Drought variability and its connection with large-scale atmospheric circulations in Haihe River Basin

Jia Wei ,Wei-guang Wang ,,Yin Huang ,c,Yi-min Ding ,Jian-yu Fu ,Ze-feng Chen ,Wan-qiu Xing

a State Key Laboratory of Hydrology-Water Resources and Hydraulic Engineering,Hohai University,Nanjing 210098,China

b College of Hydrology and Water Resources,Hohai University,Nanjing 210098,China

c China Water Resources Beifang Investigation,Design and Research Co.,Ltd.,Tianjin 300222,China

Received 22 June 2020;accepted 20 September 2020 Available online 16 December 2020

Abstract Drought is one of the most widespread and devastating extreme climate events when water availability is significantly below normal levels for a long period.In recent years,the Haihe River Basin has been threatened by intensified droughts.Therefore,characterization of droughts in the basin is of great importance for sustainable water resources management.In this study,two multi-scalar drought indices,the standardized precipitation evapotranspiration index(SPEI)with potential evapotranspiration calculated by the Penman－Monteith equation and the standardized precipitation index(SPI),were used to evaluate the spatiotemporal variations of drought characteristics from 1961 to 2017 in the Haihe River Basin.In addition,the large-scale atmospheric circulation patterns were used to further explore the potential links between drought trends and climatic anomalies.An increasing tendency in drought duration was detected over the Haihe River Basin with frequent drought events occurring in the period from 1997 to 2003.The results derived from both SPEI and SPI demonstrated that summer droughts were significantly intensified.The analysis of large-scale atmospheric circulation patterns indicated that the intensified summer droughts could be attributed to the positive geopotential height anomalies in Asian mid-high latitudes and the insufficient water vapor fluxes transported from the south.? 2021 Hohai University.Production and hosting by Elsevier B.V.This is an open access article under the CC BY-NC-ND license(http://creativecommons.org/licenses/by-nc-nd/4.0/).

Keywords:Drought;Standardized precipitation index;Standardized precipitation evapotranspiration index;Large-scale atmospheric circulation

1.Introduction

Drought,a major natural hazard resulting from prolonged water deficit,can cause significant damage to ecosystems,economies,and agriculture(Lorenzo-Lacruz et al.,2010;Potop et al.,2012;Qin et al.,2015;Yang et al.,2017).Climate change,characterized by rising temperature and shifted precipitation patterns,has caused the increase in drought frequency(Wanders and Wada,2015;Huang et al.,2017)and severity(Liu et al.,2015;Ayantobo et al.,2017),especially in China(Yu et al.,2014).In recent years,statistical data have shown that the annual average drought-related loss of grain production in China has exceeded 2.6×1010kg,which is almost the annual food demand for 60 million people(Li et al.,2010;Wang et al.,2017;Xu and Zhang,2018).Meanwhile,droughts have greatly increased the risk of water resources shortage(Montaseri and Amirataee,2017;Cao et al.,2017)and severe socio-economic impacts(Mishra and Singh,2010).Therefore,rationally understanding drought variability and determining its underlying causes can aid water resources planning and management.

Although appropriate drought assessment is necessary,it is difficult to precisely define drought because different categories of drought(e.g.,meteorological drought,agricultural drought,and hydrological drought)are related to various factors.For instance,meteorological drought is mainly related to a prolonged period of insufficient precipitation,while evapotranspiration deficit can lead to agricultural drought and hydrological drought(Corti et al.,2009;Easterling et al.,2007;Ma et al.,2018).The characterized time scale is another important drought feature.For instance,precipitation shortage in one month can cause damage to rain-fed crops,but it may have hidden impacts on a large reservoir system.Thus,given that different systems(e.g.,agricultural,hydrological,and socio-economic systems)have different response times to the accumulated precipitation deficit,drought impacts have a multi-scalar nature(Potop et al.,2012).Thus,it is difficult to precisely assess drought.For the reasons mentioned above,drought indices must have multi-scalar characteristics and be associated with specific time scales for drought assessment(Du et al.,2013;Zhang et al.,2015).

The standardized precipitation index(SPI),a typical multiscalar drought index developed by Mckee et al.(1993),is a useful tool for identifying drought based only on monthly precipitation data.Currently,SPI has been broadly used and compared with many other indices,such as the Palmer drought severity index(PDSI)(Palmer,1965),ChinaZindex(Ju et al.,1997),and deciles index(Gibbs and Maher,1967).Extensive advantages of SPI were thus defined,focusing on its simplicity,reliability,and more accurate drought forecasting ability at different time scales(Keyantash and Dracup,2002;Montaseri and Amirataee,2017).Due to these advantages,SPI has been highly recommended by the World Meteorological Organization as the reference drought index to assess drought and identify drought period.

