Analysis of Drought-Sensitive Areas and Evolution Patterns through Statistical Simulations of the Indian Ocean Dipole Mode

: In this study, we investigated extreme droughts in the Indochina peninsula and their relationship with the Indian Ocean Dipole (IOD) mode. Areas most vulnerable to drought were analyzed via statistical simulations of the IOD based on historical observations. Results of the long-term trend analysis indicate that areas with increasing spring (March–May) rainfall are mainly distributed along the eastern coast (Vietnam) and the northwestern portions of the Indochina Peninsula (ICP), while Central and Northern Laos and Northern Cambodia have witnessed a reduction in spring rainfall over the past few decades. This trend is similar to that of extreme drought. During positive IOD years, the frequency of extreme droughts was reduced throughout Vietnam and in the southwestern parts of China, while increased drought was observed in Cambodia, Central Laos, and along the coastline adjacent to the Myanmar Sea. Results for negative IOD years were similar to changes observed for positive IOD years; however, the eastern and northern parts of the ICP experienced reduced droughts. In addition, the results of the statistical simulations proposed in this study successfully simulate drought-sensitive areas and evolution patterns of various IOD changes. The results of this study can help improve diagnostic techniques for extreme droughts in the ICP. Niño Niño running average positive and IODs. Myanmar, and For extraction of two types of El Niño events, a simple transformation proposed by Ren and Jin [26] was applied using the Niño 3 and Niño 4 indices. To classify the IOD events, we used the monthly SST data obtained from the National Oceanic and Atmospheric Administration in the TIO region. The three-month running average was applied to the peak phase of the IOD mode during August–October, with ±1 σ (standard deviation) to identify positive and negative patterns of the IODs. Finally, 12 events (1961, 1963, 1967, 1972, 1977, 1982, 1994, 1997, 2006, 2012, 2015, and 2017) and 11 events (1954, 1958, 1960, 1964, 1971, 1975, 1989, 1992, 1996, 1998, and 2010) were selected as the strongest positive IOD years and negative IOD years, respectively. The selected IOD years were used for composite analysis


Introduction
Since the 1950s, an increase in atmospheric moisture demand and changes in atmospheric circulation patterns due to global warming have contributed to increased aridity in many areas [1,2]. Furthermore, drought risk in the twenty-first century, indicated by model-simulated drought indices [3] and soil moisture [4], is expected to increase.
Drought is often divided into four types [5]: meteorological or climatological, agricultural, hydrological, and socio-economic. Of these four types, we chose to analyze meteorological drought because this type of drought results in water shortages due to reduced rainfall [6]. To better anticipate drought-related disasters in Southeast Asia, we studied the Indochina Peninsula (ICP). Several studies focusing on the impacts of droughts have been conducted in the ICP. Recently, Le et al. [7] attempted to ascertain the relationship between meteorological drought in the Khánh Hòa Province (Vietnam) and climate signals, such as the Southern Oscillation Index and the Bivariate El Niño Southern Oscillation (ENSO) time series. Their work shed light on the importance of climate signals of oceanic and Atmospheric Administration Earth System Research Laboratory, Boulder, Colorado, USA, to identify spring (March-May) precipitation variability over the ICP region, which consists of six countries: Myanmar, Thailand, Laos, Cambodia, Vietnam, and Malaysia ( Figure 1). For extraction of two types of El Niño events, a simple transformation proposed by Ren and Jin [26] was applied using the Niño 3 and Niño 4 indices. To classify the IOD events, we used the monthly SST data obtained from the National Oceanic and Atmospheric Administration in the TIO region. The three-month running average was applied to the peak phase of the IOD mode during August-October, with ±1σ (standard deviation) to identify positive and negative patterns of the IODs. Finally, 12 events (1961, 1963, 1967, 1972, 1977, 1982, 1994, 1997, 2006, 2012, 2015, and 2017) and 11 events (1954, 1958, 1960, 1964, 1971, 1975, 1989, 1992, 1996, 1998, and 2010) were selected as the strongest positive IOD years and negative IOD years, respectively. The selected IOD years were used for composite analysis to investigate changes in the long-term mean (1981-2010) of seasonal precipitation and extreme drought in the ICP.
