Two Severe Prolonged Hydrological Droughts Analysis over Mainland Australia Using GRACE Satellite Data

: In recent years, many droughts have happened over mainland Australia, especially the two severe prolonged droughts, from 2006 to 2009 and 2018 to 2020, resulting in serious water scarcity. Therefore, using the Total Storage Deﬁcit Index (TSDI) from the Gravity Recovery and Climate Experiment (GRACE), we analyzed the two severe prolonged droughts from the perspective of the affected area, spatial evolution, frequency, severity and drought driving factors. The results show that the affected area of Drought 2006–2009 ranged from 57% to 95%, and that of Drought 2018–2020 ranged from 45% to 95%. Drought 2006–2009 took its rise in southeastern Australia and gradually spread to the central part. Drought 2018–2020 originated in the southwest corner of the Northern Territory and northern New South Wales, and gradually expanded to Western Australia and the whole New South Wales respectively. During Drought 2006–2009, Victoria suffered drought all months, including 59% mild drought and 41% moderate drought, North Territory had the highest drought severity of 44.26 and Victoria ranked the second high with the severity of 35.51 (cm months). For Drought 2018–2020, Northern Territory was also dominated by drought all months, including 92% mild drought and 8% moderate drought, the drought severities were in North Territory and Western Australia with 52.19 and 31.44 (cm months), respectively. Finally, the correlation coefﬁcients between the two droughts and Indo-Paciﬁc climate variability including El Niño-Southern Oscillation and Indian Ocean Dipole (IOD) are computed. By comparing the correlation coefﬁcients of Drought 2018–2020 with Drought 2006–2009, we ﬁnd that the impact of the El Niño on the hydrological drought becomes weaker while IOD is stronger, and the role of Southern Oscillation on droughts is diverse with the quite different spatial patterns. The results from Fourier analysis conﬁrm that the two hydrological droughts are all related to Indo-Paciﬁc climate variability but with slightly different driving mechanisms.


Introduction
Drought will potentially cause drastic consequences for global water resources management [1] agriculture production [2], ecological protection [3], as well as economic and social development [4]. Detecting and understanding the characteristics of the previous droughts are the preconditions to predict drought and of great importance for water resource management [5]. In recent decades, the high vulnerability and sensitivity to climate change have made Australia become a hotspot [6]. The Australia Millennium Drought from the mid-1990s to 2009, also known as the Big Dry, was the most severe in the historical record, especially for the two severe prolonged hydrological droughts (during 2006-2009 and 2017-2019) [7][8][9]. Therefore, it is meaningful to determine the drought driving factors and assess the affected area, spatial evolution, frequency and severity of the two prolonged droughts.
The GRACE and GRACE-Follow (GRACE-FO) level-3 products from April 2002 to May 2020, released by the Center for Space Research at University of Texas (CSR), are used to analyze the hydrological drought. The CSR-Mascon RL06 product with 0.25 • spatial resolution is downloaded from the website (https://grace.jpl.nasa.gov/data/get-data/, accessed on 22 February 2021) in the form of EWHs relative to the background field of 2004-2009 mean [47,48]. The gap between GRACE and GRACE-FO is interpolated using the least-squares fit with a trend and annual and semi-annual cycles [49,50]. And the other missing data of GRACE and GRACE-FO are filled by linear interpolation [51].

ENSO and IOD Indices
ENSO is usually associated with tropical and extra-tropical responses [26,52]. The strength of El Niño can be measured by Sea Surface Temperature (SST) anomalies in the equatorial Pacific and here we use the Niño-3.4 index to measure El Niño [33]. El Niño events will happen when the SST anomalies exceed ±0.4 • C for six months or more [26,53]. In addition, SOI is used to measure the Southern Oscillation phenomenon by reflecting atmosphere pressure enhancement and attenuation between the eastern and western tropical Pacific. And it is calculated from the pressure differences between Tahiti and Darwin [54].
IOD is another ocean-atmospheric phenomenon in the tropical Indian Ocean, which can strengthen or reduce the impacts of ENSO. IOD is commonly measured by the Dipole Mode Index (DMI), which is defined by the difference in SST anomaly between the tropical western Indian Ocean (10 • S-10 • N, 50 • -70 • E) and the tropical southeastern Indian Ocean (10 • S-10 • N, 90 • -110 • E) [32]. The warm/cool western/eastern Indian Ocean is associated with enhanced/deficient seasonal rainfall over the areas neighboring the Indian Ocean [55] resulting from the anomalous changes in Walker circulation that lead to anomalous moisture transport and convergence [55]. The details for ENSO and IOD indices are shown in Table 1.