Another widely used multi-scalar index for drought assessment is the standardized precipitation evapotranspiration index(SPEI),which was developed by Vicente-Serrano et al.(2010)on the basis of monthly climatic water balance.SPEI combines the sensitivity of PDSI to the change in evaporation demand with the simple calculation procedure and multi-scalar characteristics of SPI.Similar to SPI,SPEI can also be used to analyze meteorological and agricultural droughts on shorter time scales(e.g.,one to three months),and to evaluate hydrological drought on longer time scales(e.g.,12 months and longer)(Mckee et al.,1995;Szalai et al.,2000).More importantly,this index is able to identify the role of precipitation and potential evapotranspiration(PET)variability in drought assessment in the context of global warming(Dubrovsky et al.,2009;Potop et al.,2012;Vicente-Serrano et al.,2010).

Originally,Vicente-Serrano et al.(2010)only used a temperature-based method,the Thornthwaite equation,to calculate PET in SPEI.However,given that PET is driven by climatic factors and affected by vegetation and geographical locations(Arnell et al.,2001;Wang et al.,2011),it is essential to improve PET calculation in SPEI for rational investigation of actual drought variability.Thus,a more suitable PET method,the Penman－Monteith(PM)equation(Allen et al.,1998),recommended by the Food and Agriculture Organization of the United Nations,was used to improve SPEI parameterization(Sheffield et al.,2012;Ayantobo et al.,2017).Recently,the improved SPEI has been adopted in many regions around the world,including the United States,the Pacific Northwest(Abatzoglou et al.,2014),Southwest China,the North China Plain(Ming et al.,2015),the Loess Plateau(She and Xia,2018),and the Tarim River Basin(Tao et al.,2014).

On the other hand,investigating the links between regional drought and climate modes is a useful method to determine the causes of droughts.In Southwest China,the decadal winter drought is mainly caused by the decadal transition of Arctic Oscillation(AO)and frequent El Ni~no-Southern Oscillation(ENSO),especially the central Pacific ENSO(Tan et al.,2017).Moreover,previous studies have proven the important roles of ENSO,AO,North Atlantic Oscillation(NAO),Atlantic Multi-Decadal Oscillation(AMO),and Pacific Decadal Oscillation(PDO)in triggering the climate extremes in North China(Gao and Wang,2017;Wang et al.,2019).However,these previous studies mainly focused on the connection of climate anomalies with a single anomalous drought or extreme precipitation event.It is necessary to further investigate the association between wet/dry conditions and climatic patterns.

The Haihe River Basin,located in the semi-arid regions of North China,is a greatly important region for water supply to the capital city,Beijing.Meanwhile,this basin occupies approximately 11% of the total national arable land,thereby requiring sufficient irrigation water resources to sustain stable crop production.However,due to the highly uniform spatiotemporal distribution of precipitation caused by the Asian monsoon climate,the basin has experienced frequent droughts,with the droughts becoming more intensified over recent decades(Bao et al.,2012;Qin et al.,2015;Gao and Wang,2017).Thus,comprehensive investigation of spatiotemporal variability of droughts is helpful to improving the regional strategies for watershed water resources management(e.g.,He et al.,2015;Liu et al.,2016;Gao et al.,2018).Most of the previous studies on drought assessment have only adopted a single drought index.In fact,using several drought indices based on different principles could be more rational for drought assessment(Zargar et al.,2011;Paulo et al.,2012).For example,Qin et al.(2015)used the soil moisture drought severity and SPI to analyze droughts and improve the identification of drought-affected regions and the quantification of drought duration.It has been proven that using multiple drought indices can provide more reliable and comprehensive analysis.

In this study,the variability of droughts in the Haihe River Basin was assessed using two drought indices,SPEI based on the PM method and SPI,at multiple time scales.Afterwards,the association between drought variability and climate modes was examined.Furthermore,the possible teleconnection of drought processes with simultaneous large-scale atmospheric circulation was analyzed as well.This study can provide valuable references for the application of drought indices to regional drought mitigation and water resources management in different regions.

2.Study area and data

2.1.Study area

The Haihe River Basin is located between 112°E and 120°E and 35°N and 43°N,and is characterized as a temperate continental monsoon region(Fig.1).As one of the largest basins in China,the basin has an area of 318 000 km2,accounting for 3.3% of the national total area.Due to the existence of several megacities such as Beijing and Tianjin,the population of the basin accounts for approximately 10%of the national total population.Meanwhile,the Haihe River Basin is also one of the main grain-producing regions of China.However,it suffers from physical water shortage and the local water resources comprise merely 1.5% of the national total amount(Xiong et al.,2006).The mean annual precipitation is 526 mm with uneven spatial distribution across the basin.In particular,high annual precipitation mainly appears in the plain region and on the windward side of the mountains(Yang et al.,2006).Influenced by the continental high circulation system in mid-latitude Asia and the western Pacific subtropical high(WPSH)circulation system,precipitation is mainly concentrated in summer(from June to August).According to the air temperature data from 1961 to 2017,the mean annual air temperature ranges from 8.66°C to 11.55°C.Over the last several decades,the air temperature increase combined with the precipitation decrease has led to an obvious reduction in streamflow in the Haihe River Basin(Bao et al.,2012;Wang et al.,2016)and might further result in the intensification of droughts(Qin et al.,2015).