Water 2019, 11, x FOR PEER REVIEW  3 of 10 Myanmar, Thailand, Laos, Cambodia, Vietnam, and Malaysia ( Figure 1). For extraction of two types of El Niño events, a simple transformation proposed by Ren and Jin [26] was applied using the Niño 3 and Niño 4 indices. To classify the IOD events, we used the monthly SST data obtained from the National Oceanic and Atmospheric Administration in the TIO region. The three-month running average was applied to the peak phase of the IOD mode during August-October, with ±1σ (standard deviation) to identify positive and negative patterns of the IODs. Finally, 12 events (1961, 1963, 1967, 1972, 1977, 1982, 1994, 1997, 2006, 2012, 2015, and 2017) and 11 events (1954, 1958, 1960, 1964, 1971, 1975, 1989, 1992, 1996, 1998, and 2010) were selected as the strongest positive IOD years and negative IOD years, respectively. The selected IOD years were used for composite analysis to investigate changes in the long-term mean (1981-2010) of seasonal precipitation and extreme drought in the ICP. In this study, the Standardized Precipitation Index (SPI) [27] was adopted to extract drought information on a daily basis for each drought duration, using precipitation data from 1979-2018. During the calculation of the SPI, n-day precipitation data was aggregated after each day from March 1 to May 31 to obtain the probability fitted to the gamma distribution, which was then converted to a standardized normal distribution with a mean of 0 and a standard deviation of 1 [28]. In this study, the extreme drought is defined as the number of drought events occurring during March-May by applying a −1.5 threshold to the SPI calculated on a 30-day basis. The technique of approximate conversion that transforms the cumulative probability of precipitation into a standardized variable Z was developed by McKee et al. [27].
where x is precipitation, and H(x) is the cumulative probability of the observed precipitation. In this study, the Standardized Precipitation Index (SPI) [27] was adopted to extract drought information on a daily basis for each drought duration, using precipitation data from 1979-2018. During the calculation of the SPI, n-day precipitation data was aggregated after each day from March 1 to May 31 to obtain the probability fitted to the gamma distribution, which was then converted to a standardized normal distribution with a mean of 0 and a standard deviation of 1 [28]. In this study, the extreme drought is defined as the number of drought events occurring during March-May by applying a −1.5 threshold to the SPI calculated on a 30-day basis. The technique of approximate conversion that transforms the cumulative probability of precipitation into a standardized variable Z was developed by McKee et al. [27]. where (1−H(x)) 2 , f or 0.5 < H(x) < 1.
(2) where x is precipitation, and H(x) is the cumulative probability of the observed precipitation. Trend testing aims to specify whether an increasing or decreasing trend exists in time series data. Since parametric tests require several assumptions involving normality, stationarity, and/or independence of variables, non-parametric methods are preferred in meteorological and hydrological studies. One of the widely used nonparametric trend tests is the Mann-Kendall (M-K) test [29,30]. Herein, we adopted a modified M-K test to consider autocorrelation in time-series data [31]. The accuracy of the modified M-K test was significantly improved by using autocorrelated data to test the null hypothesis (H 0 ) that no trend exists in the sample data. Using the modified M-K test, we classified the confidence levels of the trends for the 90% and 95% intervals. In this study, trend analysis was performed to identify spatio-temporal variations in seasonal precipitation and extreme droughts in the ICP.
The Intentionally Biased Bootstrapping (IBB) method was developed by Hall and Presnell [32] to adjust a "uniform" bootstrap by reducing bias or variance, or by setting some characteristics as predetermined quantities. The IBB method makes it possible to apply the nonlinear relationship between two parameters as the resampling probability to predict one of the parameters. Using the IBB method, the change in the mean of the predicted parameter can be estimated according to the change in the mean of the base parameter. Recently, Lee [33] employed the IBB method to assess the impact of regional temperature on rainfall based on historical data before evaluating future changes in the global warming scenarios of seasonal rainfall over the Korean Peninsula. The IBB approach can assess the impact of rainfall as temperatures increase with the current climate horizon. Thus, the IBB method is suitable for determining the sensitivity of regional rainfall with changes in SST in the Indian Ocean.