Study Area
Mainland Australia experiences various climatic conditions ranging from wet tropical conditions in the north, arid conditions in the interior, to temperate sub-humid to humid conditions in the south. There are six climate zones over mainland Australia, which were identified based on Köppen classification (Figure 1) [38]. The amount of precipitation over mainland Australia is less than that of other countries. Thus, it is easy for Australia to be affected by severe drought. The climate over mainland Australia is vulnerable to changes in temperature and pressure of the surrounding open oceans. Therefore, drought over Australia is closely related to ENSO and IOD events [60]. In this study, we focus on six larger regions of Australian states and territories, and they are Queensland (QLD), New South Wales (NSW), Victoria (VIC), Northern Territory (NT), South Australia (SA) as well as Western Australia (WA). The smaller regions are not studied due to the spatial resolution limit in the GRACE Mascon product with 0.25 • spatial resolution, and their drought results can refer to the neighboring states or territories.
Remote Sens. 2021, 13, x FOR PEER REVIEW 4 of (SA) as well as Western Australia (WA). The smaller regions are not studied due to th spatial resolution limit in the GRACE Mascon product with 0.25° spatial resolution, an their drought results can refer to the neighboring states or territories.

Methods
The two severe prolonged droughts are assessed from the aspects of the affected are evolution, severity and driving factors. As one of the most important drought indice TSDI reflects drought from the perspective of TWSC. When we take mainland Austral or one region as a whole, the TWSC time series of latitude weighted regional mean terre trial water storage change are calculated based on the 0.25° × 0.25° gridded EWHs to kee consistent with the actual area of the grids. The TSDI is calculated from monthly TWS as follows [61], where TSD , and TWSC , denote the Total Storage Deficit (TSD) and Terrestrial Wat Storage Changes (TWSC) in jth month of the ith year relative to the background field 2004-2009 mean respectively. MTWSC is the mean value of the TWSC series for the j month of all the study years, TSDI represents the hydrological drought index, μ is th mean value of TSD and σ is the standard deviation from the mean field of 2002-04~202 05. The covered area with TSDI less than 0 is defined as a drought-affected area, the exte of drought for a given month is indicated with four drought grades that are shown Table 2 [19]. The TSDI reflects a proxy of drought severity, the larger the TSDI, the mo severe the drought will be.

Methods
The two severe prolonged droughts are assessed from the aspects of the affected area, evolution, severity and driving factors. As one of the most important drought indices, TSDI reflects drought from the perspective of TWSC. When we take mainland Australia or one region as a whole, the TWSC time series of latitude weighted regional mean terrestrial water storage change are calculated based on the 0.25 • × 0.25 • gridded EWHs to keep consistent with the actual area of the grids. The TSDI is calculated from monthly TWSC as follows [61], where TSD i,j and TWSC i,j denote the Total Storage Deficit (TSD) and Terrestrial Water Storage Changes (TWSC) in jth month of the ith year relative to the background field of 2004-2009 mean respectively. MTWSC j is the mean value of the TWSC series for the jth month of all the study years, TSDI represents the hydrological drought index, µ is the mean value of TSD and σ is the standard deviation from the mean field of 2002-04~2020-05. The covered area with TSDI less than 0 is defined as a drought-affected area, the extent of drought for a given month is indicated with four drought grades that are shown in Table 2 [19]. The TSDI reflects a proxy of drought severity, the larger the TSDI, the more severe the drought will be. A continual negative TSD lasting for three or more months means a period of drought, and the drought severity is estimated with the following expression [11], where Se(t) represents the severity of the hydrological drought period t, Me(t) and D(t) are the mean TSD and duration for a drought period, respectively. Therefore, the unit of drought severity is (cm months).
Since hydrological drought is driven by different factors, the contribution of each drought driving factor can be determined by the partial correlation technique [62], and the correlation coefficient is determined as follows [63], where ρ xy is the correlation coefficient between the time series of TSDI x(t i ) and drought driving index y(t i ), and τ is the lagging time, x and y are the mean of x(t i ) and y(t i ) of the n epochs of the time series. The lagging time is determined based on the maximum correlation of the two series and its threshold is set to be 7 months [26,28,64]. The GRACE-TWS time series is separated into the non-parametric seasonal components, trends and residuals [65], where TWSC total (t i ), TWSC seasonal (t i ), TWSC trend (t i ) and Residuals(t i ) denote the GRACE-TWS time series, non-parametric seasonal components, trends and residuals respectively. The unknown parameters in the seasonal and trend components are estimated with leastsquares fitting. After the seasonal and trend components are removed from the GRACE-TWS time series, the fitting residuals are then used to compute RMS (Root Mean Square) error [18].