2.2.Data

Fig.1.Location of Haihe River Basin and meteorological stations.

Meteorological data from 40 meteorological stations in the period of 1961－2017 were obtained from the National Meteorological Information Centre of China Meteorological Administration,and these data included daily records of precipitation,temperature,wind speed,relative humidity,and sunshine duration(see supplementary data).The quality of these data was controlled by the national standard quality management system.The missing daily precipitation data at a given station were estimated by averaging the precipitation values of the two nearest stations.

In this study,ENSO,WPSH,NAO,PDO,and AMO were selected for the correlation analysis between climatic patterns and drought indices in the Haihe River Basin.WPSH data represented by the area index(WPSHA),the ridge line index(WPSHR),and the western extension index(WPSHW),were obtained from the Climate Diagnostics and Prediction Division of the National Climate Center.The ENSO index used in this study was derived from the monthly sea surface temperature data of the National Oceanic and Atmospheric Administration(NOAA)Ni~no 3.4 in the domain of 5°N to 5°S and 170°W to 120°W.The monthly indices of NAO,PDO,and AMO were provided by the NOAA Climate Prediction Center.Additionally,the monthly geopotential height,winds,and specific humidity at a 2.5°×2.5°resolution were acquired from the National Centers for Environmental Prediction and National Centre for Atmospheric Research reanalysis dataset(Kalnay et al.,1996).

3.Methodology

3.1.SPI

Formulated by Mckee et al.(1993),SPI is a probabilitybased index meant to quantify precipitation deficit for drought assessment.The probability is standardized to make an index of zero mean precipitation amount.The long-term precipitation observations are the only required data for SPI calculation.First,the monthly precipitation(denoted asx)is accumulated at specified time scales,both the short-term scales(one month and three months)and the long-term scales(12 months and 24 months).Afterward,the best probability density function is used to describe the distribution of precipitation at a given time scale.The original formulation of SPI uses a two-parameter Gamma distribution,and many related studies(e.g.,Qin et al.,2015;Zhang et al.,2017b;Gao et al.,2018)have proven the feasibility of its cumulative probability distribution.In this study,the monthly precipitation data from the 40 meteorological stations were accurately fitted by the Gamma distribution function.The two-parameter Gamma distribution is defined by the cumulative distribution function as follows:

whereαis the scale parameter,βis the shape parameter,and Γ（α）is the gamma function.Because Eq.(1)is undefined forx=0,and precipitation may contain zero values.In this case,the mixed cumulative distribution is used to fit the monthly precipitation data,which is expressed by

whereqis the probability of zero precipitation,which is given by

wheremis the number of zero-precipitation data in the data series,andnis the sample size of precipitation data.Finally,based on the classical standardized equation(Abramowitz and Stegun,1965),H(x)converges into the standard normal SPI variable.The program provided by the National Drought Mitigation Center was used to calculate SPI in the study area.Table 1 shows the standards for categorizing the SPI-based dryness/wetness levels and the corresponding cumulative probabilities.

3.2.SPEI

Introduced by Vicente-Serrano et al.(2010),SPEI is mathematically similar to SPI,but it considers the effect of PET on drought in addition to precipitation.Initially,PET is calculated with a simple temperature-based empirical approach via the Thornthwaite equation(Thornthwaite,1948).However,many previous studies(Jensen et al.,1990;Chen and Sun,2015;Zhang et al.,2017a)have shown that the Thornthwaite equation tends to underestimate PET.This is because the Thornthwaite equation is very sensitive to temperature that acts as a driver of PET,and it ignores other variables associated with atmospheric water demand(Zhang et al.,2017a).Developed by Penman(1948),the PM equation is more physically realistic than the Thornthwaite equation,which accounts for both thermodynamic and aerodynamic effects.In this study,the PM equation was used to calculate PET(E0),which is expressed by

whereΔis the slope of the vapor pressure curve(kPa/°C);Rnis the net radiation at the land surface(MJ/(m2˙d));G0is the soil heat flux density,which is commonly considered to be 0 for daily PET estimation(kPa/°C);γis the psychrometric constant(6.77 Pa/°C);Tis the mean daily air temperature(°C);u2is the average daily wind speed at a 2-m height(m/s);esis the saturation vapor pressure(kPa);andeais the actual vapor pressure(kPa).Afterward,the difference(Di)between monthly precipitation(Pi)andE0for monthi(E0i)is calculated as follows:

Table 1Classification of SPEI and SPI and their cumulative probabilities of occurrence.

The calculatedDiseries is aggregated at predefined time scales and fitted by the three-parameter log-logistic distribution.The probability distribution function(F(x))is expressed as

The standardization procedure of SPEI is as same as that of SPI.For detailed information on SPEI,please refer to Vicente-Serrano et al.(2010).The drought and wetness classification standard for SPEI is as same as that for SPI(Table 1).