Briefly, after obtaining expected increase or decrease (∆µ) of the mean of the base parameter, we resampled the parameter pairs with weights based on the sequence of the base parameter (arranged from small to large) as follows: where n is the number of parameter pairs, i = 1, . . . , n, and the weight order (r) is used to generalize the weights. Then, using the equation below, we estimated r, and hence the resample weights.

Trends in Spring Precipitation and Extreme Drought
Trends in precipitation during spring (March-May) and extreme drought  in the ICP are depicted in Figure 2. As shown in Figure 2a, the regions exhibiting increased spring rainfall are mainly distributed along the east coast (Vietnam) and the northwest regions of the ICP, while central and Northern Laos and Northern Cambodia experienced a decrease in spring precipitation during the past four decades. This trend is similar to that of the extreme drought shown in Figure 2b, in which Vietnam experienced fewer extreme droughts while the central and northern parts of Laos suffered more extreme droughts. The increasing trend in extreme droughts appears across the ICP, except for Vietnam, but this is not statistically significant.  Figure 3 shows the composite anomalies of extreme drought during the spring season (March-May) for positive and negative IOD years. During the positive IOD years, negative anomaly patterns of drought were dominant throughout Vietnam and the southwestern region of the Yunnan Province in China. However, a positive anomaly became evident over central Cambodia and Laos, with distinct abnormalities occurring along the coastline adjacent to the Myanmar Sea; in some areas, abnormalities recorded were more than 160% higher than usual. Furthermore, if both positive IOD years and El Niño events occurred simultaneously, the spatial distribution of extreme drought was observed to intensify only compared to drought anomaly patterns during the positive IOD years. The change in extreme drought during negative IOD years was similar to the change in positive IOD years. Positive anomaly of extreme drought around Myanmar's coastline was weakened, and positive anomalies were more evident across Cambodia. The results for negative IOD years were similar to the changes for positive IOD years; however, the anomalies increased towards Northern Laos, Northeastern Myanmar, and Northern Thailand. These changes were more evident when the negative IOD years coincided with La Niña events. The modified Mann-Kendall (M-K) test was applied; statistically significant changes are highlighted in red (increasing trend) and blue (decreasing trend) for 90% (significant) and 95% (highly significant) confidence intervals. Figure 3 shows the composite anomalies of extreme drought during the spring season (March-May) for positive and negative IOD years. During the positive IOD years, negative anomaly patterns of drought were dominant throughout Vietnam and the southwestern region of the Yunnan Province in China. However, a positive anomaly became evident over central Cambodia and Laos, with distinct abnormalities occurring along the coastline adjacent to the Myanmar Sea; in some areas, abnormalities recorded were more than 160% higher than usual. Furthermore, if both positive IOD years and El Niño events occurred simultaneously, the spatial distribution of extreme drought was observed to intensify only compared to drought anomaly patterns during the positive IOD years. The change in extreme drought during negative IOD years was similar to the change in positive IOD years. Positive anomaly of extreme drought around Myanmar's coastline was weakened, and positive anomalies were more evident across Cambodia. The results for negative IOD years were similar to the changes for positive IOD years; however, the anomalies increased towards Northern Laos, Northeastern Myanmar, and Northern Thailand. These changes were more evident when the negative IOD years coincided with La Niña events.  Figure 3 shows the composite anomalies of extreme drought during the spring season (March-May) for positive and negative IOD years. During the positive IOD years, negative anomaly patterns of drought were dominant throughout Vietnam and the southwestern region of the Yunnan Province in China. However, a positive anomaly became evident over central Cambodia and Laos, with distinct abnormalities occurring along the coastline adjacent to the Myanmar Sea; in some areas, abnormalities recorded were more than 160% higher than usual. Furthermore, if both positive IOD years and El Niño events occurred simultaneously, the spatial distribution of extreme drought was observed to intensify only compared to drought anomaly patterns during the positive IOD years. The change in extreme drought during negative IOD years was similar to the change in positive IOD years. Positive anomaly of extreme drought around Myanmar's coastline was weakened, and positive anomalies were more evident across Cambodia. The results for negative IOD years were similar to the changes for positive IOD years; however, the anomalies increased towards Northern Laos, Northeastern Myanmar, and Northern Thailand. These changes were more evident when the negative IOD years coincided with La Niña events.   (Figure 4) resulted in more extreme drought across Laos and along the coastline adjacent to the Myanmar Sea, and less extreme droughts than usual across Southern Cambodia, Vietnam, and Northwest Myanmar, adjacent to the Bay of Bengal. As IOD increases, drought changes appear more severe. The northwestern region of Myanmar was observed to have a statistically significant increase in extreme drought for more than 0.5σ of the IOD. Some parts of Southern Thailand and Western Cambodia showed a statistically significant negative anomaly pattern of extreme drought. These patterns tended to be reinforced when drought symptoms appeared from +0.75σ to +1σ of the IOD. In addition, statistically significant patterns of positive anomalies in extreme droughts in some areas along the coastline of the Myanmar Sea were observed in cases of more than +1σ of IOD changes.    (Figure 4) resulted in more extreme drought across Laos and along the coastline adjacent to the Myanmar Sea, and less extreme droughts than usual across Southern Cambodia, Vietnam, and Northwest Myanmar, adjacent to the Bay of Bengal. As IOD increases, drought changes appear more severe. The northwestern region of Myanmar was observed to have a statistically significant increase in extreme drought for more than 0.5 of the IOD. Some parts of Southern Thailand and Western Cambodia showed a statistically significant negative anomaly pattern of extreme drought. These patterns tended to be reinforced when drought symptoms appeared from +0.75 to +1 of the IOD. In addition, statistically significant patterns of positive anomalies in extreme droughts in some areas along the coastline of the Myanmar Sea were observed in cases of more than +1 of IOD changes.    1981-2010 norms). The climatology mean of IOD (1981-2010) was deliberately increased by 0.25σ through IBB simulations to analyze the sensitivity of extreme drought to the warming phase of IOD in the tropical Indian Ocean. In each figure, the median results derived from 1000 simulations are shown and the statistically significant area of change at 95% significance level is represented by "x". Figure 5 shows a 0.25σ decrease in the IOD. When negative IODs were strengthened, it was found that extreme droughts were less frequent throughout Laos, Vietnam, and the southwestern region of the Yunnan Province in China. However, in Northwestern Myanmar, some parts of Thailand, and in regions adjacent to the Andaman Sea, extreme drought increased, leading to greater drought vulnerability. In addition, IOD-sensitive areas with more extreme droughts emerged in the northwestern parts of Myanmar, especially as changes in negative IOD became apparent. However, the eastern and northern parts of the ICP showed a contrasting response to negative IOD. The simulated results indicated that extreme droughts throughout Northwestern Myanmar and Thailand were more frequent than usual.

Changes in Extreme Drought Using the IBB method
Water 2019, 11, x FOR PEER REVIEW 7 of 10 Thailand, and in regions adjacent to the Andaman Sea, extreme drought increased, leading to greater drought vulnerability. In addition, IOD-sensitive areas with more extreme droughts emerged in the northwestern parts of Myanmar, especially as changes in negative IOD became apparent. However, the eastern and northern parts of the ICP showed a contrasting response to negative IOD. The simulated results indicated that extreme droughts throughout Northwestern Myanmar and Thailand were more frequent than usual.

Discussion and Conclusions
IOD in the Indian Ocean and ENSO in the Pacific Ocean are the results of atmospheric-ocean interactions in each region. Both phenomena can be identified by SST anomalies and are reported to have a direct impact on rainfall occurring globally [14,17,[34][35][36]. The ENSO phenomenon has been officially observed since 1985 [37] and is relatively well known for its anomalous convection, wind patterns, and predictable short-term behaviors [38][39][40][41] within the global climate system. However, there is still a lack of understanding of interactions, evolutionary patterns, and predictability of new forms of El Niño [42].