Drought Conditions over Mainland Australia
As shown in Figure 2, there are eight drought periods detected from the GRACE-based TSD time series over mainland Australia from April 2002 to May 2020, which have caused severe effects and got more attention at present [7,39]. The onset, termination and duration are summarized in Table 3 for the droughts over mainland Australia from April 2002 to May 2020. The first two most severe droughts that lasted for 32 months (from June 2006 to January 2009) and 24 months (from June 2018 to May 2020) are marked by the orange and green dotted line box in Figure 2  duration are summarized in Table 3 for the droughts over mainland Australia from April 2002 to May 2020. The first two most severe droughts that lasted for 32 months (from June 2006 to January 2009) and 24 months (from June 2018 to May 2020) are marked by the orange and green dotted line box in Figure 2, the correspondent drought severities are 59.50 and 50.82 (cm months) respectively. To facilitate the description, the two droughts are named Drought 2006-2009 and Drought 2018-2020 respectively.

Drought Affected Area
The percentage of the drought-affected area relative to the total area over mainland Australia [66] is used to confirm whether the drought is continent-wide or regional. The percentages of the drought-affected area are presented in

Drought Affected Area
The percentage of the drought-affected area relative to the total area over mainland Australia [66] is used to confirm whether the drought is continent-wide or regional. The percentages of the drought-affected area are presented in     Australia and gradually spread to the central part. Victoria and the Northern Territory suffered the most severe drought in terms of drought-affected areas and duration. In the six bottom sub-figures, Drought 2018-2020 originated in the southwest corner of the Northern Territory and northern New South Wales, and gradually expanded to Western Australia and the whole New South Wales respectively. The drought over Western Australia had been increasing until May 2020, but that of New South Wales disappeared in early 2020. The two hydrological droughts evolved and developed continually, yet the main affected areas of the two hydrological droughts were quite different. suffered the most severe drought in terms of drought-affected areas and duration. In the six bottom sub-figures, Drought 2018-2020 originated in the southwest corner of the Northern Territory and northern New South Wales, and gradually expanded to Western Australia and the whole New South Wales respectively. The drought over Western Australia had been increasing until May 2020, but that of New South Wales disappeared in early 2020. The two hydrological droughts evolved and developed continually, yet the main affected areas of the two hydrological droughts were quite different.

Frequency of Different Drought Grades
The drought frequency is defined as the ratio of the months at a drought grade to the total months of the drought period. According to the drought grads in Table 2

Frequency of Different Drought Grades
The drought frequency is defined as the ratio of the months at a drought grade to the total months of the drought period. According to the drought grads in Table 2 regions. Victoria and South Australia suffered drought all the months. Specifically, Victoria had 59% mild drought and 41% moderate drought while those of South Australia were 88% and 12% respectively. Since Drought 2006-2009 originated in the east and gradually developed to the center of mainland Australia, Western Australia had the least drought frequency. For Drought 2018-2020, Northern Territory was dominated by drought all the months, including 8% moderate drought and 92% mild drought and Victoria had the least drought frequency, while it had the highest drought frequency for Drought 2006-2009 (see the third pie in the upper of Figure 5).  Figure 5).