3.3.Correlation analysis

Correlation coefficients are usually used to evaluate the relationship between two variables.In previous studies,the Pearson"s correlation and Spearman"s rank test methods were extensively used(Tabari et al.,2014).The Pearson"s approach aims at detecting linear correlation,and it requires that time series have a normal distribution.Commonly used as an alternative correlation method,the Spearman"s rank test measures the rank correlation between two factors(Helsel and Hirsch,1992).In this study,the Spearman"s rank method was employed to analyze the nonlinear correlation between drought and climate patterns based on a statistical significance test.The Spearman"s rank correlation coefficient(rs)is defined as

whereYiandXirefer to the rank of drought and climatic pattern indices in monthi,respectively;andNis the sample size of data pairs.

3.4.Data processing and evaluation schemes

Due to the multi-scalar characteristics of SPI and SPEI,it is possible to choose suitable time scales to monitor droughts generated by the cumulative antecedent climate conditions at different time scales.To detect the recorded drought events,SPI and SPEI were calculated at a 1-month scale(hereafter referred to as SPI-1 and SPEI-1,respectively).Meanwhile,SPI and SPEI at a 3-month scale(hereafter referred to as SPI-3 and SPEI-3,respectively)were computed to evaluate seasonal droughts.To reflect annual drought conditions,SPI and SPEI at a 12-month scale(hereafter referred to as SPI-12 and SPEI-12,respectively)were calculated in the study area.SPI and SPEI at a 24-month scale(hereafter referred to as SPI-24 and SPEI-24,respectively),which are suitable to investigate the long-term aspects of droughts(Keyantash and Dracup,2002),were calculated to indicate the persistence of dry conditions for a duration of years.The drought frequency was defined as the number of months with the SPI and SPEI values no higher than－1.0,and the wetness frequency was denoted as the number of months with the SPI and SPEI values no lower than 1.0.As one of the most widely-used nonparametric tests,the Mann－Kendall(M－K)test(Kendall,1970;Mann,1945)was used to test the trends of SPI and SPEI in this study.The M－K test was also used in this study to detect the mutation of climate conditions.To investigate the possible physical causes of drought processes,the relationships between drought indices and atmospheric circulation pattern indices were quantified by the Spearman"s rank correlation method at both annual and seasonal scales.Moreover,the average 500-hPa geopotential height anomalies and 850-hPa water vapor flux were analyzed in this study.

4.Results

4.1.Drought processes detected by SPEI and SPI

As shown in Fig.2,SPI and SPEI were able to capture the notable droughts over the study area and demonstrated similar drought processes.Fig.2(a)through(d)displays the short-term droughts at 1-month and 3-month scales.SPEI presented more frequent drought events with a higher intensity than those identified by SPI at short-term scales.Meanwhile,at the 1-month scale,the wet events captured by SPI were more intense than those captured by SPEI.This was caused by the abnormal precipitation in certain individual months.Fig.2(e)through(h)shows the long-term drought processes at 12-month and 24-month scales.At long-term time scales,the temporal variations of drought detected by the two indices were very similar.However,before approximately 1980,SPEI detected more severe droughts than SPI did.These results indicate that PET and precipitation play more important roles in intensifying drought and wet events at short-term scales,respectively.

The variation of drought indices at long-term scales showed lower magnitudes of fluctuation than that at shortterm scales.For instance,at the 12-month scale,significant drought events were detected in the periods of 1965－1966,1968－1969,1972－1973,1997－1998,1999－2000,and 2001－2003.However,at the 24-month scale,these drought events were identified more clearly,owing to the severe and persistent dry conditions occurring in the mid-1960s,the early 1970s,the first half of the 1980s,and the period from the late 1990s to the early 2000s.In addition,the period of 1997－2003 was regarded as one of the major drought periods,with the highest degree of severity captured by both indices.Overall,long-term drought events were more likely to occur after the 1990s.

Fig.2.Time series of SPEI and SPI at different time scales.

At the short-term scales,seasonal drought processes were investigated with the drought indices at a 3-month scale.Fig.3 illustrates the seasonal drought variations detected by SPEI and SPI in spring(March to May),summer(June to August),autumn(September to November),and winter(December to February),and theZvalue in each sub-figure denotes the statistics of the M－K test.The seasonal SPEI demonstrated a significant decreasing trend in summer and a significant increase in autumn.This indicates the intensification of summer droughts and the alleviation of autumn droughts.Meanwhile,the seasonal SPEI time series shows that droughts in spring and winter had non-significant trends.Furthermore,the trends of seasonal SPI in summer,autumn,and winter were consistent with those of the seasonal SPEI,showing a significant decrease in summer,a significant increase in autumn,and a non-significant trend in winter.However,SPI in spring presented a significant increase.Additionally,the seasonal SPI showed frequent fluctuations exceeding its standard deviation.This reveals that the drought condition is mainly dominated by the large seasonal precipitation variations.

Fig.3.Seasonal drought processes detected by SPEI and SPI(the red solid lines represent the linear trend,the black dash lines represent the mean value,and the gray shadings denote the standard deviations of seasonal SPEI and SPI around their mean values).

