knippertz et al. 2009), however, and more observation and analysis is needed over the Tarim Basin in the future.

Frequency of the regional anomalous meteorological patterns
In the analysis above, we identified regional anomalous meteorological patterns associated with extreme wet events in the Tarim Basin. These patterns may also lead to many wet days that do not have RH as high as the 95th percentile. The increase in frequency of high RH events is not guaranteed to be due to an increase in the occurrence of this pattern. The intensifying hydrological cycle could potentially explain the change without any change to the statistics of the weather patterns. To determine if the weather pattern we identified above is related to the increase in high RH events, we look to see if the pattern is happening more often in later years of our time series.
To identify the specific regional anomalous meteorological patterns over the whole period, we first define the patterns for the extreme wet days (i.e., day 0) shown in Fig. 8c as the featured patterns. In Fig. 8, the study area is Fig. 9 Composite time-evolution maps for a day 0, b day −1 , c day −2 , d day −3 and e day −4 for the extreme RH events (same definition as in Fig. 8). The study region is 74 • E−90 • E and 35 • N−43 • N . The three columns of composites describe the local tendency changes of (ln RH) � , −(ln e s ) � and (ln e) � , respectively. Non-dimensionalization has been applied before transforming to the log form of these climate variables. Daily data are used here and the units are 1/day for all local tendencies. These local tendencies also can be regarded as the daily change of (ln RH) � , −(ln e s ) � and (ln e) � before the extreme wet events. The ratios of the shaded-region averages of − 70 • E−100 • E and 30 • N−50 • N , and there is a pattern of cooling, which shows distinct differences from the climatological mean state. Anomalous structure is also visible in the Z500 and Z850 fields. For efficient comparison and identification of days with similar patterns, the Pearson correlation is checked between the featured patterns and each day of summers 1981-2018. Due to the small amount of precipitation, only temperature, Z500 and Z850 are used. Days that meet the requirement of all three fields being significantly correlated (correlation coefficient larger than 0.6) with their counterparts in the composite extreme wet day are considered to have the same regional anomalous meteorological patterns as the extreme wet day. (Note that similar results can be obtained by choosing other threshold values for the correlation coefficient.) The number of such days per summer is then counted (Fig. 10). There is a large amount of interannual variability in the occurrence of this pattern during the whole period, but nevertheless the frequency of occurrence has a roughly increasing tendency in the recent decade (Fig. 10). This is in line with our conjecture. In the past ∼ 10 years, as the 95th percentile of RH has increased, the regional anomalous meteorological patterns that occur simultaneously have also appeared more frequently. Comparing Figs. 6a and 10, we see that the increasing summer RH tendency coincides with an increasing number of regional anomalous meteorological pattern occurrences during 2006-2018, which suggests that the regional anomalous meteorological patterns are not only associated with extreme wet days, but also play an important role in maintaining the seasonal mean RH. The decadal-scale variability occurs both in the mean summer RH time series and in the frequency of the regional anomalous meteorological patterns.