Drought Severity
A drought with a short duration (less than three months) usually has a relatively small impact on water storage and agriculture, because the short-term deficit of water storage can be replenished by timely rainfall. However, continuous drought often leads to serious consequences, especially for agricultural production. The severity and duration of drought are comprehensively considered in the drought severity calculation and the drought severities for all six regions are shown in Table 4

Drought Severity
A drought with a short duration (less than three months) usually has a relatively small impact on water storage and agriculture, because the short-term deficit of water storage can be replenished by timely rainfall. However, continuous drought often leads to serious consequences, especially for agricultural production. The severity and duration of drought are comprehensively considered in the drought severity calculation and the drought severities for all six regions are shown in Table 4, with corresponding time and austral season (September to November for spring, December to February for summer, March to May for autumn, and June to August for winter).  Temporal variability of drought severities in six regions is shown in Figure 6 for the two prolonged droughts. Same as that in Table 4, Northern Territory and Western Australia suffered severe drought while South Australia was affected the weakest. Besides, For Drought 2006-2009, Queensland had four drought periods, and other regions only had two or three periods. For Drought 2018-2020, all regions except for South Australia had two drought periods. Because drought severity is the cumulation of TSD during the drought period, longer drought duration and larger TSD will cause higher drought severity. For Drought 2018-2020, Northern Territory suffered the highest drought severity (52.19 cm months) because of very larger TSD, although the duration shorter than that in Western Australia. In other words, the hydrological drought severity in Northern Territory developed more violent than in West Australia. Temporal variability of drought severities in six regions is shown in Figure 6 for the two prolonged droughts. Same as that in Table 4, Northern Territory and Western Australia suffered severe drought while South Australia was affected the weakest. Besides, For Drought 2006-2009, Queensland had four drought periods, and other regions only had two or three periods. For Drought 2018-2020, all regions except for South Australia had two drought periods. Because drought severity is the cumulation of TSD during the drought period, longer drought duration and larger TSD will cause higher drought severity. For Drought 2018-2020, Northern Territory suffered the highest drought severity (52.19 cm months) because of very larger TSD, although the duration shorter than that in Western Australia. In other words, the hydrological drought severity in Northern Territory developed more violent than in West Australia.

Relationships between the Two Hydrological Droughts and Indo-Pacific Climate Variability
Temporal variability of Indo-Pacific climate indices with TSDI from April 2002 to May 2020 are shown in Figure 7 where the Drought 2006-2009 and Drought 2018-2020

Relationships between the Two Hydrological Droughts and Indo-Pacific Climate Variability
Temporal variability of Indo-Pacific climate indices with TSDI from April 2002 to May 2020 are shown in Figure 7 where the Drought 2006-2009 and Drought 2018-2020 are highlighted with orange and green backgrounds, respectively. From Figure 7 we can find that Nino3.4 and SOI during Drought 2006-2009 are more changeable and stronger than those during Drought 2018-2020. IOD is positive during almost all of the two droughts, consistent with that of Ummenhofer et al. [29] who found Australia usually experiences anomalous dry conditions during positive IOD events.
To quantify the influence of ENSO and IOD on Drought 2006-2009 and Drought 2018-2020 over mainland Australia, we first determine the lagging times between TSDI and the corresponding climate indices (Nino3.4, SOI and IOD) by maximizing the mean absolute value of correlation coefficients over mainland Australia. Then we fix the lagging times to compute the correlation coefficients over mainland Australia. Since the lagging times between TSDI and climate indices are not larger than seven months [26,28,64], we present the absolute correlation coefficients from zero to seven months in Figure 8 and summarize the determined lagging times in Table 5.
are highlighted with orange and green backgrounds, respectively. From Figure 7 we can find that Nino3.4 and SOI during Drought 2006-2009 are more changeable and stronger than those during Drought 2018-2020. IOD is positive during almost all of the two droughts, consistent with that of Ummenhofer et al. [29] who found Australia usually experiences anomalous dry conditions during positive IOD events. To quantify the influence of ENSO and IOD on Drought 2006-2009 and Drought 2018-2020 over mainland Australia, we first determine the lagging times between TSDI and the corresponding climate indices (Nino3.4, SOI and IOD) by maximizing the mean absolute value of correlation coefficients over mainland Australia. Then we fix the lagging times to compute the correlation coefficients over mainland Australia. Since the lagging times between TSDI and climate indices are not larger than seven months [26,28,64], we  present the absolute correlation coefficients from zero to seven months in Figure 8 and summarize the determined lagging times in Table 5.  From Figure 8 we can observe that three correlation coefficients vary more obviously with different lagging times in the left figure than that in the right figure, which means that the lagging times determined for Drought 2006-2009 are probably better than that for Drought 2018-2020. Considering the differences of the three lagging times in Table 5 are only one month, the determined lagging times are reliable since TSDI is served from monthly GRACE data. Moreover, the correlation coefficients of Nino3.4 are larger than that of IOD in Figure 8a, which, however, is opposite in Figure 8b  From Figure 8 we can observe that three correlation coefficients vary more obviously with different lagging times in the left figure than that in the right figure, which means that the lagging times determined for Drought 2006-2009 are probably better than that for Drought 2018-2020. Considering the differences of the three lagging times in Table 5 are only one month, the determined lagging times are reliable since TSDI is served from monthly GRACE data. Moreover, the correlation coefficients of Nino3.4 are larger than that of IOD in Figure 8a, which, however, is opposite in Figure 8b Table 5 that the correlation coefficients of Nino3.4 reduce from 0.39 to 0.26, however those of IOD increase from 0.27 to 0.31. Therefore, over mainland Australia the impact of ENSO on the hydrological droughts may have decreased from 2002 to 2020, on the contrary, the impact of IOD has increased.
The correlation coefficients between TSDI and the indices for modes of Indo-Pacific climate variability (Nino3.4, SOI and IOD) over mainland Australia are shown in the upper and bottom of Figure 9 for The mean absolute correlation coefficients between TSDI and three climate indices are shown in Figure 10  The mean absolute correlation coefficients between TSDI and three climate indices are shown in Figure 10