According to the drought severity classification standard shown in Table 1,drought events at different severity levels were identified.In the Haihe River Basin,the long-term moderate drought events occurred in the periods of 1965－1966,1968－1969,1972－1973,1997－1998,1999－2000,and 2001－2003.At seasonal scales,severe spring drought events detected by SPEI occurred in 1962,1968,1972,1981,and 2001,whereas the SPI-detected moderate spring drought events occurred in 1962,1972,and 1993,and a severe drought event occurred in 2001.

4.2.Spatiotemporal variation of drought indices

To investigate the spatiotemporal variations of droughts,the SPEI-1 and SPI-1 time series at each meteorological station were analyzed with the M－K test(Fig.4).In most regions of the Haihe River Basin,a significant intensified drought trend exceeding the 95%confidence level was observed in July and August.By contrast,a significant alleviated drought trend was found in April and May at most stations.However,the drought indices from November to February exhibited different trends at several stations.Spatially,the significant decrease in SPEI-1 mainly occurred at the DT,WTS,YS,and CZ stations,which are located in the western mountainous area of the basin.In comparison,the significant intensified drought trend detected by SPI-1 was mainly distributed at the WTS and JX stations,which are located in the southwestern region of the basin.Moreover,the two drought indices showed the similar proportions of area where a significant change in droughts occurred.For instance,approximately 13% of the basin experienced a significant alleviated drought trend based on the statistics of the SPEI time series,and the percentage in the case of SPI was 16%.In general,the two drought indices consistently showed that summer droughts had a significant rising trend in the basin.

4.3.Difference between SPEI and SPI

Given that the interannual drought processes can be easily demonstrated by drought indices at a 12-month scale,SPEI-12 and SPI-12 were adopted in many previous studies.In this study,the difference between SPEI-12 and SPI-12 was analyzed in the study period.As shown in Figs.5 and 6,the difference between SPEI and SPI gradually increased with a significant increase in annual average air temperature and decreases in annual precipitation andE0.The M－K and linear regression methods were respectively used to detect the abrupt change points and change trends of annual average air temperature,annual precipitation,andE0.Fig.6 shows that annual average air temperature in the study area increased with a rate of 0.33°C per decade,and annual precipitation andE0decreased by 9.2 mm per decade and 6.3 mm per decade,respectively.A significant change point for the difference of the two drought indices was found in 1977,which is consistent with the change point for annual precipitation.Moreover,the change point in 1977 agreed with those for annual SPEI and SPI(not shown in this paper).This indicates that precipitation is the predominant factor controlling drought variations.In addition,this study compared the drought processes in the two periods of before and after 1977.Fig.5 shows that the drought events detected by SPEI-12 had lower drought severity than those identified by SPI-12 in the 57-year historical period,whereas the SPEI-12-based wet events had higher severity than the SPI-12-based ones.The results from Figs.5 and 6 reveal thatE0plays an important role in intensifying or mitigating the wetness/drought events.In addition,the change points of annual SPEI and SPI time series at each meteorological station are shown in supplementary data.The SPEI and SPI time series showed a change point in 1977 at a few stations,some stations of which had consistent change points as detected by SPEI and SPI.Other stations showed different change points in comparison with those of the entire basin,mainly due to the variations of local precipitation andE0.

Fig.4.Z statistics of M－K test of SPEI-1 and SPI-1 in period of 1961－2017(the blue and red colors denote significant increasing and decreasing trends at a 95% confidence level,respectively,and the green color denotes an insignificant trend).

Fig.5.Time series of difference between SPEI-12 and SPI-12 in period of 1961－2017.

Fig.6.Time series of climate variables in period of 1961－2017.

4.4.Variability of drought and wetness frequencies

Based on the detected significant abrupt change point in 1977,drought variations before and after the change point were compared by dividing the study period into two subperiods:the pre-change period(1961－1977)and the post-change period(1978－2017).Fig.7(a)compares the frequency of the SPEI-12-based moderate wetness events in the pre-change period with that in the post-change period.It shows that the wetness events demonstrated a large decrease in their frequency in the western and southern regions but presented a significant increase in frequency in the northern region.As shown in Fig.7(b),the frequency of moderate drought events detected by SPEI-12 tended to consistently decrease in the entire basin,with a relatively lower magnitude of decrease in the south and southeast.The comparison of Fig.7(a)and(b)illustrates that the frequency of wetness events exhibited an obvious decrease in the southern Haihe River Basin,and that of drought events slightly decreased in the same region.This indicates that the southern Haihe River Basin might still experience droughts in the post-change period.To explore the variation of drought frequency across the entire Haihe River Basin,the frequencies for each drought and wetness category in the pre-and post-change periods were compared,as shown in Fig.8.The drought frequency with SPEI lower than－1.0 decreased from 14.58% in the prechange period(1961－1977)to 4.78% in the post-change period(1978－2017).The wetness frequency with SPEI higher than 1.0 dropped from 8.85% in the pre-change period to 3.11% in the post-change period.