Discussion and conclusions
In the past half century, the impact of global warming has greatly increased the water holding capacity of the atmosphere (Dai 2006). Correspondingly, the atmosphere over land is drying, since the speed of water vapor transport from ocean to land cannot keep up with the speed of the increase in temperature over land (Byrne and O'Gorman 2018). This large-scale behavior is a clear example showing that global warming can impact the long-term change of RH over land. The magnitude of the internal variability within the Earth's climate system is important for understanding interannual variability and trends in RH. Regional behavior may also be quite different from the overall expectation of decreasing RH over land.
We analyzed daily summer RH data over the Tarim Basin, which is an extremely arid region dominated by a large desert. Over the length of the record at the stations in the Tarim Basin, the RH has exhibited a large amount of decadal-scale variability, including an upward tendency in the summertime mean RH anomaly during 2006-2018. In light of the large-scale, long term drying trend predicted, it is important to understand the internal variability that can cause regional deviations. The desert provides a relatively simple case study, and as moisture is critical in this region, it is also an important area for predicting local environmental and economic impacts.
Studies exploring why the Tarim Basin is getting wetter (whether measured by precipitation or humidity) find different results (Huang et al. 2015;Li et al. 2019b;Peng et al. 2020), and this disagreement implies that more work needs to be done. We provide a novel perspective on the recent increase in RH. We studied the changes in the PDF and extreme values of summer RH. The results using quantile regression show that there was an increasing trend for the 95th percentile of RH during 2006-2018, while there is no consistent regional tendency for the 5th percentile RH. This coincides with increasing variance and PDF changes of the summer RH anomalies during the most recent decade. High values become higher, while the low values do not have a consistent change. The average value of RH in the Tarim Basin becomes higher, likely contributed by the upper part of the PDF.
To explore why the 95th percentile is increasing, we identified the regional anomalous meteorological patterns that occur at the same time as extreme wet events. The patterns show abnormal southwesterly airflow at 500 hPa that transports water vapor into the basin and abnormal low temperature and high pressure near the surface with Fig. 10 Frequency of occurrence of the regional anomalous meteorological pattern corresponding to the 95th percentile extreme wet events for each year from 1981 to 2018. The blue shaded area represents the period 2006-2018. We see that during 2006-2018 the anomalous weather pattern tends to occur more frequently over time local precipitation. These processes cause the water vapor pressure to increase with more atmospheric water vapor and the saturation water vapor pressure to decrease with low surface temperature, resulting in an extremely high RH. The contributions of water vapor pressure and temperature show equal importance to the mechanism. In the development of this regional anomalous meteorological pattern, water vapor transport has a greater impact in the early stage, and temperature has a greater impact in the later stage.
In addition to examining the progression of the events themselves, we quantified variations in the frequency of the regional anomalous meteorological pattern. We find that the regional meteorological pattern has occurred more frequently over the past decade, providing a reasonable qualitative explanation for the increase of summer RH extreme wet days. This and the asymmetric changes to the distribution both suggest that dynamical changes are important for the recent change and that the changes are not simply a thermodynamically-driven intensification of the water cycle.
The Tarim Basin is a clear demonstration of large interannual variability in RH that complicates the detection of any forced trend without much longer time series. The decadalscale variability in RH extremes in this specific region is not consistent with the overall expectation of drier land areas with climate change. This recent change is likely dynamical in nature, and we see an increase in variance as well as an increase in the mean. We have focused on the most recent period as an example, and more research is needed to identify the mechanisms for the inter-decadal variability in summer RH in the Tarim Basin. More regional studies of the variability and tendency of RH in different areas are also necessary to investigate the nuances in predictions of decreased RH over land in the future.

Sequential Mann-Kendall test
To identify a change point in the summer RH time series, the sequential Mann-Kendall (SQMK) test is adopted. Based on Mann-Kendall test, Sneyers (1991) introduced sequential values to help determine the approximate year of the beginning of a significant trend. This method calculates forward and backward sequences of the test statistic and enables detection of the approximate change point of a trend from the intersection point of the two sequences. The SQMK method is frequently used to identify trend start points (Yang and Tian 2009). For more details and algorithms see Nasri and Modarres (2009).

Standard normal homogeneity test
We also use another method, the standard normal homogeneity test (SNHT), to check the change point. This technique identifies 2005 as the change point year. Our other results are not sensitive to this distinction; selecting 2005 as the starting year yields similar patterns and tendencies. The SNHT was first applied in climate science by Alexandersson (1986). The non-parametric variant of the SNHT is a popular and effective way to detect a change point (Salehi et al. 2020). Under the null hypothesis, the annual means of summer RH are assumed independent and identically distributed and thus the series is homogeneous. Then the test can detect the year where a break occurs (Kang and Yusof 2012). The details of this method can be found in Pohlert (2020).