RMS of the Hydrological Drought Analysis
The RMS presented in Figure 11 is derived from the residuals of Equation (5) after removing the trends and seasonal terms, and the latitude weighted RMS values of the six regions are shown in Table 6. Since the residuals may still contain sub-seasonal and interannual signals, these RMS values may overestimate the actual uncertainties associated

RMS of the Hydrological Drought Analysis
The RMS presented in Figure 11 is derived from the residuals of Equation (5) after removing the trends and seasonal terms, and the latitude weighted RMS values of the six regions are shown in Table 6. Since the residuals may still contain sub-seasonal and inter-annual signals, these RMS values may overestimate the actual uncertainties associated with the CSR mascon. As a whole, the RMS values over mainland Australia range from 0 to 3 cm. And they are relatively larger in northern mainland Australia and relatively smaller in the southwest. We can see from Figure 11 that the RMS near the equator is larger because the GRACE satellites are polar-orbiting and the spatial sampling in the equator is relatively sparse [67,68]. Besides, it may also because the larger inter-annual variation of terrestrial water storage in the Northern Territory makes its RMS larger. For the region scale, the RMS in the Northern Territory is 2.11 cm, and those of other regions are less than 1.5 cm, which is consistent with the present accuracy of the GRACE data (cm level).
regions are shown in Table 6. Since the residuals annual signals, these RMS values may overesti with the CSR mascon. As a whole, the RMS valu to 3 cm. And they are relatively larger in nor smaller in the southwest. We can see from Fig larger because the GRACE satellites are polarequator is relatively sparse [67,68]. Besides, it m variation of terrestrial water storage in the Nort the region scale, the RMS in the Northern Territ are less than 1.5 cm, which is consistent with the level).   First, we compared the results of the linkage between the two droughts and the Indo-Pacific climate variability with the previous study. The two hydrological droughts in northern and eastern mainland Australia are influenced by ENSO since significant correlations are between TSDI and both of Nino3.4 and SOI over the two droughts (see Figure 9), consistent with Risbey et al. [33] who found ENSO is related to rainfall over much of the continent at different times, particularly in the north and east [33]. Moreover, Forootan et al. [69] found a strong presence of ENSO-induced TWS derived from GRACE data over Australia is increasing from 1990 to 2010 (Drought 2006-2009 in this work) and IOD-induced TWS is increasing from 2004 to 2014. We find that the hydrological drought in the southwestern corner of mainland Australia is primarily influenced by Pacific climate variability during Drought 2006-2009, but with the stronger impact of IOD for Drought 2018-2020. The large precipitation variations associated with ENSO and IOD were found over the northern and northeast coast during 2002-2011 [26], but we found that the TSDI in the northeast coast has little linkage with ENSO and IOD. One reason to accounts for this is that TSDI is also influenced by evapotranspiration and runoff, which may lead to the differences between precipitation and hydrological drought derived from GRACE-based TWS [70]. Besides, the study periods here are determined based on the drought severity over mainland Australia and they are different from that of [26], which may also lead to the difference.
Furthermore, we analyzed the spatial pattern of the correlation characteristic and tried to give the prediction of the impact of ENSO and IOD on both of the two droughts, northern and eastern mainland Australia were impacted by El Niño and IOD, indicating the relatively stable impacts of them. So, we can carefully deduce that the hydrological drought in northern and eastern mainland Australia may still be related to the changes of El Niño and IOD variability in the future. The role of Southern Oscillation is diverse, for Drought 2006-2009, the correlation coefficients of TSDI and SOI in western and central parts of Australia are negative and that in the northeastern part are positive. However, for Drought 2018-2020, the pattern is opposite, indicating the more complex and diverse driving mechanism of Southern Oscillation. And these conclusions are consistent with the findings that ocean-based ENSO indices (e.g., Nino3.4) are more skillful as predictors of Australian rainfall than the SOI at a longer lead time [71]. And the correlation coefficients between TSDI and both of Nino3.4 and IOD in western, northern and central parts of mainland Australia are negative, and those of the other places are positive, this phenomenon was also shown in [26,33]. It should be noted that isolating the physical mechanisms through which a specific global climate index might influence the hydrological droughts is beyond the scope of the present study. Instead, the study focused on isolating the possible influence of global climate indices and teleconnections on TSDI over mainland Australia based on the correlation analysis for the two hydrological droughts.
To explore the non-linear relationships between the two hydrological droughts and Indo-Pacific climate variability, we further use the Fourier analysis to analyze the four indices (TSDI, Nino3.4, SOI and IOD) over the periods of the two hydrological droughts. The corresponding power spectra are shown in Figure 12 shows that Drought 2006-2009 was driven by Indo-Pacific climate variability to some extent. However, for Drought 2018-2020, the power spectrum of TSDI matches with that of Nino3.4 and IOD well at the period about 0.90 years but approximately matches with that of SOI at the period about 0.42 years, confirming a slight difference of driving mechanism between Drought 2006-2009 and Drought 2018-2020.