4.5.Teleconnection with atmospheric patterns

As climatic patterns may influence annual and seasonal droughts,the teleconnections between the atmospheric patterns(WPSH,ENSO,NAO,PDO,and AMO)and drought indices were explored.The Spearman correlation analysis was conducted to examine the links between drought indices and seven circulation indices(WPSHA,WPSHR,WPSHW,Ni~no 3.4,NAO,PDO,and AMO)(Table 2).

Fig.7.Changes in wetness and drought frequencies in post-change period(1978－2017)relative to those in pre-change period(1961－1977).

The physical mechanisms of climate variability in the Haihe River Basin were further explored at both annual and seasonal scales.At annual scales,low simultaneous correlation coefficients were obtained between drought indices and these climatic indices(Table 2).However,at seasonal scales,the WPSH-related indices(WPSHA,WPSHR,and WPSHW)and ENSO(Ni~no 3.4)presented significant relationships with drought indices in spring and summer(Table 2).In particular,the impacts of ENSO and WPSH on SPI were more pronounced than those on SPEI in summer and spring.This reveals that ENSO and WPSH have more significant effects in regulating precipitation variations.The effects of ENSO on the variations of SPEI and SPI in summer were more obvious than those in spring because the significance of the correlation between Ni~no 3.4 and summer drought indices exceeded the 95% confidence level.In addition,AMO presented a significantly positive relationship with SPI in spring.In autumn,a significantly negative relationship between Ni~no 3.4 and SPEI was found.Except for the spring SPI,the significance of correlation between seasonal/annual drought indices and the three climatic indices(NAO,PDO,and AMO)did not exceed the 90% confidence level.This manifests that the effect of these three climate variables on droughts was not statistically significant.

Fig.8.Frequency distribution curves of SPEI in pre-and post-change periods in entire basin.

Both SPEI and SPI demonstrated an increasing trend in summer droughts in the period of 1961－2017.Therefore,to investigate the possible links between drought tendency and interdecadal variation of atmospheric circulation,the synchronous changes in large-scale circulation patterns were quantitatively analyzed,and the entire study period was divided into six sub-periods:1961－1970,1971－1980,1981－1990,1991－2000,2001－2010,and 2011－2017.At first,the 500-hPa geopotential height anomalies in summer(June to August)in the six sub-periods were analyzed.As shown in Fig.9,a significant increase in 500-hPa geopotential height anomalies mainly appeared in the middleand high-latitude regions,including Baikal Lake and the Mongolia regions.In the first two decades(1961－1980),the 500-hPa geopotential height changed from negative to positive anomalies in the Haihe River Basin.The positive geopotential height anomalies continuously increased in the subsequent two decades(1981－2000)and reached the peak values after the 2000s.In the last two decades(2001－2017),significantly positive anomalies dominated the Haihe River Basin and were expanded to the middleand high-latitude regions.The anomalous 500-hPa geopotential height would lead to the increase of air temperature and the decrease of water vapor flux.In addition,with the intensification of 500-hPa geopotential height anomalies,the location of WPSH in low latitudes moved westward in the period of 1961－2017,which also favored the occurrence of hot summers.

To investigate the synchronous changes of water vapor flux patterns,the 850-hPa water vapor flux in summer in the period of 1961－2017 was calculated(Fig.10).In the period of 1981－2017,the southwesterly and southeasterly water vapor fluxes were significantly weakened over the Haihe River Basin,in comparison with the situation in the period of 1961－1980.In addition,the route of water vapor transport in the period of 1981－2017 was more eastward relative to the case of 1961－1980.Consequently,the northward propagation of the East Asian summer monsoon was depressed,thereby leading to lower summer rainfall than the normal level in North China,including the Haihe River Basin.This agreed with the spatial distribution pattern of geopotential height in summer.

Table 2Correlation coefficients between annual and seasonal drought indices and seven climatic indices.

To further analyze the association between wet/dry conditions and atmospheric circulations,the summer wet/dry conditions were classified into four categories according to the standards shown in Table 1.In the past 57 years,two severe wetness events,eight moderate wetness events,two moderate drought events,and 45 near-normal events occurred.Fig.11 shows the circulation patterns during the four typical summer drought or wetness events in 1964,1973,1986,and 1997 detected by both SPEI and SPI.Table 3 shows the values of circulation pattern indices that are in a significant correlation with drought indices.As shown in Figs.11 and 12,negative geopotential height anomalies were located in Baikal Lake and Mongolia in the wet summers of 1973 and 1964.Meanwhile,it is evident that a mass of water vapor converged in the basin from the southern side,thereby providing warm airflows for continuous precipitation over the basin.However,in the drought summer of 1997,positive geopotential height anomalies were centered over Mongolia,and positive geopotential height anomalies appeared in the western Pacific.Meanwhile,the water vapor flux was insufficient in comparison with that in wet conditions.In terms of climatic pattern indices,WPSHR and Ni~no 3.4 in the moderate drought year of 1997 were respectively lower and higher than those in normal years(Table 3).By contrast,WPSHR and Ni~no 3.4 in the wet years of 1964 and 1973 were respectively higher and lower than those in normal years.The climatic indices for these four typical summer drought/wetness events had significantly positive and negative correlations with summer drought and simultaneous climatic patterns.Based on the comparison of atmospheric circulation patterns under drought and wetness conditions,it can be inferred that the anomalous 500-hPa geopotential height in the middle-high latitudes and water vapor flux largely affect the occurrence of drought or wetness events in the Haihe River Basin.