Long-term drought conditions in Southeast Asia generally relate to El Niño, while wetter conditions tend to be associated with La Nina [43]. These changes are consistent with the results of the composite analysis of IOD and ENSO conducted in this study. When positive IOD and El Niño occur simultaneously, the spatial distribution of spring rainfall is significantly reduced compared to the rainfall anomaly pattern that takes into account positive IOD years only. Results for negative IOD years are similar to the changes observed for positive IOD years, but it is expanding to Northern Laos, Northeastern Myanmar, and Northern Thailand. These changes are more evident when the negative IOD years coincide with La Niña events.

Discussion and Conclusions
IOD in the Indian Ocean and ENSO in the Pacific Ocean are the results of atmospheric-ocean interactions in each region. Both phenomena can be identified by SST anomalies and are reported to have a direct impact on rainfall occurring globally [14,17,[34][35][36]. The ENSO phenomenon has been officially observed since 1985 [37] and is relatively well known for its anomalous convection, wind patterns, and predictable short-term behaviors [38][39][40][41] within the global climate system. However, there is still a lack of understanding of interactions, evolutionary patterns, and predictability of new forms of El Niño [42].
Long-term drought conditions in Southeast Asia generally relate to El Niño, while wetter conditions tend to be associated with La Nina [43]. These changes are consistent with the results of the composite analysis of IOD and ENSO conducted in this study. When positive IOD and El Niño occur simultaneously, the spatial distribution of spring rainfall is significantly reduced compared to the rainfall anomaly pattern that takes into account positive IOD years only. Results for negative IOD years are similar to the changes observed for positive IOD years, but it is expanding to Northern Laos, Northeastern Myanmar, and Northern Thailand. These changes are more evident when the negative IOD years coincide with La Niña events.
Although different forms of El Niño have been reported to occur with increasing frequency in the central Pacific since the 2000s [36,40], the exact causes of these occurrences and their impacts on the Indian Ocean and Pacific Rim countries-according to the evolution patterns of different types of El Niño events-are still unknown. Identification of different types of El Niño is evidence that ENSO events are not well understood and are certainly unpredictable [42]. Furthermore, ENSO and IOD-rainfall linkages in the ICP show significant spatial and seasonal variations. Recent studies suggested that the slowdown in the observed global average atmospheric temperature rise is closely related to substantial heat transfer from the Pacific Ocean to the Indian Ocean through Indonesian throughflow [44][45][46]. Despite similarities with ENSO, the IOD is an even more recent discovery with even less understood implications [14,18].
The spatio-temporal connection between SST and winds shows a strong coupling of precipitation and ocean dynamics in the Indian Ocean. However, with regards to the IOD phenomena, few quantitative studies provide a complete picture of extreme drought fluctuations across the ICP. Therefore, in this study, areas most vulnerable to drought were analyzed through statistical simulations of IODs using the IBB method based on historical observations.
The results of the long-term trend analysis show that the areas where spring rainfall has increased are mainly distributed along the eastern coast (Vietnam) and the northwestern portions of the ICP, while central and Northern Laos and Northern Cambodia have experienced a reduction in spring rainfall over the past few decades. This is because of the Truong-Son Mountains in Vietnam, which are adjacent to the South China Sea, have relatively abundant rainfall, whereas Laos has seen relatively little rainfall. This trend is similar to that of extreme droughts. Vietnam experienced less severe droughts, but central and Northern Laos suffered more severe droughts. During positive IOD years, extreme drought occurred less than usual throughout Vietnam and Southwestern China, while more droughts were observed in Cambodia and central Laos, as well as along the coastline adjacent to the Myanmar Sea. Results for negative IOD years are similar to changes observed for positive IOD years; however, the eastern and northern parts of the ICP are experiencing a decrease in droughts. In addition, the results of the statistical simulations proposed in this study successfully simulate drought-sensitive areas and evolution patterns for various IOD changes.
Investigating SST changes in the Indo-Pacific Ocean can help improve the understanding of regional-scale climatic variability as well as quantitatively diagnose this variability. Moreover, the sensitivity analysis of extreme drought to IODs based on the statistical simulations demonstrates the importance of further investigation of the mechanisms of IOD impacts on drought evolution in the ICP, which may include complex coupled oceanic-atmospheric processes. Although the results obtained herein were based on limited observations, the study's findings have strong implications for improving extreme drought predictions in the ICP.