Conclusions
Focusing on two severe and prolonged droughts over mainland Australia, this paper analyzed the hydrological droughts from the perspective of the affected area, spatial evolution, frequency, severity as well as the relationship between hydrological droughts and Indo-Pacific climate indices.
The percentages of affected areas of total mainland Australia for Drought 2006-2009 range from 57% to 95%, while those of Drought 2018-2020 were from 45% to 95%. As for

Conclusions
Focusing on two severe and prolonged droughts over mainland Australia, this paper analyzed the hydrological droughts from the perspective of the affected area, spatial evolution, frequency, severity as well as the relationship between hydrological droughts and Indo-Pacific climate indices.
The percentages of affected areas of total mainland Australia for Drought 2006-2009 range from 57% to 95%, while those of Drought 2018-2020 were from 45% to 95%. As for the spatial evolutions of the two droughts, Drought 2006-2009 took its rise in southeastern mainland Australia and gradually spread to the interior. Drought 2018-2020 originated in the southwest corner of the Northern Territory and northern New South Wales, and gradually expanded to Western Australia and New South Wales respectively. For Drought 2006-2009, Victoria and South Australia suffered drought for all 32 months. The mild and moderate droughts in Victoria were 59% and 41%, while those in South Australia were 88% and 12% respectively. Yet Western Australia had the least number of drought months among the six regions. For Drought 2018-2020, Northern Territory was dry all the months, including 8% moderate and 92% mild drought respectively. The drought severities of North Territory during Drought 2006-2009 and 2018-2020 were both the highest than the other regions, with the severities of 44.26 and 52.19 (cm months) respectively. For Drought 2006-2009, the drought severity of Victoria ranked the second, while that of South Australia was the lowest. For Drought 2018-2020, the drought severity of Western Australia was the second high, while those of Queensland and Victoria were both the lowest. Considering the drought season, we found that summer is usually the turning season with the most severe droughts, and the hydrological conditions over mainland Australia usually change from drought to wet at the end of summer. The correlation patterns of El Niño are similar in the northwest and southeast areas for the two droughts, while that of SOI are significantly different. For IOD, correlation patterns are similar in the northwest and eastern edge, but with opposite correlations in the southwest corner. The impacts of Pacific climate variability on the hydrological droughts have decreased from Drought2006-2009 to Drought 2018-2020, on the contrary, the impact of Indo climate variability has increased. The results of Fourier analysis show that the two hydrological droughts are all related to Indo-Pacific climate variability, while the driving mechanism of Drought 2018-2020 is slightly different from that of Drought 2018-2020 to some extent.