Fig.9.Decadal mean(contour lines)and anomalies(shaded areas)of 500-hPa geopotential height(m)in summer(June to August)in period of 1961－2017(the shaded areas with dots denote the anomalies exceeding the 95% confidence level based on a standard t-test,and the thick red solid lines represent the boundary of the Haihe River Basin).

Fig.10.Decadal mean 850-hPa water vapor flux in summer(June to August)in period of 1961－2017.

5.Discussion

5.1.Drought processes in Haihe River Basin

In this study,SPI and SPEI that considered different climatic factors were adopted to investigate the spatiotemporal variation of droughts in the Haihe River Basin in the period of 1961－2017.Although SPEI and SPI have different mechanisms,the temporal variations of drought periods detected by these two indices were similar in the study area(Fig.2).This finding agrees with previous studies in different regions(Gurrapu et al.,2014;Labudov?a et al.,2016).Our results showed that drought duration increased with the prolonged time scales.For instance,the long-term drought condition investigated by SPEI-24 and SPI-24,which represented the hydrological drought,significantly increased with an evident decrease in streamflow(Yang and Tian,2009;Cong et al.,2010;Bao et al.,2012;Wang et al.,2013).Thus,severe hydrological drought has been one of the significant challenges for the Haihe River Basin,and if this intensified drought trend continues it will result in ecological disasters in the future.

At seasonal scales,the spring drought tendency detected by SPEI agreed with the findings of previous studies(Qin et al.,2015;Zong et al.,2013).The significantly alleviated spring droughts identified by SPI were consistent with the increasing spring precipitation at almost all stations in the basin(Wang et al.,2011).However,the insignificant trend in spring droughts captured by SPEI was also supported by a few previous relevant studies,which used drought indices driven by multiple climatic factors(Yin et al.,2014;Zong et al.,2013).This study found that autumn droughts had a decreasing tendency,as detected by both SPEI and SPI.However,Yang et al.(2016)concluded that droughts in autumn had an intensified trend in the period of 1961－2010.Although both this study and Yang et al.(2016)used the same database,the difference in the findings is mainly caused by the remarkable precipitation increase in the period of 2011－2017.As for the variability of drought severity,Yang et al.(2016)used the Thornthwaite equation to calculate PET for SPEI and concluded that drought severity in the Haihe River Basin had an increasing trend in the period of 1961－2010.However,this study found that drought severity in the study area presented a weakened trend in the period of 1961－2017.Therefore,the different PET estimation methods may explain the different conclusions derived from this study and Yang et al.(2016).

Fig.11.Decadal mean(contour lines)and anomalies(shaded areas)of 500-hPa geopotential height(m)in summer of four typical years.

Table 3Climatic indices in typical summer drought and wetness conditions.

Drought occurrence was mainly affected by precipitation.However,the variability of drought severity might have been mainly affected by PET(Figs.2 and 6).Meanwhile,the SPIbased seasonal drought trend was more significant than the SPEI-based one(Fig.3).A possible explanation is that PET might weaken the impact of precipitation on droughts.As time went by,the difference between SPEI and SPI became more pronounced,and SPEI identified more wetness events and less drought events than SPI did.This implies a decreasing PET in the same evaluation period.Fig.6 demonstrates the decreasing PET that was accompanied with air temperature rise.This is known as the“evaporation paradox”(Roderick and Farquhar,2002).Several previous studies confirmed that this phenomenon occurred in the basin because PET is not only controlled by air temperature(Wang et al.,2011;Tang et al.,2011;Xing et al.,2014).Furthermore,PET is very sensitive to a few meteorological factors,including wind speed,air humidity,and solar radiation in the semi-arid regions of China(Zhang et al.,2013).Thus,the Thornthwaite method originally used to calculate PET in SPEI is not feasible for drought assessment in regions with similar conditions as the Haihe River Basin.

PET and precipitation have synergistic effects on drought severity and drought processes.Drought severity is susceptible to PET.At most stations in the Haihe River Basin,PET had a significantly negative trend(Xing et al.,2014).Meanwhile,air temperature increased in the period of 1961－2017.Considering these facts,SPEI with the PM equation-based PET calculation should be more feasible for drought assessment.Therefore,SPEI combined with the PM equation is an appropriate drought index for drought severity assessment in the Haihe River Basin.Although this study concluded that both SPI and SPEI are effective in drought detection,the drought severities identified by both indices were different because these two indices require different climatic data.In practice,the choice of drought indices largely depends on data availability and research purpose(Masud et al.,2015;Svoboda and Fuchs,2017).SPEI is more effective in agriculture drought detection and has been broadly used in agricultural research(e.g.,Ming et al.,2015;Labudov?a et al.,2016;P?ascoa et al.,2016).By contrast,SPI is more appropriate for meteorological droughts,and it has been widely used because it only requires precipitation data.

Fig.12.Decadal mean 850-hPa water vapor flux in summer of four typical years.

5.2.Links between meteorological circulation and summer droughts

This study found that the summer droughts detected by SPEI and SPI were significantly intensified,which is consistent with the findings of previous studies in the Haihe River Basin(Wei et al.,2003;Wang et al.,2015).Large-scale atmospheric circulation patterns,to some extent,induce changes in drought processes because they trigger the changes in water vapor for precipitation formation and further affect the key components of the hydrological cycle.The trend of summer droughts in the Haihe River Basin can be attributed to the anomalies in the synchronous large-scale atmospheric circulation,which controls the evolution of dry and wet conditions through time and space by transporting water vapor required for precipitation formation(Liu et al.,2013).In the middle and high latitudes,the pattern of positive geopotential height anomalies is opposite to the northward propagation of the East Asian summer monsoon.Generally,the East Asian summer monsoon that transports water vapor mainly affects summer precipitation over the basin.A weak East Asian summer monsoon would limit the southwesterly moisture-rich air flows to North China(Gao and Wang,2017).The monsoon airflows have remarkably changed in direction and magnitude since the 1970s(Fig.10).Therefore,the transported moisture might not be able to reach the study area as easily as before.Previous studies(Zhang et al.,2003;Huang et al.,2006;Xu et al.,2015)have confirmed the effects of positive geopotential height anomalies in the mid-high latitudes on drought processes in North China.Hence,the positive geopotential height anomalies play an important role by blocking and weakening the water vapor transported to the Haihe River Basin in summer.In low latitudes,the remarkable effects of ENSO and WPSH on summer droughts in North China have been found by several previous studies(Luo,2000;Ouyang et al.,2014;Wang et al.,2015)as well.The correlation between summer droughts and climatic indices indicated that ENSO(Ni~no 3.4)was negatively correlated with SPEI and SPI in summer(Table 2).The intensified summer droughts over the study basin could be attributed to the trend of ENSO-like modes(Yang and Lau,2004).Previous studies(Hui and Sun 2003;Duan et al.,2008)have found that summer precipitation in North China is closely correlated with the position and intensity of WPSH.The westward movement and southward expansion of WPSH might weaken the water vapor that is transported to the Haihe River Basin,which is consistent with the findings of Qian et al.(2009).Ding et al.(2008),Zhang et al.(2008),and Zhang et al.(2011)confirmed that the weakened East Asian summer monsoon and the change of WPSH likely control summer droughts in the Haihe River Basin.The findings of this study help to further explore the mechanisms of the changing feature of droughts associated with large-scale atmospheric circulations and provide a better understanding of hydrometeorological changes in the Haihe River Basin.

6.Conclusions

In this study,two multi-scalar drought indices(SPEI and SPI,respectively)were calculated at 40 meteorological stations in the Haihe River Basin in the period of 1961－2017.The temporal trend and spatial distribution of droughts were thoroughly investigated using a few widely-used statistical methods.Meanwhile,the variability of drought frequency was assessed for the two periods,before and after the abrupt change point in 1977.The correlation between drought indices and climatic patterns was quantified with the Spearman"s rank correlation method at both annual and seasonal scales.Furthermore,the possible causes of change in drought tendency were analyzed.The main conclusions are summarized as follows:

(1)Both SPI and SPEI demonstrated similar temporal variation patterns of droughts.Frequent drought events occurred in the period of 1997－2003.At seasonal scales,an intensified trend of summer droughts was found,and autumn droughts demonstrated an alleviated tendency.In the Haihe River Basin,SPEI presented a larger area with wetting tendency than SPI and identified a smaller area with drying tendency in comparison with SPI.

(2)At annual scales,the frequency of moderate drought events has decreased in the basin since 1977.In terms of spatial distribution,the southern region of the basin might still be threatened by droughts due to the evident decrease in wetness events.By contrast,the northern basin experienced wetter conditions in the period of 2011－2017 than in the previous decades.

(3)The correlation analysis between drought indices and climatic indices indicated that spring and summer droughts were significantly affected by ENSO and WPSH.The largescale atmospheric circulation anomalies largely controlled the tendency of summer droughts.The positive geopotential height anomalies in mid-high latitudes and insufficient water vapor flux play an important role in intensifying the drying conditions.

Acknowledgements

Sincere thanks are given to the following organizations for kindly providing the data for this work:China Meteorological Administration,and the NOAA Oceanic and Atmospheric Research,Earth System Research Laboratories,Physical Sciences Division,and Climate Prediction Center of the United States.Meanwhile,cordial thanks are extended to anonymous reviewers for their critical and constructive comments,which highly improved this manuscript.

Appendix A.Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.wse.2020.12.007.

Declaration of competing interest

The authors declare no conflicts of interest.